diff --git "a/UnetChunk1.mlmodelc/model.mil" "b/UnetChunk1.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/UnetChunk1.mlmodelc/model.mil" @@ -0,0 +1,28911 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})] +{ + func main(tensor encoder_hidden_states, tensor sample, tensor text_embeds, tensor time_ids, tensor timestep) { + tensor var_24 = const()[name = tensor("op_24"), val = tensor(-1)]; + tensor var_41_axes_0 = const()[name = tensor("op_41_axes_0"), val = tensor([1])]; + tensor var_41_cast_fp16 = expand_dims(axes = var_41_axes_0, x = timestep)[name = tensor("op_41_cast_fp16")]; + tensor var_43_to_fp16 = const()[name = tensor("op_43_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor emb_3_cast_fp16 = mul(x = var_41_cast_fp16, y = var_43_to_fp16)[name = tensor("emb_3_cast_fp16")]; + tensor var_48_cast_fp16 = sin(x = emb_3_cast_fp16)[name = tensor("op_48_cast_fp16")]; + tensor var_49_cast_fp16 = cos(x = emb_3_cast_fp16)[name = tensor("op_49_cast_fp16")]; + tensor emb_7_interleave_0 = const()[name = tensor("emb_7_interleave_0"), val = tensor(false)]; + tensor emb_7_cast_fp16 = concat(axis = var_24, interleave = emb_7_interleave_0, values = (var_48_cast_fp16, var_49_cast_fp16))[name = tensor("emb_7_cast_fp16")]; + tensor var_53_begin_0 = const()[name = tensor("op_53_begin_0"), val = tensor([0, 160])]; + tensor var_53_end_0 = const()[name = tensor("op_53_end_0"), val = tensor([2, 320])]; + tensor var_53_end_mask_0 = const()[name = tensor("op_53_end_mask_0"), val = tensor([true, true])]; + tensor var_53_cast_fp16 = slice_by_index(begin = var_53_begin_0, end = var_53_end_0, end_mask = var_53_end_mask_0, x = emb_7_cast_fp16)[name = tensor("op_53_cast_fp16")]; + tensor var_55_begin_0 = const()[name = tensor("op_55_begin_0"), val = tensor([0, 0])]; + tensor var_55_end_0 = const()[name = tensor("op_55_end_0"), val = tensor([2, 160])]; + tensor var_55_end_mask_0 = const()[name = tensor("op_55_end_mask_0"), val = tensor([true, false])]; + tensor var_55_cast_fp16 = slice_by_index(begin = var_55_begin_0, end = var_55_end_0, end_mask = var_55_end_mask_0, x = emb_7_cast_fp16)[name = tensor("op_55_cast_fp16")]; + tensor sample_3_interleave_0 = const()[name = tensor("sample_3_interleave_0"), val = tensor(false)]; + tensor sample_3_cast_fp16 = concat(axis = var_24, interleave = sample_3_interleave_0, values = (var_53_cast_fp16, var_55_cast_fp16))[name = tensor("sample_3_cast_fp16")]; + tensor var_65_axes_0 = const()[name = tensor("op_65_axes_0"), val = tensor([-1])]; + tensor var_65_cast_fp16 = expand_dims(axes = var_65_axes_0, x = sample_3_cast_fp16)[name = tensor("op_65_cast_fp16")]; + tensor input_1_axes_0 = const()[name = tensor("input_1_axes_0"), val = tensor([-1])]; + tensor input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = var_65_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor input_3_pad_type_0 = const()[name = tensor("input_3_pad_type_0"), val = tensor("valid")]; + tensor input_3_strides_0 = const()[name = tensor("input_3_strides_0"), val = tensor([1, 1])]; + tensor input_3_pad_0 = const()[name = tensor("input_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_3_dilations_0 = const()[name = tensor("input_3_dilations_0"), val = tensor([1, 1])]; + tensor input_3_groups_0 = const()[name = tensor("input_3_groups_0"), val = tensor(1)]; + tensor time_embedding_linear_1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307712))), name = tensor("time_embedding_linear_1_weight_to_fp16_palettized"), shape = tensor([1280, 320, 1, 1])]; + tensor time_embedding_linear_1_bias_to_fp16 = const()[name = tensor("time_embedding_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307904)))]; + tensor input_3_cast_fp16 = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = time_embedding_linear_1_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = tensor("input_3_cast_fp16")]; + tensor input_5_cast_fp16 = silu(x = input_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor emb_pad_type_0 = const()[name = tensor("emb_pad_type_0"), val = tensor("valid")]; + tensor emb_strides_0 = const()[name = tensor("emb_strides_0"), val = tensor([1, 1])]; + tensor emb_pad_0 = const()[name = tensor("emb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor emb_dilations_0 = const()[name = tensor("emb_dilations_0"), val = tensor([1, 1])]; + tensor emb_groups_0 = const()[name = tensor("emb_groups_0"), val = tensor(1)]; + tensor time_embedding_linear_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1539392))), name = tensor("time_embedding_linear_2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor time_embedding_linear_2_bias_to_fp16 = const()[name = tensor("time_embedding_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1539584)))]; + tensor emb_cast_fp16 = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = emb_dilations_0, groups = emb_groups_0, pad = emb_pad_0, pad_type = emb_pad_type_0, strides = emb_strides_0, weight = time_embedding_linear_2_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor("emb_cast_fp16")]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([12])]; + tensor timesteps_cast_fp16 = reshape(shape = concat_0, x = time_ids)[name = tensor("timesteps_cast_fp16")]; + tensor var_85 = const()[name = tensor("op_85"), val = tensor(-1)]; + tensor var_102_axes_0 = const()[name = tensor("op_102_axes_0"), val = tensor([1])]; + tensor var_102_cast_fp16 = expand_dims(axes = var_102_axes_0, x = timesteps_cast_fp16)[name = tensor("op_102_cast_fp16")]; + tensor var_104_to_fp16 = const()[name = tensor("op_104_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1542208)))]; + tensor emb_11_cast_fp16 = mul(x = var_102_cast_fp16, y = var_104_to_fp16)[name = tensor("emb_11_cast_fp16")]; + tensor var_109_cast_fp16 = sin(x = emb_11_cast_fp16)[name = tensor("op_109_cast_fp16")]; + tensor var_110_cast_fp16 = cos(x = emb_11_cast_fp16)[name = tensor("op_110_cast_fp16")]; + tensor emb_15_interleave_0 = const()[name = tensor("emb_15_interleave_0"), val = tensor(false)]; + tensor emb_15_cast_fp16 = concat(axis = var_85, interleave = emb_15_interleave_0, values = (var_109_cast_fp16, var_110_cast_fp16))[name = tensor("emb_15_cast_fp16")]; + tensor var_114_begin_0 = const()[name = tensor("op_114_begin_0"), val = tensor([0, 128])]; + tensor var_114_end_0 = const()[name = tensor("op_114_end_0"), val = tensor([12, 256])]; + tensor var_114_end_mask_0 = const()[name = tensor("op_114_end_mask_0"), val = tensor([true, true])]; + tensor var_114_cast_fp16 = slice_by_index(begin = var_114_begin_0, end = var_114_end_0, end_mask = var_114_end_mask_0, x = emb_15_cast_fp16)[name = tensor("op_114_cast_fp16")]; + tensor var_116_begin_0 = const()[name = tensor("op_116_begin_0"), val = tensor([0, 0])]; + tensor var_116_end_0 = const()[name = tensor("op_116_end_0"), val = tensor([12, 128])]; + tensor var_116_end_mask_0 = const()[name = tensor("op_116_end_mask_0"), val = tensor([true, false])]; + tensor var_116_cast_fp16 = slice_by_index(begin = var_116_begin_0, end = var_116_end_0, end_mask = var_116_end_mask_0, x = emb_15_cast_fp16)[name = tensor("op_116_cast_fp16")]; + tensor time_embeds_1_interleave_0 = const()[name = tensor("time_embeds_1_interleave_0"), val = tensor(false)]; + tensor time_embeds_1_cast_fp16 = concat(axis = var_85, interleave = time_embeds_1_interleave_0, values = (var_114_cast_fp16, var_116_cast_fp16))[name = tensor("time_embeds_1_cast_fp16")]; + tensor var_124 = const()[name = tensor("op_124"), val = tensor([2, -1])]; + tensor time_embeds_cast_fp16 = reshape(shape = var_124, x = time_embeds_1_cast_fp16)[name = tensor("time_embeds_cast_fp16")]; + tensor var_127 = const()[name = tensor("op_127"), val = tensor(-1)]; + tensor sample_interleave_0 = const()[name = tensor("sample_interleave_0"), val = tensor(false)]; + tensor sample_cast_fp16 = concat(axis = var_127, interleave = sample_interleave_0, values = (text_embeds, time_embeds_cast_fp16))[name = tensor("sample_cast_fp16")]; + tensor var_136_axes_0 = const()[name = tensor("op_136_axes_0"), val = tensor([-1])]; + tensor var_136_cast_fp16 = expand_dims(axes = var_136_axes_0, x = sample_cast_fp16)[name = tensor("op_136_cast_fp16")]; + tensor input_7_axes_0 = const()[name = tensor("input_7_axes_0"), val = tensor([-1])]; + tensor input_7_cast_fp16 = expand_dims(axes = input_7_axes_0, x = var_136_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor input_9_pad_type_0 = const()[name = tensor("input_9_pad_type_0"), val = tensor("valid")]; + tensor input_9_strides_0 = const()[name = tensor("input_9_strides_0"), val = tensor([1, 1])]; + tensor input_9_pad_0 = const()[name = tensor("input_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_9_dilations_0 = const()[name = tensor("input_9_dilations_0"), val = tensor([1, 1])]; + tensor input_9_groups_0 = const()[name = tensor("input_9_groups_0"), val = tensor(1)]; + tensor add_embedding_linear_1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1542528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4245952))), name = tensor("add_embedding_linear_1_weight_to_fp16_palettized"), shape = tensor([1280, 2816, 1, 1])]; + tensor add_embedding_linear_1_bias_to_fp16 = const()[name = tensor("add_embedding_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4246144)))]; + tensor input_9_cast_fp16 = conv(bias = add_embedding_linear_1_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = add_embedding_linear_1_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor input_11_cast_fp16 = silu(x = input_9_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor aug_emb_pad_type_0 = const()[name = tensor("aug_emb_pad_type_0"), val = tensor("valid")]; + tensor aug_emb_strides_0 = const()[name = tensor("aug_emb_strides_0"), val = tensor([1, 1])]; + tensor aug_emb_pad_0 = const()[name = tensor("aug_emb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor aug_emb_dilations_0 = const()[name = tensor("aug_emb_dilations_0"), val = tensor([1, 1])]; + tensor aug_emb_groups_0 = const()[name = tensor("aug_emb_groups_0"), val = tensor(1)]; + tensor add_embedding_linear_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4248768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5477632))), name = tensor("add_embedding_linear_2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor add_embedding_linear_2_bias_to_fp16 = const()[name = tensor("add_embedding_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5477824)))]; + tensor aug_emb_cast_fp16 = conv(bias = add_embedding_linear_2_bias_to_fp16, dilations = aug_emb_dilations_0, groups = aug_emb_groups_0, pad = aug_emb_pad_0, pad_type = aug_emb_pad_type_0, strides = aug_emb_strides_0, weight = add_embedding_linear_2_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = tensor("aug_emb_cast_fp16")]; + tensor input_19_cast_fp16 = add(x = emb_cast_fp16, y = aug_emb_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor input_13_pad_type_0 = const()[name = tensor("input_13_pad_type_0"), val = tensor("custom")]; + tensor input_13_pad_0 = const()[name = tensor("input_13_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_13_strides_0 = const()[name = tensor("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_dilations_0 = const()[name = tensor("input_13_dilations_0"), val = tensor([1, 1])]; + tensor input_13_groups_0 = const()[name = tensor("input_13_groups_0"), val = tensor(1)]; + tensor conv_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5480448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5489152))), name = tensor("conv_in_weight_to_fp16_palettized"), shape = tensor([320, 4, 3, 3])]; + tensor conv_in_bias_to_fp16 = const()[name = tensor("conv_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5489344)))]; + tensor input_13_cast_fp16_1 = conv(bias = conv_in_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = conv_in_weight_to_fp16_palettized, x = sample)[name = tensor("input_13_cast_fp16")]; + tensor reshape_0_shape_0 = const()[name = tensor("reshape_0_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_0_cast_fp16 = reshape(shape = reshape_0_shape_0, x = input_13_cast_fp16_1)[name = tensor("reshape_0_cast_fp16")]; + tensor reduce_mean_0_axes_0 = const()[name = tensor("reduce_mean_0_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_0_keep_dims_0 = const()[name = tensor("reduce_mean_0_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_0_cast_fp16 = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast_fp16)[name = tensor("reduce_mean_0_cast_fp16")]; + tensor sub_0_cast_fp16 = sub(x = reshape_0_cast_fp16, y = reduce_mean_0_cast_fp16)[name = tensor("sub_0_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = sub_0_cast_fp16)[name = tensor("square_0_cast_fp16")]; + tensor reduce_mean_2_axes_0 = const()[name = tensor("reduce_mean_2_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_2_keep_dims_0 = const()[name = tensor("reduce_mean_2_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_2_cast_fp16 = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast_fp16)[name = tensor("reduce_mean_2_cast_fp16")]; + tensor add_0_y_0_to_fp16 = const()[name = tensor("add_0_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_0_cast_fp16 = add(x = reduce_mean_2_cast_fp16, y = add_0_y_0_to_fp16)[name = tensor("add_0_cast_fp16")]; + tensor sqrt_0_cast_fp16 = sqrt(x = add_0_cast_fp16)[name = tensor("sqrt_0_cast_fp16")]; + tensor real_div_0_cast_fp16 = real_div(x = sub_0_cast_fp16, y = sqrt_0_cast_fp16)[name = tensor("real_div_0_cast_fp16")]; + tensor reshape_1_shape_0 = const()[name = tensor("reshape_1_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_1_cast_fp16 = reshape(shape = reshape_1_shape_0, x = real_div_0_cast_fp16)[name = tensor("reshape_1_cast_fp16")]; + tensor add_1_mean_0_to_fp16 = const()[name = tensor("add_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5490048)))]; + tensor add_1_variance_0_to_fp16 = const()[name = tensor("add_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5490752)))]; + tensor add_1_gamma_0_to_fp16 = const()[name = tensor("add_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5491456)))]; + tensor add_1_beta_0_to_fp16 = const()[name = tensor("add_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5492160)))]; + tensor add_1_epsilon_0_to_fp16 = const()[name = tensor("add_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_1_cast_fp16 = batch_norm(beta = add_1_beta_0_to_fp16, epsilon = add_1_epsilon_0_to_fp16, gamma = add_1_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_1_cast_fp16)[name = tensor("add_1_cast_fp16")]; + tensor input_17_cast_fp16 = silu(x = add_1_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor hidden_states_1_pad_type_0 = const()[name = tensor("hidden_states_1_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_1_pad_0 = const()[name = tensor("hidden_states_1_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_1_strides_0 = const()[name = tensor("hidden_states_1_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_1_dilations_0 = const()[name = tensor("hidden_states_1_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_1_groups_0 = const()[name = tensor("hidden_states_1_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5492864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6184128))), name = tensor("down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6184320)))]; + tensor hidden_states_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_1_dilations_0, groups = hidden_states_1_groups_0, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = hidden_states_1_strides_0, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; + tensor input_21_cast_fp16_1 = silu(x = input_19_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor temb_1_pad_type_0 = const()[name = tensor("temb_1_pad_type_0"), val = tensor("valid")]; + tensor temb_1_strides_0 = const()[name = tensor("temb_1_strides_0"), val = tensor([1, 1])]; + tensor temb_1_pad_0 = const()[name = tensor("temb_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_1_dilations_0 = const()[name = tensor("temb_1_dilations_0"), val = tensor([1, 1])]; + tensor temb_1_groups_0 = const()[name = tensor("temb_1_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6185024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6492288))), name = tensor("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6492480)))]; + tensor temb_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_1_dilations_0, groups = temb_1_groups_0, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = temb_1_strides_0, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_1_cast_fp16")]; + tensor input_23_cast_fp16 = add(x = hidden_states_1_cast_fp16, y = temb_1_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor reshape_4_shape_0 = const()[name = tensor("reshape_4_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_4_cast_fp16 = reshape(shape = reshape_4_shape_0, x = input_23_cast_fp16)[name = tensor("reshape_4_cast_fp16")]; + tensor reduce_mean_3_axes_0 = const()[name = tensor("reduce_mean_3_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_3_keep_dims_0 = const()[name = tensor("reduce_mean_3_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_3_cast_fp16 = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast_fp16)[name = tensor("reduce_mean_3_cast_fp16")]; + tensor sub_2_cast_fp16 = sub(x = reshape_4_cast_fp16, y = reduce_mean_3_cast_fp16)[name = tensor("sub_2_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = sub_2_cast_fp16)[name = tensor("square_1_cast_fp16")]; + tensor reduce_mean_5_axes_0 = const()[name = tensor("reduce_mean_5_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_5_keep_dims_0 = const()[name = tensor("reduce_mean_5_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_5_cast_fp16 = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast_fp16)[name = tensor("reduce_mean_5_cast_fp16")]; + tensor add_2_y_0_to_fp16 = const()[name = tensor("add_2_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_2_cast_fp16 = add(x = reduce_mean_5_cast_fp16, y = add_2_y_0_to_fp16)[name = tensor("add_2_cast_fp16")]; + tensor sqrt_1_cast_fp16 = sqrt(x = add_2_cast_fp16)[name = tensor("sqrt_1_cast_fp16")]; + tensor real_div_1_cast_fp16 = real_div(x = sub_2_cast_fp16, y = sqrt_1_cast_fp16)[name = tensor("real_div_1_cast_fp16")]; + tensor reshape_5_shape_0 = const()[name = tensor("reshape_5_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_5_cast_fp16 = reshape(shape = reshape_5_shape_0, x = real_div_1_cast_fp16)[name = tensor("reshape_5_cast_fp16")]; + tensor add_3_gamma_0_to_fp16 = const()[name = tensor("add_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6493184)))]; + tensor add_3_beta_0_to_fp16 = const()[name = tensor("add_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6493888)))]; + tensor add_3_epsilon_0_to_fp16 = const()[name = tensor("add_3_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_3_cast_fp16 = batch_norm(beta = add_3_beta_0_to_fp16, epsilon = add_3_epsilon_0_to_fp16, gamma = add_3_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_5_cast_fp16)[name = tensor("add_3_cast_fp16")]; + tensor input_27_cast_fp16 = silu(x = add_3_cast_fp16)[name = tensor("input_27_cast_fp16")]; + tensor hidden_states_3_pad_type_0 = const()[name = tensor("hidden_states_3_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_3_pad_0 = const()[name = tensor("hidden_states_3_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_3_strides_0 = const()[name = tensor("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_dilations_0 = const()[name = tensor("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_3_groups_0 = const()[name = tensor("hidden_states_3_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6494592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7185856))), name = tensor("down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7186048)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_27_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; + tensor input_29_cast_fp16_1 = add(x = input_13_cast_fp16_1, y = hidden_states_3_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor reshape_8_shape_0 = const()[name = tensor("reshape_8_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_8_cast_fp16 = reshape(shape = reshape_8_shape_0, x = input_29_cast_fp16_1)[name = tensor("reshape_8_cast_fp16")]; + tensor reduce_mean_6_axes_0 = const()[name = tensor("reduce_mean_6_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_6_keep_dims_0 = const()[name = tensor("reduce_mean_6_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_6_cast_fp16 = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast_fp16)[name = tensor("reduce_mean_6_cast_fp16")]; + tensor sub_4_cast_fp16 = sub(x = reshape_8_cast_fp16, y = reduce_mean_6_cast_fp16)[name = tensor("sub_4_cast_fp16")]; + tensor square_2_cast_fp16 = square(x = sub_4_cast_fp16)[name = tensor("square_2_cast_fp16")]; + tensor reduce_mean_8_axes_0 = const()[name = tensor("reduce_mean_8_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_8_keep_dims_0 = const()[name = tensor("reduce_mean_8_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_8_cast_fp16 = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast_fp16)[name = tensor("reduce_mean_8_cast_fp16")]; + tensor add_4_y_0_to_fp16 = const()[name = tensor("add_4_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_4_cast_fp16 = add(x = reduce_mean_8_cast_fp16, y = add_4_y_0_to_fp16)[name = tensor("add_4_cast_fp16")]; + tensor sqrt_2_cast_fp16 = sqrt(x = add_4_cast_fp16)[name = tensor("sqrt_2_cast_fp16")]; + tensor real_div_2_cast_fp16 = real_div(x = sub_4_cast_fp16, y = sqrt_2_cast_fp16)[name = tensor("real_div_2_cast_fp16")]; + tensor reshape_9_shape_0 = const()[name = tensor("reshape_9_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_9_cast_fp16 = reshape(shape = reshape_9_shape_0, x = real_div_2_cast_fp16)[name = tensor("reshape_9_cast_fp16")]; + tensor add_5_gamma_0_to_fp16 = const()[name = tensor("add_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7186752)))]; + tensor add_5_beta_0_to_fp16 = const()[name = tensor("add_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7187456)))]; + tensor add_5_epsilon_0_to_fp16 = const()[name = tensor("add_5_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_5_cast_fp16 = batch_norm(beta = add_5_beta_0_to_fp16, epsilon = add_5_epsilon_0_to_fp16, gamma = add_5_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_9_cast_fp16)[name = tensor("add_5_cast_fp16")]; + tensor input_33_cast_fp16 = silu(x = add_5_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor hidden_states_5_pad_type_0 = const()[name = tensor("hidden_states_5_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_5_pad_0 = const()[name = tensor("hidden_states_5_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_5_strides_0 = const()[name = tensor("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_dilations_0 = const()[name = tensor("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_5_groups_0 = const()[name = tensor("hidden_states_5_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7188160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7879424))), name = tensor("down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7879616)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor("hidden_states_5_cast_fp16")]; + tensor temb_3_pad_type_0 = const()[name = tensor("temb_3_pad_type_0"), val = tensor("valid")]; + tensor temb_3_strides_0 = const()[name = tensor("temb_3_strides_0"), val = tensor([1, 1])]; + tensor temb_3_pad_0 = const()[name = tensor("temb_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_3_dilations_0 = const()[name = tensor("temb_3_dilations_0"), val = tensor([1, 1])]; + tensor temb_3_groups_0 = const()[name = tensor("temb_3_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7880320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8187584))), name = tensor("down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8187776)))]; + tensor temb_3_cast_fp16 = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_3_dilations_0, groups = temb_3_groups_0, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = temb_3_strides_0, weight = down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_3_cast_fp16")]; + tensor input_37_cast_fp16 = add(x = hidden_states_5_cast_fp16, y = temb_3_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor reshape_12_shape_0 = const()[name = tensor("reshape_12_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_12_cast_fp16 = reshape(shape = reshape_12_shape_0, x = input_37_cast_fp16)[name = tensor("reshape_12_cast_fp16")]; + tensor reduce_mean_9_axes_0 = const()[name = tensor("reduce_mean_9_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_9_keep_dims_0 = const()[name = tensor("reduce_mean_9_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_9_cast_fp16 = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast_fp16)[name = tensor("reduce_mean_9_cast_fp16")]; + tensor sub_6_cast_fp16 = sub(x = reshape_12_cast_fp16, y = reduce_mean_9_cast_fp16)[name = tensor("sub_6_cast_fp16")]; + tensor square_3_cast_fp16 = square(x = sub_6_cast_fp16)[name = tensor("square_3_cast_fp16")]; + tensor reduce_mean_11_axes_0 = const()[name = tensor("reduce_mean_11_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_11_keep_dims_0 = const()[name = tensor("reduce_mean_11_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_11_cast_fp16 = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast_fp16)[name = tensor("reduce_mean_11_cast_fp16")]; + tensor add_6_y_0_to_fp16 = const()[name = tensor("add_6_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_6_cast_fp16 = add(x = reduce_mean_11_cast_fp16, y = add_6_y_0_to_fp16)[name = tensor("add_6_cast_fp16")]; + tensor sqrt_3_cast_fp16 = sqrt(x = add_6_cast_fp16)[name = tensor("sqrt_3_cast_fp16")]; + tensor real_div_3_cast_fp16 = real_div(x = sub_6_cast_fp16, y = sqrt_3_cast_fp16)[name = tensor("real_div_3_cast_fp16")]; + tensor reshape_13_shape_0 = const()[name = tensor("reshape_13_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_13_cast_fp16 = reshape(shape = reshape_13_shape_0, x = real_div_3_cast_fp16)[name = tensor("reshape_13_cast_fp16")]; + tensor add_7_gamma_0_to_fp16 = const()[name = tensor("add_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8188480)))]; + tensor add_7_beta_0_to_fp16 = const()[name = tensor("add_7_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8189184)))]; + tensor add_7_epsilon_0_to_fp16 = const()[name = tensor("add_7_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_7_cast_fp16 = batch_norm(beta = add_7_beta_0_to_fp16, epsilon = add_7_epsilon_0_to_fp16, gamma = add_7_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_13_cast_fp16)[name = tensor("add_7_cast_fp16")]; + tensor input_41_cast_fp16 = silu(x = add_7_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor hidden_states_7_pad_type_0 = const()[name = tensor("hidden_states_7_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_7_pad_0 = const()[name = tensor("hidden_states_7_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_7_strides_0 = const()[name = tensor("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_dilations_0 = const()[name = tensor("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_7_groups_0 = const()[name = tensor("hidden_states_7_groups_0"), val = tensor(1)]; + tensor down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8189888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8881152))), name = tensor("down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8881344)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; + tensor input_43_cast_fp16_1 = add(x = input_29_cast_fp16_1, y = hidden_states_7_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor input_45_pad_type_0 = const()[name = tensor("input_45_pad_type_0"), val = tensor("custom")]; + tensor input_45_pad_0 = const()[name = tensor("input_45_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_45_strides_0 = const()[name = tensor("input_45_strides_0"), val = tensor([2, 2])]; + tensor input_45_dilations_0 = const()[name = tensor("input_45_dilations_0"), val = tensor([1, 1])]; + tensor input_45_groups_0 = const()[name = tensor("input_45_groups_0"), val = tensor(1)]; + tensor down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8882048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9573312))), name = tensor("down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_0_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9573504)))]; + tensor input_45_cast_fp16_1 = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized, x = input_43_cast_fp16_1)[name = tensor("input_45_cast_fp16")]; + tensor var_288 = const()[name = tensor("op_288"), val = tensor(1)]; + tensor reshape_16_shape_0 = const()[name = tensor("reshape_16_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_16_cast_fp16 = reshape(shape = reshape_16_shape_0, x = input_45_cast_fp16_1)[name = tensor("reshape_16_cast_fp16")]; + tensor reduce_mean_12_axes_0 = const()[name = tensor("reduce_mean_12_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_12_keep_dims_0 = const()[name = tensor("reduce_mean_12_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_12_cast_fp16 = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast_fp16)[name = tensor("reduce_mean_12_cast_fp16")]; + tensor sub_8_cast_fp16 = sub(x = reshape_16_cast_fp16, y = reduce_mean_12_cast_fp16)[name = tensor("sub_8_cast_fp16")]; + tensor square_4_cast_fp16 = square(x = sub_8_cast_fp16)[name = tensor("square_4_cast_fp16")]; + tensor reduce_mean_14_axes_0 = const()[name = tensor("reduce_mean_14_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_14_keep_dims_0 = const()[name = tensor("reduce_mean_14_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_14_cast_fp16 = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast_fp16)[name = tensor("reduce_mean_14_cast_fp16")]; + tensor add_8_y_0_to_fp16 = const()[name = tensor("add_8_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_8_cast_fp16 = add(x = reduce_mean_14_cast_fp16, y = add_8_y_0_to_fp16)[name = tensor("add_8_cast_fp16")]; + tensor sqrt_4_cast_fp16 = sqrt(x = add_8_cast_fp16)[name = tensor("sqrt_4_cast_fp16")]; + tensor real_div_4_cast_fp16 = real_div(x = sub_8_cast_fp16, y = sqrt_4_cast_fp16)[name = tensor("real_div_4_cast_fp16")]; + tensor reshape_17_shape_0 = const()[name = tensor("reshape_17_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_17_cast_fp16 = reshape(shape = reshape_17_shape_0, x = real_div_4_cast_fp16)[name = tensor("reshape_17_cast_fp16")]; + tensor add_9_gamma_0_to_fp16 = const()[name = tensor("add_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9574208)))]; + tensor add_9_beta_0_to_fp16 = const()[name = tensor("add_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9574912)))]; + tensor add_9_epsilon_0_to_fp16 = const()[name = tensor("add_9_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_9_cast_fp16 = batch_norm(beta = add_9_beta_0_to_fp16, epsilon = add_9_epsilon_0_to_fp16, gamma = add_9_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_17_cast_fp16)[name = tensor("add_9_cast_fp16")]; + tensor input_49_cast_fp16 = silu(x = add_9_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor hidden_states_9_pad_type_0 = const()[name = tensor("hidden_states_9_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_9_pad_0 = const()[name = tensor("hidden_states_9_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_9_strides_0 = const()[name = tensor("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_dilations_0 = const()[name = tensor("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_9_groups_0 = const()[name = tensor("hidden_states_9_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9575616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10958080))), name = tensor("down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([640, 320, 3, 3])]; + tensor down_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10958272)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("hidden_states_9_cast_fp16")]; + tensor temb_5_pad_type_0 = const()[name = tensor("temb_5_pad_type_0"), val = tensor("valid")]; + tensor temb_5_strides_0 = const()[name = tensor("temb_5_strides_0"), val = tensor([1, 1])]; + tensor temb_5_pad_0 = const()[name = tensor("temb_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_5_dilations_0 = const()[name = tensor("temb_5_dilations_0"), val = tensor([1, 1])]; + tensor temb_5_groups_0 = const()[name = tensor("temb_5_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10959616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11574080))), name = tensor("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11574272)))]; + tensor temb_5_cast_fp16 = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_5_dilations_0, groups = temb_5_groups_0, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = temb_5_strides_0, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_5_cast_fp16")]; + tensor input_53_cast_fp16 = add(x = hidden_states_9_cast_fp16, y = temb_5_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor reshape_20_shape_0 = const()[name = tensor("reshape_20_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_20_cast_fp16 = reshape(shape = reshape_20_shape_0, x = input_53_cast_fp16)[name = tensor("reshape_20_cast_fp16")]; + tensor reduce_mean_15_axes_0 = const()[name = tensor("reduce_mean_15_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_15_keep_dims_0 = const()[name = tensor("reduce_mean_15_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_15_cast_fp16 = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast_fp16)[name = tensor("reduce_mean_15_cast_fp16")]; + tensor sub_10_cast_fp16 = sub(x = reshape_20_cast_fp16, y = reduce_mean_15_cast_fp16)[name = tensor("sub_10_cast_fp16")]; + tensor square_5_cast_fp16 = square(x = sub_10_cast_fp16)[name = tensor("square_5_cast_fp16")]; + tensor reduce_mean_17_axes_0 = const()[name = tensor("reduce_mean_17_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_17_keep_dims_0 = const()[name = tensor("reduce_mean_17_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_17_cast_fp16 = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast_fp16)[name = tensor("reduce_mean_17_cast_fp16")]; + tensor add_10_y_0_to_fp16 = const()[name = tensor("add_10_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_10_cast_fp16 = add(x = reduce_mean_17_cast_fp16, y = add_10_y_0_to_fp16)[name = tensor("add_10_cast_fp16")]; + tensor sqrt_5_cast_fp16 = sqrt(x = add_10_cast_fp16)[name = tensor("sqrt_5_cast_fp16")]; + tensor real_div_5_cast_fp16 = real_div(x = sub_10_cast_fp16, y = sqrt_5_cast_fp16)[name = tensor("real_div_5_cast_fp16")]; + tensor reshape_21_shape_0 = const()[name = tensor("reshape_21_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_21_cast_fp16 = reshape(shape = reshape_21_shape_0, x = real_div_5_cast_fp16)[name = tensor("reshape_21_cast_fp16")]; + tensor add_11_mean_0_to_fp16 = const()[name = tensor("add_11_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11575616)))]; + tensor add_11_variance_0_to_fp16 = const()[name = tensor("add_11_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11576960)))]; + tensor add_11_gamma_0_to_fp16 = const()[name = tensor("add_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11578304)))]; + tensor add_11_beta_0_to_fp16 = const()[name = tensor("add_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11579648)))]; + tensor add_11_epsilon_0_to_fp16 = const()[name = tensor("add_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_11_cast_fp16 = batch_norm(beta = add_11_beta_0_to_fp16, epsilon = add_11_epsilon_0_to_fp16, gamma = add_11_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_21_cast_fp16)[name = tensor("add_11_cast_fp16")]; + tensor input_57_cast_fp16 = silu(x = add_11_cast_fp16)[name = tensor("input_57_cast_fp16")]; + tensor hidden_states_11_pad_type_0 = const()[name = tensor("hidden_states_11_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_11_pad_0 = const()[name = tensor("hidden_states_11_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_11_strides_0 = const()[name = tensor("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_dilations_0 = const()[name = tensor("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_11_groups_0 = const()[name = tensor("hidden_states_11_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11580992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14345856))), name = tensor("down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14346048)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor("hidden_states_11_cast_fp16")]; + tensor x_1_pad_type_0 = const()[name = tensor("x_1_pad_type_0"), val = tensor("valid")]; + tensor x_1_strides_0 = const()[name = tensor("x_1_strides_0"), val = tensor([1, 1])]; + tensor x_1_pad_0 = const()[name = tensor("x_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_1_dilations_0 = const()[name = tensor("x_1_dilations_0"), val = tensor([1, 1])]; + tensor x_1_groups_0 = const()[name = tensor("x_1_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14347392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14501056))), name = tensor("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([640, 320, 1, 1])]; + tensor down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14501248)))]; + tensor x_1_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_45_cast_fp16_1)[name = tensor("x_1_cast_fp16")]; + tensor hidden_states_13_cast_fp16 = add(x = x_1_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; + tensor reshape_24_shape_0 = const()[name = tensor("reshape_24_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_24_cast_fp16 = reshape(shape = reshape_24_shape_0, x = hidden_states_13_cast_fp16)[name = tensor("reshape_24_cast_fp16")]; + tensor reduce_mean_18_axes_0 = const()[name = tensor("reduce_mean_18_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_18_keep_dims_0 = const()[name = tensor("reduce_mean_18_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_18_cast_fp16 = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast_fp16)[name = tensor("reduce_mean_18_cast_fp16")]; + tensor sub_12_cast_fp16 = sub(x = reshape_24_cast_fp16, y = reduce_mean_18_cast_fp16)[name = tensor("sub_12_cast_fp16")]; + tensor square_6_cast_fp16 = square(x = sub_12_cast_fp16)[name = tensor("square_6_cast_fp16")]; + tensor reduce_mean_20_axes_0 = const()[name = tensor("reduce_mean_20_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_20_keep_dims_0 = const()[name = tensor("reduce_mean_20_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_20_cast_fp16 = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast_fp16)[name = tensor("reduce_mean_20_cast_fp16")]; + tensor add_12_y_0_to_fp16 = const()[name = tensor("add_12_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_12_cast_fp16 = add(x = reduce_mean_20_cast_fp16, y = add_12_y_0_to_fp16)[name = tensor("add_12_cast_fp16")]; + tensor sqrt_6_cast_fp16 = sqrt(x = add_12_cast_fp16)[name = tensor("sqrt_6_cast_fp16")]; + tensor real_div_6_cast_fp16 = real_div(x = sub_12_cast_fp16, y = sqrt_6_cast_fp16)[name = tensor("real_div_6_cast_fp16")]; + tensor reshape_25_shape_0 = const()[name = tensor("reshape_25_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_25_cast_fp16 = reshape(shape = reshape_25_shape_0, x = real_div_6_cast_fp16)[name = tensor("reshape_25_cast_fp16")]; + tensor add_13_gamma_0_to_fp16 = const()[name = tensor("add_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14502592)))]; + tensor add_13_beta_0_to_fp16 = const()[name = tensor("add_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14503936)))]; + tensor add_13_epsilon_0_to_fp16 = const()[name = tensor("add_13_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_13_cast_fp16 = batch_norm(beta = add_13_beta_0_to_fp16, epsilon = add_13_epsilon_0_to_fp16, gamma = add_13_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_25_cast_fp16)[name = tensor("add_13_cast_fp16")]; + tensor hidden_states_15_pad_type_0 = const()[name = tensor("hidden_states_15_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_15_strides_0 = const()[name = tensor("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = tensor("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = tensor("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_15_groups_0 = const()[name = tensor("hidden_states_15_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14505280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14812544))), name = tensor("down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14812736)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_13_cast_fp16)[name = tensor("hidden_states_15_cast_fp16")]; + tensor var_369 = const()[name = tensor("op_369"), val = tensor([2, 640, 1, 4096])]; + tensor inputs_1_cast_fp16 = reshape(shape = var_369, x = hidden_states_15_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; + tensor hidden_states_17_axes_0 = const()[name = tensor("hidden_states_17_axes_0"), val = tensor([1])]; + tensor hidden_states_17_gamma_0_to_fp16 = const()[name = tensor("hidden_states_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14814080)))]; + tensor hidden_states_17_beta_0_to_fp16 = const()[name = tensor("hidden_states_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14815424)))]; + tensor var_385_to_fp16 = const()[name = tensor("op_385_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_17_cast_fp16 = layer_norm(axes = hidden_states_17_axes_0, beta = hidden_states_17_beta_0_to_fp16, epsilon = var_385_to_fp16, gamma = hidden_states_17_gamma_0_to_fp16, x = inputs_1_cast_fp16)[name = tensor("hidden_states_17_cast_fp16")]; + tensor q_1_pad_type_0 = const()[name = tensor("q_1_pad_type_0"), val = tensor("valid")]; + tensor q_1_strides_0 = const()[name = tensor("q_1_strides_0"), val = tensor([1, 1])]; + tensor q_1_pad_0 = const()[name = tensor("q_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_1_dilations_0 = const()[name = tensor("q_1_dilations_0"), val = tensor([1, 1])]; + tensor q_1_groups_0 = const()[name = tensor("q_1_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14816768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15124032))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_1_cast_fp16 = conv(dilations = q_1_dilations_0, groups = q_1_groups_0, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = q_1_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_17_cast_fp16)[name = tensor("q_1_cast_fp16")]; + tensor k_1_pad_type_0 = const()[name = tensor("k_1_pad_type_0"), val = tensor("valid")]; + tensor k_1_strides_0 = const()[name = tensor("k_1_strides_0"), val = tensor([1, 1])]; + tensor k_1_pad_0 = const()[name = tensor("k_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_1_dilations_0 = const()[name = tensor("k_1_dilations_0"), val = tensor([1, 1])]; + tensor k_1_groups_0 = const()[name = tensor("k_1_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15124224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15431488))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_1_cast_fp16 = conv(dilations = k_1_dilations_0, groups = k_1_groups_0, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = k_1_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_17_cast_fp16)[name = tensor("k_1_cast_fp16")]; + tensor v_1_pad_type_0 = const()[name = tensor("v_1_pad_type_0"), val = tensor("valid")]; + tensor v_1_strides_0 = const()[name = tensor("v_1_strides_0"), val = tensor([1, 1])]; + tensor v_1_pad_0 = const()[name = tensor("v_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_1_dilations_0 = const()[name = tensor("v_1_dilations_0"), val = tensor([1, 1])]; + tensor v_1_groups_0 = const()[name = tensor("v_1_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15431680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15738944))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_1_cast_fp16 = conv(dilations = v_1_dilations_0, groups = v_1_groups_0, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = v_1_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_17_cast_fp16)[name = tensor("v_1_cast_fp16")]; + tensor var_418_begin_0 = const()[name = tensor("op_418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_418_end_0 = const()[name = tensor("op_418_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_418_end_mask_0 = const()[name = tensor("op_418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_418_cast_fp16 = slice_by_index(begin = var_418_begin_0, end = var_418_end_0, end_mask = var_418_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_418_cast_fp16")]; + tensor var_422_begin_0 = const()[name = tensor("op_422_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_422_end_0 = const()[name = tensor("op_422_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_422_end_mask_0 = const()[name = tensor("op_422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_422_cast_fp16 = slice_by_index(begin = var_422_begin_0, end = var_422_end_0, end_mask = var_422_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_422_cast_fp16")]; + tensor var_426_begin_0 = const()[name = tensor("op_426_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_426_end_0 = const()[name = tensor("op_426_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_426_end_mask_0 = const()[name = tensor("op_426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_426_cast_fp16 = slice_by_index(begin = var_426_begin_0, end = var_426_end_0, end_mask = var_426_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_426_cast_fp16")]; + tensor var_430_begin_0 = const()[name = tensor("op_430_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_430_end_0 = const()[name = tensor("op_430_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_430_end_mask_0 = const()[name = tensor("op_430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_430_cast_fp16 = slice_by_index(begin = var_430_begin_0, end = var_430_end_0, end_mask = var_430_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_430_cast_fp16")]; + tensor var_434_begin_0 = const()[name = tensor("op_434_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_434_end_0 = const()[name = tensor("op_434_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_434_end_mask_0 = const()[name = tensor("op_434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_434_cast_fp16 = slice_by_index(begin = var_434_begin_0, end = var_434_end_0, end_mask = var_434_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_434_cast_fp16")]; + tensor var_438_begin_0 = const()[name = tensor("op_438_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_438_end_0 = const()[name = tensor("op_438_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_438_end_mask_0 = const()[name = tensor("op_438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_438_cast_fp16 = slice_by_index(begin = var_438_begin_0, end = var_438_end_0, end_mask = var_438_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_438_cast_fp16")]; + tensor var_442_begin_0 = const()[name = tensor("op_442_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_442_end_0 = const()[name = tensor("op_442_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_442_end_mask_0 = const()[name = tensor("op_442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_442_cast_fp16 = slice_by_index(begin = var_442_begin_0, end = var_442_end_0, end_mask = var_442_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_442_cast_fp16")]; + tensor var_446_begin_0 = const()[name = tensor("op_446_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_446_end_0 = const()[name = tensor("op_446_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_446_end_mask_0 = const()[name = tensor("op_446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_446_cast_fp16 = slice_by_index(begin = var_446_begin_0, end = var_446_end_0, end_mask = var_446_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_446_cast_fp16")]; + tensor var_450_begin_0 = const()[name = tensor("op_450_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_450_end_0 = const()[name = tensor("op_450_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_450_end_mask_0 = const()[name = tensor("op_450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_450_cast_fp16 = slice_by_index(begin = var_450_begin_0, end = var_450_end_0, end_mask = var_450_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_450_cast_fp16")]; + tensor var_454_begin_0 = const()[name = tensor("op_454_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_454_end_0 = const()[name = tensor("op_454_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_454_end_mask_0 = const()[name = tensor("op_454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_454_cast_fp16 = slice_by_index(begin = var_454_begin_0, end = var_454_end_0, end_mask = var_454_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_454_cast_fp16")]; + tensor k_3_perm_0 = const()[name = tensor("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_461_begin_0 = const()[name = tensor("op_461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_461_end_0 = const()[name = tensor("op_461_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_461_end_mask_0 = const()[name = tensor("op_461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = k_1_cast_fp16)[name = tensor("transpose_67")]; + tensor var_461_cast_fp16 = slice_by_index(begin = var_461_begin_0, end = var_461_end_0, end_mask = var_461_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_461_cast_fp16")]; + tensor var_465_begin_0 = const()[name = tensor("op_465_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_465_end_0 = const()[name = tensor("op_465_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_465_end_mask_0 = const()[name = tensor("op_465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_465_cast_fp16")]; + tensor var_469_begin_0 = const()[name = tensor("op_469_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_469_end_0 = const()[name = tensor("op_469_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_469_end_mask_0 = const()[name = tensor("op_469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_469_cast_fp16")]; + tensor var_473_begin_0 = const()[name = tensor("op_473_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_473_end_0 = const()[name = tensor("op_473_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_473_end_mask_0 = const()[name = tensor("op_473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_473_cast_fp16")]; + tensor var_477_begin_0 = const()[name = tensor("op_477_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_477_end_0 = const()[name = tensor("op_477_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_477_end_mask_0 = const()[name = tensor("op_477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_477_cast_fp16 = slice_by_index(begin = var_477_begin_0, end = var_477_end_0, end_mask = var_477_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_477_cast_fp16")]; + tensor var_481_begin_0 = const()[name = tensor("op_481_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_481_end_0 = const()[name = tensor("op_481_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_481_end_mask_0 = const()[name = tensor("op_481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_481_cast_fp16 = slice_by_index(begin = var_481_begin_0, end = var_481_end_0, end_mask = var_481_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_481_cast_fp16")]; + tensor var_485_begin_0 = const()[name = tensor("op_485_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_485_end_0 = const()[name = tensor("op_485_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_485_end_mask_0 = const()[name = tensor("op_485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_485_cast_fp16 = slice_by_index(begin = var_485_begin_0, end = var_485_end_0, end_mask = var_485_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_485_cast_fp16")]; + tensor var_489_begin_0 = const()[name = tensor("op_489_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_489_end_0 = const()[name = tensor("op_489_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_489_end_mask_0 = const()[name = tensor("op_489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_489_cast_fp16 = slice_by_index(begin = var_489_begin_0, end = var_489_end_0, end_mask = var_489_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_489_cast_fp16")]; + tensor var_493_begin_0 = const()[name = tensor("op_493_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_493_end_0 = const()[name = tensor("op_493_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_493_end_mask_0 = const()[name = tensor("op_493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_493_cast_fp16 = slice_by_index(begin = var_493_begin_0, end = var_493_end_0, end_mask = var_493_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_493_cast_fp16")]; + tensor var_497_begin_0 = const()[name = tensor("op_497_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_497_end_0 = const()[name = tensor("op_497_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_497_end_mask_0 = const()[name = tensor("op_497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_497_cast_fp16 = slice_by_index(begin = var_497_begin_0, end = var_497_end_0, end_mask = var_497_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_497_cast_fp16")]; + tensor var_499_begin_0 = const()[name = tensor("op_499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_499_end_0 = const()[name = tensor("op_499_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_499_end_mask_0 = const()[name = tensor("op_499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_499_cast_fp16 = slice_by_index(begin = var_499_begin_0, end = var_499_end_0, end_mask = var_499_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_499_cast_fp16")]; + tensor var_503_begin_0 = const()[name = tensor("op_503_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_503_end_0 = const()[name = tensor("op_503_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_503_end_mask_0 = const()[name = tensor("op_503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_503_cast_fp16 = slice_by_index(begin = var_503_begin_0, end = var_503_end_0, end_mask = var_503_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_503_cast_fp16")]; + tensor var_507_begin_0 = const()[name = tensor("op_507_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_507_end_0 = const()[name = tensor("op_507_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_507_end_mask_0 = const()[name = tensor("op_507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_507_cast_fp16")]; + tensor var_511_begin_0 = const()[name = tensor("op_511_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_511_end_0 = const()[name = tensor("op_511_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_511_end_mask_0 = const()[name = tensor("op_511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_511_cast_fp16")]; + tensor var_515_begin_0 = const()[name = tensor("op_515_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_515_end_0 = const()[name = tensor("op_515_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_515_end_mask_0 = const()[name = tensor("op_515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_515_cast_fp16")]; + tensor var_519_begin_0 = const()[name = tensor("op_519_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_519_end_0 = const()[name = tensor("op_519_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_519_end_mask_0 = const()[name = tensor("op_519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_519_cast_fp16")]; + tensor var_523_begin_0 = const()[name = tensor("op_523_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_523_end_0 = const()[name = tensor("op_523_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_523_end_mask_0 = const()[name = tensor("op_523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_523_cast_fp16 = slice_by_index(begin = var_523_begin_0, end = var_523_end_0, end_mask = var_523_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_523_cast_fp16")]; + tensor var_527_begin_0 = const()[name = tensor("op_527_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_527_end_0 = const()[name = tensor("op_527_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_527_end_mask_0 = const()[name = tensor("op_527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_527_cast_fp16 = slice_by_index(begin = var_527_begin_0, end = var_527_end_0, end_mask = var_527_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_527_cast_fp16")]; + tensor var_531_begin_0 = const()[name = tensor("op_531_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_531_end_0 = const()[name = tensor("op_531_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_531_end_mask_0 = const()[name = tensor("op_531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_531_cast_fp16 = slice_by_index(begin = var_531_begin_0, end = var_531_end_0, end_mask = var_531_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_531_cast_fp16")]; + tensor var_535_begin_0 = const()[name = tensor("op_535_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_535_end_0 = const()[name = tensor("op_535_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_535_end_mask_0 = const()[name = tensor("op_535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_535_cast_fp16 = slice_by_index(begin = var_535_begin_0, end = var_535_end_0, end_mask = var_535_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_535_cast_fp16")]; + tensor var_539_equation_0 = const()[name = tensor("op_539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_539_cast_fp16 = einsum(equation = var_539_equation_0, values = (var_461_cast_fp16, var_418_cast_fp16))[name = tensor("op_539_cast_fp16")]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1_cast_fp16 = mul(x = var_539_cast_fp16, y = var_540_to_fp16)[name = tensor("aw_1_cast_fp16")]; + tensor var_543_equation_0 = const()[name = tensor("op_543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_543_cast_fp16 = einsum(equation = var_543_equation_0, values = (var_465_cast_fp16, var_422_cast_fp16))[name = tensor("op_543_cast_fp16")]; + tensor var_544_to_fp16 = const()[name = tensor("op_544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3_cast_fp16 = mul(x = var_543_cast_fp16, y = var_544_to_fp16)[name = tensor("aw_3_cast_fp16")]; + tensor var_547_equation_0 = const()[name = tensor("op_547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_547_cast_fp16 = einsum(equation = var_547_equation_0, values = (var_469_cast_fp16, var_426_cast_fp16))[name = tensor("op_547_cast_fp16")]; + tensor var_548_to_fp16 = const()[name = tensor("op_548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5_cast_fp16 = mul(x = var_547_cast_fp16, y = var_548_to_fp16)[name = tensor("aw_5_cast_fp16")]; + tensor var_551_equation_0 = const()[name = tensor("op_551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_551_cast_fp16 = einsum(equation = var_551_equation_0, values = (var_473_cast_fp16, var_430_cast_fp16))[name = tensor("op_551_cast_fp16")]; + tensor var_552_to_fp16 = const()[name = tensor("op_552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_7_cast_fp16 = mul(x = var_551_cast_fp16, y = var_552_to_fp16)[name = tensor("aw_7_cast_fp16")]; + tensor var_555_equation_0 = const()[name = tensor("op_555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_477_cast_fp16, var_434_cast_fp16))[name = tensor("op_555_cast_fp16")]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_9_cast_fp16 = mul(x = var_555_cast_fp16, y = var_556_to_fp16)[name = tensor("aw_9_cast_fp16")]; + tensor var_559_equation_0 = const()[name = tensor("op_559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_559_cast_fp16 = einsum(equation = var_559_equation_0, values = (var_481_cast_fp16, var_438_cast_fp16))[name = tensor("op_559_cast_fp16")]; + tensor var_560_to_fp16 = const()[name = tensor("op_560_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_11_cast_fp16 = mul(x = var_559_cast_fp16, y = var_560_to_fp16)[name = tensor("aw_11_cast_fp16")]; + tensor var_563_equation_0 = const()[name = tensor("op_563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_563_cast_fp16 = einsum(equation = var_563_equation_0, values = (var_485_cast_fp16, var_442_cast_fp16))[name = tensor("op_563_cast_fp16")]; + tensor var_564_to_fp16 = const()[name = tensor("op_564_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_13_cast_fp16 = mul(x = var_563_cast_fp16, y = var_564_to_fp16)[name = tensor("aw_13_cast_fp16")]; + tensor var_567_equation_0 = const()[name = tensor("op_567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_489_cast_fp16, var_446_cast_fp16))[name = tensor("op_567_cast_fp16")]; + tensor var_568_to_fp16 = const()[name = tensor("op_568_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_15_cast_fp16 = mul(x = var_567_cast_fp16, y = var_568_to_fp16)[name = tensor("aw_15_cast_fp16")]; + tensor var_571_equation_0 = const()[name = tensor("op_571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_571_cast_fp16 = einsum(equation = var_571_equation_0, values = (var_493_cast_fp16, var_450_cast_fp16))[name = tensor("op_571_cast_fp16")]; + tensor var_572_to_fp16 = const()[name = tensor("op_572_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_17_cast_fp16 = mul(x = var_571_cast_fp16, y = var_572_to_fp16)[name = tensor("aw_17_cast_fp16")]; + tensor var_575_equation_0 = const()[name = tensor("op_575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_575_cast_fp16 = einsum(equation = var_575_equation_0, values = (var_497_cast_fp16, var_454_cast_fp16))[name = tensor("op_575_cast_fp16")]; + tensor var_576_to_fp16 = const()[name = tensor("op_576_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_19_cast_fp16 = mul(x = var_575_cast_fp16, y = var_576_to_fp16)[name = tensor("aw_19_cast_fp16")]; + tensor var_578_cast_fp16 = softmax(axis = var_288, x = aw_1_cast_fp16)[name = tensor("op_578_cast_fp16")]; + tensor var_579_cast_fp16 = softmax(axis = var_288, x = aw_3_cast_fp16)[name = tensor("op_579_cast_fp16")]; + tensor var_580_cast_fp16 = softmax(axis = var_288, x = aw_5_cast_fp16)[name = tensor("op_580_cast_fp16")]; + tensor var_581_cast_fp16 = softmax(axis = var_288, x = aw_7_cast_fp16)[name = tensor("op_581_cast_fp16")]; + tensor var_582_cast_fp16 = softmax(axis = var_288, x = aw_9_cast_fp16)[name = tensor("op_582_cast_fp16")]; + tensor var_583_cast_fp16 = softmax(axis = var_288, x = aw_11_cast_fp16)[name = tensor("op_583_cast_fp16")]; + tensor var_584_cast_fp16 = softmax(axis = var_288, x = aw_13_cast_fp16)[name = tensor("op_584_cast_fp16")]; + tensor var_585_cast_fp16 = softmax(axis = var_288, x = aw_15_cast_fp16)[name = tensor("op_585_cast_fp16")]; + tensor var_586_cast_fp16 = softmax(axis = var_288, x = aw_17_cast_fp16)[name = tensor("op_586_cast_fp16")]; + tensor var_587_cast_fp16 = softmax(axis = var_288, x = aw_19_cast_fp16)[name = tensor("op_587_cast_fp16")]; + tensor var_589_equation_0 = const()[name = tensor("op_589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_589_cast_fp16 = einsum(equation = var_589_equation_0, values = (var_499_cast_fp16, var_578_cast_fp16))[name = tensor("op_589_cast_fp16")]; + tensor var_591_equation_0 = const()[name = tensor("op_591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_591_cast_fp16 = einsum(equation = var_591_equation_0, values = (var_503_cast_fp16, var_579_cast_fp16))[name = tensor("op_591_cast_fp16")]; + tensor var_593_equation_0 = const()[name = tensor("op_593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_593_cast_fp16 = einsum(equation = var_593_equation_0, values = (var_507_cast_fp16, var_580_cast_fp16))[name = tensor("op_593_cast_fp16")]; + tensor var_595_equation_0 = const()[name = tensor("op_595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_595_cast_fp16 = einsum(equation = var_595_equation_0, values = (var_511_cast_fp16, var_581_cast_fp16))[name = tensor("op_595_cast_fp16")]; + tensor var_597_equation_0 = const()[name = tensor("op_597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_597_cast_fp16 = einsum(equation = var_597_equation_0, values = (var_515_cast_fp16, var_582_cast_fp16))[name = tensor("op_597_cast_fp16")]; + tensor var_599_equation_0 = const()[name = tensor("op_599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_599_cast_fp16 = einsum(equation = var_599_equation_0, values = (var_519_cast_fp16, var_583_cast_fp16))[name = tensor("op_599_cast_fp16")]; + tensor var_601_equation_0 = const()[name = tensor("op_601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_601_cast_fp16 = einsum(equation = var_601_equation_0, values = (var_523_cast_fp16, var_584_cast_fp16))[name = tensor("op_601_cast_fp16")]; + tensor var_603_equation_0 = const()[name = tensor("op_603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_603_cast_fp16 = einsum(equation = var_603_equation_0, values = (var_527_cast_fp16, var_585_cast_fp16))[name = tensor("op_603_cast_fp16")]; + tensor var_605_equation_0 = const()[name = tensor("op_605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_605_cast_fp16 = einsum(equation = var_605_equation_0, values = (var_531_cast_fp16, var_586_cast_fp16))[name = tensor("op_605_cast_fp16")]; + tensor var_607_equation_0 = const()[name = tensor("op_607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_607_cast_fp16 = einsum(equation = var_607_equation_0, values = (var_535_cast_fp16, var_587_cast_fp16))[name = tensor("op_607_cast_fp16")]; + tensor input_61_interleave_0 = const()[name = tensor("input_61_interleave_0"), val = tensor(false)]; + tensor input_61_cast_fp16 = concat(axis = var_288, interleave = input_61_interleave_0, values = (var_589_cast_fp16, var_591_cast_fp16, var_593_cast_fp16, var_595_cast_fp16, var_597_cast_fp16, var_599_cast_fp16, var_601_cast_fp16, var_603_cast_fp16, var_605_cast_fp16, var_607_cast_fp16))[name = tensor("input_61_cast_fp16")]; + tensor var_617_pad_type_0 = const()[name = tensor("op_617_pad_type_0"), val = tensor("valid")]; + tensor var_617_strides_0 = const()[name = tensor("op_617_strides_0"), val = tensor([1, 1])]; + tensor var_617_pad_0 = const()[name = tensor("op_617_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_617_dilations_0 = const()[name = tensor("op_617_dilations_0"), val = tensor([1, 1])]; + tensor var_617_groups_0 = const()[name = tensor("op_617_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15739136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16046400))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16046592)))]; + tensor var_617_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_617_dilations_0, groups = var_617_groups_0, pad = var_617_pad_0, pad_type = var_617_pad_type_0, strides = var_617_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor("op_617_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = var_617_cast_fp16, y = inputs_1_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; + tensor hidden_states_19_axes_0 = const()[name = tensor("hidden_states_19_axes_0"), val = tensor([1])]; + tensor hidden_states_19_gamma_0_to_fp16 = const()[name = tensor("hidden_states_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16047936)))]; + tensor hidden_states_19_beta_0_to_fp16 = const()[name = tensor("hidden_states_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16049280)))]; + tensor var_627_to_fp16 = const()[name = tensor("op_627_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = hidden_states_19_beta_0_to_fp16, epsilon = var_627_to_fp16, gamma = hidden_states_19_gamma_0_to_fp16, x = inputs_3_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; + tensor q_3_pad_type_0 = const()[name = tensor("q_3_pad_type_0"), val = tensor("valid")]; + tensor q_3_strides_0 = const()[name = tensor("q_3_strides_0"), val = tensor([1, 1])]; + tensor q_3_pad_0 = const()[name = tensor("q_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_3_dilations_0 = const()[name = tensor("q_3_dilations_0"), val = tensor([1, 1])]; + tensor q_3_groups_0 = const()[name = tensor("q_3_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16050624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16357888))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_3_cast_fp16 = conv(dilations = q_3_dilations_0, groups = q_3_groups_0, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = q_3_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor("q_3_cast_fp16")]; + tensor k_5_pad_type_0 = const()[name = tensor("k_5_pad_type_0"), val = tensor("valid")]; + tensor k_5_strides_0 = const()[name = tensor("k_5_strides_0"), val = tensor([1, 1])]; + tensor k_5_pad_0 = const()[name = tensor("k_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_5_dilations_0 = const()[name = tensor("k_5_dilations_0"), val = tensor([1, 1])]; + tensor k_5_groups_0 = const()[name = tensor("k_5_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16358080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17341184))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor k_5_cast_fp16 = conv(dilations = k_5_dilations_0, groups = k_5_groups_0, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = k_5_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_5_cast_fp16")]; + tensor v_3_pad_type_0 = const()[name = tensor("v_3_pad_type_0"), val = tensor("valid")]; + tensor v_3_strides_0 = const()[name = tensor("v_3_strides_0"), val = tensor([1, 1])]; + tensor v_3_pad_0 = const()[name = tensor("v_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_3_dilations_0 = const()[name = tensor("v_3_dilations_0"), val = tensor([1, 1])]; + tensor v_3_groups_0 = const()[name = tensor("v_3_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17341376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18324480))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor v_3_cast_fp16 = conv(dilations = v_3_dilations_0, groups = v_3_groups_0, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = v_3_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_3_cast_fp16")]; + tensor var_660_begin_0 = const()[name = tensor("op_660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_660_end_0 = const()[name = tensor("op_660_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_660_end_mask_0 = const()[name = tensor("op_660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_660_cast_fp16 = slice_by_index(begin = var_660_begin_0, end = var_660_end_0, end_mask = var_660_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_660_cast_fp16")]; + tensor var_664_begin_0 = const()[name = tensor("op_664_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_664_end_0 = const()[name = tensor("op_664_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_664_end_mask_0 = const()[name = tensor("op_664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_664_cast_fp16 = slice_by_index(begin = var_664_begin_0, end = var_664_end_0, end_mask = var_664_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_664_cast_fp16")]; + tensor var_668_begin_0 = const()[name = tensor("op_668_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_668_end_0 = const()[name = tensor("op_668_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_668_end_mask_0 = const()[name = tensor("op_668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_668_cast_fp16 = slice_by_index(begin = var_668_begin_0, end = var_668_end_0, end_mask = var_668_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_668_cast_fp16")]; + tensor var_672_begin_0 = const()[name = tensor("op_672_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_672_end_0 = const()[name = tensor("op_672_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_672_end_mask_0 = const()[name = tensor("op_672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_672_cast_fp16 = slice_by_index(begin = var_672_begin_0, end = var_672_end_0, end_mask = var_672_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_672_cast_fp16")]; + tensor var_676_begin_0 = const()[name = tensor("op_676_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_676_end_0 = const()[name = tensor("op_676_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_676_end_mask_0 = const()[name = tensor("op_676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_676_cast_fp16 = slice_by_index(begin = var_676_begin_0, end = var_676_end_0, end_mask = var_676_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_676_cast_fp16")]; + tensor var_680_begin_0 = const()[name = tensor("op_680_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_680_end_0 = const()[name = tensor("op_680_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_680_end_mask_0 = const()[name = tensor("op_680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_680_cast_fp16 = slice_by_index(begin = var_680_begin_0, end = var_680_end_0, end_mask = var_680_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_680_cast_fp16")]; + tensor var_684_begin_0 = const()[name = tensor("op_684_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_684_end_0 = const()[name = tensor("op_684_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_684_end_mask_0 = const()[name = tensor("op_684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_684_cast_fp16 = slice_by_index(begin = var_684_begin_0, end = var_684_end_0, end_mask = var_684_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_684_cast_fp16")]; + tensor var_688_begin_0 = const()[name = tensor("op_688_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_688_end_0 = const()[name = tensor("op_688_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_688_end_mask_0 = const()[name = tensor("op_688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_688_cast_fp16 = slice_by_index(begin = var_688_begin_0, end = var_688_end_0, end_mask = var_688_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_688_cast_fp16")]; + tensor var_692_begin_0 = const()[name = tensor("op_692_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_692_end_0 = const()[name = tensor("op_692_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_692_end_mask_0 = const()[name = tensor("op_692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_692_cast_fp16 = slice_by_index(begin = var_692_begin_0, end = var_692_end_0, end_mask = var_692_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_692_cast_fp16")]; + tensor var_696_begin_0 = const()[name = tensor("op_696_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_696_end_0 = const()[name = tensor("op_696_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_696_end_mask_0 = const()[name = tensor("op_696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_696_cast_fp16 = slice_by_index(begin = var_696_begin_0, end = var_696_end_0, end_mask = var_696_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_696_cast_fp16")]; + tensor k_7_perm_0 = const()[name = tensor("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_703_begin_0 = const()[name = tensor("op_703_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_703_end_0 = const()[name = tensor("op_703_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_703_end_mask_0 = const()[name = tensor("op_703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = k_5_cast_fp16)[name = tensor("transpose_66")]; + tensor var_703_cast_fp16 = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_703_cast_fp16")]; + tensor var_707_begin_0 = const()[name = tensor("op_707_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_707_end_0 = const()[name = tensor("op_707_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_707_end_mask_0 = const()[name = tensor("op_707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_707_cast_fp16")]; + tensor var_711_begin_0 = const()[name = tensor("op_711_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_711_end_0 = const()[name = tensor("op_711_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_711_end_mask_0 = const()[name = tensor("op_711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_711_cast_fp16 = slice_by_index(begin = var_711_begin_0, end = var_711_end_0, end_mask = var_711_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_711_cast_fp16")]; + tensor var_715_begin_0 = const()[name = tensor("op_715_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_715_end_0 = const()[name = tensor("op_715_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_715_end_mask_0 = const()[name = tensor("op_715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_715_cast_fp16 = slice_by_index(begin = var_715_begin_0, end = var_715_end_0, end_mask = var_715_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_715_cast_fp16")]; + tensor var_719_begin_0 = const()[name = tensor("op_719_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_719_end_0 = const()[name = tensor("op_719_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_719_end_mask_0 = const()[name = tensor("op_719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_719_cast_fp16 = slice_by_index(begin = var_719_begin_0, end = var_719_end_0, end_mask = var_719_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_719_cast_fp16")]; + tensor var_723_begin_0 = const()[name = tensor("op_723_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_723_end_0 = const()[name = tensor("op_723_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_723_end_mask_0 = const()[name = tensor("op_723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_723_cast_fp16 = slice_by_index(begin = var_723_begin_0, end = var_723_end_0, end_mask = var_723_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_723_cast_fp16")]; + tensor var_727_begin_0 = const()[name = tensor("op_727_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_727_end_0 = const()[name = tensor("op_727_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_727_end_mask_0 = const()[name = tensor("op_727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_727_cast_fp16 = slice_by_index(begin = var_727_begin_0, end = var_727_end_0, end_mask = var_727_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_727_cast_fp16")]; + tensor var_731_begin_0 = const()[name = tensor("op_731_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_731_end_0 = const()[name = tensor("op_731_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_731_end_mask_0 = const()[name = tensor("op_731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_731_cast_fp16 = slice_by_index(begin = var_731_begin_0, end = var_731_end_0, end_mask = var_731_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_731_cast_fp16")]; + tensor var_735_begin_0 = const()[name = tensor("op_735_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_735_end_0 = const()[name = tensor("op_735_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_735_end_mask_0 = const()[name = tensor("op_735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_735_cast_fp16 = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_735_cast_fp16")]; + tensor var_739_begin_0 = const()[name = tensor("op_739_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_739_end_0 = const()[name = tensor("op_739_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_739_end_mask_0 = const()[name = tensor("op_739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_739_cast_fp16 = slice_by_index(begin = var_739_begin_0, end = var_739_end_0, end_mask = var_739_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_739_cast_fp16")]; + tensor var_741_begin_0 = const()[name = tensor("op_741_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_741_end_0 = const()[name = tensor("op_741_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_741_end_mask_0 = const()[name = tensor("op_741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_741_cast_fp16 = slice_by_index(begin = var_741_begin_0, end = var_741_end_0, end_mask = var_741_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_741_cast_fp16")]; + tensor var_745_begin_0 = const()[name = tensor("op_745_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_745_end_0 = const()[name = tensor("op_745_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_745_end_mask_0 = const()[name = tensor("op_745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_745_cast_fp16 = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_745_cast_fp16")]; + tensor var_749_begin_0 = const()[name = tensor("op_749_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_749_end_0 = const()[name = tensor("op_749_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_749_end_mask_0 = const()[name = tensor("op_749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_749_cast_fp16 = slice_by_index(begin = var_749_begin_0, end = var_749_end_0, end_mask = var_749_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_749_cast_fp16")]; + tensor var_753_begin_0 = const()[name = tensor("op_753_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_753_end_0 = const()[name = tensor("op_753_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_753_end_mask_0 = const()[name = tensor("op_753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_753_cast_fp16")]; + tensor var_757_begin_0 = const()[name = tensor("op_757_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_757_end_0 = const()[name = tensor("op_757_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_757_end_mask_0 = const()[name = tensor("op_757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_757_cast_fp16 = slice_by_index(begin = var_757_begin_0, end = var_757_end_0, end_mask = var_757_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_757_cast_fp16")]; + tensor var_761_begin_0 = const()[name = tensor("op_761_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_761_end_0 = const()[name = tensor("op_761_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_761_end_mask_0 = const()[name = tensor("op_761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_761_cast_fp16 = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_761_cast_fp16")]; + tensor var_765_begin_0 = const()[name = tensor("op_765_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_765_end_0 = const()[name = tensor("op_765_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_765_end_mask_0 = const()[name = tensor("op_765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_765_cast_fp16 = slice_by_index(begin = var_765_begin_0, end = var_765_end_0, end_mask = var_765_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_765_cast_fp16")]; + tensor var_769_begin_0 = const()[name = tensor("op_769_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_769_end_0 = const()[name = tensor("op_769_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_769_end_mask_0 = const()[name = tensor("op_769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_769_cast_fp16 = slice_by_index(begin = var_769_begin_0, end = var_769_end_0, end_mask = var_769_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_769_cast_fp16")]; + tensor var_773_begin_0 = const()[name = tensor("op_773_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_773_end_0 = const()[name = tensor("op_773_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_773_end_mask_0 = const()[name = tensor("op_773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_773_cast_fp16 = slice_by_index(begin = var_773_begin_0, end = var_773_end_0, end_mask = var_773_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_773_cast_fp16")]; + tensor var_777_begin_0 = const()[name = tensor("op_777_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_777_end_0 = const()[name = tensor("op_777_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_777_end_mask_0 = const()[name = tensor("op_777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_777_cast_fp16 = slice_by_index(begin = var_777_begin_0, end = var_777_end_0, end_mask = var_777_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_777_cast_fp16")]; + tensor var_781_equation_0 = const()[name = tensor("op_781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_781_cast_fp16 = einsum(equation = var_781_equation_0, values = (var_703_cast_fp16, var_660_cast_fp16))[name = tensor("op_781_cast_fp16")]; + tensor var_782_to_fp16 = const()[name = tensor("op_782_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_21_cast_fp16 = mul(x = var_781_cast_fp16, y = var_782_to_fp16)[name = tensor("aw_21_cast_fp16")]; + tensor var_785_equation_0 = const()[name = tensor("op_785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_785_cast_fp16 = einsum(equation = var_785_equation_0, values = (var_707_cast_fp16, var_664_cast_fp16))[name = tensor("op_785_cast_fp16")]; + tensor var_786_to_fp16 = const()[name = tensor("op_786_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_23_cast_fp16 = mul(x = var_785_cast_fp16, y = var_786_to_fp16)[name = tensor("aw_23_cast_fp16")]; + tensor var_789_equation_0 = const()[name = tensor("op_789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_789_cast_fp16 = einsum(equation = var_789_equation_0, values = (var_711_cast_fp16, var_668_cast_fp16))[name = tensor("op_789_cast_fp16")]; + tensor var_790_to_fp16 = const()[name = tensor("op_790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_25_cast_fp16 = mul(x = var_789_cast_fp16, y = var_790_to_fp16)[name = tensor("aw_25_cast_fp16")]; + tensor var_793_equation_0 = const()[name = tensor("op_793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_793_cast_fp16 = einsum(equation = var_793_equation_0, values = (var_715_cast_fp16, var_672_cast_fp16))[name = tensor("op_793_cast_fp16")]; + tensor var_794_to_fp16 = const()[name = tensor("op_794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_27_cast_fp16 = mul(x = var_793_cast_fp16, y = var_794_to_fp16)[name = tensor("aw_27_cast_fp16")]; + tensor var_797_equation_0 = const()[name = tensor("op_797_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_797_cast_fp16 = einsum(equation = var_797_equation_0, values = (var_719_cast_fp16, var_676_cast_fp16))[name = tensor("op_797_cast_fp16")]; + tensor var_798_to_fp16 = const()[name = tensor("op_798_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_29_cast_fp16 = mul(x = var_797_cast_fp16, y = var_798_to_fp16)[name = tensor("aw_29_cast_fp16")]; + tensor var_801_equation_0 = const()[name = tensor("op_801_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_801_cast_fp16 = einsum(equation = var_801_equation_0, values = (var_723_cast_fp16, var_680_cast_fp16))[name = tensor("op_801_cast_fp16")]; + tensor var_802_to_fp16 = const()[name = tensor("op_802_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_31_cast_fp16 = mul(x = var_801_cast_fp16, y = var_802_to_fp16)[name = tensor("aw_31_cast_fp16")]; + tensor var_805_equation_0 = const()[name = tensor("op_805_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_805_cast_fp16 = einsum(equation = var_805_equation_0, values = (var_727_cast_fp16, var_684_cast_fp16))[name = tensor("op_805_cast_fp16")]; + tensor var_806_to_fp16 = const()[name = tensor("op_806_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_33_cast_fp16 = mul(x = var_805_cast_fp16, y = var_806_to_fp16)[name = tensor("aw_33_cast_fp16")]; + tensor var_809_equation_0 = const()[name = tensor("op_809_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_809_cast_fp16 = einsum(equation = var_809_equation_0, values = (var_731_cast_fp16, var_688_cast_fp16))[name = tensor("op_809_cast_fp16")]; + tensor var_810_to_fp16 = const()[name = tensor("op_810_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_35_cast_fp16 = mul(x = var_809_cast_fp16, y = var_810_to_fp16)[name = tensor("aw_35_cast_fp16")]; + tensor var_813_equation_0 = const()[name = tensor("op_813_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_813_cast_fp16 = einsum(equation = var_813_equation_0, values = (var_735_cast_fp16, var_692_cast_fp16))[name = tensor("op_813_cast_fp16")]; + tensor var_814_to_fp16 = const()[name = tensor("op_814_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_37_cast_fp16 = mul(x = var_813_cast_fp16, y = var_814_to_fp16)[name = tensor("aw_37_cast_fp16")]; + tensor var_817_equation_0 = const()[name = tensor("op_817_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_817_cast_fp16 = einsum(equation = var_817_equation_0, values = (var_739_cast_fp16, var_696_cast_fp16))[name = tensor("op_817_cast_fp16")]; + tensor var_818_to_fp16 = const()[name = tensor("op_818_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_39_cast_fp16 = mul(x = var_817_cast_fp16, y = var_818_to_fp16)[name = tensor("aw_39_cast_fp16")]; + tensor var_820_cast_fp16 = softmax(axis = var_288, x = aw_21_cast_fp16)[name = tensor("op_820_cast_fp16")]; + tensor var_821_cast_fp16 = softmax(axis = var_288, x = aw_23_cast_fp16)[name = tensor("op_821_cast_fp16")]; + tensor var_822_cast_fp16 = softmax(axis = var_288, x = aw_25_cast_fp16)[name = tensor("op_822_cast_fp16")]; + tensor var_823_cast_fp16 = softmax(axis = var_288, x = aw_27_cast_fp16)[name = tensor("op_823_cast_fp16")]; + tensor var_824_cast_fp16 = softmax(axis = var_288, x = aw_29_cast_fp16)[name = tensor("op_824_cast_fp16")]; + tensor var_825_cast_fp16 = softmax(axis = var_288, x = aw_31_cast_fp16)[name = tensor("op_825_cast_fp16")]; + tensor var_826_cast_fp16 = softmax(axis = var_288, x = aw_33_cast_fp16)[name = tensor("op_826_cast_fp16")]; + tensor var_827_cast_fp16 = softmax(axis = var_288, x = aw_35_cast_fp16)[name = tensor("op_827_cast_fp16")]; + tensor var_828_cast_fp16 = softmax(axis = var_288, x = aw_37_cast_fp16)[name = tensor("op_828_cast_fp16")]; + tensor var_829_cast_fp16 = softmax(axis = var_288, x = aw_39_cast_fp16)[name = tensor("op_829_cast_fp16")]; + tensor var_831_equation_0 = const()[name = tensor("op_831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_831_cast_fp16 = einsum(equation = var_831_equation_0, values = (var_741_cast_fp16, var_820_cast_fp16))[name = tensor("op_831_cast_fp16")]; + tensor var_833_equation_0 = const()[name = tensor("op_833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_833_cast_fp16 = einsum(equation = var_833_equation_0, values = (var_745_cast_fp16, var_821_cast_fp16))[name = tensor("op_833_cast_fp16")]; + tensor var_835_equation_0 = const()[name = tensor("op_835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_835_cast_fp16 = einsum(equation = var_835_equation_0, values = (var_749_cast_fp16, var_822_cast_fp16))[name = tensor("op_835_cast_fp16")]; + tensor var_837_equation_0 = const()[name = tensor("op_837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_837_cast_fp16 = einsum(equation = var_837_equation_0, values = (var_753_cast_fp16, var_823_cast_fp16))[name = tensor("op_837_cast_fp16")]; + tensor var_839_equation_0 = const()[name = tensor("op_839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_839_cast_fp16 = einsum(equation = var_839_equation_0, values = (var_757_cast_fp16, var_824_cast_fp16))[name = tensor("op_839_cast_fp16")]; + tensor var_841_equation_0 = const()[name = tensor("op_841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_841_cast_fp16 = einsum(equation = var_841_equation_0, values = (var_761_cast_fp16, var_825_cast_fp16))[name = tensor("op_841_cast_fp16")]; + tensor var_843_equation_0 = const()[name = tensor("op_843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_843_cast_fp16 = einsum(equation = var_843_equation_0, values = (var_765_cast_fp16, var_826_cast_fp16))[name = tensor("op_843_cast_fp16")]; + tensor var_845_equation_0 = const()[name = tensor("op_845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_845_cast_fp16 = einsum(equation = var_845_equation_0, values = (var_769_cast_fp16, var_827_cast_fp16))[name = tensor("op_845_cast_fp16")]; + tensor var_847_equation_0 = const()[name = tensor("op_847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_847_cast_fp16 = einsum(equation = var_847_equation_0, values = (var_773_cast_fp16, var_828_cast_fp16))[name = tensor("op_847_cast_fp16")]; + tensor var_849_equation_0 = const()[name = tensor("op_849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_849_cast_fp16 = einsum(equation = var_849_equation_0, values = (var_777_cast_fp16, var_829_cast_fp16))[name = tensor("op_849_cast_fp16")]; + tensor input_63_interleave_0 = const()[name = tensor("input_63_interleave_0"), val = tensor(false)]; + tensor input_63_cast_fp16 = concat(axis = var_288, interleave = input_63_interleave_0, values = (var_831_cast_fp16, var_833_cast_fp16, var_835_cast_fp16, var_837_cast_fp16, var_839_cast_fp16, var_841_cast_fp16, var_843_cast_fp16, var_845_cast_fp16, var_847_cast_fp16, var_849_cast_fp16))[name = tensor("input_63_cast_fp16")]; + tensor var_859_pad_type_0 = const()[name = tensor("op_859_pad_type_0"), val = tensor("valid")]; + tensor var_859_strides_0 = const()[name = tensor("op_859_strides_0"), val = tensor([1, 1])]; + tensor var_859_pad_0 = const()[name = tensor("op_859_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_859_dilations_0 = const()[name = tensor("op_859_dilations_0"), val = tensor([1, 1])]; + tensor var_859_groups_0 = const()[name = tensor("op_859_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18324672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18631936))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18632128)))]; + tensor var_859_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_859_dilations_0, groups = var_859_groups_0, pad = var_859_pad_0, pad_type = var_859_pad_type_0, strides = var_859_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor("op_859_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = var_859_cast_fp16, y = inputs_3_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; + tensor input_65_axes_0 = const()[name = tensor("input_65_axes_0"), val = tensor([1])]; + tensor input_65_gamma_0_to_fp16 = const()[name = tensor("input_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18633472)))]; + tensor input_65_beta_0_to_fp16 = const()[name = tensor("input_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18634816)))]; + tensor var_869_to_fp16 = const()[name = tensor("op_869_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_65_cast_fp16 = layer_norm(axes = input_65_axes_0, beta = input_65_beta_0_to_fp16, epsilon = var_869_to_fp16, gamma = input_65_gamma_0_to_fp16, x = inputs_5_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor var_889_pad_type_0 = const()[name = tensor("op_889_pad_type_0"), val = tensor("valid")]; + tensor var_889_strides_0 = const()[name = tensor("op_889_strides_0"), val = tensor([1, 1])]; + tensor var_889_pad_0 = const()[name = tensor("op_889_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_889_dilations_0 = const()[name = tensor("op_889_dilations_0"), val = tensor([1, 1])]; + tensor var_889_groups_0 = const()[name = tensor("op_889_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18636160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21093824))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21094016)))]; + tensor var_889_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_889_dilations_0, groups = var_889_groups_0, pad = var_889_pad_0, pad_type = var_889_pad_type_0, strides = var_889_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor("op_889_cast_fp16")]; + tensor var_890_split_sizes_0 = const()[name = tensor("op_890_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_890_axis_0 = const()[name = tensor("op_890_axis_0"), val = tensor(1)]; + tensor var_890_cast_fp16_0, tensor var_890_cast_fp16_1 = split(axis = var_890_axis_0, split_sizes = var_890_split_sizes_0, x = var_889_cast_fp16)[name = tensor("op_890_cast_fp16")]; + tensor var_892_mode_0 = const()[name = tensor("op_892_mode_0"), val = tensor("EXACT")]; + tensor var_892_cast_fp16 = gelu(mode = var_892_mode_0, x = var_890_cast_fp16_1)[name = tensor("op_892_cast_fp16")]; + tensor input_67_cast_fp16 = mul(x = var_890_cast_fp16_0, y = var_892_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor var_900_pad_type_0 = const()[name = tensor("op_900_pad_type_0"), val = tensor("valid")]; + tensor var_900_strides_0 = const()[name = tensor("op_900_strides_0"), val = tensor([1, 1])]; + tensor var_900_pad_0 = const()[name = tensor("op_900_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_900_dilations_0 = const()[name = tensor("op_900_dilations_0"), val = tensor([1, 1])]; + tensor var_900_groups_0 = const()[name = tensor("op_900_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21104320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22333184))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22333376)))]; + tensor var_900_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_900_dilations_0, groups = var_900_groups_0, pad = var_900_pad_0, pad_type = var_900_pad_type_0, strides = var_900_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = tensor("op_900_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = var_900_cast_fp16, y = inputs_5_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; + tensor hidden_states_23_axes_0 = const()[name = tensor("hidden_states_23_axes_0"), val = tensor([1])]; + tensor hidden_states_23_gamma_0_to_fp16 = const()[name = tensor("hidden_states_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22334720)))]; + tensor hidden_states_23_beta_0_to_fp16 = const()[name = tensor("hidden_states_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22336064)))]; + tensor var_916_to_fp16 = const()[name = tensor("op_916_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_23_cast_fp16 = layer_norm(axes = hidden_states_23_axes_0, beta = hidden_states_23_beta_0_to_fp16, epsilon = var_916_to_fp16, gamma = hidden_states_23_gamma_0_to_fp16, x = inputs_7_cast_fp16)[name = tensor("hidden_states_23_cast_fp16")]; + tensor q_5_pad_type_0 = const()[name = tensor("q_5_pad_type_0"), val = tensor("valid")]; + tensor q_5_strides_0 = const()[name = tensor("q_5_strides_0"), val = tensor([1, 1])]; + tensor q_5_pad_0 = const()[name = tensor("q_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_5_dilations_0 = const()[name = tensor("q_5_dilations_0"), val = tensor([1, 1])]; + tensor q_5_groups_0 = const()[name = tensor("q_5_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22337408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22644672))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_5_cast_fp16 = conv(dilations = q_5_dilations_0, groups = q_5_groups_0, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = q_5_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_23_cast_fp16)[name = tensor("q_5_cast_fp16")]; + tensor k_9_pad_type_0 = const()[name = tensor("k_9_pad_type_0"), val = tensor("valid")]; + tensor k_9_strides_0 = const()[name = tensor("k_9_strides_0"), val = tensor([1, 1])]; + tensor k_9_pad_0 = const()[name = tensor("k_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_9_dilations_0 = const()[name = tensor("k_9_dilations_0"), val = tensor([1, 1])]; + tensor k_9_groups_0 = const()[name = tensor("k_9_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22644864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22952128))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_9_cast_fp16 = conv(dilations = k_9_dilations_0, groups = k_9_groups_0, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = k_9_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_23_cast_fp16)[name = tensor("k_9_cast_fp16")]; + tensor v_5_pad_type_0 = const()[name = tensor("v_5_pad_type_0"), val = tensor("valid")]; + tensor v_5_strides_0 = const()[name = tensor("v_5_strides_0"), val = tensor([1, 1])]; + tensor v_5_pad_0 = const()[name = tensor("v_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_5_dilations_0 = const()[name = tensor("v_5_dilations_0"), val = tensor([1, 1])]; + tensor v_5_groups_0 = const()[name = tensor("v_5_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22952320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23259584))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_5_cast_fp16 = conv(dilations = v_5_dilations_0, groups = v_5_groups_0, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = v_5_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_23_cast_fp16)[name = tensor("v_5_cast_fp16")]; + tensor var_949_begin_0 = const()[name = tensor("op_949_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_949_end_0 = const()[name = tensor("op_949_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_949_end_mask_0 = const()[name = tensor("op_949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_949_cast_fp16")]; + tensor var_953_begin_0 = const()[name = tensor("op_953_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_953_end_0 = const()[name = tensor("op_953_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_953_end_mask_0 = const()[name = tensor("op_953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_953_cast_fp16 = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_953_cast_fp16")]; + tensor var_957_begin_0 = const()[name = tensor("op_957_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_957_end_0 = const()[name = tensor("op_957_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_957_end_mask_0 = const()[name = tensor("op_957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_957_cast_fp16 = slice_by_index(begin = var_957_begin_0, end = var_957_end_0, end_mask = var_957_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_957_cast_fp16")]; + tensor var_961_begin_0 = const()[name = tensor("op_961_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_961_end_0 = const()[name = tensor("op_961_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_961_end_mask_0 = const()[name = tensor("op_961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_961_cast_fp16 = slice_by_index(begin = var_961_begin_0, end = var_961_end_0, end_mask = var_961_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_961_cast_fp16")]; + tensor var_965_begin_0 = const()[name = tensor("op_965_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_965_end_0 = const()[name = tensor("op_965_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_965_end_mask_0 = const()[name = tensor("op_965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_965_cast_fp16 = slice_by_index(begin = var_965_begin_0, end = var_965_end_0, end_mask = var_965_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_965_cast_fp16")]; + tensor var_969_begin_0 = const()[name = tensor("op_969_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_969_end_0 = const()[name = tensor("op_969_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_969_end_mask_0 = const()[name = tensor("op_969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_969_cast_fp16 = slice_by_index(begin = var_969_begin_0, end = var_969_end_0, end_mask = var_969_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_969_cast_fp16")]; + tensor var_973_begin_0 = const()[name = tensor("op_973_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_973_end_0 = const()[name = tensor("op_973_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_973_end_mask_0 = const()[name = tensor("op_973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_973_cast_fp16")]; + tensor var_977_begin_0 = const()[name = tensor("op_977_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_977_end_0 = const()[name = tensor("op_977_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_977_end_mask_0 = const()[name = tensor("op_977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_977_cast_fp16")]; + tensor var_981_begin_0 = const()[name = tensor("op_981_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_981_end_0 = const()[name = tensor("op_981_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_981_end_mask_0 = const()[name = tensor("op_981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_981_cast_fp16 = slice_by_index(begin = var_981_begin_0, end = var_981_end_0, end_mask = var_981_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_981_cast_fp16")]; + tensor var_985_begin_0 = const()[name = tensor("op_985_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_985_end_0 = const()[name = tensor("op_985_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_985_end_mask_0 = const()[name = tensor("op_985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_985_cast_fp16 = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_985_cast_fp16")]; + tensor k_11_perm_0 = const()[name = tensor("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_992_begin_0 = const()[name = tensor("op_992_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_992_end_0 = const()[name = tensor("op_992_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_992_end_mask_0 = const()[name = tensor("op_992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = k_9_cast_fp16)[name = tensor("transpose_65")]; + tensor var_992_cast_fp16 = slice_by_index(begin = var_992_begin_0, end = var_992_end_0, end_mask = var_992_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_992_cast_fp16")]; + tensor var_996_begin_0 = const()[name = tensor("op_996_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_996_end_0 = const()[name = tensor("op_996_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_996_end_mask_0 = const()[name = tensor("op_996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_996_cast_fp16 = slice_by_index(begin = var_996_begin_0, end = var_996_end_0, end_mask = var_996_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_996_cast_fp16")]; + tensor var_1000_begin_0 = const()[name = tensor("op_1000_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1000_end_0 = const()[name = tensor("op_1000_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_1000_end_mask_0 = const()[name = tensor("op_1000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1000_cast_fp16 = slice_by_index(begin = var_1000_begin_0, end = var_1000_end_0, end_mask = var_1000_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1000_cast_fp16")]; + tensor var_1004_begin_0 = const()[name = tensor("op_1004_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1004_end_0 = const()[name = tensor("op_1004_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_1004_end_mask_0 = const()[name = tensor("op_1004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1004_cast_fp16 = slice_by_index(begin = var_1004_begin_0, end = var_1004_end_0, end_mask = var_1004_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1004_cast_fp16")]; + tensor var_1008_begin_0 = const()[name = tensor("op_1008_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1008_end_0 = const()[name = tensor("op_1008_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_1008_end_mask_0 = const()[name = tensor("op_1008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1008_cast_fp16 = slice_by_index(begin = var_1008_begin_0, end = var_1008_end_0, end_mask = var_1008_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1008_cast_fp16")]; + tensor var_1012_begin_0 = const()[name = tensor("op_1012_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1012_end_0 = const()[name = tensor("op_1012_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_1012_end_mask_0 = const()[name = tensor("op_1012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1012_cast_fp16")]; + tensor var_1016_begin_0 = const()[name = tensor("op_1016_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1016_end_0 = const()[name = tensor("op_1016_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_1016_end_mask_0 = const()[name = tensor("op_1016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1016_cast_fp16 = slice_by_index(begin = var_1016_begin_0, end = var_1016_end_0, end_mask = var_1016_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1016_cast_fp16")]; + tensor var_1020_begin_0 = const()[name = tensor("op_1020_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1020_end_0 = const()[name = tensor("op_1020_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_1020_end_mask_0 = const()[name = tensor("op_1020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1020_cast_fp16 = slice_by_index(begin = var_1020_begin_0, end = var_1020_end_0, end_mask = var_1020_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1020_cast_fp16")]; + tensor var_1024_begin_0 = const()[name = tensor("op_1024_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1024_end_0 = const()[name = tensor("op_1024_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_1024_end_mask_0 = const()[name = tensor("op_1024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1024_cast_fp16 = slice_by_index(begin = var_1024_begin_0, end = var_1024_end_0, end_mask = var_1024_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1024_cast_fp16")]; + tensor var_1028_begin_0 = const()[name = tensor("op_1028_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1028_end_0 = const()[name = tensor("op_1028_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_1028_end_mask_0 = const()[name = tensor("op_1028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1028_cast_fp16 = slice_by_index(begin = var_1028_begin_0, end = var_1028_end_0, end_mask = var_1028_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1028_cast_fp16")]; + tensor var_1030_begin_0 = const()[name = tensor("op_1030_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1030_end_0 = const()[name = tensor("op_1030_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1030_end_mask_0 = const()[name = tensor("op_1030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1030_cast_fp16 = slice_by_index(begin = var_1030_begin_0, end = var_1030_end_0, end_mask = var_1030_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1030_cast_fp16")]; + tensor var_1034_begin_0 = const()[name = tensor("op_1034_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1034_end_0 = const()[name = tensor("op_1034_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1034_end_mask_0 = const()[name = tensor("op_1034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1034_cast_fp16 = slice_by_index(begin = var_1034_begin_0, end = var_1034_end_0, end_mask = var_1034_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1034_cast_fp16")]; + tensor var_1038_begin_0 = const()[name = tensor("op_1038_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1038_end_0 = const()[name = tensor("op_1038_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1038_end_mask_0 = const()[name = tensor("op_1038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1038_cast_fp16 = slice_by_index(begin = var_1038_begin_0, end = var_1038_end_0, end_mask = var_1038_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1038_cast_fp16")]; + tensor var_1042_begin_0 = const()[name = tensor("op_1042_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1042_end_0 = const()[name = tensor("op_1042_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1042_end_mask_0 = const()[name = tensor("op_1042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1042_cast_fp16 = slice_by_index(begin = var_1042_begin_0, end = var_1042_end_0, end_mask = var_1042_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1042_cast_fp16")]; + tensor var_1046_begin_0 = const()[name = tensor("op_1046_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1046_end_0 = const()[name = tensor("op_1046_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1046_end_mask_0 = const()[name = tensor("op_1046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1046_cast_fp16 = slice_by_index(begin = var_1046_begin_0, end = var_1046_end_0, end_mask = var_1046_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1046_cast_fp16")]; + tensor var_1050_begin_0 = const()[name = tensor("op_1050_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1050_end_0 = const()[name = tensor("op_1050_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1050_end_mask_0 = const()[name = tensor("op_1050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1050_cast_fp16 = slice_by_index(begin = var_1050_begin_0, end = var_1050_end_0, end_mask = var_1050_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1050_cast_fp16")]; + tensor var_1054_begin_0 = const()[name = tensor("op_1054_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1054_end_0 = const()[name = tensor("op_1054_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1054_end_mask_0 = const()[name = tensor("op_1054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1054_cast_fp16 = slice_by_index(begin = var_1054_begin_0, end = var_1054_end_0, end_mask = var_1054_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1054_cast_fp16")]; + tensor var_1058_begin_0 = const()[name = tensor("op_1058_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1058_end_0 = const()[name = tensor("op_1058_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1058_end_mask_0 = const()[name = tensor("op_1058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1058_cast_fp16 = slice_by_index(begin = var_1058_begin_0, end = var_1058_end_0, end_mask = var_1058_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1058_cast_fp16")]; + tensor var_1062_begin_0 = const()[name = tensor("op_1062_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1062_end_0 = const()[name = tensor("op_1062_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1062_end_mask_0 = const()[name = tensor("op_1062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1062_cast_fp16 = slice_by_index(begin = var_1062_begin_0, end = var_1062_end_0, end_mask = var_1062_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1062_cast_fp16")]; + tensor var_1066_begin_0 = const()[name = tensor("op_1066_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1066_end_0 = const()[name = tensor("op_1066_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1066_end_mask_0 = const()[name = tensor("op_1066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1066_cast_fp16 = slice_by_index(begin = var_1066_begin_0, end = var_1066_end_0, end_mask = var_1066_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1066_cast_fp16")]; + tensor var_1070_equation_0 = const()[name = tensor("op_1070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1070_cast_fp16 = einsum(equation = var_1070_equation_0, values = (var_992_cast_fp16, var_949_cast_fp16))[name = tensor("op_1070_cast_fp16")]; + tensor var_1071_to_fp16 = const()[name = tensor("op_1071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_41_cast_fp16 = mul(x = var_1070_cast_fp16, y = var_1071_to_fp16)[name = tensor("aw_41_cast_fp16")]; + tensor var_1074_equation_0 = const()[name = tensor("op_1074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1074_cast_fp16 = einsum(equation = var_1074_equation_0, values = (var_996_cast_fp16, var_953_cast_fp16))[name = tensor("op_1074_cast_fp16")]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_43_cast_fp16 = mul(x = var_1074_cast_fp16, y = var_1075_to_fp16)[name = tensor("aw_43_cast_fp16")]; + tensor var_1078_equation_0 = const()[name = tensor("op_1078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1078_cast_fp16 = einsum(equation = var_1078_equation_0, values = (var_1000_cast_fp16, var_957_cast_fp16))[name = tensor("op_1078_cast_fp16")]; + tensor var_1079_to_fp16 = const()[name = tensor("op_1079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_45_cast_fp16 = mul(x = var_1078_cast_fp16, y = var_1079_to_fp16)[name = tensor("aw_45_cast_fp16")]; + tensor var_1082_equation_0 = const()[name = tensor("op_1082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1082_cast_fp16 = einsum(equation = var_1082_equation_0, values = (var_1004_cast_fp16, var_961_cast_fp16))[name = tensor("op_1082_cast_fp16")]; + tensor var_1083_to_fp16 = const()[name = tensor("op_1083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_47_cast_fp16 = mul(x = var_1082_cast_fp16, y = var_1083_to_fp16)[name = tensor("aw_47_cast_fp16")]; + tensor var_1086_equation_0 = const()[name = tensor("op_1086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1086_cast_fp16 = einsum(equation = var_1086_equation_0, values = (var_1008_cast_fp16, var_965_cast_fp16))[name = tensor("op_1086_cast_fp16")]; + tensor var_1087_to_fp16 = const()[name = tensor("op_1087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_49_cast_fp16 = mul(x = var_1086_cast_fp16, y = var_1087_to_fp16)[name = tensor("aw_49_cast_fp16")]; + tensor var_1090_equation_0 = const()[name = tensor("op_1090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1090_cast_fp16 = einsum(equation = var_1090_equation_0, values = (var_1012_cast_fp16, var_969_cast_fp16))[name = tensor("op_1090_cast_fp16")]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_51_cast_fp16 = mul(x = var_1090_cast_fp16, y = var_1091_to_fp16)[name = tensor("aw_51_cast_fp16")]; + tensor var_1094_equation_0 = const()[name = tensor("op_1094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1094_cast_fp16 = einsum(equation = var_1094_equation_0, values = (var_1016_cast_fp16, var_973_cast_fp16))[name = tensor("op_1094_cast_fp16")]; + tensor var_1095_to_fp16 = const()[name = tensor("op_1095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_53_cast_fp16 = mul(x = var_1094_cast_fp16, y = var_1095_to_fp16)[name = tensor("aw_53_cast_fp16")]; + tensor var_1098_equation_0 = const()[name = tensor("op_1098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1098_cast_fp16 = einsum(equation = var_1098_equation_0, values = (var_1020_cast_fp16, var_977_cast_fp16))[name = tensor("op_1098_cast_fp16")]; + tensor var_1099_to_fp16 = const()[name = tensor("op_1099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_55_cast_fp16 = mul(x = var_1098_cast_fp16, y = var_1099_to_fp16)[name = tensor("aw_55_cast_fp16")]; + tensor var_1102_equation_0 = const()[name = tensor("op_1102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1102_cast_fp16 = einsum(equation = var_1102_equation_0, values = (var_1024_cast_fp16, var_981_cast_fp16))[name = tensor("op_1102_cast_fp16")]; + tensor var_1103_to_fp16 = const()[name = tensor("op_1103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_57_cast_fp16 = mul(x = var_1102_cast_fp16, y = var_1103_to_fp16)[name = tensor("aw_57_cast_fp16")]; + tensor var_1106_equation_0 = const()[name = tensor("op_1106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1106_cast_fp16 = einsum(equation = var_1106_equation_0, values = (var_1028_cast_fp16, var_985_cast_fp16))[name = tensor("op_1106_cast_fp16")]; + tensor var_1107_to_fp16 = const()[name = tensor("op_1107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_59_cast_fp16 = mul(x = var_1106_cast_fp16, y = var_1107_to_fp16)[name = tensor("aw_59_cast_fp16")]; + tensor var_1109_cast_fp16 = softmax(axis = var_288, x = aw_41_cast_fp16)[name = tensor("op_1109_cast_fp16")]; + tensor var_1110_cast_fp16 = softmax(axis = var_288, x = aw_43_cast_fp16)[name = tensor("op_1110_cast_fp16")]; + tensor var_1111_cast_fp16 = softmax(axis = var_288, x = aw_45_cast_fp16)[name = tensor("op_1111_cast_fp16")]; + tensor var_1112_cast_fp16 = softmax(axis = var_288, x = aw_47_cast_fp16)[name = tensor("op_1112_cast_fp16")]; + tensor var_1113_cast_fp16 = softmax(axis = var_288, x = aw_49_cast_fp16)[name = tensor("op_1113_cast_fp16")]; + tensor var_1114_cast_fp16 = softmax(axis = var_288, x = aw_51_cast_fp16)[name = tensor("op_1114_cast_fp16")]; + tensor var_1115_cast_fp16 = softmax(axis = var_288, x = aw_53_cast_fp16)[name = tensor("op_1115_cast_fp16")]; + tensor var_1116_cast_fp16 = softmax(axis = var_288, x = aw_55_cast_fp16)[name = tensor("op_1116_cast_fp16")]; + tensor var_1117_cast_fp16 = softmax(axis = var_288, x = aw_57_cast_fp16)[name = tensor("op_1117_cast_fp16")]; + tensor var_1118_cast_fp16 = softmax(axis = var_288, x = aw_59_cast_fp16)[name = tensor("op_1118_cast_fp16")]; + tensor var_1120_equation_0 = const()[name = tensor("op_1120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1120_cast_fp16 = einsum(equation = var_1120_equation_0, values = (var_1030_cast_fp16, var_1109_cast_fp16))[name = tensor("op_1120_cast_fp16")]; + tensor var_1122_equation_0 = const()[name = tensor("op_1122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1122_cast_fp16 = einsum(equation = var_1122_equation_0, values = (var_1034_cast_fp16, var_1110_cast_fp16))[name = tensor("op_1122_cast_fp16")]; + tensor var_1124_equation_0 = const()[name = tensor("op_1124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1124_cast_fp16 = einsum(equation = var_1124_equation_0, values = (var_1038_cast_fp16, var_1111_cast_fp16))[name = tensor("op_1124_cast_fp16")]; + tensor var_1126_equation_0 = const()[name = tensor("op_1126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1126_cast_fp16 = einsum(equation = var_1126_equation_0, values = (var_1042_cast_fp16, var_1112_cast_fp16))[name = tensor("op_1126_cast_fp16")]; + tensor var_1128_equation_0 = const()[name = tensor("op_1128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1128_cast_fp16 = einsum(equation = var_1128_equation_0, values = (var_1046_cast_fp16, var_1113_cast_fp16))[name = tensor("op_1128_cast_fp16")]; + tensor var_1130_equation_0 = const()[name = tensor("op_1130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1130_cast_fp16 = einsum(equation = var_1130_equation_0, values = (var_1050_cast_fp16, var_1114_cast_fp16))[name = tensor("op_1130_cast_fp16")]; + tensor var_1132_equation_0 = const()[name = tensor("op_1132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1132_cast_fp16 = einsum(equation = var_1132_equation_0, values = (var_1054_cast_fp16, var_1115_cast_fp16))[name = tensor("op_1132_cast_fp16")]; + tensor var_1134_equation_0 = const()[name = tensor("op_1134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1134_cast_fp16 = einsum(equation = var_1134_equation_0, values = (var_1058_cast_fp16, var_1116_cast_fp16))[name = tensor("op_1134_cast_fp16")]; + tensor var_1136_equation_0 = const()[name = tensor("op_1136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1136_cast_fp16 = einsum(equation = var_1136_equation_0, values = (var_1062_cast_fp16, var_1117_cast_fp16))[name = tensor("op_1136_cast_fp16")]; + tensor var_1138_equation_0 = const()[name = tensor("op_1138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1138_cast_fp16 = einsum(equation = var_1138_equation_0, values = (var_1066_cast_fp16, var_1118_cast_fp16))[name = tensor("op_1138_cast_fp16")]; + tensor input_69_interleave_0 = const()[name = tensor("input_69_interleave_0"), val = tensor(false)]; + tensor input_69_cast_fp16 = concat(axis = var_288, interleave = input_69_interleave_0, values = (var_1120_cast_fp16, var_1122_cast_fp16, var_1124_cast_fp16, var_1126_cast_fp16, var_1128_cast_fp16, var_1130_cast_fp16, var_1132_cast_fp16, var_1134_cast_fp16, var_1136_cast_fp16, var_1138_cast_fp16))[name = tensor("input_69_cast_fp16")]; + tensor var_1148_pad_type_0 = const()[name = tensor("op_1148_pad_type_0"), val = tensor("valid")]; + tensor var_1148_strides_0 = const()[name = tensor("op_1148_strides_0"), val = tensor([1, 1])]; + tensor var_1148_pad_0 = const()[name = tensor("op_1148_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1148_dilations_0 = const()[name = tensor("op_1148_dilations_0"), val = tensor([1, 1])]; + tensor var_1148_groups_0 = const()[name = tensor("op_1148_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23259776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23567040))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23567232)))]; + tensor var_1148_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_1148_dilations_0, groups = var_1148_groups_0, pad = var_1148_pad_0, pad_type = var_1148_pad_type_0, strides = var_1148_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor("op_1148_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = var_1148_cast_fp16, y = inputs_7_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; + tensor hidden_states_25_axes_0 = const()[name = tensor("hidden_states_25_axes_0"), val = tensor([1])]; + tensor hidden_states_25_gamma_0_to_fp16 = const()[name = tensor("hidden_states_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23568576)))]; + tensor hidden_states_25_beta_0_to_fp16 = const()[name = tensor("hidden_states_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23569920)))]; + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = hidden_states_25_beta_0_to_fp16, epsilon = var_1158_to_fp16, gamma = hidden_states_25_gamma_0_to_fp16, x = inputs_9_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; + tensor q_7_pad_type_0 = const()[name = tensor("q_7_pad_type_0"), val = tensor("valid")]; + tensor q_7_strides_0 = const()[name = tensor("q_7_strides_0"), val = tensor([1, 1])]; + tensor q_7_pad_0 = const()[name = tensor("q_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_7_dilations_0 = const()[name = tensor("q_7_dilations_0"), val = tensor([1, 1])]; + tensor q_7_groups_0 = const()[name = tensor("q_7_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23571264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23878528))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_7_cast_fp16 = conv(dilations = q_7_dilations_0, groups = q_7_groups_0, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = q_7_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor("q_7_cast_fp16")]; + tensor k_13_pad_type_0 = const()[name = tensor("k_13_pad_type_0"), val = tensor("valid")]; + tensor k_13_strides_0 = const()[name = tensor("k_13_strides_0"), val = tensor([1, 1])]; + tensor k_13_pad_0 = const()[name = tensor("k_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_13_dilations_0 = const()[name = tensor("k_13_dilations_0"), val = tensor([1, 1])]; + tensor k_13_groups_0 = const()[name = tensor("k_13_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23878720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24861824))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor k_13_cast_fp16 = conv(dilations = k_13_dilations_0, groups = k_13_groups_0, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = k_13_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_13_cast_fp16")]; + tensor v_7_pad_type_0 = const()[name = tensor("v_7_pad_type_0"), val = tensor("valid")]; + tensor v_7_strides_0 = const()[name = tensor("v_7_strides_0"), val = tensor([1, 1])]; + tensor v_7_pad_0 = const()[name = tensor("v_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_7_dilations_0 = const()[name = tensor("v_7_dilations_0"), val = tensor([1, 1])]; + tensor v_7_groups_0 = const()[name = tensor("v_7_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24862016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25845120))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor v_7_cast_fp16 = conv(dilations = v_7_dilations_0, groups = v_7_groups_0, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = v_7_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_7_cast_fp16")]; + tensor var_1191_begin_0 = const()[name = tensor("op_1191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1191_end_0 = const()[name = tensor("op_1191_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1191_end_mask_0 = const()[name = tensor("op_1191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1191_cast_fp16 = slice_by_index(begin = var_1191_begin_0, end = var_1191_end_0, end_mask = var_1191_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1191_cast_fp16")]; + tensor var_1195_begin_0 = const()[name = tensor("op_1195_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1195_end_0 = const()[name = tensor("op_1195_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1195_end_mask_0 = const()[name = tensor("op_1195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1195_cast_fp16 = slice_by_index(begin = var_1195_begin_0, end = var_1195_end_0, end_mask = var_1195_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1195_cast_fp16")]; + tensor var_1199_begin_0 = const()[name = tensor("op_1199_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1199_end_0 = const()[name = tensor("op_1199_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1199_end_mask_0 = const()[name = tensor("op_1199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1199_cast_fp16 = slice_by_index(begin = var_1199_begin_0, end = var_1199_end_0, end_mask = var_1199_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1199_cast_fp16")]; + tensor var_1203_begin_0 = const()[name = tensor("op_1203_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1203_end_0 = const()[name = tensor("op_1203_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1203_end_mask_0 = const()[name = tensor("op_1203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1203_cast_fp16 = slice_by_index(begin = var_1203_begin_0, end = var_1203_end_0, end_mask = var_1203_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1203_cast_fp16")]; + tensor var_1207_begin_0 = const()[name = tensor("op_1207_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1207_end_0 = const()[name = tensor("op_1207_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1207_end_mask_0 = const()[name = tensor("op_1207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1207_cast_fp16 = slice_by_index(begin = var_1207_begin_0, end = var_1207_end_0, end_mask = var_1207_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1207_cast_fp16")]; + tensor var_1211_begin_0 = const()[name = tensor("op_1211_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1211_end_0 = const()[name = tensor("op_1211_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1211_end_mask_0 = const()[name = tensor("op_1211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1211_cast_fp16 = slice_by_index(begin = var_1211_begin_0, end = var_1211_end_0, end_mask = var_1211_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1211_cast_fp16")]; + tensor var_1215_begin_0 = const()[name = tensor("op_1215_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1215_end_0 = const()[name = tensor("op_1215_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1215_end_mask_0 = const()[name = tensor("op_1215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1215_cast_fp16 = slice_by_index(begin = var_1215_begin_0, end = var_1215_end_0, end_mask = var_1215_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1215_cast_fp16")]; + tensor var_1219_begin_0 = const()[name = tensor("op_1219_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1219_end_0 = const()[name = tensor("op_1219_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1219_end_mask_0 = const()[name = tensor("op_1219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1219_cast_fp16 = slice_by_index(begin = var_1219_begin_0, end = var_1219_end_0, end_mask = var_1219_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1219_cast_fp16")]; + tensor var_1223_begin_0 = const()[name = tensor("op_1223_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1223_end_0 = const()[name = tensor("op_1223_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1223_end_mask_0 = const()[name = tensor("op_1223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1223_cast_fp16 = slice_by_index(begin = var_1223_begin_0, end = var_1223_end_0, end_mask = var_1223_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1223_cast_fp16")]; + tensor var_1227_begin_0 = const()[name = tensor("op_1227_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1227_end_0 = const()[name = tensor("op_1227_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1227_end_mask_0 = const()[name = tensor("op_1227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1227_cast_fp16 = slice_by_index(begin = var_1227_begin_0, end = var_1227_end_0, end_mask = var_1227_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1227_cast_fp16")]; + tensor k_15_perm_0 = const()[name = tensor("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1234_begin_0 = const()[name = tensor("op_1234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1234_end_0 = const()[name = tensor("op_1234_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_1234_end_mask_0 = const()[name = tensor("op_1234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = k_13_cast_fp16)[name = tensor("transpose_64")]; + tensor var_1234_cast_fp16 = slice_by_index(begin = var_1234_begin_0, end = var_1234_end_0, end_mask = var_1234_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1234_cast_fp16")]; + tensor var_1238_begin_0 = const()[name = tensor("op_1238_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1238_end_0 = const()[name = tensor("op_1238_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_1238_end_mask_0 = const()[name = tensor("op_1238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1238_cast_fp16 = slice_by_index(begin = var_1238_begin_0, end = var_1238_end_0, end_mask = var_1238_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1238_cast_fp16")]; + tensor var_1242_begin_0 = const()[name = tensor("op_1242_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1242_end_0 = const()[name = tensor("op_1242_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_1242_end_mask_0 = const()[name = tensor("op_1242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1242_cast_fp16 = slice_by_index(begin = var_1242_begin_0, end = var_1242_end_0, end_mask = var_1242_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1242_cast_fp16")]; + tensor var_1246_begin_0 = const()[name = tensor("op_1246_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1246_end_0 = const()[name = tensor("op_1246_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_1246_end_mask_0 = const()[name = tensor("op_1246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1246_cast_fp16 = slice_by_index(begin = var_1246_begin_0, end = var_1246_end_0, end_mask = var_1246_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1246_cast_fp16")]; + tensor var_1250_begin_0 = const()[name = tensor("op_1250_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1250_end_0 = const()[name = tensor("op_1250_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_1250_end_mask_0 = const()[name = tensor("op_1250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1250_cast_fp16 = slice_by_index(begin = var_1250_begin_0, end = var_1250_end_0, end_mask = var_1250_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1250_cast_fp16")]; + tensor var_1254_begin_0 = const()[name = tensor("op_1254_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1254_end_0 = const()[name = tensor("op_1254_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_1254_end_mask_0 = const()[name = tensor("op_1254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1254_cast_fp16 = slice_by_index(begin = var_1254_begin_0, end = var_1254_end_0, end_mask = var_1254_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1254_cast_fp16")]; + tensor var_1258_begin_0 = const()[name = tensor("op_1258_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1258_end_0 = const()[name = tensor("op_1258_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_1258_end_mask_0 = const()[name = tensor("op_1258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1258_cast_fp16 = slice_by_index(begin = var_1258_begin_0, end = var_1258_end_0, end_mask = var_1258_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1258_cast_fp16")]; + tensor var_1262_begin_0 = const()[name = tensor("op_1262_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1262_end_0 = const()[name = tensor("op_1262_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_1262_end_mask_0 = const()[name = tensor("op_1262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1262_cast_fp16 = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1262_cast_fp16")]; + tensor var_1266_begin_0 = const()[name = tensor("op_1266_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1266_end_0 = const()[name = tensor("op_1266_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_1266_end_mask_0 = const()[name = tensor("op_1266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1266_cast_fp16 = slice_by_index(begin = var_1266_begin_0, end = var_1266_end_0, end_mask = var_1266_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1266_cast_fp16")]; + tensor var_1270_begin_0 = const()[name = tensor("op_1270_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1270_end_0 = const()[name = tensor("op_1270_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_1270_end_mask_0 = const()[name = tensor("op_1270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1270_cast_fp16")]; + tensor var_1272_begin_0 = const()[name = tensor("op_1272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1272_end_0 = const()[name = tensor("op_1272_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_1272_end_mask_0 = const()[name = tensor("op_1272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1272_cast_fp16 = slice_by_index(begin = var_1272_begin_0, end = var_1272_end_0, end_mask = var_1272_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1272_cast_fp16")]; + tensor var_1276_begin_0 = const()[name = tensor("op_1276_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1276_end_0 = const()[name = tensor("op_1276_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_1276_end_mask_0 = const()[name = tensor("op_1276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1276_cast_fp16 = slice_by_index(begin = var_1276_begin_0, end = var_1276_end_0, end_mask = var_1276_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1276_cast_fp16")]; + tensor var_1280_begin_0 = const()[name = tensor("op_1280_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1280_end_0 = const()[name = tensor("op_1280_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_1280_end_mask_0 = const()[name = tensor("op_1280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1280_cast_fp16 = slice_by_index(begin = var_1280_begin_0, end = var_1280_end_0, end_mask = var_1280_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1280_cast_fp16")]; + tensor var_1284_begin_0 = const()[name = tensor("op_1284_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1284_end_0 = const()[name = tensor("op_1284_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_1284_end_mask_0 = const()[name = tensor("op_1284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1284_cast_fp16 = slice_by_index(begin = var_1284_begin_0, end = var_1284_end_0, end_mask = var_1284_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1284_cast_fp16")]; + tensor var_1288_begin_0 = const()[name = tensor("op_1288_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1288_end_0 = const()[name = tensor("op_1288_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_1288_end_mask_0 = const()[name = tensor("op_1288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1288_cast_fp16 = slice_by_index(begin = var_1288_begin_0, end = var_1288_end_0, end_mask = var_1288_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1288_cast_fp16")]; + tensor var_1292_begin_0 = const()[name = tensor("op_1292_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1292_end_0 = const()[name = tensor("op_1292_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_1292_end_mask_0 = const()[name = tensor("op_1292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1292_cast_fp16 = slice_by_index(begin = var_1292_begin_0, end = var_1292_end_0, end_mask = var_1292_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1292_cast_fp16")]; + tensor var_1296_begin_0 = const()[name = tensor("op_1296_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1296_end_0 = const()[name = tensor("op_1296_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_1296_end_mask_0 = const()[name = tensor("op_1296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1296_cast_fp16 = slice_by_index(begin = var_1296_begin_0, end = var_1296_end_0, end_mask = var_1296_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1296_cast_fp16")]; + tensor var_1300_begin_0 = const()[name = tensor("op_1300_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1300_end_0 = const()[name = tensor("op_1300_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_1300_end_mask_0 = const()[name = tensor("op_1300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1300_cast_fp16 = slice_by_index(begin = var_1300_begin_0, end = var_1300_end_0, end_mask = var_1300_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1300_cast_fp16")]; + tensor var_1304_begin_0 = const()[name = tensor("op_1304_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1304_end_0 = const()[name = tensor("op_1304_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_1304_end_mask_0 = const()[name = tensor("op_1304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1304_cast_fp16 = slice_by_index(begin = var_1304_begin_0, end = var_1304_end_0, end_mask = var_1304_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1304_cast_fp16")]; + tensor var_1308_begin_0 = const()[name = tensor("op_1308_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1308_end_0 = const()[name = tensor("op_1308_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_1308_end_mask_0 = const()[name = tensor("op_1308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1308_cast_fp16 = slice_by_index(begin = var_1308_begin_0, end = var_1308_end_0, end_mask = var_1308_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1308_cast_fp16")]; + tensor var_1312_equation_0 = const()[name = tensor("op_1312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1312_cast_fp16 = einsum(equation = var_1312_equation_0, values = (var_1234_cast_fp16, var_1191_cast_fp16))[name = tensor("op_1312_cast_fp16")]; + tensor var_1313_to_fp16 = const()[name = tensor("op_1313_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_61_cast_fp16 = mul(x = var_1312_cast_fp16, y = var_1313_to_fp16)[name = tensor("aw_61_cast_fp16")]; + tensor var_1316_equation_0 = const()[name = tensor("op_1316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1316_cast_fp16 = einsum(equation = var_1316_equation_0, values = (var_1238_cast_fp16, var_1195_cast_fp16))[name = tensor("op_1316_cast_fp16")]; + tensor var_1317_to_fp16 = const()[name = tensor("op_1317_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_63_cast_fp16 = mul(x = var_1316_cast_fp16, y = var_1317_to_fp16)[name = tensor("aw_63_cast_fp16")]; + tensor var_1320_equation_0 = const()[name = tensor("op_1320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1320_cast_fp16 = einsum(equation = var_1320_equation_0, values = (var_1242_cast_fp16, var_1199_cast_fp16))[name = tensor("op_1320_cast_fp16")]; + tensor var_1321_to_fp16 = const()[name = tensor("op_1321_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_65_cast_fp16 = mul(x = var_1320_cast_fp16, y = var_1321_to_fp16)[name = tensor("aw_65_cast_fp16")]; + tensor var_1324_equation_0 = const()[name = tensor("op_1324_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1324_cast_fp16 = einsum(equation = var_1324_equation_0, values = (var_1246_cast_fp16, var_1203_cast_fp16))[name = tensor("op_1324_cast_fp16")]; + tensor var_1325_to_fp16 = const()[name = tensor("op_1325_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_67_cast_fp16 = mul(x = var_1324_cast_fp16, y = var_1325_to_fp16)[name = tensor("aw_67_cast_fp16")]; + tensor var_1328_equation_0 = const()[name = tensor("op_1328_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1328_cast_fp16 = einsum(equation = var_1328_equation_0, values = (var_1250_cast_fp16, var_1207_cast_fp16))[name = tensor("op_1328_cast_fp16")]; + tensor var_1329_to_fp16 = const()[name = tensor("op_1329_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_69_cast_fp16 = mul(x = var_1328_cast_fp16, y = var_1329_to_fp16)[name = tensor("aw_69_cast_fp16")]; + tensor var_1332_equation_0 = const()[name = tensor("op_1332_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1332_cast_fp16 = einsum(equation = var_1332_equation_0, values = (var_1254_cast_fp16, var_1211_cast_fp16))[name = tensor("op_1332_cast_fp16")]; + tensor var_1333_to_fp16 = const()[name = tensor("op_1333_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_71_cast_fp16 = mul(x = var_1332_cast_fp16, y = var_1333_to_fp16)[name = tensor("aw_71_cast_fp16")]; + tensor var_1336_equation_0 = const()[name = tensor("op_1336_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1336_cast_fp16 = einsum(equation = var_1336_equation_0, values = (var_1258_cast_fp16, var_1215_cast_fp16))[name = tensor("op_1336_cast_fp16")]; + tensor var_1337_to_fp16 = const()[name = tensor("op_1337_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_73_cast_fp16 = mul(x = var_1336_cast_fp16, y = var_1337_to_fp16)[name = tensor("aw_73_cast_fp16")]; + tensor var_1340_equation_0 = const()[name = tensor("op_1340_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1340_cast_fp16 = einsum(equation = var_1340_equation_0, values = (var_1262_cast_fp16, var_1219_cast_fp16))[name = tensor("op_1340_cast_fp16")]; + tensor var_1341_to_fp16 = const()[name = tensor("op_1341_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_75_cast_fp16 = mul(x = var_1340_cast_fp16, y = var_1341_to_fp16)[name = tensor("aw_75_cast_fp16")]; + tensor var_1344_equation_0 = const()[name = tensor("op_1344_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1344_cast_fp16 = einsum(equation = var_1344_equation_0, values = (var_1266_cast_fp16, var_1223_cast_fp16))[name = tensor("op_1344_cast_fp16")]; + tensor var_1345_to_fp16 = const()[name = tensor("op_1345_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_77_cast_fp16 = mul(x = var_1344_cast_fp16, y = var_1345_to_fp16)[name = tensor("aw_77_cast_fp16")]; + tensor var_1348_equation_0 = const()[name = tensor("op_1348_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1348_cast_fp16 = einsum(equation = var_1348_equation_0, values = (var_1270_cast_fp16, var_1227_cast_fp16))[name = tensor("op_1348_cast_fp16")]; + tensor var_1349_to_fp16 = const()[name = tensor("op_1349_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_79_cast_fp16 = mul(x = var_1348_cast_fp16, y = var_1349_to_fp16)[name = tensor("aw_79_cast_fp16")]; + tensor var_1351_cast_fp16 = softmax(axis = var_288, x = aw_61_cast_fp16)[name = tensor("op_1351_cast_fp16")]; + tensor var_1352_cast_fp16 = softmax(axis = var_288, x = aw_63_cast_fp16)[name = tensor("op_1352_cast_fp16")]; + tensor var_1353_cast_fp16 = softmax(axis = var_288, x = aw_65_cast_fp16)[name = tensor("op_1353_cast_fp16")]; + tensor var_1354_cast_fp16 = softmax(axis = var_288, x = aw_67_cast_fp16)[name = tensor("op_1354_cast_fp16")]; + tensor var_1355_cast_fp16 = softmax(axis = var_288, x = aw_69_cast_fp16)[name = tensor("op_1355_cast_fp16")]; + tensor var_1356_cast_fp16 = softmax(axis = var_288, x = aw_71_cast_fp16)[name = tensor("op_1356_cast_fp16")]; + tensor var_1357_cast_fp16 = softmax(axis = var_288, x = aw_73_cast_fp16)[name = tensor("op_1357_cast_fp16")]; + tensor var_1358_cast_fp16 = softmax(axis = var_288, x = aw_75_cast_fp16)[name = tensor("op_1358_cast_fp16")]; + tensor var_1359_cast_fp16 = softmax(axis = var_288, x = aw_77_cast_fp16)[name = tensor("op_1359_cast_fp16")]; + tensor var_1360_cast_fp16 = softmax(axis = var_288, x = aw_79_cast_fp16)[name = tensor("op_1360_cast_fp16")]; + tensor var_1362_equation_0 = const()[name = tensor("op_1362_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1362_cast_fp16 = einsum(equation = var_1362_equation_0, values = (var_1272_cast_fp16, var_1351_cast_fp16))[name = tensor("op_1362_cast_fp16")]; + tensor var_1364_equation_0 = const()[name = tensor("op_1364_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1364_cast_fp16 = einsum(equation = var_1364_equation_0, values = (var_1276_cast_fp16, var_1352_cast_fp16))[name = tensor("op_1364_cast_fp16")]; + tensor var_1366_equation_0 = const()[name = tensor("op_1366_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1366_cast_fp16 = einsum(equation = var_1366_equation_0, values = (var_1280_cast_fp16, var_1353_cast_fp16))[name = tensor("op_1366_cast_fp16")]; + tensor var_1368_equation_0 = const()[name = tensor("op_1368_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1368_cast_fp16 = einsum(equation = var_1368_equation_0, values = (var_1284_cast_fp16, var_1354_cast_fp16))[name = tensor("op_1368_cast_fp16")]; + tensor var_1370_equation_0 = const()[name = tensor("op_1370_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1370_cast_fp16 = einsum(equation = var_1370_equation_0, values = (var_1288_cast_fp16, var_1355_cast_fp16))[name = tensor("op_1370_cast_fp16")]; + tensor var_1372_equation_0 = const()[name = tensor("op_1372_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1372_cast_fp16 = einsum(equation = var_1372_equation_0, values = (var_1292_cast_fp16, var_1356_cast_fp16))[name = tensor("op_1372_cast_fp16")]; + tensor var_1374_equation_0 = const()[name = tensor("op_1374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1374_cast_fp16 = einsum(equation = var_1374_equation_0, values = (var_1296_cast_fp16, var_1357_cast_fp16))[name = tensor("op_1374_cast_fp16")]; + tensor var_1376_equation_0 = const()[name = tensor("op_1376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1376_cast_fp16 = einsum(equation = var_1376_equation_0, values = (var_1300_cast_fp16, var_1358_cast_fp16))[name = tensor("op_1376_cast_fp16")]; + tensor var_1378_equation_0 = const()[name = tensor("op_1378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1378_cast_fp16 = einsum(equation = var_1378_equation_0, values = (var_1304_cast_fp16, var_1359_cast_fp16))[name = tensor("op_1378_cast_fp16")]; + tensor var_1380_equation_0 = const()[name = tensor("op_1380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1380_cast_fp16 = einsum(equation = var_1380_equation_0, values = (var_1308_cast_fp16, var_1360_cast_fp16))[name = tensor("op_1380_cast_fp16")]; + tensor input_71_interleave_0 = const()[name = tensor("input_71_interleave_0"), val = tensor(false)]; + tensor input_71_cast_fp16 = concat(axis = var_288, interleave = input_71_interleave_0, values = (var_1362_cast_fp16, var_1364_cast_fp16, var_1366_cast_fp16, var_1368_cast_fp16, var_1370_cast_fp16, var_1372_cast_fp16, var_1374_cast_fp16, var_1376_cast_fp16, var_1378_cast_fp16, var_1380_cast_fp16))[name = tensor("input_71_cast_fp16")]; + tensor var_1390_pad_type_0 = const()[name = tensor("op_1390_pad_type_0"), val = tensor("valid")]; + tensor var_1390_strides_0 = const()[name = tensor("op_1390_strides_0"), val = tensor([1, 1])]; + tensor var_1390_pad_0 = const()[name = tensor("op_1390_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1390_dilations_0 = const()[name = tensor("op_1390_dilations_0"), val = tensor([1, 1])]; + tensor var_1390_groups_0 = const()[name = tensor("op_1390_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25845312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26152576))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26152768)))]; + tensor var_1390_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_1390_dilations_0, groups = var_1390_groups_0, pad = var_1390_pad_0, pad_type = var_1390_pad_type_0, strides = var_1390_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_71_cast_fp16)[name = tensor("op_1390_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = var_1390_cast_fp16, y = inputs_9_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; + tensor input_73_axes_0 = const()[name = tensor("input_73_axes_0"), val = tensor([1])]; + tensor input_73_gamma_0_to_fp16 = const()[name = tensor("input_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26154112)))]; + tensor input_73_beta_0_to_fp16 = const()[name = tensor("input_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26155456)))]; + tensor var_1400_to_fp16 = const()[name = tensor("op_1400_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = input_73_beta_0_to_fp16, epsilon = var_1400_to_fp16, gamma = input_73_gamma_0_to_fp16, x = inputs_11_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor var_1420_pad_type_0 = const()[name = tensor("op_1420_pad_type_0"), val = tensor("valid")]; + tensor var_1420_strides_0 = const()[name = tensor("op_1420_strides_0"), val = tensor([1, 1])]; + tensor var_1420_pad_0 = const()[name = tensor("op_1420_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1420_dilations_0 = const()[name = tensor("op_1420_dilations_0"), val = tensor([1, 1])]; + tensor var_1420_groups_0 = const()[name = tensor("op_1420_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26156800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28614464))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28614656)))]; + tensor var_1420_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_1420_dilations_0, groups = var_1420_groups_0, pad = var_1420_pad_0, pad_type = var_1420_pad_type_0, strides = var_1420_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor("op_1420_cast_fp16")]; + tensor var_1421_split_sizes_0 = const()[name = tensor("op_1421_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_1421_axis_0 = const()[name = tensor("op_1421_axis_0"), val = tensor(1)]; + tensor var_1421_cast_fp16_0, tensor var_1421_cast_fp16_1 = split(axis = var_1421_axis_0, split_sizes = var_1421_split_sizes_0, x = var_1420_cast_fp16)[name = tensor("op_1421_cast_fp16")]; + tensor var_1423_mode_0 = const()[name = tensor("op_1423_mode_0"), val = tensor("EXACT")]; + tensor var_1423_cast_fp16 = gelu(mode = var_1423_mode_0, x = var_1421_cast_fp16_1)[name = tensor("op_1423_cast_fp16")]; + tensor input_75_cast_fp16 = mul(x = var_1421_cast_fp16_0, y = var_1423_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor var_1431_pad_type_0 = const()[name = tensor("op_1431_pad_type_0"), val = tensor("valid")]; + tensor var_1431_strides_0 = const()[name = tensor("op_1431_strides_0"), val = tensor([1, 1])]; + tensor var_1431_pad_0 = const()[name = tensor("op_1431_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1431_dilations_0 = const()[name = tensor("op_1431_dilations_0"), val = tensor([1, 1])]; + tensor var_1431_groups_0 = const()[name = tensor("op_1431_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28624960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29853824))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29854016)))]; + tensor var_1431_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_1431_dilations_0, groups = var_1431_groups_0, pad = var_1431_pad_0, pad_type = var_1431_pad_type_0, strides = var_1431_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor("op_1431_cast_fp16")]; + tensor hidden_states_29_cast_fp16 = add(x = var_1431_cast_fp16, y = inputs_11_cast_fp16)[name = tensor("hidden_states_29_cast_fp16")]; + tensor var_1433 = const()[name = tensor("op_1433"), val = tensor([2, 640, 64, 64])]; + tensor input_77_cast_fp16 = reshape(shape = var_1433, x = hidden_states_29_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor hidden_states_31_pad_type_0 = const()[name = tensor("hidden_states_31_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_31_strides_0 = const()[name = tensor("hidden_states_31_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_31_pad_0 = const()[name = tensor("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_31_dilations_0 = const()[name = tensor("hidden_states_31_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_31_groups_0 = const()[name = tensor("hidden_states_31_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29855360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30162624))), name = tensor("down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30162816)))]; + tensor hidden_states_31_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; + tensor input_79_cast_fp16_1 = add(x = hidden_states_31_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor("input_79_cast_fp16")]; + tensor reshape_28_shape_0 = const()[name = tensor("reshape_28_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_28_cast_fp16 = reshape(shape = reshape_28_shape_0, x = input_79_cast_fp16_1)[name = tensor("reshape_28_cast_fp16")]; + tensor reduce_mean_21_axes_0 = const()[name = tensor("reduce_mean_21_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_21_keep_dims_0 = const()[name = tensor("reduce_mean_21_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_21_cast_fp16 = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast_fp16)[name = tensor("reduce_mean_21_cast_fp16")]; + tensor sub_14_cast_fp16 = sub(x = reshape_28_cast_fp16, y = reduce_mean_21_cast_fp16)[name = tensor("sub_14_cast_fp16")]; + tensor square_7_cast_fp16 = square(x = sub_14_cast_fp16)[name = tensor("square_7_cast_fp16")]; + tensor reduce_mean_23_axes_0 = const()[name = tensor("reduce_mean_23_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_23_keep_dims_0 = const()[name = tensor("reduce_mean_23_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_23_cast_fp16 = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast_fp16)[name = tensor("reduce_mean_23_cast_fp16")]; + tensor add_14_y_0_to_fp16 = const()[name = tensor("add_14_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_14_cast_fp16 = add(x = reduce_mean_23_cast_fp16, y = add_14_y_0_to_fp16)[name = tensor("add_14_cast_fp16")]; + tensor sqrt_7_cast_fp16 = sqrt(x = add_14_cast_fp16)[name = tensor("sqrt_7_cast_fp16")]; + tensor real_div_7_cast_fp16 = real_div(x = sub_14_cast_fp16, y = sqrt_7_cast_fp16)[name = tensor("real_div_7_cast_fp16")]; + tensor reshape_29_shape_0 = const()[name = tensor("reshape_29_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_29_cast_fp16 = reshape(shape = reshape_29_shape_0, x = real_div_7_cast_fp16)[name = tensor("reshape_29_cast_fp16")]; + tensor add_15_gamma_0_to_fp16 = const()[name = tensor("add_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30164160)))]; + tensor add_15_beta_0_to_fp16 = const()[name = tensor("add_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30165504)))]; + tensor add_15_epsilon_0_to_fp16 = const()[name = tensor("add_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_15_cast_fp16 = batch_norm(beta = add_15_beta_0_to_fp16, epsilon = add_15_epsilon_0_to_fp16, gamma = add_15_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_29_cast_fp16)[name = tensor("add_15_cast_fp16")]; + tensor input_83_cast_fp16 = silu(x = add_15_cast_fp16)[name = tensor("input_83_cast_fp16")]; + tensor hidden_states_33_pad_type_0 = const()[name = tensor("hidden_states_33_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_33_pad_0 = const()[name = tensor("hidden_states_33_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_33_strides_0 = const()[name = tensor("hidden_states_33_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_33_dilations_0 = const()[name = tensor("hidden_states_33_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_33_groups_0 = const()[name = tensor("hidden_states_33_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30166848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32931712))), name = tensor("down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32931904)))]; + tensor hidden_states_33_cast_fp16 = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = tensor("hidden_states_33_cast_fp16")]; + tensor temb_7_pad_type_0 = const()[name = tensor("temb_7_pad_type_0"), val = tensor("valid")]; + tensor temb_7_strides_0 = const()[name = tensor("temb_7_strides_0"), val = tensor([1, 1])]; + tensor temb_7_pad_0 = const()[name = tensor("temb_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_7_dilations_0 = const()[name = tensor("temb_7_dilations_0"), val = tensor([1, 1])]; + tensor temb_7_groups_0 = const()[name = tensor("temb_7_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32933248))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33547712))), name = tensor("down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33547904)))]; + tensor temb_7_cast_fp16 = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_7_dilations_0, groups = temb_7_groups_0, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = temb_7_strides_0, weight = down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_7_cast_fp16")]; + tensor input_87_cast_fp16 = add(x = hidden_states_33_cast_fp16, y = temb_7_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor reshape_32_shape_0 = const()[name = tensor("reshape_32_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_32_cast_fp16 = reshape(shape = reshape_32_shape_0, x = input_87_cast_fp16)[name = tensor("reshape_32_cast_fp16")]; + tensor reduce_mean_24_axes_0 = const()[name = tensor("reduce_mean_24_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_24_keep_dims_0 = const()[name = tensor("reduce_mean_24_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_24_cast_fp16 = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast_fp16)[name = tensor("reduce_mean_24_cast_fp16")]; + tensor sub_16_cast_fp16 = sub(x = reshape_32_cast_fp16, y = reduce_mean_24_cast_fp16)[name = tensor("sub_16_cast_fp16")]; + tensor square_8_cast_fp16 = square(x = sub_16_cast_fp16)[name = tensor("square_8_cast_fp16")]; + tensor reduce_mean_26_axes_0 = const()[name = tensor("reduce_mean_26_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_26_keep_dims_0 = const()[name = tensor("reduce_mean_26_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_26_cast_fp16 = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast_fp16)[name = tensor("reduce_mean_26_cast_fp16")]; + tensor add_16_y_0_to_fp16 = const()[name = tensor("add_16_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_16_cast_fp16 = add(x = reduce_mean_26_cast_fp16, y = add_16_y_0_to_fp16)[name = tensor("add_16_cast_fp16")]; + tensor sqrt_8_cast_fp16 = sqrt(x = add_16_cast_fp16)[name = tensor("sqrt_8_cast_fp16")]; + tensor real_div_8_cast_fp16 = real_div(x = sub_16_cast_fp16, y = sqrt_8_cast_fp16)[name = tensor("real_div_8_cast_fp16")]; + tensor reshape_33_shape_0 = const()[name = tensor("reshape_33_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_33_cast_fp16 = reshape(shape = reshape_33_shape_0, x = real_div_8_cast_fp16)[name = tensor("reshape_33_cast_fp16")]; + tensor add_17_gamma_0_to_fp16 = const()[name = tensor("add_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33549248)))]; + tensor add_17_beta_0_to_fp16 = const()[name = tensor("add_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33550592)))]; + tensor add_17_epsilon_0_to_fp16 = const()[name = tensor("add_17_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_17_cast_fp16 = batch_norm(beta = add_17_beta_0_to_fp16, epsilon = add_17_epsilon_0_to_fp16, gamma = add_17_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_33_cast_fp16)[name = tensor("add_17_cast_fp16")]; + tensor input_91_cast_fp16 = silu(x = add_17_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor hidden_states_35_pad_type_0 = const()[name = tensor("hidden_states_35_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_35_pad_0 = const()[name = tensor("hidden_states_35_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_35_strides_0 = const()[name = tensor("hidden_states_35_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_35_dilations_0 = const()[name = tensor("hidden_states_35_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_35_groups_0 = const()[name = tensor("hidden_states_35_groups_0"), val = tensor(1)]; + tensor down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33551936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36316800))), name = tensor("down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36316992)))]; + tensor hidden_states_35_cast_fp16 = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = tensor("hidden_states_35_cast_fp16")]; + tensor hidden_states_37_cast_fp16 = add(x = input_79_cast_fp16_1, y = hidden_states_35_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; + tensor reshape_36_shape_0 = const()[name = tensor("reshape_36_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_36_cast_fp16 = reshape(shape = reshape_36_shape_0, x = hidden_states_37_cast_fp16)[name = tensor("reshape_36_cast_fp16")]; + tensor reduce_mean_27_axes_0 = const()[name = tensor("reduce_mean_27_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_27_keep_dims_0 = const()[name = tensor("reduce_mean_27_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_27_cast_fp16 = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast_fp16)[name = tensor("reduce_mean_27_cast_fp16")]; + tensor sub_18_cast_fp16 = sub(x = reshape_36_cast_fp16, y = reduce_mean_27_cast_fp16)[name = tensor("sub_18_cast_fp16")]; + tensor square_9_cast_fp16 = square(x = sub_18_cast_fp16)[name = tensor("square_9_cast_fp16")]; + tensor reduce_mean_29_axes_0 = const()[name = tensor("reduce_mean_29_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_29_keep_dims_0 = const()[name = tensor("reduce_mean_29_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_29_cast_fp16 = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast_fp16)[name = tensor("reduce_mean_29_cast_fp16")]; + tensor add_18_y_0_to_fp16 = const()[name = tensor("add_18_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_18_cast_fp16 = add(x = reduce_mean_29_cast_fp16, y = add_18_y_0_to_fp16)[name = tensor("add_18_cast_fp16")]; + tensor sqrt_9_cast_fp16 = sqrt(x = add_18_cast_fp16)[name = tensor("sqrt_9_cast_fp16")]; + tensor real_div_9_cast_fp16 = real_div(x = sub_18_cast_fp16, y = sqrt_9_cast_fp16)[name = tensor("real_div_9_cast_fp16")]; + tensor reshape_37_shape_0 = const()[name = tensor("reshape_37_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_37_cast_fp16 = reshape(shape = reshape_37_shape_0, x = real_div_9_cast_fp16)[name = tensor("reshape_37_cast_fp16")]; + tensor add_19_gamma_0_to_fp16 = const()[name = tensor("add_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36318336)))]; + tensor add_19_beta_0_to_fp16 = const()[name = tensor("add_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36319680)))]; + tensor add_19_epsilon_0_to_fp16 = const()[name = tensor("add_19_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_19_cast_fp16 = batch_norm(beta = add_19_beta_0_to_fp16, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_37_cast_fp16)[name = tensor("add_19_cast_fp16")]; + tensor hidden_states_39_pad_type_0 = const()[name = tensor("hidden_states_39_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_39_strides_0 = const()[name = tensor("hidden_states_39_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_0 = const()[name = tensor("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_39_dilations_0 = const()[name = tensor("hidden_states_39_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_39_groups_0 = const()[name = tensor("hidden_states_39_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36321024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36628288))), name = tensor("down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36628480)))]; + tensor hidden_states_39_cast_fp16 = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized, x = add_19_cast_fp16)[name = tensor("hidden_states_39_cast_fp16")]; + tensor var_1505 = const()[name = tensor("op_1505"), val = tensor([2, 640, 1, 4096])]; + tensor inputs_13_cast_fp16 = reshape(shape = var_1505, x = hidden_states_39_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; + tensor hidden_states_41_axes_0 = const()[name = tensor("hidden_states_41_axes_0"), val = tensor([1])]; + tensor hidden_states_41_gamma_0_to_fp16 = const()[name = tensor("hidden_states_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36629824)))]; + tensor hidden_states_41_beta_0_to_fp16 = const()[name = tensor("hidden_states_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36631168)))]; + tensor var_1521_to_fp16 = const()[name = tensor("op_1521_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_41_cast_fp16 = layer_norm(axes = hidden_states_41_axes_0, beta = hidden_states_41_beta_0_to_fp16, epsilon = var_1521_to_fp16, gamma = hidden_states_41_gamma_0_to_fp16, x = inputs_13_cast_fp16)[name = tensor("hidden_states_41_cast_fp16")]; + tensor q_9_pad_type_0 = const()[name = tensor("q_9_pad_type_0"), val = tensor("valid")]; + tensor q_9_strides_0 = const()[name = tensor("q_9_strides_0"), val = tensor([1, 1])]; + tensor q_9_pad_0 = const()[name = tensor("q_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_9_dilations_0 = const()[name = tensor("q_9_dilations_0"), val = tensor([1, 1])]; + tensor q_9_groups_0 = const()[name = tensor("q_9_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36632512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36939776))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_9_cast_fp16 = conv(dilations = q_9_dilations_0, groups = q_9_groups_0, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = q_9_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_41_cast_fp16)[name = tensor("q_9_cast_fp16")]; + tensor k_17_pad_type_0 = const()[name = tensor("k_17_pad_type_0"), val = tensor("valid")]; + tensor k_17_strides_0 = const()[name = tensor("k_17_strides_0"), val = tensor([1, 1])]; + tensor k_17_pad_0 = const()[name = tensor("k_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_17_dilations_0 = const()[name = tensor("k_17_dilations_0"), val = tensor([1, 1])]; + tensor k_17_groups_0 = const()[name = tensor("k_17_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36939968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37247232))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_17_cast_fp16 = conv(dilations = k_17_dilations_0, groups = k_17_groups_0, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = k_17_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_41_cast_fp16)[name = tensor("k_17_cast_fp16")]; + tensor v_9_pad_type_0 = const()[name = tensor("v_9_pad_type_0"), val = tensor("valid")]; + tensor v_9_strides_0 = const()[name = tensor("v_9_strides_0"), val = tensor([1, 1])]; + tensor v_9_pad_0 = const()[name = tensor("v_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_9_dilations_0 = const()[name = tensor("v_9_dilations_0"), val = tensor([1, 1])]; + tensor v_9_groups_0 = const()[name = tensor("v_9_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37247424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37554688))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_9_cast_fp16 = conv(dilations = v_9_dilations_0, groups = v_9_groups_0, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = v_9_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_41_cast_fp16)[name = tensor("v_9_cast_fp16")]; + tensor var_1554_begin_0 = const()[name = tensor("op_1554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1554_end_0 = const()[name = tensor("op_1554_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1554_end_mask_0 = const()[name = tensor("op_1554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1554_cast_fp16 = slice_by_index(begin = var_1554_begin_0, end = var_1554_end_0, end_mask = var_1554_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1554_cast_fp16")]; + tensor var_1558_begin_0 = const()[name = tensor("op_1558_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1558_end_0 = const()[name = tensor("op_1558_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1558_end_mask_0 = const()[name = tensor("op_1558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1558_cast_fp16 = slice_by_index(begin = var_1558_begin_0, end = var_1558_end_0, end_mask = var_1558_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1558_cast_fp16")]; + tensor var_1562_begin_0 = const()[name = tensor("op_1562_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1562_end_0 = const()[name = tensor("op_1562_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1562_end_mask_0 = const()[name = tensor("op_1562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1562_cast_fp16 = slice_by_index(begin = var_1562_begin_0, end = var_1562_end_0, end_mask = var_1562_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1562_cast_fp16")]; + tensor var_1566_begin_0 = const()[name = tensor("op_1566_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1566_end_0 = const()[name = tensor("op_1566_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1566_end_mask_0 = const()[name = tensor("op_1566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1566_cast_fp16 = slice_by_index(begin = var_1566_begin_0, end = var_1566_end_0, end_mask = var_1566_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1566_cast_fp16")]; + tensor var_1570_begin_0 = const()[name = tensor("op_1570_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1570_end_0 = const()[name = tensor("op_1570_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1570_end_mask_0 = const()[name = tensor("op_1570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1570_cast_fp16 = slice_by_index(begin = var_1570_begin_0, end = var_1570_end_0, end_mask = var_1570_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1570_cast_fp16")]; + tensor var_1574_begin_0 = const()[name = tensor("op_1574_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1574_end_0 = const()[name = tensor("op_1574_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1574_end_mask_0 = const()[name = tensor("op_1574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1574_cast_fp16 = slice_by_index(begin = var_1574_begin_0, end = var_1574_end_0, end_mask = var_1574_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1574_cast_fp16")]; + tensor var_1578_begin_0 = const()[name = tensor("op_1578_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1578_end_0 = const()[name = tensor("op_1578_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1578_end_mask_0 = const()[name = tensor("op_1578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1578_cast_fp16 = slice_by_index(begin = var_1578_begin_0, end = var_1578_end_0, end_mask = var_1578_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1578_cast_fp16")]; + tensor var_1582_begin_0 = const()[name = tensor("op_1582_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1582_end_0 = const()[name = tensor("op_1582_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1582_end_mask_0 = const()[name = tensor("op_1582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1582_cast_fp16 = slice_by_index(begin = var_1582_begin_0, end = var_1582_end_0, end_mask = var_1582_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1582_cast_fp16")]; + tensor var_1586_begin_0 = const()[name = tensor("op_1586_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1586_end_0 = const()[name = tensor("op_1586_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1586_end_mask_0 = const()[name = tensor("op_1586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1586_cast_fp16 = slice_by_index(begin = var_1586_begin_0, end = var_1586_end_0, end_mask = var_1586_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1586_cast_fp16")]; + tensor var_1590_begin_0 = const()[name = tensor("op_1590_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1590_end_0 = const()[name = tensor("op_1590_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1590_end_mask_0 = const()[name = tensor("op_1590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1590_cast_fp16 = slice_by_index(begin = var_1590_begin_0, end = var_1590_end_0, end_mask = var_1590_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1590_cast_fp16")]; + tensor k_19_perm_0 = const()[name = tensor("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1597_begin_0 = const()[name = tensor("op_1597_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1597_end_0 = const()[name = tensor("op_1597_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_1597_end_mask_0 = const()[name = tensor("op_1597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = k_17_cast_fp16)[name = tensor("transpose_63")]; + tensor var_1597_cast_fp16 = slice_by_index(begin = var_1597_begin_0, end = var_1597_end_0, end_mask = var_1597_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1597_cast_fp16")]; + tensor var_1601_begin_0 = const()[name = tensor("op_1601_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1601_end_0 = const()[name = tensor("op_1601_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_1601_end_mask_0 = const()[name = tensor("op_1601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1601_cast_fp16 = slice_by_index(begin = var_1601_begin_0, end = var_1601_end_0, end_mask = var_1601_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1601_cast_fp16")]; + tensor var_1605_begin_0 = const()[name = tensor("op_1605_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1605_end_0 = const()[name = tensor("op_1605_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_1605_end_mask_0 = const()[name = tensor("op_1605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1605_cast_fp16 = slice_by_index(begin = var_1605_begin_0, end = var_1605_end_0, end_mask = var_1605_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1605_cast_fp16")]; + tensor var_1609_begin_0 = const()[name = tensor("op_1609_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1609_end_0 = const()[name = tensor("op_1609_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_1609_end_mask_0 = const()[name = tensor("op_1609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1609_cast_fp16 = slice_by_index(begin = var_1609_begin_0, end = var_1609_end_0, end_mask = var_1609_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1609_cast_fp16")]; + tensor var_1613_begin_0 = const()[name = tensor("op_1613_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1613_end_0 = const()[name = tensor("op_1613_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_1613_end_mask_0 = const()[name = tensor("op_1613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1613_cast_fp16 = slice_by_index(begin = var_1613_begin_0, end = var_1613_end_0, end_mask = var_1613_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1613_cast_fp16")]; + tensor var_1617_begin_0 = const()[name = tensor("op_1617_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1617_end_0 = const()[name = tensor("op_1617_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_1617_end_mask_0 = const()[name = tensor("op_1617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1617_cast_fp16 = slice_by_index(begin = var_1617_begin_0, end = var_1617_end_0, end_mask = var_1617_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1617_cast_fp16")]; + tensor var_1621_begin_0 = const()[name = tensor("op_1621_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1621_end_0 = const()[name = tensor("op_1621_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_1621_end_mask_0 = const()[name = tensor("op_1621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1621_cast_fp16 = slice_by_index(begin = var_1621_begin_0, end = var_1621_end_0, end_mask = var_1621_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1621_cast_fp16")]; + tensor var_1625_begin_0 = const()[name = tensor("op_1625_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1625_end_0 = const()[name = tensor("op_1625_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_1625_end_mask_0 = const()[name = tensor("op_1625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1625_cast_fp16 = slice_by_index(begin = var_1625_begin_0, end = var_1625_end_0, end_mask = var_1625_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1625_cast_fp16")]; + tensor var_1629_begin_0 = const()[name = tensor("op_1629_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1629_end_0 = const()[name = tensor("op_1629_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_1629_end_mask_0 = const()[name = tensor("op_1629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1629_cast_fp16 = slice_by_index(begin = var_1629_begin_0, end = var_1629_end_0, end_mask = var_1629_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1629_cast_fp16")]; + tensor var_1633_begin_0 = const()[name = tensor("op_1633_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1633_end_0 = const()[name = tensor("op_1633_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_1633_end_mask_0 = const()[name = tensor("op_1633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1633_cast_fp16 = slice_by_index(begin = var_1633_begin_0, end = var_1633_end_0, end_mask = var_1633_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1633_cast_fp16")]; + tensor var_1635_begin_0 = const()[name = tensor("op_1635_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1635_end_0 = const()[name = tensor("op_1635_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1635_end_mask_0 = const()[name = tensor("op_1635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1635_cast_fp16 = slice_by_index(begin = var_1635_begin_0, end = var_1635_end_0, end_mask = var_1635_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1635_cast_fp16")]; + tensor var_1639_begin_0 = const()[name = tensor("op_1639_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1639_end_0 = const()[name = tensor("op_1639_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1639_end_mask_0 = const()[name = tensor("op_1639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1639_cast_fp16 = slice_by_index(begin = var_1639_begin_0, end = var_1639_end_0, end_mask = var_1639_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1639_cast_fp16")]; + tensor var_1643_begin_0 = const()[name = tensor("op_1643_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1643_end_0 = const()[name = tensor("op_1643_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1643_end_mask_0 = const()[name = tensor("op_1643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1643_cast_fp16 = slice_by_index(begin = var_1643_begin_0, end = var_1643_end_0, end_mask = var_1643_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1643_cast_fp16")]; + tensor var_1647_begin_0 = const()[name = tensor("op_1647_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1647_end_0 = const()[name = tensor("op_1647_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1647_end_mask_0 = const()[name = tensor("op_1647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1647_cast_fp16 = slice_by_index(begin = var_1647_begin_0, end = var_1647_end_0, end_mask = var_1647_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1647_cast_fp16")]; + tensor var_1651_begin_0 = const()[name = tensor("op_1651_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1651_end_0 = const()[name = tensor("op_1651_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1651_end_mask_0 = const()[name = tensor("op_1651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1651_cast_fp16 = slice_by_index(begin = var_1651_begin_0, end = var_1651_end_0, end_mask = var_1651_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1651_cast_fp16")]; + tensor var_1655_begin_0 = const()[name = tensor("op_1655_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1655_end_0 = const()[name = tensor("op_1655_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1655_end_mask_0 = const()[name = tensor("op_1655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1655_cast_fp16 = slice_by_index(begin = var_1655_begin_0, end = var_1655_end_0, end_mask = var_1655_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1655_cast_fp16")]; + tensor var_1659_begin_0 = const()[name = tensor("op_1659_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1659_end_0 = const()[name = tensor("op_1659_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1659_end_mask_0 = const()[name = tensor("op_1659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1659_cast_fp16 = slice_by_index(begin = var_1659_begin_0, end = var_1659_end_0, end_mask = var_1659_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1659_cast_fp16")]; + tensor var_1663_begin_0 = const()[name = tensor("op_1663_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1663_end_0 = const()[name = tensor("op_1663_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1663_end_mask_0 = const()[name = tensor("op_1663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1663_cast_fp16 = slice_by_index(begin = var_1663_begin_0, end = var_1663_end_0, end_mask = var_1663_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1663_cast_fp16")]; + tensor var_1667_begin_0 = const()[name = tensor("op_1667_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1667_end_0 = const()[name = tensor("op_1667_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1667_end_mask_0 = const()[name = tensor("op_1667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1667_cast_fp16 = slice_by_index(begin = var_1667_begin_0, end = var_1667_end_0, end_mask = var_1667_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1667_cast_fp16")]; + tensor var_1671_begin_0 = const()[name = tensor("op_1671_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1671_end_0 = const()[name = tensor("op_1671_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1671_end_mask_0 = const()[name = tensor("op_1671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1671_cast_fp16 = slice_by_index(begin = var_1671_begin_0, end = var_1671_end_0, end_mask = var_1671_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1671_cast_fp16")]; + tensor var_1675_equation_0 = const()[name = tensor("op_1675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1675_cast_fp16 = einsum(equation = var_1675_equation_0, values = (var_1597_cast_fp16, var_1554_cast_fp16))[name = tensor("op_1675_cast_fp16")]; + tensor var_1676_to_fp16 = const()[name = tensor("op_1676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_81_cast_fp16 = mul(x = var_1675_cast_fp16, y = var_1676_to_fp16)[name = tensor("aw_81_cast_fp16")]; + tensor var_1679_equation_0 = const()[name = tensor("op_1679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1679_cast_fp16 = einsum(equation = var_1679_equation_0, values = (var_1601_cast_fp16, var_1558_cast_fp16))[name = tensor("op_1679_cast_fp16")]; + tensor var_1680_to_fp16 = const()[name = tensor("op_1680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_83_cast_fp16 = mul(x = var_1679_cast_fp16, y = var_1680_to_fp16)[name = tensor("aw_83_cast_fp16")]; + tensor var_1683_equation_0 = const()[name = tensor("op_1683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1683_cast_fp16 = einsum(equation = var_1683_equation_0, values = (var_1605_cast_fp16, var_1562_cast_fp16))[name = tensor("op_1683_cast_fp16")]; + tensor var_1684_to_fp16 = const()[name = tensor("op_1684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_85_cast_fp16 = mul(x = var_1683_cast_fp16, y = var_1684_to_fp16)[name = tensor("aw_85_cast_fp16")]; + tensor var_1687_equation_0 = const()[name = tensor("op_1687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1687_cast_fp16 = einsum(equation = var_1687_equation_0, values = (var_1609_cast_fp16, var_1566_cast_fp16))[name = tensor("op_1687_cast_fp16")]; + tensor var_1688_to_fp16 = const()[name = tensor("op_1688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_87_cast_fp16 = mul(x = var_1687_cast_fp16, y = var_1688_to_fp16)[name = tensor("aw_87_cast_fp16")]; + tensor var_1691_equation_0 = const()[name = tensor("op_1691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1691_cast_fp16 = einsum(equation = var_1691_equation_0, values = (var_1613_cast_fp16, var_1570_cast_fp16))[name = tensor("op_1691_cast_fp16")]; + tensor var_1692_to_fp16 = const()[name = tensor("op_1692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_89_cast_fp16 = mul(x = var_1691_cast_fp16, y = var_1692_to_fp16)[name = tensor("aw_89_cast_fp16")]; + tensor var_1695_equation_0 = const()[name = tensor("op_1695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1695_cast_fp16 = einsum(equation = var_1695_equation_0, values = (var_1617_cast_fp16, var_1574_cast_fp16))[name = tensor("op_1695_cast_fp16")]; + tensor var_1696_to_fp16 = const()[name = tensor("op_1696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_91_cast_fp16 = mul(x = var_1695_cast_fp16, y = var_1696_to_fp16)[name = tensor("aw_91_cast_fp16")]; + tensor var_1699_equation_0 = const()[name = tensor("op_1699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1699_cast_fp16 = einsum(equation = var_1699_equation_0, values = (var_1621_cast_fp16, var_1578_cast_fp16))[name = tensor("op_1699_cast_fp16")]; + tensor var_1700_to_fp16 = const()[name = tensor("op_1700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_93_cast_fp16 = mul(x = var_1699_cast_fp16, y = var_1700_to_fp16)[name = tensor("aw_93_cast_fp16")]; + tensor var_1703_equation_0 = const()[name = tensor("op_1703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1703_cast_fp16 = einsum(equation = var_1703_equation_0, values = (var_1625_cast_fp16, var_1582_cast_fp16))[name = tensor("op_1703_cast_fp16")]; + tensor var_1704_to_fp16 = const()[name = tensor("op_1704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_95_cast_fp16 = mul(x = var_1703_cast_fp16, y = var_1704_to_fp16)[name = tensor("aw_95_cast_fp16")]; + tensor var_1707_equation_0 = const()[name = tensor("op_1707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1707_cast_fp16 = einsum(equation = var_1707_equation_0, values = (var_1629_cast_fp16, var_1586_cast_fp16))[name = tensor("op_1707_cast_fp16")]; + tensor var_1708_to_fp16 = const()[name = tensor("op_1708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_97_cast_fp16 = mul(x = var_1707_cast_fp16, y = var_1708_to_fp16)[name = tensor("aw_97_cast_fp16")]; + tensor var_1711_equation_0 = const()[name = tensor("op_1711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1711_cast_fp16 = einsum(equation = var_1711_equation_0, values = (var_1633_cast_fp16, var_1590_cast_fp16))[name = tensor("op_1711_cast_fp16")]; + tensor var_1712_to_fp16 = const()[name = tensor("op_1712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_99_cast_fp16 = mul(x = var_1711_cast_fp16, y = var_1712_to_fp16)[name = tensor("aw_99_cast_fp16")]; + tensor var_1714_cast_fp16 = softmax(axis = var_288, x = aw_81_cast_fp16)[name = tensor("op_1714_cast_fp16")]; + tensor var_1715_cast_fp16 = softmax(axis = var_288, x = aw_83_cast_fp16)[name = tensor("op_1715_cast_fp16")]; + tensor var_1716_cast_fp16 = softmax(axis = var_288, x = aw_85_cast_fp16)[name = tensor("op_1716_cast_fp16")]; + tensor var_1717_cast_fp16 = softmax(axis = var_288, x = aw_87_cast_fp16)[name = tensor("op_1717_cast_fp16")]; + tensor var_1718_cast_fp16 = softmax(axis = var_288, x = aw_89_cast_fp16)[name = tensor("op_1718_cast_fp16")]; + tensor var_1719_cast_fp16 = softmax(axis = var_288, x = aw_91_cast_fp16)[name = tensor("op_1719_cast_fp16")]; + tensor var_1720_cast_fp16 = softmax(axis = var_288, x = aw_93_cast_fp16)[name = tensor("op_1720_cast_fp16")]; + tensor var_1721_cast_fp16 = softmax(axis = var_288, x = aw_95_cast_fp16)[name = tensor("op_1721_cast_fp16")]; + tensor var_1722_cast_fp16 = softmax(axis = var_288, x = aw_97_cast_fp16)[name = tensor("op_1722_cast_fp16")]; + tensor var_1723_cast_fp16 = softmax(axis = var_288, x = aw_99_cast_fp16)[name = tensor("op_1723_cast_fp16")]; + tensor var_1725_equation_0 = const()[name = tensor("op_1725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1725_cast_fp16 = einsum(equation = var_1725_equation_0, values = (var_1635_cast_fp16, var_1714_cast_fp16))[name = tensor("op_1725_cast_fp16")]; + tensor var_1727_equation_0 = const()[name = tensor("op_1727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1727_cast_fp16 = einsum(equation = var_1727_equation_0, values = (var_1639_cast_fp16, var_1715_cast_fp16))[name = tensor("op_1727_cast_fp16")]; + tensor var_1729_equation_0 = const()[name = tensor("op_1729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1729_cast_fp16 = einsum(equation = var_1729_equation_0, values = (var_1643_cast_fp16, var_1716_cast_fp16))[name = tensor("op_1729_cast_fp16")]; + tensor var_1731_equation_0 = const()[name = tensor("op_1731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1731_cast_fp16 = einsum(equation = var_1731_equation_0, values = (var_1647_cast_fp16, var_1717_cast_fp16))[name = tensor("op_1731_cast_fp16")]; + tensor var_1733_equation_0 = const()[name = tensor("op_1733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1733_cast_fp16 = einsum(equation = var_1733_equation_0, values = (var_1651_cast_fp16, var_1718_cast_fp16))[name = tensor("op_1733_cast_fp16")]; + tensor var_1735_equation_0 = const()[name = tensor("op_1735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1735_cast_fp16 = einsum(equation = var_1735_equation_0, values = (var_1655_cast_fp16, var_1719_cast_fp16))[name = tensor("op_1735_cast_fp16")]; + tensor var_1737_equation_0 = const()[name = tensor("op_1737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1737_cast_fp16 = einsum(equation = var_1737_equation_0, values = (var_1659_cast_fp16, var_1720_cast_fp16))[name = tensor("op_1737_cast_fp16")]; + tensor var_1739_equation_0 = const()[name = tensor("op_1739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1739_cast_fp16 = einsum(equation = var_1739_equation_0, values = (var_1663_cast_fp16, var_1721_cast_fp16))[name = tensor("op_1739_cast_fp16")]; + tensor var_1741_equation_0 = const()[name = tensor("op_1741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1741_cast_fp16 = einsum(equation = var_1741_equation_0, values = (var_1667_cast_fp16, var_1722_cast_fp16))[name = tensor("op_1741_cast_fp16")]; + tensor var_1743_equation_0 = const()[name = tensor("op_1743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1743_cast_fp16 = einsum(equation = var_1743_equation_0, values = (var_1671_cast_fp16, var_1723_cast_fp16))[name = tensor("op_1743_cast_fp16")]; + tensor input_95_interleave_0 = const()[name = tensor("input_95_interleave_0"), val = tensor(false)]; + tensor input_95_cast_fp16 = concat(axis = var_288, interleave = input_95_interleave_0, values = (var_1725_cast_fp16, var_1727_cast_fp16, var_1729_cast_fp16, var_1731_cast_fp16, var_1733_cast_fp16, var_1735_cast_fp16, var_1737_cast_fp16, var_1739_cast_fp16, var_1741_cast_fp16, var_1743_cast_fp16))[name = tensor("input_95_cast_fp16")]; + tensor var_1753_pad_type_0 = const()[name = tensor("op_1753_pad_type_0"), val = tensor("valid")]; + tensor var_1753_strides_0 = const()[name = tensor("op_1753_strides_0"), val = tensor([1, 1])]; + tensor var_1753_pad_0 = const()[name = tensor("op_1753_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1753_dilations_0 = const()[name = tensor("op_1753_dilations_0"), val = tensor([1, 1])]; + tensor var_1753_groups_0 = const()[name = tensor("op_1753_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37554880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37862144))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37862336)))]; + tensor var_1753_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1753_dilations_0, groups = var_1753_groups_0, pad = var_1753_pad_0, pad_type = var_1753_pad_type_0, strides = var_1753_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor("op_1753_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = var_1753_cast_fp16, y = inputs_13_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; + tensor hidden_states_43_axes_0 = const()[name = tensor("hidden_states_43_axes_0"), val = tensor([1])]; + tensor hidden_states_43_gamma_0_to_fp16 = const()[name = tensor("hidden_states_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37863680)))]; + tensor hidden_states_43_beta_0_to_fp16 = const()[name = tensor("hidden_states_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37865024)))]; + tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = hidden_states_43_beta_0_to_fp16, epsilon = var_1763_to_fp16, gamma = hidden_states_43_gamma_0_to_fp16, x = inputs_15_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; + tensor q_11_pad_type_0 = const()[name = tensor("q_11_pad_type_0"), val = tensor("valid")]; + tensor q_11_strides_0 = const()[name = tensor("q_11_strides_0"), val = tensor([1, 1])]; + tensor q_11_pad_0 = const()[name = tensor("q_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_11_dilations_0 = const()[name = tensor("q_11_dilations_0"), val = tensor([1, 1])]; + tensor q_11_groups_0 = const()[name = tensor("q_11_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37866368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38173632))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_11_cast_fp16 = conv(dilations = q_11_dilations_0, groups = q_11_groups_0, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = q_11_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor("q_11_cast_fp16")]; + tensor k_21_pad_type_0 = const()[name = tensor("k_21_pad_type_0"), val = tensor("valid")]; + tensor k_21_strides_0 = const()[name = tensor("k_21_strides_0"), val = tensor([1, 1])]; + tensor k_21_pad_0 = const()[name = tensor("k_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_21_dilations_0 = const()[name = tensor("k_21_dilations_0"), val = tensor([1, 1])]; + tensor k_21_groups_0 = const()[name = tensor("k_21_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38173824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39156928))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor k_21_cast_fp16 = conv(dilations = k_21_dilations_0, groups = k_21_groups_0, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = k_21_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_21_cast_fp16")]; + tensor v_11_pad_type_0 = const()[name = tensor("v_11_pad_type_0"), val = tensor("valid")]; + tensor v_11_strides_0 = const()[name = tensor("v_11_strides_0"), val = tensor([1, 1])]; + tensor v_11_pad_0 = const()[name = tensor("v_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_11_dilations_0 = const()[name = tensor("v_11_dilations_0"), val = tensor([1, 1])]; + tensor v_11_groups_0 = const()[name = tensor("v_11_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39157120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40140224))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor v_11_cast_fp16 = conv(dilations = v_11_dilations_0, groups = v_11_groups_0, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = v_11_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_11_cast_fp16")]; + tensor var_1796_begin_0 = const()[name = tensor("op_1796_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1796_end_0 = const()[name = tensor("op_1796_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1796_end_mask_0 = const()[name = tensor("op_1796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1796_cast_fp16 = slice_by_index(begin = var_1796_begin_0, end = var_1796_end_0, end_mask = var_1796_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1796_cast_fp16")]; + tensor var_1800_begin_0 = const()[name = tensor("op_1800_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1800_end_0 = const()[name = tensor("op_1800_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1800_end_mask_0 = const()[name = tensor("op_1800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1800_cast_fp16 = slice_by_index(begin = var_1800_begin_0, end = var_1800_end_0, end_mask = var_1800_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1800_cast_fp16")]; + tensor var_1804_begin_0 = const()[name = tensor("op_1804_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1804_end_0 = const()[name = tensor("op_1804_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1804_end_mask_0 = const()[name = tensor("op_1804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1804_cast_fp16 = slice_by_index(begin = var_1804_begin_0, end = var_1804_end_0, end_mask = var_1804_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1804_cast_fp16")]; + tensor var_1808_begin_0 = const()[name = tensor("op_1808_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1808_end_0 = const()[name = tensor("op_1808_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1808_end_mask_0 = const()[name = tensor("op_1808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1808_cast_fp16 = slice_by_index(begin = var_1808_begin_0, end = var_1808_end_0, end_mask = var_1808_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1808_cast_fp16")]; + tensor var_1812_begin_0 = const()[name = tensor("op_1812_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1812_end_0 = const()[name = tensor("op_1812_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1812_end_mask_0 = const()[name = tensor("op_1812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1812_cast_fp16 = slice_by_index(begin = var_1812_begin_0, end = var_1812_end_0, end_mask = var_1812_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1812_cast_fp16")]; + tensor var_1816_begin_0 = const()[name = tensor("op_1816_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1816_end_0 = const()[name = tensor("op_1816_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1816_end_mask_0 = const()[name = tensor("op_1816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1816_cast_fp16 = slice_by_index(begin = var_1816_begin_0, end = var_1816_end_0, end_mask = var_1816_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1816_cast_fp16")]; + tensor var_1820_begin_0 = const()[name = tensor("op_1820_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1820_end_0 = const()[name = tensor("op_1820_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1820_end_mask_0 = const()[name = tensor("op_1820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1820_cast_fp16 = slice_by_index(begin = var_1820_begin_0, end = var_1820_end_0, end_mask = var_1820_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1820_cast_fp16")]; + tensor var_1824_begin_0 = const()[name = tensor("op_1824_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1824_end_0 = const()[name = tensor("op_1824_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1824_end_mask_0 = const()[name = tensor("op_1824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1824_cast_fp16 = slice_by_index(begin = var_1824_begin_0, end = var_1824_end_0, end_mask = var_1824_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1824_cast_fp16")]; + tensor var_1828_begin_0 = const()[name = tensor("op_1828_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1828_end_0 = const()[name = tensor("op_1828_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1828_end_mask_0 = const()[name = tensor("op_1828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1828_cast_fp16 = slice_by_index(begin = var_1828_begin_0, end = var_1828_end_0, end_mask = var_1828_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1828_cast_fp16")]; + tensor var_1832_begin_0 = const()[name = tensor("op_1832_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1832_end_0 = const()[name = tensor("op_1832_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1832_end_mask_0 = const()[name = tensor("op_1832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1832_cast_fp16 = slice_by_index(begin = var_1832_begin_0, end = var_1832_end_0, end_mask = var_1832_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1832_cast_fp16")]; + tensor k_23_perm_0 = const()[name = tensor("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1839_begin_0 = const()[name = tensor("op_1839_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1839_end_0 = const()[name = tensor("op_1839_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_1839_end_mask_0 = const()[name = tensor("op_1839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = k_21_cast_fp16)[name = tensor("transpose_62")]; + tensor var_1839_cast_fp16 = slice_by_index(begin = var_1839_begin_0, end = var_1839_end_0, end_mask = var_1839_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1839_cast_fp16")]; + tensor var_1843_begin_0 = const()[name = tensor("op_1843_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1843_end_0 = const()[name = tensor("op_1843_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_1843_end_mask_0 = const()[name = tensor("op_1843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1843_cast_fp16 = slice_by_index(begin = var_1843_begin_0, end = var_1843_end_0, end_mask = var_1843_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1843_cast_fp16")]; + tensor var_1847_begin_0 = const()[name = tensor("op_1847_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1847_end_0 = const()[name = tensor("op_1847_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_1847_end_mask_0 = const()[name = tensor("op_1847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1847_cast_fp16 = slice_by_index(begin = var_1847_begin_0, end = var_1847_end_0, end_mask = var_1847_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1847_cast_fp16")]; + tensor var_1851_begin_0 = const()[name = tensor("op_1851_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1851_end_0 = const()[name = tensor("op_1851_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_1851_end_mask_0 = const()[name = tensor("op_1851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1851_cast_fp16 = slice_by_index(begin = var_1851_begin_0, end = var_1851_end_0, end_mask = var_1851_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1851_cast_fp16")]; + tensor var_1855_begin_0 = const()[name = tensor("op_1855_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1855_end_0 = const()[name = tensor("op_1855_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_1855_end_mask_0 = const()[name = tensor("op_1855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1855_cast_fp16")]; + tensor var_1859_begin_0 = const()[name = tensor("op_1859_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1859_end_0 = const()[name = tensor("op_1859_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_1859_end_mask_0 = const()[name = tensor("op_1859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1859_cast_fp16 = slice_by_index(begin = var_1859_begin_0, end = var_1859_end_0, end_mask = var_1859_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1859_cast_fp16")]; + tensor var_1863_begin_0 = const()[name = tensor("op_1863_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1863_end_0 = const()[name = tensor("op_1863_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_1863_end_mask_0 = const()[name = tensor("op_1863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1863_cast_fp16 = slice_by_index(begin = var_1863_begin_0, end = var_1863_end_0, end_mask = var_1863_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1863_cast_fp16")]; + tensor var_1867_begin_0 = const()[name = tensor("op_1867_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1867_end_0 = const()[name = tensor("op_1867_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_1867_end_mask_0 = const()[name = tensor("op_1867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1867_cast_fp16 = slice_by_index(begin = var_1867_begin_0, end = var_1867_end_0, end_mask = var_1867_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1867_cast_fp16")]; + tensor var_1871_begin_0 = const()[name = tensor("op_1871_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1871_end_0 = const()[name = tensor("op_1871_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_1871_end_mask_0 = const()[name = tensor("op_1871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1871_cast_fp16 = slice_by_index(begin = var_1871_begin_0, end = var_1871_end_0, end_mask = var_1871_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1871_cast_fp16")]; + tensor var_1875_begin_0 = const()[name = tensor("op_1875_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1875_end_0 = const()[name = tensor("op_1875_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_1875_end_mask_0 = const()[name = tensor("op_1875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1875_cast_fp16 = slice_by_index(begin = var_1875_begin_0, end = var_1875_end_0, end_mask = var_1875_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1875_cast_fp16")]; + tensor var_1877_begin_0 = const()[name = tensor("op_1877_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1877_end_0 = const()[name = tensor("op_1877_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_1877_end_mask_0 = const()[name = tensor("op_1877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1877_cast_fp16 = slice_by_index(begin = var_1877_begin_0, end = var_1877_end_0, end_mask = var_1877_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1877_cast_fp16")]; + tensor var_1881_begin_0 = const()[name = tensor("op_1881_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1881_end_0 = const()[name = tensor("op_1881_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_1881_end_mask_0 = const()[name = tensor("op_1881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1881_cast_fp16 = slice_by_index(begin = var_1881_begin_0, end = var_1881_end_0, end_mask = var_1881_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1881_cast_fp16")]; + tensor var_1885_begin_0 = const()[name = tensor("op_1885_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1885_end_0 = const()[name = tensor("op_1885_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_1885_end_mask_0 = const()[name = tensor("op_1885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1885_cast_fp16 = slice_by_index(begin = var_1885_begin_0, end = var_1885_end_0, end_mask = var_1885_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1885_cast_fp16")]; + tensor var_1889_begin_0 = const()[name = tensor("op_1889_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1889_end_0 = const()[name = tensor("op_1889_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_1889_end_mask_0 = const()[name = tensor("op_1889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1889_cast_fp16 = slice_by_index(begin = var_1889_begin_0, end = var_1889_end_0, end_mask = var_1889_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1889_cast_fp16")]; + tensor var_1893_begin_0 = const()[name = tensor("op_1893_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1893_end_0 = const()[name = tensor("op_1893_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_1893_end_mask_0 = const()[name = tensor("op_1893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1893_cast_fp16 = slice_by_index(begin = var_1893_begin_0, end = var_1893_end_0, end_mask = var_1893_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1893_cast_fp16")]; + tensor var_1897_begin_0 = const()[name = tensor("op_1897_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1897_end_0 = const()[name = tensor("op_1897_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_1897_end_mask_0 = const()[name = tensor("op_1897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1897_cast_fp16 = slice_by_index(begin = var_1897_begin_0, end = var_1897_end_0, end_mask = var_1897_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1897_cast_fp16")]; + tensor var_1901_begin_0 = const()[name = tensor("op_1901_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1901_end_0 = const()[name = tensor("op_1901_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_1901_end_mask_0 = const()[name = tensor("op_1901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1901_cast_fp16 = slice_by_index(begin = var_1901_begin_0, end = var_1901_end_0, end_mask = var_1901_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1901_cast_fp16")]; + tensor var_1905_begin_0 = const()[name = tensor("op_1905_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1905_end_0 = const()[name = tensor("op_1905_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_1905_end_mask_0 = const()[name = tensor("op_1905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1905_cast_fp16 = slice_by_index(begin = var_1905_begin_0, end = var_1905_end_0, end_mask = var_1905_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1905_cast_fp16")]; + tensor var_1909_begin_0 = const()[name = tensor("op_1909_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1909_end_0 = const()[name = tensor("op_1909_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_1909_end_mask_0 = const()[name = tensor("op_1909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1909_cast_fp16 = slice_by_index(begin = var_1909_begin_0, end = var_1909_end_0, end_mask = var_1909_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1909_cast_fp16")]; + tensor var_1913_begin_0 = const()[name = tensor("op_1913_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1913_end_0 = const()[name = tensor("op_1913_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_1913_end_mask_0 = const()[name = tensor("op_1913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1913_cast_fp16 = slice_by_index(begin = var_1913_begin_0, end = var_1913_end_0, end_mask = var_1913_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1913_cast_fp16")]; + tensor var_1917_equation_0 = const()[name = tensor("op_1917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1917_cast_fp16 = einsum(equation = var_1917_equation_0, values = (var_1839_cast_fp16, var_1796_cast_fp16))[name = tensor("op_1917_cast_fp16")]; + tensor var_1918_to_fp16 = const()[name = tensor("op_1918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_101_cast_fp16 = mul(x = var_1917_cast_fp16, y = var_1918_to_fp16)[name = tensor("aw_101_cast_fp16")]; + tensor var_1921_equation_0 = const()[name = tensor("op_1921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1921_cast_fp16 = einsum(equation = var_1921_equation_0, values = (var_1843_cast_fp16, var_1800_cast_fp16))[name = tensor("op_1921_cast_fp16")]; + tensor var_1922_to_fp16 = const()[name = tensor("op_1922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_103_cast_fp16 = mul(x = var_1921_cast_fp16, y = var_1922_to_fp16)[name = tensor("aw_103_cast_fp16")]; + tensor var_1925_equation_0 = const()[name = tensor("op_1925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1925_cast_fp16 = einsum(equation = var_1925_equation_0, values = (var_1847_cast_fp16, var_1804_cast_fp16))[name = tensor("op_1925_cast_fp16")]; + tensor var_1926_to_fp16 = const()[name = tensor("op_1926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_105_cast_fp16 = mul(x = var_1925_cast_fp16, y = var_1926_to_fp16)[name = tensor("aw_105_cast_fp16")]; + tensor var_1929_equation_0 = const()[name = tensor("op_1929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1929_cast_fp16 = einsum(equation = var_1929_equation_0, values = (var_1851_cast_fp16, var_1808_cast_fp16))[name = tensor("op_1929_cast_fp16")]; + tensor var_1930_to_fp16 = const()[name = tensor("op_1930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_107_cast_fp16 = mul(x = var_1929_cast_fp16, y = var_1930_to_fp16)[name = tensor("aw_107_cast_fp16")]; + tensor var_1933_equation_0 = const()[name = tensor("op_1933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1933_cast_fp16 = einsum(equation = var_1933_equation_0, values = (var_1855_cast_fp16, var_1812_cast_fp16))[name = tensor("op_1933_cast_fp16")]; + tensor var_1934_to_fp16 = const()[name = tensor("op_1934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_109_cast_fp16 = mul(x = var_1933_cast_fp16, y = var_1934_to_fp16)[name = tensor("aw_109_cast_fp16")]; + tensor var_1937_equation_0 = const()[name = tensor("op_1937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1937_cast_fp16 = einsum(equation = var_1937_equation_0, values = (var_1859_cast_fp16, var_1816_cast_fp16))[name = tensor("op_1937_cast_fp16")]; + tensor var_1938_to_fp16 = const()[name = tensor("op_1938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_111_cast_fp16 = mul(x = var_1937_cast_fp16, y = var_1938_to_fp16)[name = tensor("aw_111_cast_fp16")]; + tensor var_1941_equation_0 = const()[name = tensor("op_1941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1941_cast_fp16 = einsum(equation = var_1941_equation_0, values = (var_1863_cast_fp16, var_1820_cast_fp16))[name = tensor("op_1941_cast_fp16")]; + tensor var_1942_to_fp16 = const()[name = tensor("op_1942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_113_cast_fp16 = mul(x = var_1941_cast_fp16, y = var_1942_to_fp16)[name = tensor("aw_113_cast_fp16")]; + tensor var_1945_equation_0 = const()[name = tensor("op_1945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1945_cast_fp16 = einsum(equation = var_1945_equation_0, values = (var_1867_cast_fp16, var_1824_cast_fp16))[name = tensor("op_1945_cast_fp16")]; + tensor var_1946_to_fp16 = const()[name = tensor("op_1946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_115_cast_fp16 = mul(x = var_1945_cast_fp16, y = var_1946_to_fp16)[name = tensor("aw_115_cast_fp16")]; + tensor var_1949_equation_0 = const()[name = tensor("op_1949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1949_cast_fp16 = einsum(equation = var_1949_equation_0, values = (var_1871_cast_fp16, var_1828_cast_fp16))[name = tensor("op_1949_cast_fp16")]; + tensor var_1950_to_fp16 = const()[name = tensor("op_1950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_117_cast_fp16 = mul(x = var_1949_cast_fp16, y = var_1950_to_fp16)[name = tensor("aw_117_cast_fp16")]; + tensor var_1953_equation_0 = const()[name = tensor("op_1953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1953_cast_fp16 = einsum(equation = var_1953_equation_0, values = (var_1875_cast_fp16, var_1832_cast_fp16))[name = tensor("op_1953_cast_fp16")]; + tensor var_1954_to_fp16 = const()[name = tensor("op_1954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_119_cast_fp16 = mul(x = var_1953_cast_fp16, y = var_1954_to_fp16)[name = tensor("aw_119_cast_fp16")]; + tensor var_1956_cast_fp16 = softmax(axis = var_288, x = aw_101_cast_fp16)[name = tensor("op_1956_cast_fp16")]; + tensor var_1957_cast_fp16 = softmax(axis = var_288, x = aw_103_cast_fp16)[name = tensor("op_1957_cast_fp16")]; + tensor var_1958_cast_fp16 = softmax(axis = var_288, x = aw_105_cast_fp16)[name = tensor("op_1958_cast_fp16")]; + tensor var_1959_cast_fp16 = softmax(axis = var_288, x = aw_107_cast_fp16)[name = tensor("op_1959_cast_fp16")]; + tensor var_1960_cast_fp16 = softmax(axis = var_288, x = aw_109_cast_fp16)[name = tensor("op_1960_cast_fp16")]; + tensor var_1961_cast_fp16 = softmax(axis = var_288, x = aw_111_cast_fp16)[name = tensor("op_1961_cast_fp16")]; + tensor var_1962_cast_fp16 = softmax(axis = var_288, x = aw_113_cast_fp16)[name = tensor("op_1962_cast_fp16")]; + tensor var_1963_cast_fp16 = softmax(axis = var_288, x = aw_115_cast_fp16)[name = tensor("op_1963_cast_fp16")]; + tensor var_1964_cast_fp16 = softmax(axis = var_288, x = aw_117_cast_fp16)[name = tensor("op_1964_cast_fp16")]; + tensor var_1965_cast_fp16 = softmax(axis = var_288, x = aw_119_cast_fp16)[name = tensor("op_1965_cast_fp16")]; + tensor var_1967_equation_0 = const()[name = tensor("op_1967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1967_cast_fp16 = einsum(equation = var_1967_equation_0, values = (var_1877_cast_fp16, var_1956_cast_fp16))[name = tensor("op_1967_cast_fp16")]; + tensor var_1969_equation_0 = const()[name = tensor("op_1969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1969_cast_fp16 = einsum(equation = var_1969_equation_0, values = (var_1881_cast_fp16, var_1957_cast_fp16))[name = tensor("op_1969_cast_fp16")]; + tensor var_1971_equation_0 = const()[name = tensor("op_1971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1971_cast_fp16 = einsum(equation = var_1971_equation_0, values = (var_1885_cast_fp16, var_1958_cast_fp16))[name = tensor("op_1971_cast_fp16")]; + tensor var_1973_equation_0 = const()[name = tensor("op_1973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1973_cast_fp16 = einsum(equation = var_1973_equation_0, values = (var_1889_cast_fp16, var_1959_cast_fp16))[name = tensor("op_1973_cast_fp16")]; + tensor var_1975_equation_0 = const()[name = tensor("op_1975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1975_cast_fp16 = einsum(equation = var_1975_equation_0, values = (var_1893_cast_fp16, var_1960_cast_fp16))[name = tensor("op_1975_cast_fp16")]; + tensor var_1977_equation_0 = const()[name = tensor("op_1977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1977_cast_fp16 = einsum(equation = var_1977_equation_0, values = (var_1897_cast_fp16, var_1961_cast_fp16))[name = tensor("op_1977_cast_fp16")]; + tensor var_1979_equation_0 = const()[name = tensor("op_1979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1979_cast_fp16 = einsum(equation = var_1979_equation_0, values = (var_1901_cast_fp16, var_1962_cast_fp16))[name = tensor("op_1979_cast_fp16")]; + tensor var_1981_equation_0 = const()[name = tensor("op_1981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1981_cast_fp16 = einsum(equation = var_1981_equation_0, values = (var_1905_cast_fp16, var_1963_cast_fp16))[name = tensor("op_1981_cast_fp16")]; + tensor var_1983_equation_0 = const()[name = tensor("op_1983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1983_cast_fp16 = einsum(equation = var_1983_equation_0, values = (var_1909_cast_fp16, var_1964_cast_fp16))[name = tensor("op_1983_cast_fp16")]; + tensor var_1985_equation_0 = const()[name = tensor("op_1985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1985_cast_fp16 = einsum(equation = var_1985_equation_0, values = (var_1913_cast_fp16, var_1965_cast_fp16))[name = tensor("op_1985_cast_fp16")]; + tensor input_97_interleave_0 = const()[name = tensor("input_97_interleave_0"), val = tensor(false)]; + tensor input_97_cast_fp16 = concat(axis = var_288, interleave = input_97_interleave_0, values = (var_1967_cast_fp16, var_1969_cast_fp16, var_1971_cast_fp16, var_1973_cast_fp16, var_1975_cast_fp16, var_1977_cast_fp16, var_1979_cast_fp16, var_1981_cast_fp16, var_1983_cast_fp16, var_1985_cast_fp16))[name = tensor("input_97_cast_fp16")]; + tensor var_1995_pad_type_0 = const()[name = tensor("op_1995_pad_type_0"), val = tensor("valid")]; + tensor var_1995_strides_0 = const()[name = tensor("op_1995_strides_0"), val = tensor([1, 1])]; + tensor var_1995_pad_0 = const()[name = tensor("op_1995_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1995_dilations_0 = const()[name = tensor("op_1995_dilations_0"), val = tensor([1, 1])]; + tensor var_1995_groups_0 = const()[name = tensor("op_1995_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40140416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40447680))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40447872)))]; + tensor var_1995_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1995_dilations_0, groups = var_1995_groups_0, pad = var_1995_pad_0, pad_type = var_1995_pad_type_0, strides = var_1995_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor("op_1995_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = var_1995_cast_fp16, y = inputs_15_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; + tensor input_99_axes_0 = const()[name = tensor("input_99_axes_0"), val = tensor([1])]; + tensor input_99_gamma_0_to_fp16 = const()[name = tensor("input_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40449216)))]; + tensor input_99_beta_0_to_fp16 = const()[name = tensor("input_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40450560)))]; + tensor var_2005_to_fp16 = const()[name = tensor("op_2005_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_99_cast_fp16 = layer_norm(axes = input_99_axes_0, beta = input_99_beta_0_to_fp16, epsilon = var_2005_to_fp16, gamma = input_99_gamma_0_to_fp16, x = inputs_17_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor var_2025_pad_type_0 = const()[name = tensor("op_2025_pad_type_0"), val = tensor("valid")]; + tensor var_2025_strides_0 = const()[name = tensor("op_2025_strides_0"), val = tensor([1, 1])]; + tensor var_2025_pad_0 = const()[name = tensor("op_2025_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2025_dilations_0 = const()[name = tensor("op_2025_dilations_0"), val = tensor([1, 1])]; + tensor var_2025_groups_0 = const()[name = tensor("op_2025_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40451904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42909568))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42909760)))]; + tensor var_2025_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_2025_dilations_0, groups = var_2025_groups_0, pad = var_2025_pad_0, pad_type = var_2025_pad_type_0, strides = var_2025_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = tensor("op_2025_cast_fp16")]; + tensor var_2026_split_sizes_0 = const()[name = tensor("op_2026_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_2026_axis_0 = const()[name = tensor("op_2026_axis_0"), val = tensor(1)]; + tensor var_2026_cast_fp16_0, tensor var_2026_cast_fp16_1 = split(axis = var_2026_axis_0, split_sizes = var_2026_split_sizes_0, x = var_2025_cast_fp16)[name = tensor("op_2026_cast_fp16")]; + tensor var_2028_mode_0 = const()[name = tensor("op_2028_mode_0"), val = tensor("EXACT")]; + tensor var_2028_cast_fp16 = gelu(mode = var_2028_mode_0, x = var_2026_cast_fp16_1)[name = tensor("op_2028_cast_fp16")]; + tensor input_101_cast_fp16 = mul(x = var_2026_cast_fp16_0, y = var_2028_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor var_2036_pad_type_0 = const()[name = tensor("op_2036_pad_type_0"), val = tensor("valid")]; + tensor var_2036_strides_0 = const()[name = tensor("op_2036_strides_0"), val = tensor([1, 1])]; + tensor var_2036_pad_0 = const()[name = tensor("op_2036_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2036_dilations_0 = const()[name = tensor("op_2036_dilations_0"), val = tensor([1, 1])]; + tensor var_2036_groups_0 = const()[name = tensor("op_2036_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42920064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44148928))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44149120)))]; + tensor var_2036_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2036_dilations_0, groups = var_2036_groups_0, pad = var_2036_pad_0, pad_type = var_2036_pad_type_0, strides = var_2036_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_101_cast_fp16)[name = tensor("op_2036_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = var_2036_cast_fp16, y = inputs_17_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; + tensor hidden_states_47_axes_0 = const()[name = tensor("hidden_states_47_axes_0"), val = tensor([1])]; + tensor hidden_states_47_gamma_0_to_fp16 = const()[name = tensor("hidden_states_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44150464)))]; + tensor hidden_states_47_beta_0_to_fp16 = const()[name = tensor("hidden_states_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44151808)))]; + tensor var_2052_to_fp16 = const()[name = tensor("op_2052_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_47_cast_fp16 = layer_norm(axes = hidden_states_47_axes_0, beta = hidden_states_47_beta_0_to_fp16, epsilon = var_2052_to_fp16, gamma = hidden_states_47_gamma_0_to_fp16, x = inputs_19_cast_fp16)[name = tensor("hidden_states_47_cast_fp16")]; + tensor q_13_pad_type_0 = const()[name = tensor("q_13_pad_type_0"), val = tensor("valid")]; + tensor q_13_strides_0 = const()[name = tensor("q_13_strides_0"), val = tensor([1, 1])]; + tensor q_13_pad_0 = const()[name = tensor("q_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_13_dilations_0 = const()[name = tensor("q_13_dilations_0"), val = tensor([1, 1])]; + tensor q_13_groups_0 = const()[name = tensor("q_13_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44153152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44460416))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_13_cast_fp16 = conv(dilations = q_13_dilations_0, groups = q_13_groups_0, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = q_13_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_47_cast_fp16)[name = tensor("q_13_cast_fp16")]; + tensor k_25_pad_type_0 = const()[name = tensor("k_25_pad_type_0"), val = tensor("valid")]; + tensor k_25_strides_0 = const()[name = tensor("k_25_strides_0"), val = tensor([1, 1])]; + tensor k_25_pad_0 = const()[name = tensor("k_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_25_dilations_0 = const()[name = tensor("k_25_dilations_0"), val = tensor([1, 1])]; + tensor k_25_groups_0 = const()[name = tensor("k_25_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44460608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44767872))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_25_cast_fp16 = conv(dilations = k_25_dilations_0, groups = k_25_groups_0, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = k_25_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_47_cast_fp16)[name = tensor("k_25_cast_fp16")]; + tensor v_13_pad_type_0 = const()[name = tensor("v_13_pad_type_0"), val = tensor("valid")]; + tensor v_13_strides_0 = const()[name = tensor("v_13_strides_0"), val = tensor([1, 1])]; + tensor v_13_pad_0 = const()[name = tensor("v_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_13_dilations_0 = const()[name = tensor("v_13_dilations_0"), val = tensor([1, 1])]; + tensor v_13_groups_0 = const()[name = tensor("v_13_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44768064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45075328))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_13_cast_fp16 = conv(dilations = v_13_dilations_0, groups = v_13_groups_0, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = v_13_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_47_cast_fp16)[name = tensor("v_13_cast_fp16")]; + tensor var_2085_begin_0 = const()[name = tensor("op_2085_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2085_end_0 = const()[name = tensor("op_2085_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_2085_end_mask_0 = const()[name = tensor("op_2085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2085_cast_fp16 = slice_by_index(begin = var_2085_begin_0, end = var_2085_end_0, end_mask = var_2085_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2085_cast_fp16")]; + tensor var_2089_begin_0 = const()[name = tensor("op_2089_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2089_end_0 = const()[name = tensor("op_2089_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_2089_end_mask_0 = const()[name = tensor("op_2089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2089_cast_fp16 = slice_by_index(begin = var_2089_begin_0, end = var_2089_end_0, end_mask = var_2089_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2089_cast_fp16")]; + tensor var_2093_begin_0 = const()[name = tensor("op_2093_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2093_end_0 = const()[name = tensor("op_2093_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_2093_end_mask_0 = const()[name = tensor("op_2093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2093_cast_fp16 = slice_by_index(begin = var_2093_begin_0, end = var_2093_end_0, end_mask = var_2093_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2093_cast_fp16")]; + tensor var_2097_begin_0 = const()[name = tensor("op_2097_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2097_end_0 = const()[name = tensor("op_2097_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_2097_end_mask_0 = const()[name = tensor("op_2097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2097_cast_fp16 = slice_by_index(begin = var_2097_begin_0, end = var_2097_end_0, end_mask = var_2097_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2097_cast_fp16")]; + tensor var_2101_begin_0 = const()[name = tensor("op_2101_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2101_end_0 = const()[name = tensor("op_2101_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_2101_end_mask_0 = const()[name = tensor("op_2101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2101_cast_fp16 = slice_by_index(begin = var_2101_begin_0, end = var_2101_end_0, end_mask = var_2101_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2101_cast_fp16")]; + tensor var_2105_begin_0 = const()[name = tensor("op_2105_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2105_end_0 = const()[name = tensor("op_2105_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_2105_end_mask_0 = const()[name = tensor("op_2105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2105_cast_fp16 = slice_by_index(begin = var_2105_begin_0, end = var_2105_end_0, end_mask = var_2105_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2105_cast_fp16")]; + tensor var_2109_begin_0 = const()[name = tensor("op_2109_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2109_end_0 = const()[name = tensor("op_2109_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_2109_end_mask_0 = const()[name = tensor("op_2109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2109_cast_fp16 = slice_by_index(begin = var_2109_begin_0, end = var_2109_end_0, end_mask = var_2109_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2109_cast_fp16")]; + tensor var_2113_begin_0 = const()[name = tensor("op_2113_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2113_end_0 = const()[name = tensor("op_2113_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_2113_end_mask_0 = const()[name = tensor("op_2113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2113_cast_fp16 = slice_by_index(begin = var_2113_begin_0, end = var_2113_end_0, end_mask = var_2113_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2113_cast_fp16")]; + tensor var_2117_begin_0 = const()[name = tensor("op_2117_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2117_end_0 = const()[name = tensor("op_2117_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_2117_end_mask_0 = const()[name = tensor("op_2117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2117_cast_fp16 = slice_by_index(begin = var_2117_begin_0, end = var_2117_end_0, end_mask = var_2117_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2117_cast_fp16")]; + tensor var_2121_begin_0 = const()[name = tensor("op_2121_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2121_end_0 = const()[name = tensor("op_2121_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_2121_end_mask_0 = const()[name = tensor("op_2121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2121_cast_fp16 = slice_by_index(begin = var_2121_begin_0, end = var_2121_end_0, end_mask = var_2121_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2121_cast_fp16")]; + tensor k_27_perm_0 = const()[name = tensor("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2128_begin_0 = const()[name = tensor("op_2128_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2128_end_0 = const()[name = tensor("op_2128_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_2128_end_mask_0 = const()[name = tensor("op_2128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = k_25_cast_fp16)[name = tensor("transpose_61")]; + tensor var_2128_cast_fp16 = slice_by_index(begin = var_2128_begin_0, end = var_2128_end_0, end_mask = var_2128_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2128_cast_fp16")]; + tensor var_2132_begin_0 = const()[name = tensor("op_2132_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2132_end_0 = const()[name = tensor("op_2132_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_2132_end_mask_0 = const()[name = tensor("op_2132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2132_cast_fp16 = slice_by_index(begin = var_2132_begin_0, end = var_2132_end_0, end_mask = var_2132_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2132_cast_fp16")]; + tensor var_2136_begin_0 = const()[name = tensor("op_2136_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2136_end_0 = const()[name = tensor("op_2136_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_2136_end_mask_0 = const()[name = tensor("op_2136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2136_cast_fp16 = slice_by_index(begin = var_2136_begin_0, end = var_2136_end_0, end_mask = var_2136_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2136_cast_fp16")]; + tensor var_2140_begin_0 = const()[name = tensor("op_2140_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2140_end_0 = const()[name = tensor("op_2140_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_2140_end_mask_0 = const()[name = tensor("op_2140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2140_cast_fp16 = slice_by_index(begin = var_2140_begin_0, end = var_2140_end_0, end_mask = var_2140_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2140_cast_fp16")]; + tensor var_2144_begin_0 = const()[name = tensor("op_2144_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2144_end_0 = const()[name = tensor("op_2144_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_2144_end_mask_0 = const()[name = tensor("op_2144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2144_cast_fp16 = slice_by_index(begin = var_2144_begin_0, end = var_2144_end_0, end_mask = var_2144_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2144_cast_fp16")]; + tensor var_2148_begin_0 = const()[name = tensor("op_2148_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2148_end_0 = const()[name = tensor("op_2148_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_2148_end_mask_0 = const()[name = tensor("op_2148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2148_cast_fp16 = slice_by_index(begin = var_2148_begin_0, end = var_2148_end_0, end_mask = var_2148_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2148_cast_fp16")]; + tensor var_2152_begin_0 = const()[name = tensor("op_2152_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2152_end_0 = const()[name = tensor("op_2152_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_2152_end_mask_0 = const()[name = tensor("op_2152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2152_cast_fp16 = slice_by_index(begin = var_2152_begin_0, end = var_2152_end_0, end_mask = var_2152_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2152_cast_fp16")]; + tensor var_2156_begin_0 = const()[name = tensor("op_2156_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2156_end_0 = const()[name = tensor("op_2156_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_2156_end_mask_0 = const()[name = tensor("op_2156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2156_cast_fp16 = slice_by_index(begin = var_2156_begin_0, end = var_2156_end_0, end_mask = var_2156_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2156_cast_fp16")]; + tensor var_2160_begin_0 = const()[name = tensor("op_2160_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2160_end_0 = const()[name = tensor("op_2160_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_2160_end_mask_0 = const()[name = tensor("op_2160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2160_cast_fp16")]; + tensor var_2164_begin_0 = const()[name = tensor("op_2164_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2164_end_0 = const()[name = tensor("op_2164_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_2164_end_mask_0 = const()[name = tensor("op_2164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2164_cast_fp16")]; + tensor var_2166_begin_0 = const()[name = tensor("op_2166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2166_end_0 = const()[name = tensor("op_2166_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_2166_end_mask_0 = const()[name = tensor("op_2166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2166_cast_fp16 = slice_by_index(begin = var_2166_begin_0, end = var_2166_end_0, end_mask = var_2166_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2166_cast_fp16")]; + tensor var_2170_begin_0 = const()[name = tensor("op_2170_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2170_end_0 = const()[name = tensor("op_2170_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_2170_end_mask_0 = const()[name = tensor("op_2170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2170_cast_fp16 = slice_by_index(begin = var_2170_begin_0, end = var_2170_end_0, end_mask = var_2170_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2170_cast_fp16")]; + tensor var_2174_begin_0 = const()[name = tensor("op_2174_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2174_end_0 = const()[name = tensor("op_2174_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_2174_end_mask_0 = const()[name = tensor("op_2174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2174_cast_fp16 = slice_by_index(begin = var_2174_begin_0, end = var_2174_end_0, end_mask = var_2174_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2174_cast_fp16")]; + tensor var_2178_begin_0 = const()[name = tensor("op_2178_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2178_end_0 = const()[name = tensor("op_2178_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_2178_end_mask_0 = const()[name = tensor("op_2178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2178_cast_fp16 = slice_by_index(begin = var_2178_begin_0, end = var_2178_end_0, end_mask = var_2178_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2178_cast_fp16")]; + tensor var_2182_begin_0 = const()[name = tensor("op_2182_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2182_end_0 = const()[name = tensor("op_2182_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_2182_end_mask_0 = const()[name = tensor("op_2182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2182_cast_fp16 = slice_by_index(begin = var_2182_begin_0, end = var_2182_end_0, end_mask = var_2182_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2182_cast_fp16")]; + tensor var_2186_begin_0 = const()[name = tensor("op_2186_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2186_end_0 = const()[name = tensor("op_2186_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_2186_end_mask_0 = const()[name = tensor("op_2186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2186_cast_fp16 = slice_by_index(begin = var_2186_begin_0, end = var_2186_end_0, end_mask = var_2186_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2186_cast_fp16")]; + tensor var_2190_begin_0 = const()[name = tensor("op_2190_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2190_end_0 = const()[name = tensor("op_2190_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_2190_end_mask_0 = const()[name = tensor("op_2190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2190_cast_fp16 = slice_by_index(begin = var_2190_begin_0, end = var_2190_end_0, end_mask = var_2190_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2190_cast_fp16")]; + tensor var_2194_begin_0 = const()[name = tensor("op_2194_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2194_end_0 = const()[name = tensor("op_2194_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_2194_end_mask_0 = const()[name = tensor("op_2194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2194_cast_fp16 = slice_by_index(begin = var_2194_begin_0, end = var_2194_end_0, end_mask = var_2194_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2194_cast_fp16")]; + tensor var_2198_begin_0 = const()[name = tensor("op_2198_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2198_end_0 = const()[name = tensor("op_2198_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_2198_end_mask_0 = const()[name = tensor("op_2198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2198_cast_fp16 = slice_by_index(begin = var_2198_begin_0, end = var_2198_end_0, end_mask = var_2198_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2198_cast_fp16")]; + tensor var_2202_begin_0 = const()[name = tensor("op_2202_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2202_end_0 = const()[name = tensor("op_2202_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_2202_end_mask_0 = const()[name = tensor("op_2202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2202_cast_fp16 = slice_by_index(begin = var_2202_begin_0, end = var_2202_end_0, end_mask = var_2202_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2202_cast_fp16")]; + tensor var_2206_equation_0 = const()[name = tensor("op_2206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2206_cast_fp16 = einsum(equation = var_2206_equation_0, values = (var_2128_cast_fp16, var_2085_cast_fp16))[name = tensor("op_2206_cast_fp16")]; + tensor var_2207_to_fp16 = const()[name = tensor("op_2207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_121_cast_fp16 = mul(x = var_2206_cast_fp16, y = var_2207_to_fp16)[name = tensor("aw_121_cast_fp16")]; + tensor var_2210_equation_0 = const()[name = tensor("op_2210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2210_cast_fp16 = einsum(equation = var_2210_equation_0, values = (var_2132_cast_fp16, var_2089_cast_fp16))[name = tensor("op_2210_cast_fp16")]; + tensor var_2211_to_fp16 = const()[name = tensor("op_2211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_123_cast_fp16 = mul(x = var_2210_cast_fp16, y = var_2211_to_fp16)[name = tensor("aw_123_cast_fp16")]; + tensor var_2214_equation_0 = const()[name = tensor("op_2214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2214_cast_fp16 = einsum(equation = var_2214_equation_0, values = (var_2136_cast_fp16, var_2093_cast_fp16))[name = tensor("op_2214_cast_fp16")]; + tensor var_2215_to_fp16 = const()[name = tensor("op_2215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_125_cast_fp16 = mul(x = var_2214_cast_fp16, y = var_2215_to_fp16)[name = tensor("aw_125_cast_fp16")]; + tensor var_2218_equation_0 = const()[name = tensor("op_2218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2218_cast_fp16 = einsum(equation = var_2218_equation_0, values = (var_2140_cast_fp16, var_2097_cast_fp16))[name = tensor("op_2218_cast_fp16")]; + tensor var_2219_to_fp16 = const()[name = tensor("op_2219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_127_cast_fp16 = mul(x = var_2218_cast_fp16, y = var_2219_to_fp16)[name = tensor("aw_127_cast_fp16")]; + tensor var_2222_equation_0 = const()[name = tensor("op_2222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2222_cast_fp16 = einsum(equation = var_2222_equation_0, values = (var_2144_cast_fp16, var_2101_cast_fp16))[name = tensor("op_2222_cast_fp16")]; + tensor var_2223_to_fp16 = const()[name = tensor("op_2223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_129_cast_fp16 = mul(x = var_2222_cast_fp16, y = var_2223_to_fp16)[name = tensor("aw_129_cast_fp16")]; + tensor var_2226_equation_0 = const()[name = tensor("op_2226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2226_cast_fp16 = einsum(equation = var_2226_equation_0, values = (var_2148_cast_fp16, var_2105_cast_fp16))[name = tensor("op_2226_cast_fp16")]; + tensor var_2227_to_fp16 = const()[name = tensor("op_2227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_131_cast_fp16 = mul(x = var_2226_cast_fp16, y = var_2227_to_fp16)[name = tensor("aw_131_cast_fp16")]; + tensor var_2230_equation_0 = const()[name = tensor("op_2230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2230_cast_fp16 = einsum(equation = var_2230_equation_0, values = (var_2152_cast_fp16, var_2109_cast_fp16))[name = tensor("op_2230_cast_fp16")]; + tensor var_2231_to_fp16 = const()[name = tensor("op_2231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_133_cast_fp16 = mul(x = var_2230_cast_fp16, y = var_2231_to_fp16)[name = tensor("aw_133_cast_fp16")]; + tensor var_2234_equation_0 = const()[name = tensor("op_2234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2234_cast_fp16 = einsum(equation = var_2234_equation_0, values = (var_2156_cast_fp16, var_2113_cast_fp16))[name = tensor("op_2234_cast_fp16")]; + tensor var_2235_to_fp16 = const()[name = tensor("op_2235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_135_cast_fp16 = mul(x = var_2234_cast_fp16, y = var_2235_to_fp16)[name = tensor("aw_135_cast_fp16")]; + tensor var_2238_equation_0 = const()[name = tensor("op_2238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2238_cast_fp16 = einsum(equation = var_2238_equation_0, values = (var_2160_cast_fp16, var_2117_cast_fp16))[name = tensor("op_2238_cast_fp16")]; + tensor var_2239_to_fp16 = const()[name = tensor("op_2239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_137_cast_fp16 = mul(x = var_2238_cast_fp16, y = var_2239_to_fp16)[name = tensor("aw_137_cast_fp16")]; + tensor var_2242_equation_0 = const()[name = tensor("op_2242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2242_cast_fp16 = einsum(equation = var_2242_equation_0, values = (var_2164_cast_fp16, var_2121_cast_fp16))[name = tensor("op_2242_cast_fp16")]; + tensor var_2243_to_fp16 = const()[name = tensor("op_2243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_139_cast_fp16 = mul(x = var_2242_cast_fp16, y = var_2243_to_fp16)[name = tensor("aw_139_cast_fp16")]; + tensor var_2245_cast_fp16 = softmax(axis = var_288, x = aw_121_cast_fp16)[name = tensor("op_2245_cast_fp16")]; + tensor var_2246_cast_fp16 = softmax(axis = var_288, x = aw_123_cast_fp16)[name = tensor("op_2246_cast_fp16")]; + tensor var_2247_cast_fp16 = softmax(axis = var_288, x = aw_125_cast_fp16)[name = tensor("op_2247_cast_fp16")]; + tensor var_2248_cast_fp16 = softmax(axis = var_288, x = aw_127_cast_fp16)[name = tensor("op_2248_cast_fp16")]; + tensor var_2249_cast_fp16 = softmax(axis = var_288, x = aw_129_cast_fp16)[name = tensor("op_2249_cast_fp16")]; + tensor var_2250_cast_fp16 = softmax(axis = var_288, x = aw_131_cast_fp16)[name = tensor("op_2250_cast_fp16")]; + tensor var_2251_cast_fp16 = softmax(axis = var_288, x = aw_133_cast_fp16)[name = tensor("op_2251_cast_fp16")]; + tensor var_2252_cast_fp16 = softmax(axis = var_288, x = aw_135_cast_fp16)[name = tensor("op_2252_cast_fp16")]; + tensor var_2253_cast_fp16 = softmax(axis = var_288, x = aw_137_cast_fp16)[name = tensor("op_2253_cast_fp16")]; + tensor var_2254_cast_fp16 = softmax(axis = var_288, x = aw_139_cast_fp16)[name = tensor("op_2254_cast_fp16")]; + tensor var_2256_equation_0 = const()[name = tensor("op_2256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2256_cast_fp16 = einsum(equation = var_2256_equation_0, values = (var_2166_cast_fp16, var_2245_cast_fp16))[name = tensor("op_2256_cast_fp16")]; + tensor var_2258_equation_0 = const()[name = tensor("op_2258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2258_cast_fp16 = einsum(equation = var_2258_equation_0, values = (var_2170_cast_fp16, var_2246_cast_fp16))[name = tensor("op_2258_cast_fp16")]; + tensor var_2260_equation_0 = const()[name = tensor("op_2260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2260_cast_fp16 = einsum(equation = var_2260_equation_0, values = (var_2174_cast_fp16, var_2247_cast_fp16))[name = tensor("op_2260_cast_fp16")]; + tensor var_2262_equation_0 = const()[name = tensor("op_2262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2262_cast_fp16 = einsum(equation = var_2262_equation_0, values = (var_2178_cast_fp16, var_2248_cast_fp16))[name = tensor("op_2262_cast_fp16")]; + tensor var_2264_equation_0 = const()[name = tensor("op_2264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2264_cast_fp16 = einsum(equation = var_2264_equation_0, values = (var_2182_cast_fp16, var_2249_cast_fp16))[name = tensor("op_2264_cast_fp16")]; + tensor var_2266_equation_0 = const()[name = tensor("op_2266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2266_cast_fp16 = einsum(equation = var_2266_equation_0, values = (var_2186_cast_fp16, var_2250_cast_fp16))[name = tensor("op_2266_cast_fp16")]; + tensor var_2268_equation_0 = const()[name = tensor("op_2268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2268_cast_fp16 = einsum(equation = var_2268_equation_0, values = (var_2190_cast_fp16, var_2251_cast_fp16))[name = tensor("op_2268_cast_fp16")]; + tensor var_2270_equation_0 = const()[name = tensor("op_2270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2270_cast_fp16 = einsum(equation = var_2270_equation_0, values = (var_2194_cast_fp16, var_2252_cast_fp16))[name = tensor("op_2270_cast_fp16")]; + tensor var_2272_equation_0 = const()[name = tensor("op_2272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2272_cast_fp16 = einsum(equation = var_2272_equation_0, values = (var_2198_cast_fp16, var_2253_cast_fp16))[name = tensor("op_2272_cast_fp16")]; + tensor var_2274_equation_0 = const()[name = tensor("op_2274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2274_cast_fp16 = einsum(equation = var_2274_equation_0, values = (var_2202_cast_fp16, var_2254_cast_fp16))[name = tensor("op_2274_cast_fp16")]; + tensor input_103_interleave_0 = const()[name = tensor("input_103_interleave_0"), val = tensor(false)]; + tensor input_103_cast_fp16 = concat(axis = var_288, interleave = input_103_interleave_0, values = (var_2256_cast_fp16, var_2258_cast_fp16, var_2260_cast_fp16, var_2262_cast_fp16, var_2264_cast_fp16, var_2266_cast_fp16, var_2268_cast_fp16, var_2270_cast_fp16, var_2272_cast_fp16, var_2274_cast_fp16))[name = tensor("input_103_cast_fp16")]; + tensor var_2284_pad_type_0 = const()[name = tensor("op_2284_pad_type_0"), val = tensor("valid")]; + tensor var_2284_strides_0 = const()[name = tensor("op_2284_strides_0"), val = tensor([1, 1])]; + tensor var_2284_pad_0 = const()[name = tensor("op_2284_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2284_dilations_0 = const()[name = tensor("op_2284_dilations_0"), val = tensor([1, 1])]; + tensor var_2284_groups_0 = const()[name = tensor("op_2284_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45075520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45382784))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45382976)))]; + tensor var_2284_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_2284_dilations_0, groups = var_2284_groups_0, pad = var_2284_pad_0, pad_type = var_2284_pad_type_0, strides = var_2284_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = tensor("op_2284_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = var_2284_cast_fp16, y = inputs_19_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; + tensor hidden_states_49_axes_0 = const()[name = tensor("hidden_states_49_axes_0"), val = tensor([1])]; + tensor hidden_states_49_gamma_0_to_fp16 = const()[name = tensor("hidden_states_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45384320)))]; + tensor hidden_states_49_beta_0_to_fp16 = const()[name = tensor("hidden_states_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45385664)))]; + tensor var_2294_to_fp16 = const()[name = tensor("op_2294_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = hidden_states_49_beta_0_to_fp16, epsilon = var_2294_to_fp16, gamma = hidden_states_49_gamma_0_to_fp16, x = inputs_21_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; + tensor q_15_pad_type_0 = const()[name = tensor("q_15_pad_type_0"), val = tensor("valid")]; + tensor q_15_strides_0 = const()[name = tensor("q_15_strides_0"), val = tensor([1, 1])]; + tensor q_15_pad_0 = const()[name = tensor("q_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_15_dilations_0 = const()[name = tensor("q_15_dilations_0"), val = tensor([1, 1])]; + tensor q_15_groups_0 = const()[name = tensor("q_15_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45387008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45694272))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_15_cast_fp16 = conv(dilations = q_15_dilations_0, groups = q_15_groups_0, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = q_15_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor("q_15_cast_fp16")]; + tensor k_29_pad_type_0 = const()[name = tensor("k_29_pad_type_0"), val = tensor("valid")]; + tensor k_29_strides_0 = const()[name = tensor("k_29_strides_0"), val = tensor([1, 1])]; + tensor k_29_pad_0 = const()[name = tensor("k_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_29_dilations_0 = const()[name = tensor("k_29_dilations_0"), val = tensor([1, 1])]; + tensor k_29_groups_0 = const()[name = tensor("k_29_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45694464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46677568))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor k_29_cast_fp16 = conv(dilations = k_29_dilations_0, groups = k_29_groups_0, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = k_29_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_29_cast_fp16")]; + tensor v_15_pad_type_0 = const()[name = tensor("v_15_pad_type_0"), val = tensor("valid")]; + tensor v_15_strides_0 = const()[name = tensor("v_15_strides_0"), val = tensor([1, 1])]; + tensor v_15_pad_0 = const()[name = tensor("v_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_15_dilations_0 = const()[name = tensor("v_15_dilations_0"), val = tensor([1, 1])]; + tensor v_15_groups_0 = const()[name = tensor("v_15_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46677760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47660864))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 2048, 1, 1])]; + tensor v_15_cast_fp16 = conv(dilations = v_15_dilations_0, groups = v_15_groups_0, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = v_15_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_15_cast_fp16")]; + tensor var_2327_begin_0 = const()[name = tensor("op_2327_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2327_end_0 = const()[name = tensor("op_2327_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_2327_end_mask_0 = const()[name = tensor("op_2327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2327_cast_fp16")]; + tensor var_2331_begin_0 = const()[name = tensor("op_2331_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2331_end_0 = const()[name = tensor("op_2331_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_2331_end_mask_0 = const()[name = tensor("op_2331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2331_cast_fp16 = slice_by_index(begin = var_2331_begin_0, end = var_2331_end_0, end_mask = var_2331_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2331_cast_fp16")]; + tensor var_2335_begin_0 = const()[name = tensor("op_2335_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2335_end_0 = const()[name = tensor("op_2335_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_2335_end_mask_0 = const()[name = tensor("op_2335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2335_cast_fp16 = slice_by_index(begin = var_2335_begin_0, end = var_2335_end_0, end_mask = var_2335_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2335_cast_fp16")]; + tensor var_2339_begin_0 = const()[name = tensor("op_2339_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2339_end_0 = const()[name = tensor("op_2339_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_2339_end_mask_0 = const()[name = tensor("op_2339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2339_cast_fp16 = slice_by_index(begin = var_2339_begin_0, end = var_2339_end_0, end_mask = var_2339_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2339_cast_fp16")]; + tensor var_2343_begin_0 = const()[name = tensor("op_2343_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2343_end_0 = const()[name = tensor("op_2343_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_2343_end_mask_0 = const()[name = tensor("op_2343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2343_cast_fp16 = slice_by_index(begin = var_2343_begin_0, end = var_2343_end_0, end_mask = var_2343_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2343_cast_fp16")]; + tensor var_2347_begin_0 = const()[name = tensor("op_2347_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2347_end_0 = const()[name = tensor("op_2347_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_2347_end_mask_0 = const()[name = tensor("op_2347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2347_cast_fp16 = slice_by_index(begin = var_2347_begin_0, end = var_2347_end_0, end_mask = var_2347_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2347_cast_fp16")]; + tensor var_2351_begin_0 = const()[name = tensor("op_2351_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2351_end_0 = const()[name = tensor("op_2351_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_2351_end_mask_0 = const()[name = tensor("op_2351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2351_cast_fp16 = slice_by_index(begin = var_2351_begin_0, end = var_2351_end_0, end_mask = var_2351_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2351_cast_fp16")]; + tensor var_2355_begin_0 = const()[name = tensor("op_2355_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2355_end_0 = const()[name = tensor("op_2355_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_2355_end_mask_0 = const()[name = tensor("op_2355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2355_cast_fp16")]; + tensor var_2359_begin_0 = const()[name = tensor("op_2359_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2359_end_0 = const()[name = tensor("op_2359_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_2359_end_mask_0 = const()[name = tensor("op_2359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2359_cast_fp16 = slice_by_index(begin = var_2359_begin_0, end = var_2359_end_0, end_mask = var_2359_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2359_cast_fp16")]; + tensor var_2363_begin_0 = const()[name = tensor("op_2363_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2363_end_0 = const()[name = tensor("op_2363_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_2363_end_mask_0 = const()[name = tensor("op_2363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2363_cast_fp16 = slice_by_index(begin = var_2363_begin_0, end = var_2363_end_0, end_mask = var_2363_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2363_cast_fp16")]; + tensor k_31_perm_0 = const()[name = tensor("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2370_begin_0 = const()[name = tensor("op_2370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2370_end_0 = const()[name = tensor("op_2370_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_2370_end_mask_0 = const()[name = tensor("op_2370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_31_cast_fp16 = transpose(perm = k_31_perm_0, x = k_29_cast_fp16)[name = tensor("transpose_60")]; + tensor var_2370_cast_fp16 = slice_by_index(begin = var_2370_begin_0, end = var_2370_end_0, end_mask = var_2370_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2370_cast_fp16")]; + tensor var_2374_begin_0 = const()[name = tensor("op_2374_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2374_end_0 = const()[name = tensor("op_2374_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_2374_end_mask_0 = const()[name = tensor("op_2374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2374_cast_fp16 = slice_by_index(begin = var_2374_begin_0, end = var_2374_end_0, end_mask = var_2374_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2374_cast_fp16")]; + tensor var_2378_begin_0 = const()[name = tensor("op_2378_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2378_end_0 = const()[name = tensor("op_2378_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_2378_end_mask_0 = const()[name = tensor("op_2378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2378_cast_fp16 = slice_by_index(begin = var_2378_begin_0, end = var_2378_end_0, end_mask = var_2378_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2378_cast_fp16")]; + tensor var_2382_begin_0 = const()[name = tensor("op_2382_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2382_end_0 = const()[name = tensor("op_2382_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_2382_end_mask_0 = const()[name = tensor("op_2382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2382_cast_fp16 = slice_by_index(begin = var_2382_begin_0, end = var_2382_end_0, end_mask = var_2382_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2382_cast_fp16")]; + tensor var_2386_begin_0 = const()[name = tensor("op_2386_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2386_end_0 = const()[name = tensor("op_2386_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_2386_end_mask_0 = const()[name = tensor("op_2386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2386_cast_fp16 = slice_by_index(begin = var_2386_begin_0, end = var_2386_end_0, end_mask = var_2386_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2386_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = tensor("op_2390_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2390_end_0 = const()[name = tensor("op_2390_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_2390_end_mask_0 = const()[name = tensor("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2390_cast_fp16")]; + tensor var_2394_begin_0 = const()[name = tensor("op_2394_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2394_end_0 = const()[name = tensor("op_2394_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_2394_end_mask_0 = const()[name = tensor("op_2394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2394_cast_fp16 = slice_by_index(begin = var_2394_begin_0, end = var_2394_end_0, end_mask = var_2394_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2394_cast_fp16")]; + tensor var_2398_begin_0 = const()[name = tensor("op_2398_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2398_end_0 = const()[name = tensor("op_2398_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_2398_end_mask_0 = const()[name = tensor("op_2398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2398_cast_fp16 = slice_by_index(begin = var_2398_begin_0, end = var_2398_end_0, end_mask = var_2398_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2398_cast_fp16")]; + tensor var_2402_begin_0 = const()[name = tensor("op_2402_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2402_end_0 = const()[name = tensor("op_2402_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_2402_end_mask_0 = const()[name = tensor("op_2402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2402_cast_fp16 = slice_by_index(begin = var_2402_begin_0, end = var_2402_end_0, end_mask = var_2402_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2402_cast_fp16")]; + tensor var_2406_begin_0 = const()[name = tensor("op_2406_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2406_end_0 = const()[name = tensor("op_2406_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_2406_end_mask_0 = const()[name = tensor("op_2406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2406_cast_fp16 = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2406_cast_fp16")]; + tensor var_2408_begin_0 = const()[name = tensor("op_2408_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2408_end_0 = const()[name = tensor("op_2408_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_2408_end_mask_0 = const()[name = tensor("op_2408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2408_cast_fp16 = slice_by_index(begin = var_2408_begin_0, end = var_2408_end_0, end_mask = var_2408_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2408_cast_fp16")]; + tensor var_2412_begin_0 = const()[name = tensor("op_2412_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2412_end_0 = const()[name = tensor("op_2412_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_2412_end_mask_0 = const()[name = tensor("op_2412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2412_cast_fp16 = slice_by_index(begin = var_2412_begin_0, end = var_2412_end_0, end_mask = var_2412_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2412_cast_fp16")]; + tensor var_2416_begin_0 = const()[name = tensor("op_2416_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2416_end_0 = const()[name = tensor("op_2416_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_2416_end_mask_0 = const()[name = tensor("op_2416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2416_cast_fp16 = slice_by_index(begin = var_2416_begin_0, end = var_2416_end_0, end_mask = var_2416_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2416_cast_fp16")]; + tensor var_2420_begin_0 = const()[name = tensor("op_2420_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2420_end_0 = const()[name = tensor("op_2420_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_2420_end_mask_0 = const()[name = tensor("op_2420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2420_cast_fp16 = slice_by_index(begin = var_2420_begin_0, end = var_2420_end_0, end_mask = var_2420_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2420_cast_fp16")]; + tensor var_2424_begin_0 = const()[name = tensor("op_2424_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2424_end_0 = const()[name = tensor("op_2424_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_2424_end_mask_0 = const()[name = tensor("op_2424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2424_cast_fp16 = slice_by_index(begin = var_2424_begin_0, end = var_2424_end_0, end_mask = var_2424_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2424_cast_fp16")]; + tensor var_2428_begin_0 = const()[name = tensor("op_2428_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2428_end_0 = const()[name = tensor("op_2428_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_2428_end_mask_0 = const()[name = tensor("op_2428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2428_cast_fp16 = slice_by_index(begin = var_2428_begin_0, end = var_2428_end_0, end_mask = var_2428_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2428_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = tensor("op_2432_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2432_end_0 = const()[name = tensor("op_2432_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_2432_end_mask_0 = const()[name = tensor("op_2432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2432_cast_fp16")]; + tensor var_2436_begin_0 = const()[name = tensor("op_2436_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2436_end_0 = const()[name = tensor("op_2436_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_2436_end_mask_0 = const()[name = tensor("op_2436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2436_cast_fp16 = slice_by_index(begin = var_2436_begin_0, end = var_2436_end_0, end_mask = var_2436_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2436_cast_fp16")]; + tensor var_2440_begin_0 = const()[name = tensor("op_2440_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2440_end_0 = const()[name = tensor("op_2440_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_2440_end_mask_0 = const()[name = tensor("op_2440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2440_cast_fp16 = slice_by_index(begin = var_2440_begin_0, end = var_2440_end_0, end_mask = var_2440_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2440_cast_fp16")]; + tensor var_2444_begin_0 = const()[name = tensor("op_2444_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2444_end_0 = const()[name = tensor("op_2444_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_2444_end_mask_0 = const()[name = tensor("op_2444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2444_cast_fp16 = slice_by_index(begin = var_2444_begin_0, end = var_2444_end_0, end_mask = var_2444_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2444_cast_fp16")]; + tensor var_2448_equation_0 = const()[name = tensor("op_2448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2448_cast_fp16 = einsum(equation = var_2448_equation_0, values = (var_2370_cast_fp16, var_2327_cast_fp16))[name = tensor("op_2448_cast_fp16")]; + tensor var_2449_to_fp16 = const()[name = tensor("op_2449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_141_cast_fp16 = mul(x = var_2448_cast_fp16, y = var_2449_to_fp16)[name = tensor("aw_141_cast_fp16")]; + tensor var_2452_equation_0 = const()[name = tensor("op_2452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2452_cast_fp16 = einsum(equation = var_2452_equation_0, values = (var_2374_cast_fp16, var_2331_cast_fp16))[name = tensor("op_2452_cast_fp16")]; + tensor var_2453_to_fp16 = const()[name = tensor("op_2453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_143_cast_fp16 = mul(x = var_2452_cast_fp16, y = var_2453_to_fp16)[name = tensor("aw_143_cast_fp16")]; + tensor var_2456_equation_0 = const()[name = tensor("op_2456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2456_cast_fp16 = einsum(equation = var_2456_equation_0, values = (var_2378_cast_fp16, var_2335_cast_fp16))[name = tensor("op_2456_cast_fp16")]; + tensor var_2457_to_fp16 = const()[name = tensor("op_2457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_145_cast_fp16 = mul(x = var_2456_cast_fp16, y = var_2457_to_fp16)[name = tensor("aw_145_cast_fp16")]; + tensor var_2460_equation_0 = const()[name = tensor("op_2460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2460_cast_fp16 = einsum(equation = var_2460_equation_0, values = (var_2382_cast_fp16, var_2339_cast_fp16))[name = tensor("op_2460_cast_fp16")]; + tensor var_2461_to_fp16 = const()[name = tensor("op_2461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_147_cast_fp16 = mul(x = var_2460_cast_fp16, y = var_2461_to_fp16)[name = tensor("aw_147_cast_fp16")]; + tensor var_2464_equation_0 = const()[name = tensor("op_2464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2464_cast_fp16 = einsum(equation = var_2464_equation_0, values = (var_2386_cast_fp16, var_2343_cast_fp16))[name = tensor("op_2464_cast_fp16")]; + tensor var_2465_to_fp16 = const()[name = tensor("op_2465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_149_cast_fp16 = mul(x = var_2464_cast_fp16, y = var_2465_to_fp16)[name = tensor("aw_149_cast_fp16")]; + tensor var_2468_equation_0 = const()[name = tensor("op_2468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2468_cast_fp16 = einsum(equation = var_2468_equation_0, values = (var_2390_cast_fp16, var_2347_cast_fp16))[name = tensor("op_2468_cast_fp16")]; + tensor var_2469_to_fp16 = const()[name = tensor("op_2469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_151_cast_fp16 = mul(x = var_2468_cast_fp16, y = var_2469_to_fp16)[name = tensor("aw_151_cast_fp16")]; + tensor var_2472_equation_0 = const()[name = tensor("op_2472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2472_cast_fp16 = einsum(equation = var_2472_equation_0, values = (var_2394_cast_fp16, var_2351_cast_fp16))[name = tensor("op_2472_cast_fp16")]; + tensor var_2473_to_fp16 = const()[name = tensor("op_2473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_153_cast_fp16 = mul(x = var_2472_cast_fp16, y = var_2473_to_fp16)[name = tensor("aw_153_cast_fp16")]; + tensor var_2476_equation_0 = const()[name = tensor("op_2476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2476_cast_fp16 = einsum(equation = var_2476_equation_0, values = (var_2398_cast_fp16, var_2355_cast_fp16))[name = tensor("op_2476_cast_fp16")]; + tensor var_2477_to_fp16 = const()[name = tensor("op_2477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_155_cast_fp16 = mul(x = var_2476_cast_fp16, y = var_2477_to_fp16)[name = tensor("aw_155_cast_fp16")]; + tensor var_2480_equation_0 = const()[name = tensor("op_2480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2480_cast_fp16 = einsum(equation = var_2480_equation_0, values = (var_2402_cast_fp16, var_2359_cast_fp16))[name = tensor("op_2480_cast_fp16")]; + tensor var_2481_to_fp16 = const()[name = tensor("op_2481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_157_cast_fp16 = mul(x = var_2480_cast_fp16, y = var_2481_to_fp16)[name = tensor("aw_157_cast_fp16")]; + tensor var_2484_equation_0 = const()[name = tensor("op_2484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2484_cast_fp16 = einsum(equation = var_2484_equation_0, values = (var_2406_cast_fp16, var_2363_cast_fp16))[name = tensor("op_2484_cast_fp16")]; + tensor var_2485_to_fp16 = const()[name = tensor("op_2485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_159_cast_fp16 = mul(x = var_2484_cast_fp16, y = var_2485_to_fp16)[name = tensor("aw_159_cast_fp16")]; + tensor var_2487_cast_fp16 = softmax(axis = var_288, x = aw_141_cast_fp16)[name = tensor("op_2487_cast_fp16")]; + tensor var_2488_cast_fp16 = softmax(axis = var_288, x = aw_143_cast_fp16)[name = tensor("op_2488_cast_fp16")]; + tensor var_2489_cast_fp16 = softmax(axis = var_288, x = aw_145_cast_fp16)[name = tensor("op_2489_cast_fp16")]; + tensor var_2490_cast_fp16 = softmax(axis = var_288, x = aw_147_cast_fp16)[name = tensor("op_2490_cast_fp16")]; + tensor var_2491_cast_fp16 = softmax(axis = var_288, x = aw_149_cast_fp16)[name = tensor("op_2491_cast_fp16")]; + tensor var_2492_cast_fp16 = softmax(axis = var_288, x = aw_151_cast_fp16)[name = tensor("op_2492_cast_fp16")]; + tensor var_2493_cast_fp16 = softmax(axis = var_288, x = aw_153_cast_fp16)[name = tensor("op_2493_cast_fp16")]; + tensor var_2494_cast_fp16 = softmax(axis = var_288, x = aw_155_cast_fp16)[name = tensor("op_2494_cast_fp16")]; + tensor var_2495_cast_fp16 = softmax(axis = var_288, x = aw_157_cast_fp16)[name = tensor("op_2495_cast_fp16")]; + tensor var_2496_cast_fp16 = softmax(axis = var_288, x = aw_159_cast_fp16)[name = tensor("op_2496_cast_fp16")]; + tensor var_2498_equation_0 = const()[name = tensor("op_2498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2498_cast_fp16 = einsum(equation = var_2498_equation_0, values = (var_2408_cast_fp16, var_2487_cast_fp16))[name = tensor("op_2498_cast_fp16")]; + tensor var_2500_equation_0 = const()[name = tensor("op_2500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2500_cast_fp16 = einsum(equation = var_2500_equation_0, values = (var_2412_cast_fp16, var_2488_cast_fp16))[name = tensor("op_2500_cast_fp16")]; + tensor var_2502_equation_0 = const()[name = tensor("op_2502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2502_cast_fp16 = einsum(equation = var_2502_equation_0, values = (var_2416_cast_fp16, var_2489_cast_fp16))[name = tensor("op_2502_cast_fp16")]; + tensor var_2504_equation_0 = const()[name = tensor("op_2504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2504_cast_fp16 = einsum(equation = var_2504_equation_0, values = (var_2420_cast_fp16, var_2490_cast_fp16))[name = tensor("op_2504_cast_fp16")]; + tensor var_2506_equation_0 = const()[name = tensor("op_2506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2506_cast_fp16 = einsum(equation = var_2506_equation_0, values = (var_2424_cast_fp16, var_2491_cast_fp16))[name = tensor("op_2506_cast_fp16")]; + tensor var_2508_equation_0 = const()[name = tensor("op_2508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2508_cast_fp16 = einsum(equation = var_2508_equation_0, values = (var_2428_cast_fp16, var_2492_cast_fp16))[name = tensor("op_2508_cast_fp16")]; + tensor var_2510_equation_0 = const()[name = tensor("op_2510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2510_cast_fp16 = einsum(equation = var_2510_equation_0, values = (var_2432_cast_fp16, var_2493_cast_fp16))[name = tensor("op_2510_cast_fp16")]; + tensor var_2512_equation_0 = const()[name = tensor("op_2512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2512_cast_fp16 = einsum(equation = var_2512_equation_0, values = (var_2436_cast_fp16, var_2494_cast_fp16))[name = tensor("op_2512_cast_fp16")]; + tensor var_2514_equation_0 = const()[name = tensor("op_2514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2514_cast_fp16 = einsum(equation = var_2514_equation_0, values = (var_2440_cast_fp16, var_2495_cast_fp16))[name = tensor("op_2514_cast_fp16")]; + tensor var_2516_equation_0 = const()[name = tensor("op_2516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2516_cast_fp16 = einsum(equation = var_2516_equation_0, values = (var_2444_cast_fp16, var_2496_cast_fp16))[name = tensor("op_2516_cast_fp16")]; + tensor input_105_interleave_0 = const()[name = tensor("input_105_interleave_0"), val = tensor(false)]; + tensor input_105_cast_fp16 = concat(axis = var_288, interleave = input_105_interleave_0, values = (var_2498_cast_fp16, var_2500_cast_fp16, var_2502_cast_fp16, var_2504_cast_fp16, var_2506_cast_fp16, var_2508_cast_fp16, var_2510_cast_fp16, var_2512_cast_fp16, var_2514_cast_fp16, var_2516_cast_fp16))[name = tensor("input_105_cast_fp16")]; + tensor var_2526_pad_type_0 = const()[name = tensor("op_2526_pad_type_0"), val = tensor("valid")]; + tensor var_2526_strides_0 = const()[name = tensor("op_2526_strides_0"), val = tensor([1, 1])]; + tensor var_2526_pad_0 = const()[name = tensor("op_2526_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2526_dilations_0 = const()[name = tensor("op_2526_dilations_0"), val = tensor([1, 1])]; + tensor var_2526_groups_0 = const()[name = tensor("op_2526_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47661056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47968320))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47968512)))]; + tensor var_2526_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_2526_dilations_0, groups = var_2526_groups_0, pad = var_2526_pad_0, pad_type = var_2526_pad_type_0, strides = var_2526_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor("op_2526_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = var_2526_cast_fp16, y = inputs_21_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; + tensor input_107_axes_0 = const()[name = tensor("input_107_axes_0"), val = tensor([1])]; + tensor input_107_gamma_0_to_fp16 = const()[name = tensor("input_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47969856)))]; + tensor input_107_beta_0_to_fp16 = const()[name = tensor("input_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47971200)))]; + tensor var_2536_to_fp16 = const()[name = tensor("op_2536_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_107_cast_fp16 = layer_norm(axes = input_107_axes_0, beta = input_107_beta_0_to_fp16, epsilon = var_2536_to_fp16, gamma = input_107_gamma_0_to_fp16, x = inputs_23_cast_fp16)[name = tensor("input_107_cast_fp16")]; + tensor var_2556_pad_type_0 = const()[name = tensor("op_2556_pad_type_0"), val = tensor("valid")]; + tensor var_2556_strides_0 = const()[name = tensor("op_2556_strides_0"), val = tensor([1, 1])]; + tensor var_2556_pad_0 = const()[name = tensor("op_2556_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2556_dilations_0 = const()[name = tensor("op_2556_dilations_0"), val = tensor([1, 1])]; + tensor var_2556_groups_0 = const()[name = tensor("op_2556_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47972544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50430208))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50430400)))]; + tensor var_2556_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_2556_dilations_0, groups = var_2556_groups_0, pad = var_2556_pad_0, pad_type = var_2556_pad_type_0, strides = var_2556_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_107_cast_fp16)[name = tensor("op_2556_cast_fp16")]; + tensor var_2557_split_sizes_0 = const()[name = tensor("op_2557_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_2557_axis_0 = const()[name = tensor("op_2557_axis_0"), val = tensor(1)]; + tensor var_2557_cast_fp16_0, tensor var_2557_cast_fp16_1 = split(axis = var_2557_axis_0, split_sizes = var_2557_split_sizes_0, x = var_2556_cast_fp16)[name = tensor("op_2557_cast_fp16")]; + tensor var_2559_mode_0 = const()[name = tensor("op_2559_mode_0"), val = tensor("EXACT")]; + tensor var_2559_cast_fp16 = gelu(mode = var_2559_mode_0, x = var_2557_cast_fp16_1)[name = tensor("op_2559_cast_fp16")]; + tensor input_109_cast_fp16 = mul(x = var_2557_cast_fp16_0, y = var_2559_cast_fp16)[name = tensor("input_109_cast_fp16")]; + tensor var_2567_pad_type_0 = const()[name = tensor("op_2567_pad_type_0"), val = tensor("valid")]; + tensor var_2567_strides_0 = const()[name = tensor("op_2567_strides_0"), val = tensor([1, 1])]; + tensor var_2567_pad_0 = const()[name = tensor("op_2567_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2567_dilations_0 = const()[name = tensor("op_2567_dilations_0"), val = tensor([1, 1])]; + tensor var_2567_groups_0 = const()[name = tensor("op_2567_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50440704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51669568))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51669760)))]; + tensor var_2567_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_2567_dilations_0, groups = var_2567_groups_0, pad = var_2567_pad_0, pad_type = var_2567_pad_type_0, strides = var_2567_strides_0, weight = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor("op_2567_cast_fp16")]; + tensor hidden_states_53_cast_fp16 = add(x = var_2567_cast_fp16, y = inputs_23_cast_fp16)[name = tensor("hidden_states_53_cast_fp16")]; + tensor var_2569 = const()[name = tensor("op_2569"), val = tensor([2, 640, 64, 64])]; + tensor input_111_cast_fp16 = reshape(shape = var_2569, x = hidden_states_53_cast_fp16)[name = tensor("input_111_cast_fp16")]; + tensor hidden_states_55_pad_type_0 = const()[name = tensor("hidden_states_55_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_55_strides_0 = const()[name = tensor("hidden_states_55_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_0 = const()[name = tensor("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_55_dilations_0 = const()[name = tensor("hidden_states_55_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_55_groups_0 = const()[name = tensor("hidden_states_55_groups_0"), val = tensor(1)]; + tensor down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51671104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51978368))), name = tensor("down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51978560)))]; + tensor hidden_states_55_cast_fp16 = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized, x = input_111_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; + tensor input_113_cast_fp16_1 = add(x = hidden_states_55_cast_fp16, y = hidden_states_37_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor input_115_pad_type_0 = const()[name = tensor("input_115_pad_type_0"), val = tensor("custom")]; + tensor input_115_pad_0 = const()[name = tensor("input_115_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_115_strides_0 = const()[name = tensor("input_115_strides_0"), val = tensor([2, 2])]; + tensor input_115_dilations_0 = const()[name = tensor("input_115_dilations_0"), val = tensor([1, 1])]; + tensor input_115_groups_0 = const()[name = tensor("input_115_groups_0"), val = tensor(1)]; + tensor down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51979904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54744768))), name = tensor("down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_1_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54744960)))]; + tensor input_115_cast_fp16_1 = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = input_115_dilations_0, groups = input_115_groups_0, pad = input_115_pad_0, pad_type = input_115_pad_type_0, strides = input_115_strides_0, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized, x = input_113_cast_fp16_1)[name = tensor("input_115_cast_fp16")]; + tensor var_2624 = const()[name = tensor("op_2624"), val = tensor(1)]; + tensor reshape_40_shape_0 = const()[name = tensor("reshape_40_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_40_cast_fp16 = reshape(shape = reshape_40_shape_0, x = input_115_cast_fp16_1)[name = tensor("reshape_40_cast_fp16")]; + tensor reduce_mean_30_axes_0 = const()[name = tensor("reduce_mean_30_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_30_keep_dims_0 = const()[name = tensor("reduce_mean_30_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_30_cast_fp16 = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast_fp16)[name = tensor("reduce_mean_30_cast_fp16")]; + tensor sub_20_cast_fp16 = sub(x = reshape_40_cast_fp16, y = reduce_mean_30_cast_fp16)[name = tensor("sub_20_cast_fp16")]; + tensor square_10_cast_fp16 = square(x = sub_20_cast_fp16)[name = tensor("square_10_cast_fp16")]; + tensor reduce_mean_32_axes_0 = const()[name = tensor("reduce_mean_32_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_32_keep_dims_0 = const()[name = tensor("reduce_mean_32_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_32_cast_fp16 = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast_fp16)[name = tensor("reduce_mean_32_cast_fp16")]; + tensor add_20_y_0_to_fp16 = const()[name = tensor("add_20_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_20_cast_fp16 = add(x = reduce_mean_32_cast_fp16, y = add_20_y_0_to_fp16)[name = tensor("add_20_cast_fp16")]; + tensor sqrt_10_cast_fp16 = sqrt(x = add_20_cast_fp16)[name = tensor("sqrt_10_cast_fp16")]; + tensor real_div_10_cast_fp16 = real_div(x = sub_20_cast_fp16, y = sqrt_10_cast_fp16)[name = tensor("real_div_10_cast_fp16")]; + tensor reshape_41_shape_0 = const()[name = tensor("reshape_41_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_41_cast_fp16 = reshape(shape = reshape_41_shape_0, x = real_div_10_cast_fp16)[name = tensor("reshape_41_cast_fp16")]; + tensor add_21_gamma_0_to_fp16 = const()[name = tensor("add_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54746304)))]; + tensor add_21_beta_0_to_fp16 = const()[name = tensor("add_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54747648)))]; + tensor add_21_epsilon_0_to_fp16 = const()[name = tensor("add_21_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_21_cast_fp16 = batch_norm(beta = add_21_beta_0_to_fp16, epsilon = add_21_epsilon_0_to_fp16, gamma = add_21_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_41_cast_fp16)[name = tensor("add_21_cast_fp16")]; + tensor input_119_cast_fp16 = silu(x = add_21_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor hidden_states_57_pad_type_0 = const()[name = tensor("hidden_states_57_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_57_pad_0 = const()[name = tensor("hidden_states_57_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_57_strides_0 = const()[name = tensor("hidden_states_57_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_57_dilations_0 = const()[name = tensor("hidden_states_57_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_57_groups_0 = const()[name = tensor("hidden_states_57_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54748992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60278656))), name = tensor("down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 640, 3, 3])]; + tensor down_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60278848)))]; + tensor hidden_states_57_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = tensor("hidden_states_57_cast_fp16")]; + tensor temb_9_pad_type_0 = const()[name = tensor("temb_9_pad_type_0"), val = tensor("valid")]; + tensor temb_9_strides_0 = const()[name = tensor("temb_9_strides_0"), val = tensor([1, 1])]; + tensor temb_9_pad_0 = const()[name = tensor("temb_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_9_dilations_0 = const()[name = tensor("temb_9_dilations_0"), val = tensor([1, 1])]; + tensor temb_9_groups_0 = const()[name = tensor("temb_9_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60281472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61510336))), name = tensor("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61510528)))]; + tensor temb_9_cast_fp16 = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_9_dilations_0, groups = temb_9_groups_0, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = temb_9_strides_0, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_9_cast_fp16")]; + tensor input_123_cast_fp16 = add(x = hidden_states_57_cast_fp16, y = temb_9_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor reshape_44_shape_0 = const()[name = tensor("reshape_44_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_44_cast_fp16 = reshape(shape = reshape_44_shape_0, x = input_123_cast_fp16)[name = tensor("reshape_44_cast_fp16")]; + tensor reduce_mean_33_axes_0 = const()[name = tensor("reduce_mean_33_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_33_keep_dims_0 = const()[name = tensor("reduce_mean_33_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_33_cast_fp16 = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast_fp16)[name = tensor("reduce_mean_33_cast_fp16")]; + tensor sub_22_cast_fp16 = sub(x = reshape_44_cast_fp16, y = reduce_mean_33_cast_fp16)[name = tensor("sub_22_cast_fp16")]; + tensor square_11_cast_fp16 = square(x = sub_22_cast_fp16)[name = tensor("square_11_cast_fp16")]; + tensor reduce_mean_35_axes_0 = const()[name = tensor("reduce_mean_35_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_35_keep_dims_0 = const()[name = tensor("reduce_mean_35_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_35_cast_fp16 = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast_fp16)[name = tensor("reduce_mean_35_cast_fp16")]; + tensor add_22_y_0_to_fp16 = const()[name = tensor("add_22_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_22_cast_fp16 = add(x = reduce_mean_35_cast_fp16, y = add_22_y_0_to_fp16)[name = tensor("add_22_cast_fp16")]; + tensor sqrt_11_cast_fp16 = sqrt(x = add_22_cast_fp16)[name = tensor("sqrt_11_cast_fp16")]; + tensor real_div_11_cast_fp16 = real_div(x = sub_22_cast_fp16, y = sqrt_11_cast_fp16)[name = tensor("real_div_11_cast_fp16")]; + tensor reshape_45_shape_0 = const()[name = tensor("reshape_45_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_45_cast_fp16 = reshape(shape = reshape_45_shape_0, x = real_div_11_cast_fp16)[name = tensor("reshape_45_cast_fp16")]; + tensor add_23_mean_0_to_fp16 = const()[name = tensor("add_23_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61513152)))]; + tensor add_23_variance_0_to_fp16 = const()[name = tensor("add_23_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61515776)))]; + tensor add_23_gamma_0_to_fp16 = const()[name = tensor("add_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61518400)))]; + tensor add_23_beta_0_to_fp16 = const()[name = tensor("add_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61521024)))]; + tensor add_23_epsilon_0_to_fp16 = const()[name = tensor("add_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_23_cast_fp16 = batch_norm(beta = add_23_beta_0_to_fp16, epsilon = add_23_epsilon_0_to_fp16, gamma = add_23_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_45_cast_fp16)[name = tensor("add_23_cast_fp16")]; + tensor input_127_cast_fp16 = silu(x = add_23_cast_fp16)[name = tensor("input_127_cast_fp16")]; + tensor hidden_states_59_pad_type_0 = const()[name = tensor("hidden_states_59_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_59_pad_0 = const()[name = tensor("hidden_states_59_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_59_strides_0 = const()[name = tensor("hidden_states_59_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_59_dilations_0 = const()[name = tensor("hidden_states_59_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_59_groups_0 = const()[name = tensor("hidden_states_59_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61523648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72582912))), name = tensor("down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72583104)))]; + tensor hidden_states_59_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = tensor("hidden_states_59_cast_fp16")]; + tensor x_3_pad_type_0 = const()[name = tensor("x_3_pad_type_0"), val = tensor("valid")]; + tensor x_3_strides_0 = const()[name = tensor("x_3_strides_0"), val = tensor([1, 1])]; + tensor x_3_pad_0 = const()[name = tensor("x_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_3_dilations_0 = const()[name = tensor("x_3_dilations_0"), val = tensor([1, 1])]; + tensor x_3_groups_0 = const()[name = tensor("x_3_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72585728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73200192))), name = tensor("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 640, 1, 1])]; + tensor down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73200384)))]; + tensor x_3_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_115_cast_fp16_1)[name = tensor("x_3_cast_fp16")]; + tensor hidden_states_61_cast_fp16 = add(x = x_3_cast_fp16, y = hidden_states_59_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; + tensor reshape_48_shape_0 = const()[name = tensor("reshape_48_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_48_cast_fp16 = reshape(shape = reshape_48_shape_0, x = hidden_states_61_cast_fp16)[name = tensor("reshape_48_cast_fp16")]; + tensor reduce_mean_36_axes_0 = const()[name = tensor("reduce_mean_36_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_36_keep_dims_0 = const()[name = tensor("reduce_mean_36_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_36_cast_fp16 = reduce_mean(axes = reduce_mean_36_axes_0, keep_dims = reduce_mean_36_keep_dims_0, x = reshape_48_cast_fp16)[name = tensor("reduce_mean_36_cast_fp16")]; + tensor sub_24_cast_fp16 = sub(x = reshape_48_cast_fp16, y = reduce_mean_36_cast_fp16)[name = tensor("sub_24_cast_fp16")]; + tensor square_12_cast_fp16 = square(x = sub_24_cast_fp16)[name = tensor("square_12_cast_fp16")]; + tensor reduce_mean_38_axes_0 = const()[name = tensor("reduce_mean_38_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_38_keep_dims_0 = const()[name = tensor("reduce_mean_38_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_38_cast_fp16 = reduce_mean(axes = reduce_mean_38_axes_0, keep_dims = reduce_mean_38_keep_dims_0, x = square_12_cast_fp16)[name = tensor("reduce_mean_38_cast_fp16")]; + tensor add_24_y_0_to_fp16 = const()[name = tensor("add_24_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_24_cast_fp16 = add(x = reduce_mean_38_cast_fp16, y = add_24_y_0_to_fp16)[name = tensor("add_24_cast_fp16")]; + tensor sqrt_12_cast_fp16 = sqrt(x = add_24_cast_fp16)[name = tensor("sqrt_12_cast_fp16")]; + tensor real_div_12_cast_fp16 = real_div(x = sub_24_cast_fp16, y = sqrt_12_cast_fp16)[name = tensor("real_div_12_cast_fp16")]; + tensor reshape_49_shape_0 = const()[name = tensor("reshape_49_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_49_cast_fp16 = reshape(shape = reshape_49_shape_0, x = real_div_12_cast_fp16)[name = tensor("reshape_49_cast_fp16")]; + tensor add_25_gamma_0_to_fp16 = const()[name = tensor("add_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73203008)))]; + tensor add_25_beta_0_to_fp16 = const()[name = tensor("add_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73205632)))]; + tensor add_25_epsilon_0_to_fp16 = const()[name = tensor("add_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_25_cast_fp16 = batch_norm(beta = add_25_beta_0_to_fp16, epsilon = add_25_epsilon_0_to_fp16, gamma = add_25_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_49_cast_fp16)[name = tensor("add_25_cast_fp16")]; + tensor hidden_states_63_pad_type_0 = const()[name = tensor("hidden_states_63_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_63_strides_0 = const()[name = tensor("hidden_states_63_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_63_pad_0 = const()[name = tensor("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_63_dilations_0 = const()[name = tensor("hidden_states_63_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_63_groups_0 = const()[name = tensor("hidden_states_63_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73208256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74437120))), name = tensor("down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74437312)))]; + tensor hidden_states_63_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_63_dilations_0, groups = hidden_states_63_groups_0, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = hidden_states_63_strides_0, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized, x = add_25_cast_fp16)[name = tensor("hidden_states_63_cast_fp16")]; + tensor var_2719 = const()[name = tensor("op_2719"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_25_cast_fp16 = reshape(shape = var_2719, x = hidden_states_63_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; + tensor hidden_states_65_axes_0 = const()[name = tensor("hidden_states_65_axes_0"), val = tensor([1])]; + tensor hidden_states_65_gamma_0_to_fp16 = const()[name = tensor("hidden_states_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74439936)))]; + tensor hidden_states_65_beta_0_to_fp16 = const()[name = tensor("hidden_states_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74442560)))]; + tensor var_2735_to_fp16 = const()[name = tensor("op_2735_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_65_cast_fp16 = layer_norm(axes = hidden_states_65_axes_0, beta = hidden_states_65_beta_0_to_fp16, epsilon = var_2735_to_fp16, gamma = hidden_states_65_gamma_0_to_fp16, x = inputs_25_cast_fp16)[name = tensor("hidden_states_65_cast_fp16")]; + tensor q_17_pad_type_0 = const()[name = tensor("q_17_pad_type_0"), val = tensor("valid")]; + tensor q_17_strides_0 = const()[name = tensor("q_17_strides_0"), val = tensor([1, 1])]; + tensor q_17_pad_0 = const()[name = tensor("q_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_17_dilations_0 = const()[name = tensor("q_17_dilations_0"), val = tensor([1, 1])]; + tensor q_17_groups_0 = const()[name = tensor("q_17_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74445184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75674048))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_17_cast_fp16 = conv(dilations = q_17_dilations_0, groups = q_17_groups_0, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = q_17_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_65_cast_fp16)[name = tensor("q_17_cast_fp16")]; + tensor k_33_pad_type_0 = const()[name = tensor("k_33_pad_type_0"), val = tensor("valid")]; + tensor k_33_strides_0 = const()[name = tensor("k_33_strides_0"), val = tensor([1, 1])]; + tensor k_33_pad_0 = const()[name = tensor("k_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_33_dilations_0 = const()[name = tensor("k_33_dilations_0"), val = tensor([1, 1])]; + tensor k_33_groups_0 = const()[name = tensor("k_33_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75674240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76903104))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_33_cast_fp16 = conv(dilations = k_33_dilations_0, groups = k_33_groups_0, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = k_33_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_65_cast_fp16)[name = tensor("k_33_cast_fp16")]; + tensor v_17_pad_type_0 = const()[name = tensor("v_17_pad_type_0"), val = tensor("valid")]; + tensor v_17_strides_0 = const()[name = tensor("v_17_strides_0"), val = tensor([1, 1])]; + tensor v_17_pad_0 = const()[name = tensor("v_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_17_dilations_0 = const()[name = tensor("v_17_dilations_0"), val = tensor([1, 1])]; + tensor v_17_groups_0 = const()[name = tensor("v_17_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76903296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78132160))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_17_cast_fp16 = conv(dilations = v_17_dilations_0, groups = v_17_groups_0, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = v_17_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_65_cast_fp16)[name = tensor("v_17_cast_fp16")]; + tensor var_2768_begin_0 = const()[name = tensor("op_2768_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2768_end_0 = const()[name = tensor("op_2768_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2768_end_mask_0 = const()[name = tensor("op_2768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2768_cast_fp16 = slice_by_index(begin = var_2768_begin_0, end = var_2768_end_0, end_mask = var_2768_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2768_cast_fp16")]; + tensor var_2772_begin_0 = const()[name = tensor("op_2772_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2772_end_0 = const()[name = tensor("op_2772_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_2772_end_mask_0 = const()[name = tensor("op_2772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2772_cast_fp16 = slice_by_index(begin = var_2772_begin_0, end = var_2772_end_0, end_mask = var_2772_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2772_cast_fp16")]; + tensor var_2776_begin_0 = const()[name = tensor("op_2776_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2776_end_0 = const()[name = tensor("op_2776_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_2776_end_mask_0 = const()[name = tensor("op_2776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2776_cast_fp16 = slice_by_index(begin = var_2776_begin_0, end = var_2776_end_0, end_mask = var_2776_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2776_cast_fp16")]; + tensor var_2780_begin_0 = const()[name = tensor("op_2780_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2780_end_0 = const()[name = tensor("op_2780_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_2780_end_mask_0 = const()[name = tensor("op_2780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2780_cast_fp16 = slice_by_index(begin = var_2780_begin_0, end = var_2780_end_0, end_mask = var_2780_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2780_cast_fp16")]; + tensor var_2784_begin_0 = const()[name = tensor("op_2784_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2784_end_0 = const()[name = tensor("op_2784_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_2784_end_mask_0 = const()[name = tensor("op_2784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2784_cast_fp16 = slice_by_index(begin = var_2784_begin_0, end = var_2784_end_0, end_mask = var_2784_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2784_cast_fp16")]; + tensor var_2788_begin_0 = const()[name = tensor("op_2788_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2788_end_0 = const()[name = tensor("op_2788_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_2788_end_mask_0 = const()[name = tensor("op_2788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2788_cast_fp16 = slice_by_index(begin = var_2788_begin_0, end = var_2788_end_0, end_mask = var_2788_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2788_cast_fp16")]; + tensor var_2792_begin_0 = const()[name = tensor("op_2792_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2792_end_0 = const()[name = tensor("op_2792_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_2792_end_mask_0 = const()[name = tensor("op_2792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2792_cast_fp16 = slice_by_index(begin = var_2792_begin_0, end = var_2792_end_0, end_mask = var_2792_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2792_cast_fp16")]; + tensor var_2796_begin_0 = const()[name = tensor("op_2796_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2796_end_0 = const()[name = tensor("op_2796_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_2796_end_mask_0 = const()[name = tensor("op_2796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2796_cast_fp16 = slice_by_index(begin = var_2796_begin_0, end = var_2796_end_0, end_mask = var_2796_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2796_cast_fp16")]; + tensor var_2800_begin_0 = const()[name = tensor("op_2800_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2800_end_0 = const()[name = tensor("op_2800_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_2800_end_mask_0 = const()[name = tensor("op_2800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2800_cast_fp16 = slice_by_index(begin = var_2800_begin_0, end = var_2800_end_0, end_mask = var_2800_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2800_cast_fp16")]; + tensor var_2804_begin_0 = const()[name = tensor("op_2804_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2804_end_0 = const()[name = tensor("op_2804_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_2804_end_mask_0 = const()[name = tensor("op_2804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2804_cast_fp16 = slice_by_index(begin = var_2804_begin_0, end = var_2804_end_0, end_mask = var_2804_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2804_cast_fp16")]; + tensor var_2808_begin_0 = const()[name = tensor("op_2808_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2808_end_0 = const()[name = tensor("op_2808_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_2808_end_mask_0 = const()[name = tensor("op_2808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2808_cast_fp16 = slice_by_index(begin = var_2808_begin_0, end = var_2808_end_0, end_mask = var_2808_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2808_cast_fp16")]; + tensor var_2812_begin_0 = const()[name = tensor("op_2812_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2812_end_0 = const()[name = tensor("op_2812_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_2812_end_mask_0 = const()[name = tensor("op_2812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2812_cast_fp16 = slice_by_index(begin = var_2812_begin_0, end = var_2812_end_0, end_mask = var_2812_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2812_cast_fp16")]; + tensor var_2816_begin_0 = const()[name = tensor("op_2816_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_2816_end_0 = const()[name = tensor("op_2816_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_2816_end_mask_0 = const()[name = tensor("op_2816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2816_cast_fp16 = slice_by_index(begin = var_2816_begin_0, end = var_2816_end_0, end_mask = var_2816_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2816_cast_fp16")]; + tensor var_2820_begin_0 = const()[name = tensor("op_2820_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_2820_end_0 = const()[name = tensor("op_2820_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_2820_end_mask_0 = const()[name = tensor("op_2820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2820_cast_fp16 = slice_by_index(begin = var_2820_begin_0, end = var_2820_end_0, end_mask = var_2820_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2820_cast_fp16")]; + tensor var_2824_begin_0 = const()[name = tensor("op_2824_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_2824_end_0 = const()[name = tensor("op_2824_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_2824_end_mask_0 = const()[name = tensor("op_2824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2824_cast_fp16 = slice_by_index(begin = var_2824_begin_0, end = var_2824_end_0, end_mask = var_2824_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2824_cast_fp16")]; + tensor var_2828_begin_0 = const()[name = tensor("op_2828_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2828_end_0 = const()[name = tensor("op_2828_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_2828_end_mask_0 = const()[name = tensor("op_2828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2828_cast_fp16 = slice_by_index(begin = var_2828_begin_0, end = var_2828_end_0, end_mask = var_2828_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2828_cast_fp16")]; + tensor var_2832_begin_0 = const()[name = tensor("op_2832_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2832_end_0 = const()[name = tensor("op_2832_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_2832_end_mask_0 = const()[name = tensor("op_2832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2832_cast_fp16 = slice_by_index(begin = var_2832_begin_0, end = var_2832_end_0, end_mask = var_2832_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2832_cast_fp16")]; + tensor var_2836_begin_0 = const()[name = tensor("op_2836_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2836_end_0 = const()[name = tensor("op_2836_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_2836_end_mask_0 = const()[name = tensor("op_2836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2836_cast_fp16 = slice_by_index(begin = var_2836_begin_0, end = var_2836_end_0, end_mask = var_2836_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2836_cast_fp16")]; + tensor var_2840_begin_0 = const()[name = tensor("op_2840_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_2840_end_0 = const()[name = tensor("op_2840_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_2840_end_mask_0 = const()[name = tensor("op_2840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2840_cast_fp16 = slice_by_index(begin = var_2840_begin_0, end = var_2840_end_0, end_mask = var_2840_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2840_cast_fp16")]; + tensor var_2844_begin_0 = const()[name = tensor("op_2844_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_2844_end_0 = const()[name = tensor("op_2844_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_2844_end_mask_0 = const()[name = tensor("op_2844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2844_cast_fp16 = slice_by_index(begin = var_2844_begin_0, end = var_2844_end_0, end_mask = var_2844_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2844_cast_fp16")]; + tensor k_35_perm_0 = const()[name = tensor("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2851_begin_0 = const()[name = tensor("op_2851_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2851_end_0 = const()[name = tensor("op_2851_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_2851_end_mask_0 = const()[name = tensor("op_2851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_35_cast_fp16 = transpose(perm = k_35_perm_0, x = k_33_cast_fp16)[name = tensor("transpose_59")]; + tensor var_2851_cast_fp16 = slice_by_index(begin = var_2851_begin_0, end = var_2851_end_0, end_mask = var_2851_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2851_cast_fp16")]; + tensor var_2855_begin_0 = const()[name = tensor("op_2855_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2855_end_0 = const()[name = tensor("op_2855_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_2855_end_mask_0 = const()[name = tensor("op_2855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2855_cast_fp16 = slice_by_index(begin = var_2855_begin_0, end = var_2855_end_0, end_mask = var_2855_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2855_cast_fp16")]; + tensor var_2859_begin_0 = const()[name = tensor("op_2859_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2859_end_0 = const()[name = tensor("op_2859_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_2859_end_mask_0 = const()[name = tensor("op_2859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2859_cast_fp16 = slice_by_index(begin = var_2859_begin_0, end = var_2859_end_0, end_mask = var_2859_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2859_cast_fp16")]; + tensor var_2863_begin_0 = const()[name = tensor("op_2863_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2863_end_0 = const()[name = tensor("op_2863_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_2863_end_mask_0 = const()[name = tensor("op_2863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2863_cast_fp16 = slice_by_index(begin = var_2863_begin_0, end = var_2863_end_0, end_mask = var_2863_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2863_cast_fp16")]; + tensor var_2867_begin_0 = const()[name = tensor("op_2867_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2867_end_0 = const()[name = tensor("op_2867_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_2867_end_mask_0 = const()[name = tensor("op_2867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2867_cast_fp16 = slice_by_index(begin = var_2867_begin_0, end = var_2867_end_0, end_mask = var_2867_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2867_cast_fp16")]; + tensor var_2871_begin_0 = const()[name = tensor("op_2871_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2871_end_0 = const()[name = tensor("op_2871_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_2871_end_mask_0 = const()[name = tensor("op_2871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2871_cast_fp16 = slice_by_index(begin = var_2871_begin_0, end = var_2871_end_0, end_mask = var_2871_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2871_cast_fp16")]; + tensor var_2875_begin_0 = const()[name = tensor("op_2875_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2875_end_0 = const()[name = tensor("op_2875_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_2875_end_mask_0 = const()[name = tensor("op_2875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2875_cast_fp16 = slice_by_index(begin = var_2875_begin_0, end = var_2875_end_0, end_mask = var_2875_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2875_cast_fp16")]; + tensor var_2879_begin_0 = const()[name = tensor("op_2879_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2879_end_0 = const()[name = tensor("op_2879_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_2879_end_mask_0 = const()[name = tensor("op_2879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2879_cast_fp16 = slice_by_index(begin = var_2879_begin_0, end = var_2879_end_0, end_mask = var_2879_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2879_cast_fp16")]; + tensor var_2883_begin_0 = const()[name = tensor("op_2883_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2883_end_0 = const()[name = tensor("op_2883_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_2883_end_mask_0 = const()[name = tensor("op_2883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2883_cast_fp16 = slice_by_index(begin = var_2883_begin_0, end = var_2883_end_0, end_mask = var_2883_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2883_cast_fp16")]; + tensor var_2887_begin_0 = const()[name = tensor("op_2887_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2887_end_0 = const()[name = tensor("op_2887_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_2887_end_mask_0 = const()[name = tensor("op_2887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2887_cast_fp16 = slice_by_index(begin = var_2887_begin_0, end = var_2887_end_0, end_mask = var_2887_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2887_cast_fp16")]; + tensor var_2891_begin_0 = const()[name = tensor("op_2891_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_2891_end_0 = const()[name = tensor("op_2891_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_2891_end_mask_0 = const()[name = tensor("op_2891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2891_cast_fp16 = slice_by_index(begin = var_2891_begin_0, end = var_2891_end_0, end_mask = var_2891_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2891_cast_fp16")]; + tensor var_2895_begin_0 = const()[name = tensor("op_2895_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_2895_end_0 = const()[name = tensor("op_2895_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_2895_end_mask_0 = const()[name = tensor("op_2895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2895_cast_fp16 = slice_by_index(begin = var_2895_begin_0, end = var_2895_end_0, end_mask = var_2895_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2895_cast_fp16")]; + tensor var_2899_begin_0 = const()[name = tensor("op_2899_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_2899_end_0 = const()[name = tensor("op_2899_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_2899_end_mask_0 = const()[name = tensor("op_2899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2899_cast_fp16 = slice_by_index(begin = var_2899_begin_0, end = var_2899_end_0, end_mask = var_2899_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2899_cast_fp16")]; + tensor var_2903_begin_0 = const()[name = tensor("op_2903_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_2903_end_0 = const()[name = tensor("op_2903_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_2903_end_mask_0 = const()[name = tensor("op_2903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2903_cast_fp16 = slice_by_index(begin = var_2903_begin_0, end = var_2903_end_0, end_mask = var_2903_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2903_cast_fp16")]; + tensor var_2907_begin_0 = const()[name = tensor("op_2907_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_2907_end_0 = const()[name = tensor("op_2907_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_2907_end_mask_0 = const()[name = tensor("op_2907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2907_cast_fp16 = slice_by_index(begin = var_2907_begin_0, end = var_2907_end_0, end_mask = var_2907_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2907_cast_fp16")]; + tensor var_2911_begin_0 = const()[name = tensor("op_2911_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_2911_end_0 = const()[name = tensor("op_2911_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_2911_end_mask_0 = const()[name = tensor("op_2911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2911_cast_fp16 = slice_by_index(begin = var_2911_begin_0, end = var_2911_end_0, end_mask = var_2911_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2911_cast_fp16")]; + tensor var_2915_begin_0 = const()[name = tensor("op_2915_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2915_end_0 = const()[name = tensor("op_2915_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_2915_end_mask_0 = const()[name = tensor("op_2915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2915_cast_fp16 = slice_by_index(begin = var_2915_begin_0, end = var_2915_end_0, end_mask = var_2915_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2915_cast_fp16")]; + tensor var_2919_begin_0 = const()[name = tensor("op_2919_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_2919_end_0 = const()[name = tensor("op_2919_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_2919_end_mask_0 = const()[name = tensor("op_2919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2919_cast_fp16 = slice_by_index(begin = var_2919_begin_0, end = var_2919_end_0, end_mask = var_2919_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2919_cast_fp16")]; + tensor var_2923_begin_0 = const()[name = tensor("op_2923_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_2923_end_0 = const()[name = tensor("op_2923_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_2923_end_mask_0 = const()[name = tensor("op_2923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2923_cast_fp16 = slice_by_index(begin = var_2923_begin_0, end = var_2923_end_0, end_mask = var_2923_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2923_cast_fp16")]; + tensor var_2927_begin_0 = const()[name = tensor("op_2927_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_2927_end_0 = const()[name = tensor("op_2927_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_2927_end_mask_0 = const()[name = tensor("op_2927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2927_cast_fp16 = slice_by_index(begin = var_2927_begin_0, end = var_2927_end_0, end_mask = var_2927_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2927_cast_fp16")]; + tensor var_2929_begin_0 = const()[name = tensor("op_2929_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2929_end_0 = const()[name = tensor("op_2929_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2929_end_mask_0 = const()[name = tensor("op_2929_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2929_cast_fp16 = slice_by_index(begin = var_2929_begin_0, end = var_2929_end_0, end_mask = var_2929_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2929_cast_fp16")]; + tensor var_2933_begin_0 = const()[name = tensor("op_2933_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2933_end_0 = const()[name = tensor("op_2933_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_2933_end_mask_0 = const()[name = tensor("op_2933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2933_cast_fp16 = slice_by_index(begin = var_2933_begin_0, end = var_2933_end_0, end_mask = var_2933_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2933_cast_fp16")]; + tensor var_2937_begin_0 = const()[name = tensor("op_2937_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2937_end_0 = const()[name = tensor("op_2937_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_2937_end_mask_0 = const()[name = tensor("op_2937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2937_cast_fp16 = slice_by_index(begin = var_2937_begin_0, end = var_2937_end_0, end_mask = var_2937_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2937_cast_fp16")]; + tensor var_2941_begin_0 = const()[name = tensor("op_2941_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2941_end_0 = const()[name = tensor("op_2941_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_2941_end_mask_0 = const()[name = tensor("op_2941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2941_cast_fp16 = slice_by_index(begin = var_2941_begin_0, end = var_2941_end_0, end_mask = var_2941_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2941_cast_fp16")]; + tensor var_2945_begin_0 = const()[name = tensor("op_2945_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2945_end_0 = const()[name = tensor("op_2945_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_2945_end_mask_0 = const()[name = tensor("op_2945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2945_cast_fp16 = slice_by_index(begin = var_2945_begin_0, end = var_2945_end_0, end_mask = var_2945_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2945_cast_fp16")]; + tensor var_2949_begin_0 = const()[name = tensor("op_2949_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2949_end_0 = const()[name = tensor("op_2949_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_2949_end_mask_0 = const()[name = tensor("op_2949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2949_cast_fp16 = slice_by_index(begin = var_2949_begin_0, end = var_2949_end_0, end_mask = var_2949_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2949_cast_fp16")]; + tensor var_2953_begin_0 = const()[name = tensor("op_2953_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2953_end_0 = const()[name = tensor("op_2953_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_2953_end_mask_0 = const()[name = tensor("op_2953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2953_cast_fp16 = slice_by_index(begin = var_2953_begin_0, end = var_2953_end_0, end_mask = var_2953_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2953_cast_fp16")]; + tensor var_2957_begin_0 = const()[name = tensor("op_2957_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2957_end_0 = const()[name = tensor("op_2957_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_2957_end_mask_0 = const()[name = tensor("op_2957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2957_cast_fp16 = slice_by_index(begin = var_2957_begin_0, end = var_2957_end_0, end_mask = var_2957_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2957_cast_fp16")]; + tensor var_2961_begin_0 = const()[name = tensor("op_2961_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2961_end_0 = const()[name = tensor("op_2961_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_2961_end_mask_0 = const()[name = tensor("op_2961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2961_cast_fp16 = slice_by_index(begin = var_2961_begin_0, end = var_2961_end_0, end_mask = var_2961_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2961_cast_fp16")]; + tensor var_2965_begin_0 = const()[name = tensor("op_2965_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2965_end_0 = const()[name = tensor("op_2965_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_2965_end_mask_0 = const()[name = tensor("op_2965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2965_cast_fp16 = slice_by_index(begin = var_2965_begin_0, end = var_2965_end_0, end_mask = var_2965_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2965_cast_fp16")]; + tensor var_2969_begin_0 = const()[name = tensor("op_2969_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2969_end_0 = const()[name = tensor("op_2969_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_2969_end_mask_0 = const()[name = tensor("op_2969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2969_cast_fp16 = slice_by_index(begin = var_2969_begin_0, end = var_2969_end_0, end_mask = var_2969_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2969_cast_fp16")]; + tensor var_2973_begin_0 = const()[name = tensor("op_2973_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2973_end_0 = const()[name = tensor("op_2973_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_2973_end_mask_0 = const()[name = tensor("op_2973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2973_cast_fp16 = slice_by_index(begin = var_2973_begin_0, end = var_2973_end_0, end_mask = var_2973_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2973_cast_fp16")]; + tensor var_2977_begin_0 = const()[name = tensor("op_2977_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_2977_end_0 = const()[name = tensor("op_2977_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_2977_end_mask_0 = const()[name = tensor("op_2977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2977_cast_fp16 = slice_by_index(begin = var_2977_begin_0, end = var_2977_end_0, end_mask = var_2977_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2977_cast_fp16")]; + tensor var_2981_begin_0 = const()[name = tensor("op_2981_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_2981_end_0 = const()[name = tensor("op_2981_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_2981_end_mask_0 = const()[name = tensor("op_2981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2981_cast_fp16 = slice_by_index(begin = var_2981_begin_0, end = var_2981_end_0, end_mask = var_2981_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2981_cast_fp16")]; + tensor var_2985_begin_0 = const()[name = tensor("op_2985_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_2985_end_0 = const()[name = tensor("op_2985_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_2985_end_mask_0 = const()[name = tensor("op_2985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2985_cast_fp16 = slice_by_index(begin = var_2985_begin_0, end = var_2985_end_0, end_mask = var_2985_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2985_cast_fp16")]; + tensor var_2989_begin_0 = const()[name = tensor("op_2989_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2989_end_0 = const()[name = tensor("op_2989_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_2989_end_mask_0 = const()[name = tensor("op_2989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2989_cast_fp16 = slice_by_index(begin = var_2989_begin_0, end = var_2989_end_0, end_mask = var_2989_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2989_cast_fp16")]; + tensor var_2993_begin_0 = const()[name = tensor("op_2993_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2993_end_0 = const()[name = tensor("op_2993_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_2993_end_mask_0 = const()[name = tensor("op_2993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2993_cast_fp16 = slice_by_index(begin = var_2993_begin_0, end = var_2993_end_0, end_mask = var_2993_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2993_cast_fp16")]; + tensor var_2997_begin_0 = const()[name = tensor("op_2997_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2997_end_0 = const()[name = tensor("op_2997_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_2997_end_mask_0 = const()[name = tensor("op_2997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2997_cast_fp16 = slice_by_index(begin = var_2997_begin_0, end = var_2997_end_0, end_mask = var_2997_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2997_cast_fp16")]; + tensor var_3001_begin_0 = const()[name = tensor("op_3001_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3001_end_0 = const()[name = tensor("op_3001_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3001_end_mask_0 = const()[name = tensor("op_3001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3001_cast_fp16 = slice_by_index(begin = var_3001_begin_0, end = var_3001_end_0, end_mask = var_3001_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3001_cast_fp16")]; + tensor var_3005_begin_0 = const()[name = tensor("op_3005_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3005_end_0 = const()[name = tensor("op_3005_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3005_end_mask_0 = const()[name = tensor("op_3005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3005_cast_fp16 = slice_by_index(begin = var_3005_begin_0, end = var_3005_end_0, end_mask = var_3005_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3005_cast_fp16")]; + tensor var_3009_equation_0 = const()[name = tensor("op_3009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3009_cast_fp16 = einsum(equation = var_3009_equation_0, values = (var_2851_cast_fp16, var_2768_cast_fp16))[name = tensor("op_3009_cast_fp16")]; + tensor var_3010_to_fp16 = const()[name = tensor("op_3010_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_161_cast_fp16 = mul(x = var_3009_cast_fp16, y = var_3010_to_fp16)[name = tensor("aw_161_cast_fp16")]; + tensor var_3013_equation_0 = const()[name = tensor("op_3013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3013_cast_fp16 = einsum(equation = var_3013_equation_0, values = (var_2855_cast_fp16, var_2772_cast_fp16))[name = tensor("op_3013_cast_fp16")]; + tensor var_3014_to_fp16 = const()[name = tensor("op_3014_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_163_cast_fp16 = mul(x = var_3013_cast_fp16, y = var_3014_to_fp16)[name = tensor("aw_163_cast_fp16")]; + tensor var_3017_equation_0 = const()[name = tensor("op_3017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3017_cast_fp16 = einsum(equation = var_3017_equation_0, values = (var_2859_cast_fp16, var_2776_cast_fp16))[name = tensor("op_3017_cast_fp16")]; + tensor var_3018_to_fp16 = const()[name = tensor("op_3018_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_165_cast_fp16 = mul(x = var_3017_cast_fp16, y = var_3018_to_fp16)[name = tensor("aw_165_cast_fp16")]; + tensor var_3021_equation_0 = const()[name = tensor("op_3021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3021_cast_fp16 = einsum(equation = var_3021_equation_0, values = (var_2863_cast_fp16, var_2780_cast_fp16))[name = tensor("op_3021_cast_fp16")]; + tensor var_3022_to_fp16 = const()[name = tensor("op_3022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_167_cast_fp16 = mul(x = var_3021_cast_fp16, y = var_3022_to_fp16)[name = tensor("aw_167_cast_fp16")]; + tensor var_3025_equation_0 = const()[name = tensor("op_3025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3025_cast_fp16 = einsum(equation = var_3025_equation_0, values = (var_2867_cast_fp16, var_2784_cast_fp16))[name = tensor("op_3025_cast_fp16")]; + tensor var_3026_to_fp16 = const()[name = tensor("op_3026_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_169_cast_fp16 = mul(x = var_3025_cast_fp16, y = var_3026_to_fp16)[name = tensor("aw_169_cast_fp16")]; + tensor var_3029_equation_0 = const()[name = tensor("op_3029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3029_cast_fp16 = einsum(equation = var_3029_equation_0, values = (var_2871_cast_fp16, var_2788_cast_fp16))[name = tensor("op_3029_cast_fp16")]; + tensor var_3030_to_fp16 = const()[name = tensor("op_3030_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_171_cast_fp16 = mul(x = var_3029_cast_fp16, y = var_3030_to_fp16)[name = tensor("aw_171_cast_fp16")]; + tensor var_3033_equation_0 = const()[name = tensor("op_3033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3033_cast_fp16 = einsum(equation = var_3033_equation_0, values = (var_2875_cast_fp16, var_2792_cast_fp16))[name = tensor("op_3033_cast_fp16")]; + tensor var_3034_to_fp16 = const()[name = tensor("op_3034_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_173_cast_fp16 = mul(x = var_3033_cast_fp16, y = var_3034_to_fp16)[name = tensor("aw_173_cast_fp16")]; + tensor var_3037_equation_0 = const()[name = tensor("op_3037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3037_cast_fp16 = einsum(equation = var_3037_equation_0, values = (var_2879_cast_fp16, var_2796_cast_fp16))[name = tensor("op_3037_cast_fp16")]; + tensor var_3038_to_fp16 = const()[name = tensor("op_3038_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_175_cast_fp16 = mul(x = var_3037_cast_fp16, y = var_3038_to_fp16)[name = tensor("aw_175_cast_fp16")]; + tensor var_3041_equation_0 = const()[name = tensor("op_3041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3041_cast_fp16 = einsum(equation = var_3041_equation_0, values = (var_2883_cast_fp16, var_2800_cast_fp16))[name = tensor("op_3041_cast_fp16")]; + tensor var_3042_to_fp16 = const()[name = tensor("op_3042_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_177_cast_fp16 = mul(x = var_3041_cast_fp16, y = var_3042_to_fp16)[name = tensor("aw_177_cast_fp16")]; + tensor var_3045_equation_0 = const()[name = tensor("op_3045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3045_cast_fp16 = einsum(equation = var_3045_equation_0, values = (var_2887_cast_fp16, var_2804_cast_fp16))[name = tensor("op_3045_cast_fp16")]; + tensor var_3046_to_fp16 = const()[name = tensor("op_3046_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_179_cast_fp16 = mul(x = var_3045_cast_fp16, y = var_3046_to_fp16)[name = tensor("aw_179_cast_fp16")]; + tensor var_3049_equation_0 = const()[name = tensor("op_3049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3049_cast_fp16 = einsum(equation = var_3049_equation_0, values = (var_2891_cast_fp16, var_2808_cast_fp16))[name = tensor("op_3049_cast_fp16")]; + tensor var_3050_to_fp16 = const()[name = tensor("op_3050_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_181_cast_fp16 = mul(x = var_3049_cast_fp16, y = var_3050_to_fp16)[name = tensor("aw_181_cast_fp16")]; + tensor var_3053_equation_0 = const()[name = tensor("op_3053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3053_cast_fp16 = einsum(equation = var_3053_equation_0, values = (var_2895_cast_fp16, var_2812_cast_fp16))[name = tensor("op_3053_cast_fp16")]; + tensor var_3054_to_fp16 = const()[name = tensor("op_3054_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_183_cast_fp16 = mul(x = var_3053_cast_fp16, y = var_3054_to_fp16)[name = tensor("aw_183_cast_fp16")]; + tensor var_3057_equation_0 = const()[name = tensor("op_3057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3057_cast_fp16 = einsum(equation = var_3057_equation_0, values = (var_2899_cast_fp16, var_2816_cast_fp16))[name = tensor("op_3057_cast_fp16")]; + tensor var_3058_to_fp16 = const()[name = tensor("op_3058_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_185_cast_fp16 = mul(x = var_3057_cast_fp16, y = var_3058_to_fp16)[name = tensor("aw_185_cast_fp16")]; + tensor var_3061_equation_0 = const()[name = tensor("op_3061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3061_cast_fp16 = einsum(equation = var_3061_equation_0, values = (var_2903_cast_fp16, var_2820_cast_fp16))[name = tensor("op_3061_cast_fp16")]; + tensor var_3062_to_fp16 = const()[name = tensor("op_3062_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_187_cast_fp16 = mul(x = var_3061_cast_fp16, y = var_3062_to_fp16)[name = tensor("aw_187_cast_fp16")]; + tensor var_3065_equation_0 = const()[name = tensor("op_3065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3065_cast_fp16 = einsum(equation = var_3065_equation_0, values = (var_2907_cast_fp16, var_2824_cast_fp16))[name = tensor("op_3065_cast_fp16")]; + tensor var_3066_to_fp16 = const()[name = tensor("op_3066_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_189_cast_fp16 = mul(x = var_3065_cast_fp16, y = var_3066_to_fp16)[name = tensor("aw_189_cast_fp16")]; + tensor var_3069_equation_0 = const()[name = tensor("op_3069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3069_cast_fp16 = einsum(equation = var_3069_equation_0, values = (var_2911_cast_fp16, var_2828_cast_fp16))[name = tensor("op_3069_cast_fp16")]; + tensor var_3070_to_fp16 = const()[name = tensor("op_3070_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_191_cast_fp16 = mul(x = var_3069_cast_fp16, y = var_3070_to_fp16)[name = tensor("aw_191_cast_fp16")]; + tensor var_3073_equation_0 = const()[name = tensor("op_3073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3073_cast_fp16 = einsum(equation = var_3073_equation_0, values = (var_2915_cast_fp16, var_2832_cast_fp16))[name = tensor("op_3073_cast_fp16")]; + tensor var_3074_to_fp16 = const()[name = tensor("op_3074_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_193_cast_fp16 = mul(x = var_3073_cast_fp16, y = var_3074_to_fp16)[name = tensor("aw_193_cast_fp16")]; + tensor var_3077_equation_0 = const()[name = tensor("op_3077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3077_cast_fp16 = einsum(equation = var_3077_equation_0, values = (var_2919_cast_fp16, var_2836_cast_fp16))[name = tensor("op_3077_cast_fp16")]; + tensor var_3078_to_fp16 = const()[name = tensor("op_3078_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_195_cast_fp16 = mul(x = var_3077_cast_fp16, y = var_3078_to_fp16)[name = tensor("aw_195_cast_fp16")]; + tensor var_3081_equation_0 = const()[name = tensor("op_3081_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3081_cast_fp16 = einsum(equation = var_3081_equation_0, values = (var_2923_cast_fp16, var_2840_cast_fp16))[name = tensor("op_3081_cast_fp16")]; + tensor var_3082_to_fp16 = const()[name = tensor("op_3082_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_197_cast_fp16 = mul(x = var_3081_cast_fp16, y = var_3082_to_fp16)[name = tensor("aw_197_cast_fp16")]; + tensor var_3085_equation_0 = const()[name = tensor("op_3085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3085_cast_fp16 = einsum(equation = var_3085_equation_0, values = (var_2927_cast_fp16, var_2844_cast_fp16))[name = tensor("op_3085_cast_fp16")]; + tensor var_3086_to_fp16 = const()[name = tensor("op_3086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_199_cast_fp16 = mul(x = var_3085_cast_fp16, y = var_3086_to_fp16)[name = tensor("aw_199_cast_fp16")]; + tensor var_3088_cast_fp16 = softmax(axis = var_2624, x = aw_161_cast_fp16)[name = tensor("op_3088_cast_fp16")]; + tensor var_3089_cast_fp16 = softmax(axis = var_2624, x = aw_163_cast_fp16)[name = tensor("op_3089_cast_fp16")]; + tensor var_3090_cast_fp16 = softmax(axis = var_2624, x = aw_165_cast_fp16)[name = tensor("op_3090_cast_fp16")]; + tensor var_3091_cast_fp16 = softmax(axis = var_2624, x = aw_167_cast_fp16)[name = tensor("op_3091_cast_fp16")]; + tensor var_3092_cast_fp16 = softmax(axis = var_2624, x = aw_169_cast_fp16)[name = tensor("op_3092_cast_fp16")]; + tensor var_3093_cast_fp16 = softmax(axis = var_2624, x = aw_171_cast_fp16)[name = tensor("op_3093_cast_fp16")]; + tensor var_3094_cast_fp16 = softmax(axis = var_2624, x = aw_173_cast_fp16)[name = tensor("op_3094_cast_fp16")]; + tensor var_3095_cast_fp16 = softmax(axis = var_2624, x = aw_175_cast_fp16)[name = tensor("op_3095_cast_fp16")]; + tensor var_3096_cast_fp16 = softmax(axis = var_2624, x = aw_177_cast_fp16)[name = tensor("op_3096_cast_fp16")]; + tensor var_3097_cast_fp16 = softmax(axis = var_2624, x = aw_179_cast_fp16)[name = tensor("op_3097_cast_fp16")]; + tensor var_3098_cast_fp16 = softmax(axis = var_2624, x = aw_181_cast_fp16)[name = tensor("op_3098_cast_fp16")]; + tensor var_3099_cast_fp16 = softmax(axis = var_2624, x = aw_183_cast_fp16)[name = tensor("op_3099_cast_fp16")]; + tensor var_3100_cast_fp16 = softmax(axis = var_2624, x = aw_185_cast_fp16)[name = tensor("op_3100_cast_fp16")]; + tensor var_3101_cast_fp16 = softmax(axis = var_2624, x = aw_187_cast_fp16)[name = tensor("op_3101_cast_fp16")]; + tensor var_3102_cast_fp16 = softmax(axis = var_2624, x = aw_189_cast_fp16)[name = tensor("op_3102_cast_fp16")]; + tensor var_3103_cast_fp16 = softmax(axis = var_2624, x = aw_191_cast_fp16)[name = tensor("op_3103_cast_fp16")]; + tensor var_3104_cast_fp16 = softmax(axis = var_2624, x = aw_193_cast_fp16)[name = tensor("op_3104_cast_fp16")]; + tensor var_3105_cast_fp16 = softmax(axis = var_2624, x = aw_195_cast_fp16)[name = tensor("op_3105_cast_fp16")]; + tensor var_3106_cast_fp16 = softmax(axis = var_2624, x = aw_197_cast_fp16)[name = tensor("op_3106_cast_fp16")]; + tensor var_3107_cast_fp16 = softmax(axis = var_2624, x = aw_199_cast_fp16)[name = tensor("op_3107_cast_fp16")]; + tensor var_3109_equation_0 = const()[name = tensor("op_3109_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3109_cast_fp16 = einsum(equation = var_3109_equation_0, values = (var_2929_cast_fp16, var_3088_cast_fp16))[name = tensor("op_3109_cast_fp16")]; + tensor var_3111_equation_0 = const()[name = tensor("op_3111_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3111_cast_fp16 = einsum(equation = var_3111_equation_0, values = (var_2933_cast_fp16, var_3089_cast_fp16))[name = tensor("op_3111_cast_fp16")]; + tensor var_3113_equation_0 = const()[name = tensor("op_3113_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3113_cast_fp16 = einsum(equation = var_3113_equation_0, values = (var_2937_cast_fp16, var_3090_cast_fp16))[name = tensor("op_3113_cast_fp16")]; + tensor var_3115_equation_0 = const()[name = tensor("op_3115_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3115_cast_fp16 = einsum(equation = var_3115_equation_0, values = (var_2941_cast_fp16, var_3091_cast_fp16))[name = tensor("op_3115_cast_fp16")]; + tensor var_3117_equation_0 = const()[name = tensor("op_3117_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3117_cast_fp16 = einsum(equation = var_3117_equation_0, values = (var_2945_cast_fp16, var_3092_cast_fp16))[name = tensor("op_3117_cast_fp16")]; + tensor var_3119_equation_0 = const()[name = tensor("op_3119_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3119_cast_fp16 = einsum(equation = var_3119_equation_0, values = (var_2949_cast_fp16, var_3093_cast_fp16))[name = tensor("op_3119_cast_fp16")]; + tensor var_3121_equation_0 = const()[name = tensor("op_3121_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3121_cast_fp16 = einsum(equation = var_3121_equation_0, values = (var_2953_cast_fp16, var_3094_cast_fp16))[name = tensor("op_3121_cast_fp16")]; + tensor var_3123_equation_0 = const()[name = tensor("op_3123_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3123_cast_fp16 = einsum(equation = var_3123_equation_0, values = (var_2957_cast_fp16, var_3095_cast_fp16))[name = tensor("op_3123_cast_fp16")]; + tensor var_3125_equation_0 = const()[name = tensor("op_3125_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3125_cast_fp16 = einsum(equation = var_3125_equation_0, values = (var_2961_cast_fp16, var_3096_cast_fp16))[name = tensor("op_3125_cast_fp16")]; + tensor var_3127_equation_0 = const()[name = tensor("op_3127_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3127_cast_fp16 = einsum(equation = var_3127_equation_0, values = (var_2965_cast_fp16, var_3097_cast_fp16))[name = tensor("op_3127_cast_fp16")]; + tensor var_3129_equation_0 = const()[name = tensor("op_3129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3129_cast_fp16 = einsum(equation = var_3129_equation_0, values = (var_2969_cast_fp16, var_3098_cast_fp16))[name = tensor("op_3129_cast_fp16")]; + tensor var_3131_equation_0 = const()[name = tensor("op_3131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3131_cast_fp16 = einsum(equation = var_3131_equation_0, values = (var_2973_cast_fp16, var_3099_cast_fp16))[name = tensor("op_3131_cast_fp16")]; + tensor var_3133_equation_0 = const()[name = tensor("op_3133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3133_cast_fp16 = einsum(equation = var_3133_equation_0, values = (var_2977_cast_fp16, var_3100_cast_fp16))[name = tensor("op_3133_cast_fp16")]; + tensor var_3135_equation_0 = const()[name = tensor("op_3135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3135_cast_fp16 = einsum(equation = var_3135_equation_0, values = (var_2981_cast_fp16, var_3101_cast_fp16))[name = tensor("op_3135_cast_fp16")]; + tensor var_3137_equation_0 = const()[name = tensor("op_3137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3137_cast_fp16 = einsum(equation = var_3137_equation_0, values = (var_2985_cast_fp16, var_3102_cast_fp16))[name = tensor("op_3137_cast_fp16")]; + tensor var_3139_equation_0 = const()[name = tensor("op_3139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3139_cast_fp16 = einsum(equation = var_3139_equation_0, values = (var_2989_cast_fp16, var_3103_cast_fp16))[name = tensor("op_3139_cast_fp16")]; + tensor var_3141_equation_0 = const()[name = tensor("op_3141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3141_cast_fp16 = einsum(equation = var_3141_equation_0, values = (var_2993_cast_fp16, var_3104_cast_fp16))[name = tensor("op_3141_cast_fp16")]; + tensor var_3143_equation_0 = const()[name = tensor("op_3143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3143_cast_fp16 = einsum(equation = var_3143_equation_0, values = (var_2997_cast_fp16, var_3105_cast_fp16))[name = tensor("op_3143_cast_fp16")]; + tensor var_3145_equation_0 = const()[name = tensor("op_3145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3145_cast_fp16 = einsum(equation = var_3145_equation_0, values = (var_3001_cast_fp16, var_3106_cast_fp16))[name = tensor("op_3145_cast_fp16")]; + tensor var_3147_equation_0 = const()[name = tensor("op_3147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3147_cast_fp16 = einsum(equation = var_3147_equation_0, values = (var_3005_cast_fp16, var_3107_cast_fp16))[name = tensor("op_3147_cast_fp16")]; + tensor input_131_interleave_0 = const()[name = tensor("input_131_interleave_0"), val = tensor(false)]; + tensor input_131_cast_fp16 = concat(axis = var_2624, interleave = input_131_interleave_0, values = (var_3109_cast_fp16, var_3111_cast_fp16, var_3113_cast_fp16, var_3115_cast_fp16, var_3117_cast_fp16, var_3119_cast_fp16, var_3121_cast_fp16, var_3123_cast_fp16, var_3125_cast_fp16, var_3127_cast_fp16, var_3129_cast_fp16, var_3131_cast_fp16, var_3133_cast_fp16, var_3135_cast_fp16, var_3137_cast_fp16, var_3139_cast_fp16, var_3141_cast_fp16, var_3143_cast_fp16, var_3145_cast_fp16, var_3147_cast_fp16))[name = tensor("input_131_cast_fp16")]; + tensor var_3157_pad_type_0 = const()[name = tensor("op_3157_pad_type_0"), val = tensor("valid")]; + tensor var_3157_strides_0 = const()[name = tensor("op_3157_strides_0"), val = tensor([1, 1])]; + tensor var_3157_pad_0 = const()[name = tensor("op_3157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3157_dilations_0 = const()[name = tensor("op_3157_dilations_0"), val = tensor([1, 1])]; + tensor var_3157_groups_0 = const()[name = tensor("op_3157_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78132352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79361216))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79361408)))]; + tensor var_3157_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3157_dilations_0, groups = var_3157_groups_0, pad = var_3157_pad_0, pad_type = var_3157_pad_type_0, strides = var_3157_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_131_cast_fp16)[name = tensor("op_3157_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = var_3157_cast_fp16, y = inputs_25_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; + tensor hidden_states_67_axes_0 = const()[name = tensor("hidden_states_67_axes_0"), val = tensor([1])]; + tensor hidden_states_67_gamma_0_to_fp16 = const()[name = tensor("hidden_states_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79364032)))]; + tensor hidden_states_67_beta_0_to_fp16 = const()[name = tensor("hidden_states_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79366656)))]; + tensor var_3167_to_fp16 = const()[name = tensor("op_3167_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = hidden_states_67_beta_0_to_fp16, epsilon = var_3167_to_fp16, gamma = hidden_states_67_gamma_0_to_fp16, x = inputs_27_cast_fp16)[name = tensor("hidden_states_67_cast_fp16")]; + tensor q_19_pad_type_0 = const()[name = tensor("q_19_pad_type_0"), val = tensor("valid")]; + tensor q_19_strides_0 = const()[name = tensor("q_19_strides_0"), val = tensor([1, 1])]; + tensor q_19_pad_0 = const()[name = tensor("q_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_19_dilations_0 = const()[name = tensor("q_19_dilations_0"), val = tensor([1, 1])]; + tensor q_19_groups_0 = const()[name = tensor("q_19_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79369280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80598144))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_19_cast_fp16 = conv(dilations = q_19_dilations_0, groups = q_19_groups_0, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = q_19_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("q_19_cast_fp16")]; + tensor k_37_pad_type_0 = const()[name = tensor("k_37_pad_type_0"), val = tensor("valid")]; + tensor k_37_strides_0 = const()[name = tensor("k_37_strides_0"), val = tensor([1, 1])]; + tensor k_37_pad_0 = const()[name = tensor("k_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_37_dilations_0 = const()[name = tensor("k_37_dilations_0"), val = tensor([1, 1])]; + tensor k_37_groups_0 = const()[name = tensor("k_37_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80598336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82564480))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_37_cast_fp16 = conv(dilations = k_37_dilations_0, groups = k_37_groups_0, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = k_37_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_37_cast_fp16")]; + tensor v_19_pad_type_0 = const()[name = tensor("v_19_pad_type_0"), val = tensor("valid")]; + tensor v_19_strides_0 = const()[name = tensor("v_19_strides_0"), val = tensor([1, 1])]; + tensor v_19_pad_0 = const()[name = tensor("v_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_19_dilations_0 = const()[name = tensor("v_19_dilations_0"), val = tensor([1, 1])]; + tensor v_19_groups_0 = const()[name = tensor("v_19_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82564672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84530816))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_19_cast_fp16 = conv(dilations = v_19_dilations_0, groups = v_19_groups_0, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = v_19_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_19_cast_fp16")]; + tensor var_3200_begin_0 = const()[name = tensor("op_3200_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3200_end_0 = const()[name = tensor("op_3200_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3200_end_mask_0 = const()[name = tensor("op_3200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3200_cast_fp16 = slice_by_index(begin = var_3200_begin_0, end = var_3200_end_0, end_mask = var_3200_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3200_cast_fp16")]; + tensor var_3204_begin_0 = const()[name = tensor("op_3204_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3204_end_0 = const()[name = tensor("op_3204_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3204_end_mask_0 = const()[name = tensor("op_3204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3204_cast_fp16 = slice_by_index(begin = var_3204_begin_0, end = var_3204_end_0, end_mask = var_3204_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3204_cast_fp16")]; + tensor var_3208_begin_0 = const()[name = tensor("op_3208_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3208_end_0 = const()[name = tensor("op_3208_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3208_end_mask_0 = const()[name = tensor("op_3208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3208_cast_fp16 = slice_by_index(begin = var_3208_begin_0, end = var_3208_end_0, end_mask = var_3208_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3208_cast_fp16")]; + tensor var_3212_begin_0 = const()[name = tensor("op_3212_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3212_end_0 = const()[name = tensor("op_3212_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3212_end_mask_0 = const()[name = tensor("op_3212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3212_cast_fp16 = slice_by_index(begin = var_3212_begin_0, end = var_3212_end_0, end_mask = var_3212_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3212_cast_fp16")]; + tensor var_3216_begin_0 = const()[name = tensor("op_3216_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3216_end_0 = const()[name = tensor("op_3216_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3216_end_mask_0 = const()[name = tensor("op_3216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3216_cast_fp16 = slice_by_index(begin = var_3216_begin_0, end = var_3216_end_0, end_mask = var_3216_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3216_cast_fp16")]; + tensor var_3220_begin_0 = const()[name = tensor("op_3220_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3220_end_0 = const()[name = tensor("op_3220_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3220_end_mask_0 = const()[name = tensor("op_3220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3220_cast_fp16 = slice_by_index(begin = var_3220_begin_0, end = var_3220_end_0, end_mask = var_3220_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3220_cast_fp16")]; + tensor var_3224_begin_0 = const()[name = tensor("op_3224_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3224_end_0 = const()[name = tensor("op_3224_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3224_end_mask_0 = const()[name = tensor("op_3224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3224_cast_fp16 = slice_by_index(begin = var_3224_begin_0, end = var_3224_end_0, end_mask = var_3224_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3224_cast_fp16")]; + tensor var_3228_begin_0 = const()[name = tensor("op_3228_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3228_end_0 = const()[name = tensor("op_3228_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3228_end_mask_0 = const()[name = tensor("op_3228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3228_cast_fp16 = slice_by_index(begin = var_3228_begin_0, end = var_3228_end_0, end_mask = var_3228_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3228_cast_fp16")]; + tensor var_3232_begin_0 = const()[name = tensor("op_3232_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3232_end_0 = const()[name = tensor("op_3232_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3232_end_mask_0 = const()[name = tensor("op_3232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3232_cast_fp16 = slice_by_index(begin = var_3232_begin_0, end = var_3232_end_0, end_mask = var_3232_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3232_cast_fp16")]; + tensor var_3236_begin_0 = const()[name = tensor("op_3236_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3236_end_0 = const()[name = tensor("op_3236_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3236_end_mask_0 = const()[name = tensor("op_3236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3236_cast_fp16 = slice_by_index(begin = var_3236_begin_0, end = var_3236_end_0, end_mask = var_3236_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3236_cast_fp16")]; + tensor var_3240_begin_0 = const()[name = tensor("op_3240_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3240_end_0 = const()[name = tensor("op_3240_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_3240_end_mask_0 = const()[name = tensor("op_3240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3240_cast_fp16 = slice_by_index(begin = var_3240_begin_0, end = var_3240_end_0, end_mask = var_3240_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3240_cast_fp16")]; + tensor var_3244_begin_0 = const()[name = tensor("op_3244_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3244_end_0 = const()[name = tensor("op_3244_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_3244_end_mask_0 = const()[name = tensor("op_3244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3244_cast_fp16 = slice_by_index(begin = var_3244_begin_0, end = var_3244_end_0, end_mask = var_3244_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3244_cast_fp16")]; + tensor var_3248_begin_0 = const()[name = tensor("op_3248_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3248_end_0 = const()[name = tensor("op_3248_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_3248_end_mask_0 = const()[name = tensor("op_3248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3248_cast_fp16 = slice_by_index(begin = var_3248_begin_0, end = var_3248_end_0, end_mask = var_3248_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3248_cast_fp16")]; + tensor var_3252_begin_0 = const()[name = tensor("op_3252_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3252_end_0 = const()[name = tensor("op_3252_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_3252_end_mask_0 = const()[name = tensor("op_3252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3252_cast_fp16 = slice_by_index(begin = var_3252_begin_0, end = var_3252_end_0, end_mask = var_3252_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3252_cast_fp16")]; + tensor var_3256_begin_0 = const()[name = tensor("op_3256_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3256_end_0 = const()[name = tensor("op_3256_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_3256_end_mask_0 = const()[name = tensor("op_3256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3256_cast_fp16 = slice_by_index(begin = var_3256_begin_0, end = var_3256_end_0, end_mask = var_3256_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3256_cast_fp16")]; + tensor var_3260_begin_0 = const()[name = tensor("op_3260_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3260_end_0 = const()[name = tensor("op_3260_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3260_end_mask_0 = const()[name = tensor("op_3260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3260_cast_fp16 = slice_by_index(begin = var_3260_begin_0, end = var_3260_end_0, end_mask = var_3260_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3260_cast_fp16")]; + tensor var_3264_begin_0 = const()[name = tensor("op_3264_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3264_end_0 = const()[name = tensor("op_3264_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_3264_end_mask_0 = const()[name = tensor("op_3264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3264_cast_fp16 = slice_by_index(begin = var_3264_begin_0, end = var_3264_end_0, end_mask = var_3264_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3264_cast_fp16")]; + tensor var_3268_begin_0 = const()[name = tensor("op_3268_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3268_end_0 = const()[name = tensor("op_3268_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_3268_end_mask_0 = const()[name = tensor("op_3268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3268_cast_fp16 = slice_by_index(begin = var_3268_begin_0, end = var_3268_end_0, end_mask = var_3268_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3268_cast_fp16")]; + tensor var_3272_begin_0 = const()[name = tensor("op_3272_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3272_end_0 = const()[name = tensor("op_3272_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3272_end_mask_0 = const()[name = tensor("op_3272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3272_cast_fp16 = slice_by_index(begin = var_3272_begin_0, end = var_3272_end_0, end_mask = var_3272_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3272_cast_fp16")]; + tensor var_3276_begin_0 = const()[name = tensor("op_3276_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3276_end_0 = const()[name = tensor("op_3276_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3276_end_mask_0 = const()[name = tensor("op_3276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3276_cast_fp16 = slice_by_index(begin = var_3276_begin_0, end = var_3276_end_0, end_mask = var_3276_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3276_cast_fp16")]; + tensor k_39_perm_0 = const()[name = tensor("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3283_begin_0 = const()[name = tensor("op_3283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3283_end_0 = const()[name = tensor("op_3283_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_3283_end_mask_0 = const()[name = tensor("op_3283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_39_cast_fp16 = transpose(perm = k_39_perm_0, x = k_37_cast_fp16)[name = tensor("transpose_58")]; + tensor var_3283_cast_fp16 = slice_by_index(begin = var_3283_begin_0, end = var_3283_end_0, end_mask = var_3283_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3283_cast_fp16")]; + tensor var_3287_begin_0 = const()[name = tensor("op_3287_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3287_end_0 = const()[name = tensor("op_3287_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_3287_end_mask_0 = const()[name = tensor("op_3287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3287_cast_fp16 = slice_by_index(begin = var_3287_begin_0, end = var_3287_end_0, end_mask = var_3287_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3287_cast_fp16")]; + tensor var_3291_begin_0 = const()[name = tensor("op_3291_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3291_end_0 = const()[name = tensor("op_3291_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_3291_end_mask_0 = const()[name = tensor("op_3291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3291_cast_fp16 = slice_by_index(begin = var_3291_begin_0, end = var_3291_end_0, end_mask = var_3291_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3291_cast_fp16")]; + tensor var_3295_begin_0 = const()[name = tensor("op_3295_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3295_end_0 = const()[name = tensor("op_3295_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_3295_end_mask_0 = const()[name = tensor("op_3295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3295_cast_fp16 = slice_by_index(begin = var_3295_begin_0, end = var_3295_end_0, end_mask = var_3295_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3295_cast_fp16")]; + tensor var_3299_begin_0 = const()[name = tensor("op_3299_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3299_end_0 = const()[name = tensor("op_3299_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_3299_end_mask_0 = const()[name = tensor("op_3299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3299_cast_fp16 = slice_by_index(begin = var_3299_begin_0, end = var_3299_end_0, end_mask = var_3299_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3299_cast_fp16")]; + tensor var_3303_begin_0 = const()[name = tensor("op_3303_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3303_end_0 = const()[name = tensor("op_3303_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_3303_end_mask_0 = const()[name = tensor("op_3303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3303_cast_fp16 = slice_by_index(begin = var_3303_begin_0, end = var_3303_end_0, end_mask = var_3303_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3303_cast_fp16")]; + tensor var_3307_begin_0 = const()[name = tensor("op_3307_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3307_end_0 = const()[name = tensor("op_3307_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_3307_end_mask_0 = const()[name = tensor("op_3307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3307_cast_fp16 = slice_by_index(begin = var_3307_begin_0, end = var_3307_end_0, end_mask = var_3307_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3307_cast_fp16")]; + tensor var_3311_begin_0 = const()[name = tensor("op_3311_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3311_end_0 = const()[name = tensor("op_3311_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_3311_end_mask_0 = const()[name = tensor("op_3311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3311_cast_fp16 = slice_by_index(begin = var_3311_begin_0, end = var_3311_end_0, end_mask = var_3311_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3311_cast_fp16")]; + tensor var_3315_begin_0 = const()[name = tensor("op_3315_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3315_end_0 = const()[name = tensor("op_3315_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_3315_end_mask_0 = const()[name = tensor("op_3315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3315_cast_fp16 = slice_by_index(begin = var_3315_begin_0, end = var_3315_end_0, end_mask = var_3315_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3315_cast_fp16")]; + tensor var_3319_begin_0 = const()[name = tensor("op_3319_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3319_end_0 = const()[name = tensor("op_3319_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_3319_end_mask_0 = const()[name = tensor("op_3319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3319_cast_fp16 = slice_by_index(begin = var_3319_begin_0, end = var_3319_end_0, end_mask = var_3319_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3319_cast_fp16")]; + tensor var_3323_begin_0 = const()[name = tensor("op_3323_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3323_end_0 = const()[name = tensor("op_3323_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_3323_end_mask_0 = const()[name = tensor("op_3323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3323_cast_fp16 = slice_by_index(begin = var_3323_begin_0, end = var_3323_end_0, end_mask = var_3323_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3323_cast_fp16")]; + tensor var_3327_begin_0 = const()[name = tensor("op_3327_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_3327_end_0 = const()[name = tensor("op_3327_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_3327_end_mask_0 = const()[name = tensor("op_3327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3327_cast_fp16 = slice_by_index(begin = var_3327_begin_0, end = var_3327_end_0, end_mask = var_3327_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3327_cast_fp16")]; + tensor var_3331_begin_0 = const()[name = tensor("op_3331_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_3331_end_0 = const()[name = tensor("op_3331_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_3331_end_mask_0 = const()[name = tensor("op_3331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3331_cast_fp16 = slice_by_index(begin = var_3331_begin_0, end = var_3331_end_0, end_mask = var_3331_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3331_cast_fp16")]; + tensor var_3335_begin_0 = const()[name = tensor("op_3335_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_3335_end_0 = const()[name = tensor("op_3335_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_3335_end_mask_0 = const()[name = tensor("op_3335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3335_cast_fp16 = slice_by_index(begin = var_3335_begin_0, end = var_3335_end_0, end_mask = var_3335_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3335_cast_fp16")]; + tensor var_3339_begin_0 = const()[name = tensor("op_3339_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_3339_end_0 = const()[name = tensor("op_3339_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_3339_end_mask_0 = const()[name = tensor("op_3339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3339_cast_fp16 = slice_by_index(begin = var_3339_begin_0, end = var_3339_end_0, end_mask = var_3339_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3339_cast_fp16")]; + tensor var_3343_begin_0 = const()[name = tensor("op_3343_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_3343_end_0 = const()[name = tensor("op_3343_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_3343_end_mask_0 = const()[name = tensor("op_3343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3343_cast_fp16 = slice_by_index(begin = var_3343_begin_0, end = var_3343_end_0, end_mask = var_3343_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3343_cast_fp16")]; + tensor var_3347_begin_0 = const()[name = tensor("op_3347_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_3347_end_0 = const()[name = tensor("op_3347_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_3347_end_mask_0 = const()[name = tensor("op_3347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3347_cast_fp16 = slice_by_index(begin = var_3347_begin_0, end = var_3347_end_0, end_mask = var_3347_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3347_cast_fp16")]; + tensor var_3351_begin_0 = const()[name = tensor("op_3351_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_3351_end_0 = const()[name = tensor("op_3351_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_3351_end_mask_0 = const()[name = tensor("op_3351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3351_cast_fp16 = slice_by_index(begin = var_3351_begin_0, end = var_3351_end_0, end_mask = var_3351_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3351_cast_fp16")]; + tensor var_3355_begin_0 = const()[name = tensor("op_3355_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_3355_end_0 = const()[name = tensor("op_3355_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_3355_end_mask_0 = const()[name = tensor("op_3355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3355_cast_fp16 = slice_by_index(begin = var_3355_begin_0, end = var_3355_end_0, end_mask = var_3355_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3355_cast_fp16")]; + tensor var_3359_begin_0 = const()[name = tensor("op_3359_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_3359_end_0 = const()[name = tensor("op_3359_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_3359_end_mask_0 = const()[name = tensor("op_3359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3359_cast_fp16 = slice_by_index(begin = var_3359_begin_0, end = var_3359_end_0, end_mask = var_3359_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3359_cast_fp16")]; + tensor var_3361_begin_0 = const()[name = tensor("op_3361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3361_end_0 = const()[name = tensor("op_3361_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_3361_end_mask_0 = const()[name = tensor("op_3361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3361_cast_fp16 = slice_by_index(begin = var_3361_begin_0, end = var_3361_end_0, end_mask = var_3361_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3361_cast_fp16")]; + tensor var_3365_begin_0 = const()[name = tensor("op_3365_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3365_end_0 = const()[name = tensor("op_3365_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_3365_end_mask_0 = const()[name = tensor("op_3365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3365_cast_fp16 = slice_by_index(begin = var_3365_begin_0, end = var_3365_end_0, end_mask = var_3365_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3365_cast_fp16")]; + tensor var_3369_begin_0 = const()[name = tensor("op_3369_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3369_end_0 = const()[name = tensor("op_3369_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_3369_end_mask_0 = const()[name = tensor("op_3369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3369_cast_fp16 = slice_by_index(begin = var_3369_begin_0, end = var_3369_end_0, end_mask = var_3369_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3369_cast_fp16")]; + tensor var_3373_begin_0 = const()[name = tensor("op_3373_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3373_end_0 = const()[name = tensor("op_3373_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_3373_end_mask_0 = const()[name = tensor("op_3373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3373_cast_fp16 = slice_by_index(begin = var_3373_begin_0, end = var_3373_end_0, end_mask = var_3373_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3373_cast_fp16")]; + tensor var_3377_begin_0 = const()[name = tensor("op_3377_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3377_end_0 = const()[name = tensor("op_3377_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_3377_end_mask_0 = const()[name = tensor("op_3377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3377_cast_fp16 = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3377_cast_fp16")]; + tensor var_3381_begin_0 = const()[name = tensor("op_3381_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3381_end_0 = const()[name = tensor("op_3381_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_3381_end_mask_0 = const()[name = tensor("op_3381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3381_cast_fp16 = slice_by_index(begin = var_3381_begin_0, end = var_3381_end_0, end_mask = var_3381_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3381_cast_fp16")]; + tensor var_3385_begin_0 = const()[name = tensor("op_3385_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3385_end_0 = const()[name = tensor("op_3385_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_3385_end_mask_0 = const()[name = tensor("op_3385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3385_cast_fp16 = slice_by_index(begin = var_3385_begin_0, end = var_3385_end_0, end_mask = var_3385_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3385_cast_fp16")]; + tensor var_3389_begin_0 = const()[name = tensor("op_3389_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3389_end_0 = const()[name = tensor("op_3389_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_3389_end_mask_0 = const()[name = tensor("op_3389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3389_cast_fp16 = slice_by_index(begin = var_3389_begin_0, end = var_3389_end_0, end_mask = var_3389_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3389_cast_fp16")]; + tensor var_3393_begin_0 = const()[name = tensor("op_3393_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3393_end_0 = const()[name = tensor("op_3393_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_3393_end_mask_0 = const()[name = tensor("op_3393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3393_cast_fp16 = slice_by_index(begin = var_3393_begin_0, end = var_3393_end_0, end_mask = var_3393_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3393_cast_fp16")]; + tensor var_3397_begin_0 = const()[name = tensor("op_3397_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3397_end_0 = const()[name = tensor("op_3397_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_3397_end_mask_0 = const()[name = tensor("op_3397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3397_cast_fp16 = slice_by_index(begin = var_3397_begin_0, end = var_3397_end_0, end_mask = var_3397_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3397_cast_fp16")]; + tensor var_3401_begin_0 = const()[name = tensor("op_3401_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3401_end_0 = const()[name = tensor("op_3401_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_3401_end_mask_0 = const()[name = tensor("op_3401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3401_cast_fp16 = slice_by_index(begin = var_3401_begin_0, end = var_3401_end_0, end_mask = var_3401_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3401_cast_fp16")]; + tensor var_3405_begin_0 = const()[name = tensor("op_3405_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3405_end_0 = const()[name = tensor("op_3405_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_3405_end_mask_0 = const()[name = tensor("op_3405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3405_cast_fp16 = slice_by_index(begin = var_3405_begin_0, end = var_3405_end_0, end_mask = var_3405_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3405_cast_fp16")]; + tensor var_3409_begin_0 = const()[name = tensor("op_3409_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3409_end_0 = const()[name = tensor("op_3409_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_3409_end_mask_0 = const()[name = tensor("op_3409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3409_cast_fp16 = slice_by_index(begin = var_3409_begin_0, end = var_3409_end_0, end_mask = var_3409_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3409_cast_fp16")]; + tensor var_3413_begin_0 = const()[name = tensor("op_3413_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3413_end_0 = const()[name = tensor("op_3413_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_3413_end_mask_0 = const()[name = tensor("op_3413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3413_cast_fp16 = slice_by_index(begin = var_3413_begin_0, end = var_3413_end_0, end_mask = var_3413_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3413_cast_fp16")]; + tensor var_3417_begin_0 = const()[name = tensor("op_3417_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3417_end_0 = const()[name = tensor("op_3417_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_3417_end_mask_0 = const()[name = tensor("op_3417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3417_cast_fp16 = slice_by_index(begin = var_3417_begin_0, end = var_3417_end_0, end_mask = var_3417_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3417_cast_fp16")]; + tensor var_3421_begin_0 = const()[name = tensor("op_3421_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3421_end_0 = const()[name = tensor("op_3421_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_3421_end_mask_0 = const()[name = tensor("op_3421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3421_cast_fp16 = slice_by_index(begin = var_3421_begin_0, end = var_3421_end_0, end_mask = var_3421_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3421_cast_fp16")]; + tensor var_3425_begin_0 = const()[name = tensor("op_3425_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3425_end_0 = const()[name = tensor("op_3425_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_3425_end_mask_0 = const()[name = tensor("op_3425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3425_cast_fp16 = slice_by_index(begin = var_3425_begin_0, end = var_3425_end_0, end_mask = var_3425_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3425_cast_fp16")]; + tensor var_3429_begin_0 = const()[name = tensor("op_3429_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3429_end_0 = const()[name = tensor("op_3429_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_3429_end_mask_0 = const()[name = tensor("op_3429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3429_cast_fp16 = slice_by_index(begin = var_3429_begin_0, end = var_3429_end_0, end_mask = var_3429_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3429_cast_fp16")]; + tensor var_3433_begin_0 = const()[name = tensor("op_3433_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3433_end_0 = const()[name = tensor("op_3433_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_3433_end_mask_0 = const()[name = tensor("op_3433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3433_cast_fp16 = slice_by_index(begin = var_3433_begin_0, end = var_3433_end_0, end_mask = var_3433_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3433_cast_fp16")]; + tensor var_3437_begin_0 = const()[name = tensor("op_3437_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3437_end_0 = const()[name = tensor("op_3437_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_3437_end_mask_0 = const()[name = tensor("op_3437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3437_cast_fp16 = slice_by_index(begin = var_3437_begin_0, end = var_3437_end_0, end_mask = var_3437_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3437_cast_fp16")]; + tensor var_3441_equation_0 = const()[name = tensor("op_3441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3441_cast_fp16 = einsum(equation = var_3441_equation_0, values = (var_3283_cast_fp16, var_3200_cast_fp16))[name = tensor("op_3441_cast_fp16")]; + tensor var_3442_to_fp16 = const()[name = tensor("op_3442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_201_cast_fp16 = mul(x = var_3441_cast_fp16, y = var_3442_to_fp16)[name = tensor("aw_201_cast_fp16")]; + tensor var_3445_equation_0 = const()[name = tensor("op_3445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3445_cast_fp16 = einsum(equation = var_3445_equation_0, values = (var_3287_cast_fp16, var_3204_cast_fp16))[name = tensor("op_3445_cast_fp16")]; + tensor var_3446_to_fp16 = const()[name = tensor("op_3446_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_203_cast_fp16 = mul(x = var_3445_cast_fp16, y = var_3446_to_fp16)[name = tensor("aw_203_cast_fp16")]; + tensor var_3449_equation_0 = const()[name = tensor("op_3449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3449_cast_fp16 = einsum(equation = var_3449_equation_0, values = (var_3291_cast_fp16, var_3208_cast_fp16))[name = tensor("op_3449_cast_fp16")]; + tensor var_3450_to_fp16 = const()[name = tensor("op_3450_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_205_cast_fp16 = mul(x = var_3449_cast_fp16, y = var_3450_to_fp16)[name = tensor("aw_205_cast_fp16")]; + tensor var_3453_equation_0 = const()[name = tensor("op_3453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3453_cast_fp16 = einsum(equation = var_3453_equation_0, values = (var_3295_cast_fp16, var_3212_cast_fp16))[name = tensor("op_3453_cast_fp16")]; + tensor var_3454_to_fp16 = const()[name = tensor("op_3454_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_207_cast_fp16 = mul(x = var_3453_cast_fp16, y = var_3454_to_fp16)[name = tensor("aw_207_cast_fp16")]; + tensor var_3457_equation_0 = const()[name = tensor("op_3457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3457_cast_fp16 = einsum(equation = var_3457_equation_0, values = (var_3299_cast_fp16, var_3216_cast_fp16))[name = tensor("op_3457_cast_fp16")]; + tensor var_3458_to_fp16 = const()[name = tensor("op_3458_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_209_cast_fp16 = mul(x = var_3457_cast_fp16, y = var_3458_to_fp16)[name = tensor("aw_209_cast_fp16")]; + tensor var_3461_equation_0 = const()[name = tensor("op_3461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3461_cast_fp16 = einsum(equation = var_3461_equation_0, values = (var_3303_cast_fp16, var_3220_cast_fp16))[name = tensor("op_3461_cast_fp16")]; + tensor var_3462_to_fp16 = const()[name = tensor("op_3462_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_211_cast_fp16 = mul(x = var_3461_cast_fp16, y = var_3462_to_fp16)[name = tensor("aw_211_cast_fp16")]; + tensor var_3465_equation_0 = const()[name = tensor("op_3465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3465_cast_fp16 = einsum(equation = var_3465_equation_0, values = (var_3307_cast_fp16, var_3224_cast_fp16))[name = tensor("op_3465_cast_fp16")]; + tensor var_3466_to_fp16 = const()[name = tensor("op_3466_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_213_cast_fp16 = mul(x = var_3465_cast_fp16, y = var_3466_to_fp16)[name = tensor("aw_213_cast_fp16")]; + tensor var_3469_equation_0 = const()[name = tensor("op_3469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3469_cast_fp16 = einsum(equation = var_3469_equation_0, values = (var_3311_cast_fp16, var_3228_cast_fp16))[name = tensor("op_3469_cast_fp16")]; + tensor var_3470_to_fp16 = const()[name = tensor("op_3470_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_215_cast_fp16 = mul(x = var_3469_cast_fp16, y = var_3470_to_fp16)[name = tensor("aw_215_cast_fp16")]; + tensor var_3473_equation_0 = const()[name = tensor("op_3473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3473_cast_fp16 = einsum(equation = var_3473_equation_0, values = (var_3315_cast_fp16, var_3232_cast_fp16))[name = tensor("op_3473_cast_fp16")]; + tensor var_3474_to_fp16 = const()[name = tensor("op_3474_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_217_cast_fp16 = mul(x = var_3473_cast_fp16, y = var_3474_to_fp16)[name = tensor("aw_217_cast_fp16")]; + tensor var_3477_equation_0 = const()[name = tensor("op_3477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3477_cast_fp16 = einsum(equation = var_3477_equation_0, values = (var_3319_cast_fp16, var_3236_cast_fp16))[name = tensor("op_3477_cast_fp16")]; + tensor var_3478_to_fp16 = const()[name = tensor("op_3478_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_219_cast_fp16 = mul(x = var_3477_cast_fp16, y = var_3478_to_fp16)[name = tensor("aw_219_cast_fp16")]; + tensor var_3481_equation_0 = const()[name = tensor("op_3481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3481_cast_fp16 = einsum(equation = var_3481_equation_0, values = (var_3323_cast_fp16, var_3240_cast_fp16))[name = tensor("op_3481_cast_fp16")]; + tensor var_3482_to_fp16 = const()[name = tensor("op_3482_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_221_cast_fp16 = mul(x = var_3481_cast_fp16, y = var_3482_to_fp16)[name = tensor("aw_221_cast_fp16")]; + tensor var_3485_equation_0 = const()[name = tensor("op_3485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3485_cast_fp16 = einsum(equation = var_3485_equation_0, values = (var_3327_cast_fp16, var_3244_cast_fp16))[name = tensor("op_3485_cast_fp16")]; + tensor var_3486_to_fp16 = const()[name = tensor("op_3486_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_223_cast_fp16 = mul(x = var_3485_cast_fp16, y = var_3486_to_fp16)[name = tensor("aw_223_cast_fp16")]; + tensor var_3489_equation_0 = const()[name = tensor("op_3489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3489_cast_fp16 = einsum(equation = var_3489_equation_0, values = (var_3331_cast_fp16, var_3248_cast_fp16))[name = tensor("op_3489_cast_fp16")]; + tensor var_3490_to_fp16 = const()[name = tensor("op_3490_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_225_cast_fp16 = mul(x = var_3489_cast_fp16, y = var_3490_to_fp16)[name = tensor("aw_225_cast_fp16")]; + tensor var_3493_equation_0 = const()[name = tensor("op_3493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3493_cast_fp16 = einsum(equation = var_3493_equation_0, values = (var_3335_cast_fp16, var_3252_cast_fp16))[name = tensor("op_3493_cast_fp16")]; + tensor var_3494_to_fp16 = const()[name = tensor("op_3494_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_227_cast_fp16 = mul(x = var_3493_cast_fp16, y = var_3494_to_fp16)[name = tensor("aw_227_cast_fp16")]; + tensor var_3497_equation_0 = const()[name = tensor("op_3497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3497_cast_fp16 = einsum(equation = var_3497_equation_0, values = (var_3339_cast_fp16, var_3256_cast_fp16))[name = tensor("op_3497_cast_fp16")]; + tensor var_3498_to_fp16 = const()[name = tensor("op_3498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_229_cast_fp16 = mul(x = var_3497_cast_fp16, y = var_3498_to_fp16)[name = tensor("aw_229_cast_fp16")]; + tensor var_3501_equation_0 = const()[name = tensor("op_3501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3501_cast_fp16 = einsum(equation = var_3501_equation_0, values = (var_3343_cast_fp16, var_3260_cast_fp16))[name = tensor("op_3501_cast_fp16")]; + tensor var_3502_to_fp16 = const()[name = tensor("op_3502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_231_cast_fp16 = mul(x = var_3501_cast_fp16, y = var_3502_to_fp16)[name = tensor("aw_231_cast_fp16")]; + tensor var_3505_equation_0 = const()[name = tensor("op_3505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3505_cast_fp16 = einsum(equation = var_3505_equation_0, values = (var_3347_cast_fp16, var_3264_cast_fp16))[name = tensor("op_3505_cast_fp16")]; + tensor var_3506_to_fp16 = const()[name = tensor("op_3506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_233_cast_fp16 = mul(x = var_3505_cast_fp16, y = var_3506_to_fp16)[name = tensor("aw_233_cast_fp16")]; + tensor var_3509_equation_0 = const()[name = tensor("op_3509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3509_cast_fp16 = einsum(equation = var_3509_equation_0, values = (var_3351_cast_fp16, var_3268_cast_fp16))[name = tensor("op_3509_cast_fp16")]; + tensor var_3510_to_fp16 = const()[name = tensor("op_3510_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_235_cast_fp16 = mul(x = var_3509_cast_fp16, y = var_3510_to_fp16)[name = tensor("aw_235_cast_fp16")]; + tensor var_3513_equation_0 = const()[name = tensor("op_3513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3513_cast_fp16 = einsum(equation = var_3513_equation_0, values = (var_3355_cast_fp16, var_3272_cast_fp16))[name = tensor("op_3513_cast_fp16")]; + tensor var_3514_to_fp16 = const()[name = tensor("op_3514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_237_cast_fp16 = mul(x = var_3513_cast_fp16, y = var_3514_to_fp16)[name = tensor("aw_237_cast_fp16")]; + tensor var_3517_equation_0 = const()[name = tensor("op_3517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3517_cast_fp16 = einsum(equation = var_3517_equation_0, values = (var_3359_cast_fp16, var_3276_cast_fp16))[name = tensor("op_3517_cast_fp16")]; + tensor var_3518_to_fp16 = const()[name = tensor("op_3518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_239_cast_fp16 = mul(x = var_3517_cast_fp16, y = var_3518_to_fp16)[name = tensor("aw_239_cast_fp16")]; + tensor var_3520_cast_fp16 = softmax(axis = var_2624, x = aw_201_cast_fp16)[name = tensor("op_3520_cast_fp16")]; + tensor var_3521_cast_fp16 = softmax(axis = var_2624, x = aw_203_cast_fp16)[name = tensor("op_3521_cast_fp16")]; + tensor var_3522_cast_fp16 = softmax(axis = var_2624, x = aw_205_cast_fp16)[name = tensor("op_3522_cast_fp16")]; + tensor var_3523_cast_fp16 = softmax(axis = var_2624, x = aw_207_cast_fp16)[name = tensor("op_3523_cast_fp16")]; + tensor var_3524_cast_fp16 = softmax(axis = var_2624, x = aw_209_cast_fp16)[name = tensor("op_3524_cast_fp16")]; + tensor var_3525_cast_fp16 = softmax(axis = var_2624, x = aw_211_cast_fp16)[name = tensor("op_3525_cast_fp16")]; + tensor var_3526_cast_fp16 = softmax(axis = var_2624, x = aw_213_cast_fp16)[name = tensor("op_3526_cast_fp16")]; + tensor var_3527_cast_fp16 = softmax(axis = var_2624, x = aw_215_cast_fp16)[name = tensor("op_3527_cast_fp16")]; + tensor var_3528_cast_fp16 = softmax(axis = var_2624, x = aw_217_cast_fp16)[name = tensor("op_3528_cast_fp16")]; + tensor var_3529_cast_fp16 = softmax(axis = var_2624, x = aw_219_cast_fp16)[name = tensor("op_3529_cast_fp16")]; + tensor var_3530_cast_fp16 = softmax(axis = var_2624, x = aw_221_cast_fp16)[name = tensor("op_3530_cast_fp16")]; + tensor var_3531_cast_fp16 = softmax(axis = var_2624, x = aw_223_cast_fp16)[name = tensor("op_3531_cast_fp16")]; + tensor var_3532_cast_fp16 = softmax(axis = var_2624, x = aw_225_cast_fp16)[name = tensor("op_3532_cast_fp16")]; + tensor var_3533_cast_fp16 = softmax(axis = var_2624, x = aw_227_cast_fp16)[name = tensor("op_3533_cast_fp16")]; + tensor var_3534_cast_fp16 = softmax(axis = var_2624, x = aw_229_cast_fp16)[name = tensor("op_3534_cast_fp16")]; + tensor var_3535_cast_fp16 = softmax(axis = var_2624, x = aw_231_cast_fp16)[name = tensor("op_3535_cast_fp16")]; + tensor var_3536_cast_fp16 = softmax(axis = var_2624, x = aw_233_cast_fp16)[name = tensor("op_3536_cast_fp16")]; + tensor var_3537_cast_fp16 = softmax(axis = var_2624, x = aw_235_cast_fp16)[name = tensor("op_3537_cast_fp16")]; + tensor var_3538_cast_fp16 = softmax(axis = var_2624, x = aw_237_cast_fp16)[name = tensor("op_3538_cast_fp16")]; + tensor var_3539_cast_fp16 = softmax(axis = var_2624, x = aw_239_cast_fp16)[name = tensor("op_3539_cast_fp16")]; + tensor var_3541_equation_0 = const()[name = tensor("op_3541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3541_cast_fp16 = einsum(equation = var_3541_equation_0, values = (var_3361_cast_fp16, var_3520_cast_fp16))[name = tensor("op_3541_cast_fp16")]; + tensor var_3543_equation_0 = const()[name = tensor("op_3543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3543_cast_fp16 = einsum(equation = var_3543_equation_0, values = (var_3365_cast_fp16, var_3521_cast_fp16))[name = tensor("op_3543_cast_fp16")]; + tensor var_3545_equation_0 = const()[name = tensor("op_3545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3545_cast_fp16 = einsum(equation = var_3545_equation_0, values = (var_3369_cast_fp16, var_3522_cast_fp16))[name = tensor("op_3545_cast_fp16")]; + tensor var_3547_equation_0 = const()[name = tensor("op_3547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3547_cast_fp16 = einsum(equation = var_3547_equation_0, values = (var_3373_cast_fp16, var_3523_cast_fp16))[name = tensor("op_3547_cast_fp16")]; + tensor var_3549_equation_0 = const()[name = tensor("op_3549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3549_cast_fp16 = einsum(equation = var_3549_equation_0, values = (var_3377_cast_fp16, var_3524_cast_fp16))[name = tensor("op_3549_cast_fp16")]; + tensor var_3551_equation_0 = const()[name = tensor("op_3551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3551_cast_fp16 = einsum(equation = var_3551_equation_0, values = (var_3381_cast_fp16, var_3525_cast_fp16))[name = tensor("op_3551_cast_fp16")]; + tensor var_3553_equation_0 = const()[name = tensor("op_3553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3553_cast_fp16 = einsum(equation = var_3553_equation_0, values = (var_3385_cast_fp16, var_3526_cast_fp16))[name = tensor("op_3553_cast_fp16")]; + tensor var_3555_equation_0 = const()[name = tensor("op_3555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3555_cast_fp16 = einsum(equation = var_3555_equation_0, values = (var_3389_cast_fp16, var_3527_cast_fp16))[name = tensor("op_3555_cast_fp16")]; + tensor var_3557_equation_0 = const()[name = tensor("op_3557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3557_cast_fp16 = einsum(equation = var_3557_equation_0, values = (var_3393_cast_fp16, var_3528_cast_fp16))[name = tensor("op_3557_cast_fp16")]; + tensor var_3559_equation_0 = const()[name = tensor("op_3559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3559_cast_fp16 = einsum(equation = var_3559_equation_0, values = (var_3397_cast_fp16, var_3529_cast_fp16))[name = tensor("op_3559_cast_fp16")]; + tensor var_3561_equation_0 = const()[name = tensor("op_3561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3561_cast_fp16 = einsum(equation = var_3561_equation_0, values = (var_3401_cast_fp16, var_3530_cast_fp16))[name = tensor("op_3561_cast_fp16")]; + tensor var_3563_equation_0 = const()[name = tensor("op_3563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3563_cast_fp16 = einsum(equation = var_3563_equation_0, values = (var_3405_cast_fp16, var_3531_cast_fp16))[name = tensor("op_3563_cast_fp16")]; + tensor var_3565_equation_0 = const()[name = tensor("op_3565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3565_cast_fp16 = einsum(equation = var_3565_equation_0, values = (var_3409_cast_fp16, var_3532_cast_fp16))[name = tensor("op_3565_cast_fp16")]; + tensor var_3567_equation_0 = const()[name = tensor("op_3567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3567_cast_fp16 = einsum(equation = var_3567_equation_0, values = (var_3413_cast_fp16, var_3533_cast_fp16))[name = tensor("op_3567_cast_fp16")]; + tensor var_3569_equation_0 = const()[name = tensor("op_3569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3569_cast_fp16 = einsum(equation = var_3569_equation_0, values = (var_3417_cast_fp16, var_3534_cast_fp16))[name = tensor("op_3569_cast_fp16")]; + tensor var_3571_equation_0 = const()[name = tensor("op_3571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3571_cast_fp16 = einsum(equation = var_3571_equation_0, values = (var_3421_cast_fp16, var_3535_cast_fp16))[name = tensor("op_3571_cast_fp16")]; + tensor var_3573_equation_0 = const()[name = tensor("op_3573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3573_cast_fp16 = einsum(equation = var_3573_equation_0, values = (var_3425_cast_fp16, var_3536_cast_fp16))[name = tensor("op_3573_cast_fp16")]; + tensor var_3575_equation_0 = const()[name = tensor("op_3575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3575_cast_fp16 = einsum(equation = var_3575_equation_0, values = (var_3429_cast_fp16, var_3537_cast_fp16))[name = tensor("op_3575_cast_fp16")]; + tensor var_3577_equation_0 = const()[name = tensor("op_3577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3577_cast_fp16 = einsum(equation = var_3577_equation_0, values = (var_3433_cast_fp16, var_3538_cast_fp16))[name = tensor("op_3577_cast_fp16")]; + tensor var_3579_equation_0 = const()[name = tensor("op_3579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3579_cast_fp16 = einsum(equation = var_3579_equation_0, values = (var_3437_cast_fp16, var_3539_cast_fp16))[name = tensor("op_3579_cast_fp16")]; + tensor input_133_interleave_0 = const()[name = tensor("input_133_interleave_0"), val = tensor(false)]; + tensor input_133_cast_fp16 = concat(axis = var_2624, interleave = input_133_interleave_0, values = (var_3541_cast_fp16, var_3543_cast_fp16, var_3545_cast_fp16, var_3547_cast_fp16, var_3549_cast_fp16, var_3551_cast_fp16, var_3553_cast_fp16, var_3555_cast_fp16, var_3557_cast_fp16, var_3559_cast_fp16, var_3561_cast_fp16, var_3563_cast_fp16, var_3565_cast_fp16, var_3567_cast_fp16, var_3569_cast_fp16, var_3571_cast_fp16, var_3573_cast_fp16, var_3575_cast_fp16, var_3577_cast_fp16, var_3579_cast_fp16))[name = tensor("input_133_cast_fp16")]; + tensor var_3589_pad_type_0 = const()[name = tensor("op_3589_pad_type_0"), val = tensor("valid")]; + tensor var_3589_strides_0 = const()[name = tensor("op_3589_strides_0"), val = tensor([1, 1])]; + tensor var_3589_pad_0 = const()[name = tensor("op_3589_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3589_dilations_0 = const()[name = tensor("op_3589_dilations_0"), val = tensor([1, 1])]; + tensor var_3589_groups_0 = const()[name = tensor("op_3589_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84531008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85759872))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85760064)))]; + tensor var_3589_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3589_dilations_0, groups = var_3589_groups_0, pad = var_3589_pad_0, pad_type = var_3589_pad_type_0, strides = var_3589_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = tensor("op_3589_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = var_3589_cast_fp16, y = inputs_27_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; + tensor input_135_axes_0 = const()[name = tensor("input_135_axes_0"), val = tensor([1])]; + tensor input_135_gamma_0_to_fp16 = const()[name = tensor("input_135_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85762688)))]; + tensor input_135_beta_0_to_fp16 = const()[name = tensor("input_135_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85765312)))]; + tensor var_3599_to_fp16 = const()[name = tensor("op_3599_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_135_cast_fp16 = layer_norm(axes = input_135_axes_0, beta = input_135_beta_0_to_fp16, epsilon = var_3599_to_fp16, gamma = input_135_gamma_0_to_fp16, x = inputs_29_cast_fp16)[name = tensor("input_135_cast_fp16")]; + tensor var_3619_pad_type_0 = const()[name = tensor("op_3619_pad_type_0"), val = tensor("valid")]; + tensor var_3619_strides_0 = const()[name = tensor("op_3619_strides_0"), val = tensor([1, 1])]; + tensor var_3619_pad_0 = const()[name = tensor("op_3619_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3619_dilations_0 = const()[name = tensor("op_3619_dilations_0"), val = tensor([1, 1])]; + tensor var_3619_groups_0 = const()[name = tensor("op_3619_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85767936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95598400))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95598592)))]; + tensor var_3619_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_3619_dilations_0, groups = var_3619_groups_0, pad = var_3619_pad_0, pad_type = var_3619_pad_type_0, strides = var_3619_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_135_cast_fp16)[name = tensor("op_3619_cast_fp16")]; + tensor var_3620_split_sizes_0 = const()[name = tensor("op_3620_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_3620_axis_0 = const()[name = tensor("op_3620_axis_0"), val = tensor(1)]; + tensor var_3620_cast_fp16_0, tensor var_3620_cast_fp16_1 = split(axis = var_3620_axis_0, split_sizes = var_3620_split_sizes_0, x = var_3619_cast_fp16)[name = tensor("op_3620_cast_fp16")]; + tensor var_3622_mode_0 = const()[name = tensor("op_3622_mode_0"), val = tensor("EXACT")]; + tensor var_3622_cast_fp16 = gelu(mode = var_3622_mode_0, x = var_3620_cast_fp16_1)[name = tensor("op_3622_cast_fp16")]; + tensor input_137_cast_fp16 = mul(x = var_3620_cast_fp16_0, y = var_3622_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor var_3630_pad_type_0 = const()[name = tensor("op_3630_pad_type_0"), val = tensor("valid")]; + tensor var_3630_strides_0 = const()[name = tensor("op_3630_strides_0"), val = tensor([1, 1])]; + tensor var_3630_pad_0 = const()[name = tensor("op_3630_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3630_dilations_0 = const()[name = tensor("op_3630_dilations_0"), val = tensor([1, 1])]; + tensor var_3630_groups_0 = const()[name = tensor("op_3630_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95619136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100534400))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100534592)))]; + tensor var_3630_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3630_dilations_0, groups = var_3630_groups_0, pad = var_3630_pad_0, pad_type = var_3630_pad_type_0, strides = var_3630_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor("op_3630_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = var_3630_cast_fp16, y = inputs_29_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; + tensor hidden_states_71_axes_0 = const()[name = tensor("hidden_states_71_axes_0"), val = tensor([1])]; + tensor hidden_states_71_gamma_0_to_fp16 = const()[name = tensor("hidden_states_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100537216)))]; + tensor hidden_states_71_beta_0_to_fp16 = const()[name = tensor("hidden_states_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100539840)))]; + tensor var_3646_to_fp16 = const()[name = tensor("op_3646_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_71_cast_fp16 = layer_norm(axes = hidden_states_71_axes_0, beta = hidden_states_71_beta_0_to_fp16, epsilon = var_3646_to_fp16, gamma = hidden_states_71_gamma_0_to_fp16, x = inputs_31_cast_fp16)[name = tensor("hidden_states_71_cast_fp16")]; + tensor q_21_pad_type_0 = const()[name = tensor("q_21_pad_type_0"), val = tensor("valid")]; + tensor q_21_strides_0 = const()[name = tensor("q_21_strides_0"), val = tensor([1, 1])]; + tensor q_21_pad_0 = const()[name = tensor("q_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_21_dilations_0 = const()[name = tensor("q_21_dilations_0"), val = tensor([1, 1])]; + tensor q_21_groups_0 = const()[name = tensor("q_21_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100542464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101771328))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_21_cast_fp16 = conv(dilations = q_21_dilations_0, groups = q_21_groups_0, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = q_21_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_71_cast_fp16)[name = tensor("q_21_cast_fp16")]; + tensor k_41_pad_type_0 = const()[name = tensor("k_41_pad_type_0"), val = tensor("valid")]; + tensor k_41_strides_0 = const()[name = tensor("k_41_strides_0"), val = tensor([1, 1])]; + tensor k_41_pad_0 = const()[name = tensor("k_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_41_dilations_0 = const()[name = tensor("k_41_dilations_0"), val = tensor([1, 1])]; + tensor k_41_groups_0 = const()[name = tensor("k_41_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101771520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103000384))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_41_cast_fp16 = conv(dilations = k_41_dilations_0, groups = k_41_groups_0, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = k_41_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_71_cast_fp16)[name = tensor("k_41_cast_fp16")]; + tensor v_21_pad_type_0 = const()[name = tensor("v_21_pad_type_0"), val = tensor("valid")]; + tensor v_21_strides_0 = const()[name = tensor("v_21_strides_0"), val = tensor([1, 1])]; + tensor v_21_pad_0 = const()[name = tensor("v_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_21_dilations_0 = const()[name = tensor("v_21_dilations_0"), val = tensor([1, 1])]; + tensor v_21_groups_0 = const()[name = tensor("v_21_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103000576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104229440))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_21_cast_fp16 = conv(dilations = v_21_dilations_0, groups = v_21_groups_0, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = v_21_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_71_cast_fp16)[name = tensor("v_21_cast_fp16")]; + tensor var_3679_begin_0 = const()[name = tensor("op_3679_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3679_end_0 = const()[name = tensor("op_3679_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3679_end_mask_0 = const()[name = tensor("op_3679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3679_cast_fp16 = slice_by_index(begin = var_3679_begin_0, end = var_3679_end_0, end_mask = var_3679_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3679_cast_fp16")]; + tensor var_3683_begin_0 = const()[name = tensor("op_3683_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3683_end_0 = const()[name = tensor("op_3683_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3683_end_mask_0 = const()[name = tensor("op_3683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3683_cast_fp16 = slice_by_index(begin = var_3683_begin_0, end = var_3683_end_0, end_mask = var_3683_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3683_cast_fp16")]; + tensor var_3687_begin_0 = const()[name = tensor("op_3687_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3687_end_0 = const()[name = tensor("op_3687_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3687_end_mask_0 = const()[name = tensor("op_3687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3687_cast_fp16 = slice_by_index(begin = var_3687_begin_0, end = var_3687_end_0, end_mask = var_3687_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3687_cast_fp16")]; + tensor var_3691_begin_0 = const()[name = tensor("op_3691_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3691_end_0 = const()[name = tensor("op_3691_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3691_end_mask_0 = const()[name = tensor("op_3691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3691_cast_fp16 = slice_by_index(begin = var_3691_begin_0, end = var_3691_end_0, end_mask = var_3691_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3691_cast_fp16")]; + tensor var_3695_begin_0 = const()[name = tensor("op_3695_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3695_end_0 = const()[name = tensor("op_3695_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3695_end_mask_0 = const()[name = tensor("op_3695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3695_cast_fp16 = slice_by_index(begin = var_3695_begin_0, end = var_3695_end_0, end_mask = var_3695_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3695_cast_fp16")]; + tensor var_3699_begin_0 = const()[name = tensor("op_3699_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3699_end_0 = const()[name = tensor("op_3699_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3699_end_mask_0 = const()[name = tensor("op_3699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3699_cast_fp16 = slice_by_index(begin = var_3699_begin_0, end = var_3699_end_0, end_mask = var_3699_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3699_cast_fp16")]; + tensor var_3703_begin_0 = const()[name = tensor("op_3703_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3703_end_0 = const()[name = tensor("op_3703_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3703_end_mask_0 = const()[name = tensor("op_3703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3703_cast_fp16 = slice_by_index(begin = var_3703_begin_0, end = var_3703_end_0, end_mask = var_3703_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3703_cast_fp16")]; + tensor var_3707_begin_0 = const()[name = tensor("op_3707_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3707_end_0 = const()[name = tensor("op_3707_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3707_end_mask_0 = const()[name = tensor("op_3707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3707_cast_fp16 = slice_by_index(begin = var_3707_begin_0, end = var_3707_end_0, end_mask = var_3707_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3707_cast_fp16")]; + tensor var_3711_begin_0 = const()[name = tensor("op_3711_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3711_end_0 = const()[name = tensor("op_3711_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3711_end_mask_0 = const()[name = tensor("op_3711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3711_cast_fp16 = slice_by_index(begin = var_3711_begin_0, end = var_3711_end_0, end_mask = var_3711_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3711_cast_fp16")]; + tensor var_3715_begin_0 = const()[name = tensor("op_3715_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3715_end_0 = const()[name = tensor("op_3715_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3715_end_mask_0 = const()[name = tensor("op_3715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3715_cast_fp16 = slice_by_index(begin = var_3715_begin_0, end = var_3715_end_0, end_mask = var_3715_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3715_cast_fp16")]; + tensor var_3719_begin_0 = const()[name = tensor("op_3719_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3719_end_0 = const()[name = tensor("op_3719_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_3719_end_mask_0 = const()[name = tensor("op_3719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3719_cast_fp16 = slice_by_index(begin = var_3719_begin_0, end = var_3719_end_0, end_mask = var_3719_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3719_cast_fp16")]; + tensor var_3723_begin_0 = const()[name = tensor("op_3723_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3723_end_0 = const()[name = tensor("op_3723_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_3723_end_mask_0 = const()[name = tensor("op_3723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3723_cast_fp16 = slice_by_index(begin = var_3723_begin_0, end = var_3723_end_0, end_mask = var_3723_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3723_cast_fp16")]; + tensor var_3727_begin_0 = const()[name = tensor("op_3727_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3727_end_0 = const()[name = tensor("op_3727_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_3727_end_mask_0 = const()[name = tensor("op_3727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3727_cast_fp16 = slice_by_index(begin = var_3727_begin_0, end = var_3727_end_0, end_mask = var_3727_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3727_cast_fp16")]; + tensor var_3731_begin_0 = const()[name = tensor("op_3731_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3731_end_0 = const()[name = tensor("op_3731_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_3731_end_mask_0 = const()[name = tensor("op_3731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3731_cast_fp16 = slice_by_index(begin = var_3731_begin_0, end = var_3731_end_0, end_mask = var_3731_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3731_cast_fp16")]; + tensor var_3735_begin_0 = const()[name = tensor("op_3735_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3735_end_0 = const()[name = tensor("op_3735_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_3735_end_mask_0 = const()[name = tensor("op_3735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3735_cast_fp16 = slice_by_index(begin = var_3735_begin_0, end = var_3735_end_0, end_mask = var_3735_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3735_cast_fp16")]; + tensor var_3739_begin_0 = const()[name = tensor("op_3739_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3739_end_0 = const()[name = tensor("op_3739_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3739_end_mask_0 = const()[name = tensor("op_3739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3739_cast_fp16 = slice_by_index(begin = var_3739_begin_0, end = var_3739_end_0, end_mask = var_3739_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3739_cast_fp16")]; + tensor var_3743_begin_0 = const()[name = tensor("op_3743_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3743_end_0 = const()[name = tensor("op_3743_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_3743_end_mask_0 = const()[name = tensor("op_3743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3743_cast_fp16 = slice_by_index(begin = var_3743_begin_0, end = var_3743_end_0, end_mask = var_3743_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3743_cast_fp16")]; + tensor var_3747_begin_0 = const()[name = tensor("op_3747_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3747_end_0 = const()[name = tensor("op_3747_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_3747_end_mask_0 = const()[name = tensor("op_3747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3747_cast_fp16 = slice_by_index(begin = var_3747_begin_0, end = var_3747_end_0, end_mask = var_3747_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3747_cast_fp16")]; + tensor var_3751_begin_0 = const()[name = tensor("op_3751_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3751_end_0 = const()[name = tensor("op_3751_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3751_end_mask_0 = const()[name = tensor("op_3751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3751_cast_fp16 = slice_by_index(begin = var_3751_begin_0, end = var_3751_end_0, end_mask = var_3751_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3751_cast_fp16")]; + tensor var_3755_begin_0 = const()[name = tensor("op_3755_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3755_end_0 = const()[name = tensor("op_3755_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3755_end_mask_0 = const()[name = tensor("op_3755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3755_cast_fp16 = slice_by_index(begin = var_3755_begin_0, end = var_3755_end_0, end_mask = var_3755_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3755_cast_fp16")]; + tensor k_43_perm_0 = const()[name = tensor("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3762_begin_0 = const()[name = tensor("op_3762_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3762_end_0 = const()[name = tensor("op_3762_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_3762_end_mask_0 = const()[name = tensor("op_3762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_43_cast_fp16 = transpose(perm = k_43_perm_0, x = k_41_cast_fp16)[name = tensor("transpose_57")]; + tensor var_3762_cast_fp16 = slice_by_index(begin = var_3762_begin_0, end = var_3762_end_0, end_mask = var_3762_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3762_cast_fp16")]; + tensor var_3766_begin_0 = const()[name = tensor("op_3766_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3766_end_0 = const()[name = tensor("op_3766_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_3766_end_mask_0 = const()[name = tensor("op_3766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3766_cast_fp16 = slice_by_index(begin = var_3766_begin_0, end = var_3766_end_0, end_mask = var_3766_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3766_cast_fp16")]; + tensor var_3770_begin_0 = const()[name = tensor("op_3770_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3770_end_0 = const()[name = tensor("op_3770_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_3770_end_mask_0 = const()[name = tensor("op_3770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3770_cast_fp16 = slice_by_index(begin = var_3770_begin_0, end = var_3770_end_0, end_mask = var_3770_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3770_cast_fp16")]; + tensor var_3774_begin_0 = const()[name = tensor("op_3774_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3774_end_0 = const()[name = tensor("op_3774_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_3774_end_mask_0 = const()[name = tensor("op_3774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3774_cast_fp16 = slice_by_index(begin = var_3774_begin_0, end = var_3774_end_0, end_mask = var_3774_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3774_cast_fp16")]; + tensor var_3778_begin_0 = const()[name = tensor("op_3778_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3778_end_0 = const()[name = tensor("op_3778_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_3778_end_mask_0 = const()[name = tensor("op_3778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3778_cast_fp16 = slice_by_index(begin = var_3778_begin_0, end = var_3778_end_0, end_mask = var_3778_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3778_cast_fp16")]; + tensor var_3782_begin_0 = const()[name = tensor("op_3782_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3782_end_0 = const()[name = tensor("op_3782_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_3782_end_mask_0 = const()[name = tensor("op_3782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3782_cast_fp16 = slice_by_index(begin = var_3782_begin_0, end = var_3782_end_0, end_mask = var_3782_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3782_cast_fp16")]; + tensor var_3786_begin_0 = const()[name = tensor("op_3786_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3786_end_0 = const()[name = tensor("op_3786_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_3786_end_mask_0 = const()[name = tensor("op_3786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3786_cast_fp16 = slice_by_index(begin = var_3786_begin_0, end = var_3786_end_0, end_mask = var_3786_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3786_cast_fp16")]; + tensor var_3790_begin_0 = const()[name = tensor("op_3790_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3790_end_0 = const()[name = tensor("op_3790_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_3790_end_mask_0 = const()[name = tensor("op_3790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3790_cast_fp16 = slice_by_index(begin = var_3790_begin_0, end = var_3790_end_0, end_mask = var_3790_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3790_cast_fp16")]; + tensor var_3794_begin_0 = const()[name = tensor("op_3794_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3794_end_0 = const()[name = tensor("op_3794_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_3794_end_mask_0 = const()[name = tensor("op_3794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3794_cast_fp16 = slice_by_index(begin = var_3794_begin_0, end = var_3794_end_0, end_mask = var_3794_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3794_cast_fp16")]; + tensor var_3798_begin_0 = const()[name = tensor("op_3798_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3798_end_0 = const()[name = tensor("op_3798_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_3798_end_mask_0 = const()[name = tensor("op_3798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3798_cast_fp16 = slice_by_index(begin = var_3798_begin_0, end = var_3798_end_0, end_mask = var_3798_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3798_cast_fp16")]; + tensor var_3802_begin_0 = const()[name = tensor("op_3802_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3802_end_0 = const()[name = tensor("op_3802_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_3802_end_mask_0 = const()[name = tensor("op_3802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3802_cast_fp16 = slice_by_index(begin = var_3802_begin_0, end = var_3802_end_0, end_mask = var_3802_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3802_cast_fp16")]; + tensor var_3806_begin_0 = const()[name = tensor("op_3806_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_3806_end_0 = const()[name = tensor("op_3806_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_3806_end_mask_0 = const()[name = tensor("op_3806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3806_cast_fp16 = slice_by_index(begin = var_3806_begin_0, end = var_3806_end_0, end_mask = var_3806_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3806_cast_fp16")]; + tensor var_3810_begin_0 = const()[name = tensor("op_3810_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_3810_end_0 = const()[name = tensor("op_3810_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_3810_end_mask_0 = const()[name = tensor("op_3810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3810_cast_fp16 = slice_by_index(begin = var_3810_begin_0, end = var_3810_end_0, end_mask = var_3810_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3810_cast_fp16")]; + tensor var_3814_begin_0 = const()[name = tensor("op_3814_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_3814_end_0 = const()[name = tensor("op_3814_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_3814_end_mask_0 = const()[name = tensor("op_3814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3814_cast_fp16 = slice_by_index(begin = var_3814_begin_0, end = var_3814_end_0, end_mask = var_3814_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3814_cast_fp16")]; + tensor var_3818_begin_0 = const()[name = tensor("op_3818_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_3818_end_0 = const()[name = tensor("op_3818_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_3818_end_mask_0 = const()[name = tensor("op_3818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3818_cast_fp16 = slice_by_index(begin = var_3818_begin_0, end = var_3818_end_0, end_mask = var_3818_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3818_cast_fp16")]; + tensor var_3822_begin_0 = const()[name = tensor("op_3822_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_3822_end_0 = const()[name = tensor("op_3822_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3822_end_mask_0 = const()[name = tensor("op_3822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3822_cast_fp16 = slice_by_index(begin = var_3822_begin_0, end = var_3822_end_0, end_mask = var_3822_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3822_cast_fp16")]; + tensor var_3826_begin_0 = const()[name = tensor("op_3826_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_3826_end_0 = const()[name = tensor("op_3826_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_3826_end_mask_0 = const()[name = tensor("op_3826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3826_cast_fp16 = slice_by_index(begin = var_3826_begin_0, end = var_3826_end_0, end_mask = var_3826_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3826_cast_fp16")]; + tensor var_3830_begin_0 = const()[name = tensor("op_3830_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_3830_end_0 = const()[name = tensor("op_3830_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_3830_end_mask_0 = const()[name = tensor("op_3830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3830_cast_fp16 = slice_by_index(begin = var_3830_begin_0, end = var_3830_end_0, end_mask = var_3830_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3830_cast_fp16")]; + tensor var_3834_begin_0 = const()[name = tensor("op_3834_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_3834_end_0 = const()[name = tensor("op_3834_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_3834_end_mask_0 = const()[name = tensor("op_3834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3834_cast_fp16 = slice_by_index(begin = var_3834_begin_0, end = var_3834_end_0, end_mask = var_3834_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3834_cast_fp16")]; + tensor var_3838_begin_0 = const()[name = tensor("op_3838_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_3838_end_0 = const()[name = tensor("op_3838_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_3838_end_mask_0 = const()[name = tensor("op_3838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3838_cast_fp16 = slice_by_index(begin = var_3838_begin_0, end = var_3838_end_0, end_mask = var_3838_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3838_cast_fp16")]; + tensor var_3840_begin_0 = const()[name = tensor("op_3840_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3840_end_0 = const()[name = tensor("op_3840_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3840_end_mask_0 = const()[name = tensor("op_3840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3840_cast_fp16 = slice_by_index(begin = var_3840_begin_0, end = var_3840_end_0, end_mask = var_3840_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3840_cast_fp16")]; + tensor var_3844_begin_0 = const()[name = tensor("op_3844_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3844_end_0 = const()[name = tensor("op_3844_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3844_end_mask_0 = const()[name = tensor("op_3844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3844_cast_fp16 = slice_by_index(begin = var_3844_begin_0, end = var_3844_end_0, end_mask = var_3844_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3844_cast_fp16")]; + tensor var_3848_begin_0 = const()[name = tensor("op_3848_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3848_end_0 = const()[name = tensor("op_3848_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3848_end_mask_0 = const()[name = tensor("op_3848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3848_cast_fp16 = slice_by_index(begin = var_3848_begin_0, end = var_3848_end_0, end_mask = var_3848_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3848_cast_fp16")]; + tensor var_3852_begin_0 = const()[name = tensor("op_3852_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3852_end_0 = const()[name = tensor("op_3852_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3852_end_mask_0 = const()[name = tensor("op_3852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3852_cast_fp16")]; + tensor var_3856_begin_0 = const()[name = tensor("op_3856_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3856_end_0 = const()[name = tensor("op_3856_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3856_end_mask_0 = const()[name = tensor("op_3856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3856_cast_fp16")]; + tensor var_3860_begin_0 = const()[name = tensor("op_3860_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3860_end_0 = const()[name = tensor("op_3860_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3860_end_mask_0 = const()[name = tensor("op_3860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3860_cast_fp16 = slice_by_index(begin = var_3860_begin_0, end = var_3860_end_0, end_mask = var_3860_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3860_cast_fp16")]; + tensor var_3864_begin_0 = const()[name = tensor("op_3864_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3864_end_0 = const()[name = tensor("op_3864_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3864_end_mask_0 = const()[name = tensor("op_3864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3864_cast_fp16 = slice_by_index(begin = var_3864_begin_0, end = var_3864_end_0, end_mask = var_3864_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3864_cast_fp16")]; + tensor var_3868_begin_0 = const()[name = tensor("op_3868_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3868_end_0 = const()[name = tensor("op_3868_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3868_end_mask_0 = const()[name = tensor("op_3868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3868_cast_fp16 = slice_by_index(begin = var_3868_begin_0, end = var_3868_end_0, end_mask = var_3868_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3868_cast_fp16")]; + tensor var_3872_begin_0 = const()[name = tensor("op_3872_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3872_end_0 = const()[name = tensor("op_3872_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3872_end_mask_0 = const()[name = tensor("op_3872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3872_cast_fp16 = slice_by_index(begin = var_3872_begin_0, end = var_3872_end_0, end_mask = var_3872_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3872_cast_fp16")]; + tensor var_3876_begin_0 = const()[name = tensor("op_3876_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3876_end_0 = const()[name = tensor("op_3876_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3876_end_mask_0 = const()[name = tensor("op_3876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3876_cast_fp16 = slice_by_index(begin = var_3876_begin_0, end = var_3876_end_0, end_mask = var_3876_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3876_cast_fp16")]; + tensor var_3880_begin_0 = const()[name = tensor("op_3880_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3880_end_0 = const()[name = tensor("op_3880_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_3880_end_mask_0 = const()[name = tensor("op_3880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3880_cast_fp16 = slice_by_index(begin = var_3880_begin_0, end = var_3880_end_0, end_mask = var_3880_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3880_cast_fp16")]; + tensor var_3884_begin_0 = const()[name = tensor("op_3884_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3884_end_0 = const()[name = tensor("op_3884_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_3884_end_mask_0 = const()[name = tensor("op_3884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3884_cast_fp16 = slice_by_index(begin = var_3884_begin_0, end = var_3884_end_0, end_mask = var_3884_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3884_cast_fp16")]; + tensor var_3888_begin_0 = const()[name = tensor("op_3888_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3888_end_0 = const()[name = tensor("op_3888_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_3888_end_mask_0 = const()[name = tensor("op_3888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3888_cast_fp16 = slice_by_index(begin = var_3888_begin_0, end = var_3888_end_0, end_mask = var_3888_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3888_cast_fp16")]; + tensor var_3892_begin_0 = const()[name = tensor("op_3892_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3892_end_0 = const()[name = tensor("op_3892_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_3892_end_mask_0 = const()[name = tensor("op_3892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3892_cast_fp16 = slice_by_index(begin = var_3892_begin_0, end = var_3892_end_0, end_mask = var_3892_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3892_cast_fp16")]; + tensor var_3896_begin_0 = const()[name = tensor("op_3896_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3896_end_0 = const()[name = tensor("op_3896_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_3896_end_mask_0 = const()[name = tensor("op_3896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3896_cast_fp16 = slice_by_index(begin = var_3896_begin_0, end = var_3896_end_0, end_mask = var_3896_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3896_cast_fp16")]; + tensor var_3900_begin_0 = const()[name = tensor("op_3900_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3900_end_0 = const()[name = tensor("op_3900_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3900_end_mask_0 = const()[name = tensor("op_3900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3900_cast_fp16 = slice_by_index(begin = var_3900_begin_0, end = var_3900_end_0, end_mask = var_3900_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3900_cast_fp16")]; + tensor var_3904_begin_0 = const()[name = tensor("op_3904_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3904_end_0 = const()[name = tensor("op_3904_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_3904_end_mask_0 = const()[name = tensor("op_3904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3904_cast_fp16 = slice_by_index(begin = var_3904_begin_0, end = var_3904_end_0, end_mask = var_3904_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3904_cast_fp16")]; + tensor var_3908_begin_0 = const()[name = tensor("op_3908_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3908_end_0 = const()[name = tensor("op_3908_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_3908_end_mask_0 = const()[name = tensor("op_3908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3908_cast_fp16 = slice_by_index(begin = var_3908_begin_0, end = var_3908_end_0, end_mask = var_3908_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3908_cast_fp16")]; + tensor var_3912_begin_0 = const()[name = tensor("op_3912_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3912_end_0 = const()[name = tensor("op_3912_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3912_end_mask_0 = const()[name = tensor("op_3912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3912_cast_fp16 = slice_by_index(begin = var_3912_begin_0, end = var_3912_end_0, end_mask = var_3912_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3912_cast_fp16")]; + tensor var_3916_begin_0 = const()[name = tensor("op_3916_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3916_end_0 = const()[name = tensor("op_3916_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3916_end_mask_0 = const()[name = tensor("op_3916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3916_cast_fp16 = slice_by_index(begin = var_3916_begin_0, end = var_3916_end_0, end_mask = var_3916_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3916_cast_fp16")]; + tensor var_3920_equation_0 = const()[name = tensor("op_3920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3920_cast_fp16 = einsum(equation = var_3920_equation_0, values = (var_3762_cast_fp16, var_3679_cast_fp16))[name = tensor("op_3920_cast_fp16")]; + tensor var_3921_to_fp16 = const()[name = tensor("op_3921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_241_cast_fp16 = mul(x = var_3920_cast_fp16, y = var_3921_to_fp16)[name = tensor("aw_241_cast_fp16")]; + tensor var_3924_equation_0 = const()[name = tensor("op_3924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3924_cast_fp16 = einsum(equation = var_3924_equation_0, values = (var_3766_cast_fp16, var_3683_cast_fp16))[name = tensor("op_3924_cast_fp16")]; + tensor var_3925_to_fp16 = const()[name = tensor("op_3925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_243_cast_fp16 = mul(x = var_3924_cast_fp16, y = var_3925_to_fp16)[name = tensor("aw_243_cast_fp16")]; + tensor var_3928_equation_0 = const()[name = tensor("op_3928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3928_cast_fp16 = einsum(equation = var_3928_equation_0, values = (var_3770_cast_fp16, var_3687_cast_fp16))[name = tensor("op_3928_cast_fp16")]; + tensor var_3929_to_fp16 = const()[name = tensor("op_3929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_245_cast_fp16 = mul(x = var_3928_cast_fp16, y = var_3929_to_fp16)[name = tensor("aw_245_cast_fp16")]; + tensor var_3932_equation_0 = const()[name = tensor("op_3932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3932_cast_fp16 = einsum(equation = var_3932_equation_0, values = (var_3774_cast_fp16, var_3691_cast_fp16))[name = tensor("op_3932_cast_fp16")]; + tensor var_3933_to_fp16 = const()[name = tensor("op_3933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_247_cast_fp16 = mul(x = var_3932_cast_fp16, y = var_3933_to_fp16)[name = tensor("aw_247_cast_fp16")]; + tensor var_3936_equation_0 = const()[name = tensor("op_3936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3936_cast_fp16 = einsum(equation = var_3936_equation_0, values = (var_3778_cast_fp16, var_3695_cast_fp16))[name = tensor("op_3936_cast_fp16")]; + tensor var_3937_to_fp16 = const()[name = tensor("op_3937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_249_cast_fp16 = mul(x = var_3936_cast_fp16, y = var_3937_to_fp16)[name = tensor("aw_249_cast_fp16")]; + tensor var_3940_equation_0 = const()[name = tensor("op_3940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3940_cast_fp16 = einsum(equation = var_3940_equation_0, values = (var_3782_cast_fp16, var_3699_cast_fp16))[name = tensor("op_3940_cast_fp16")]; + tensor var_3941_to_fp16 = const()[name = tensor("op_3941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_251_cast_fp16 = mul(x = var_3940_cast_fp16, y = var_3941_to_fp16)[name = tensor("aw_251_cast_fp16")]; + tensor var_3944_equation_0 = const()[name = tensor("op_3944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3944_cast_fp16 = einsum(equation = var_3944_equation_0, values = (var_3786_cast_fp16, var_3703_cast_fp16))[name = tensor("op_3944_cast_fp16")]; + tensor var_3945_to_fp16 = const()[name = tensor("op_3945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_253_cast_fp16 = mul(x = var_3944_cast_fp16, y = var_3945_to_fp16)[name = tensor("aw_253_cast_fp16")]; + tensor var_3948_equation_0 = const()[name = tensor("op_3948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3948_cast_fp16 = einsum(equation = var_3948_equation_0, values = (var_3790_cast_fp16, var_3707_cast_fp16))[name = tensor("op_3948_cast_fp16")]; + tensor var_3949_to_fp16 = const()[name = tensor("op_3949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_255_cast_fp16 = mul(x = var_3948_cast_fp16, y = var_3949_to_fp16)[name = tensor("aw_255_cast_fp16")]; + tensor var_3952_equation_0 = const()[name = tensor("op_3952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3952_cast_fp16 = einsum(equation = var_3952_equation_0, values = (var_3794_cast_fp16, var_3711_cast_fp16))[name = tensor("op_3952_cast_fp16")]; + tensor var_3953_to_fp16 = const()[name = tensor("op_3953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_257_cast_fp16 = mul(x = var_3952_cast_fp16, y = var_3953_to_fp16)[name = tensor("aw_257_cast_fp16")]; + tensor var_3956_equation_0 = const()[name = tensor("op_3956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3956_cast_fp16 = einsum(equation = var_3956_equation_0, values = (var_3798_cast_fp16, var_3715_cast_fp16))[name = tensor("op_3956_cast_fp16")]; + tensor var_3957_to_fp16 = const()[name = tensor("op_3957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_259_cast_fp16 = mul(x = var_3956_cast_fp16, y = var_3957_to_fp16)[name = tensor("aw_259_cast_fp16")]; + tensor var_3960_equation_0 = const()[name = tensor("op_3960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3960_cast_fp16 = einsum(equation = var_3960_equation_0, values = (var_3802_cast_fp16, var_3719_cast_fp16))[name = tensor("op_3960_cast_fp16")]; + tensor var_3961_to_fp16 = const()[name = tensor("op_3961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_261_cast_fp16 = mul(x = var_3960_cast_fp16, y = var_3961_to_fp16)[name = tensor("aw_261_cast_fp16")]; + tensor var_3964_equation_0 = const()[name = tensor("op_3964_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3964_cast_fp16 = einsum(equation = var_3964_equation_0, values = (var_3806_cast_fp16, var_3723_cast_fp16))[name = tensor("op_3964_cast_fp16")]; + tensor var_3965_to_fp16 = const()[name = tensor("op_3965_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_263_cast_fp16 = mul(x = var_3964_cast_fp16, y = var_3965_to_fp16)[name = tensor("aw_263_cast_fp16")]; + tensor var_3968_equation_0 = const()[name = tensor("op_3968_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3968_cast_fp16 = einsum(equation = var_3968_equation_0, values = (var_3810_cast_fp16, var_3727_cast_fp16))[name = tensor("op_3968_cast_fp16")]; + tensor var_3969_to_fp16 = const()[name = tensor("op_3969_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_265_cast_fp16 = mul(x = var_3968_cast_fp16, y = var_3969_to_fp16)[name = tensor("aw_265_cast_fp16")]; + tensor var_3972_equation_0 = const()[name = tensor("op_3972_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3972_cast_fp16 = einsum(equation = var_3972_equation_0, values = (var_3814_cast_fp16, var_3731_cast_fp16))[name = tensor("op_3972_cast_fp16")]; + tensor var_3973_to_fp16 = const()[name = tensor("op_3973_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_267_cast_fp16 = mul(x = var_3972_cast_fp16, y = var_3973_to_fp16)[name = tensor("aw_267_cast_fp16")]; + tensor var_3976_equation_0 = const()[name = tensor("op_3976_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3976_cast_fp16 = einsum(equation = var_3976_equation_0, values = (var_3818_cast_fp16, var_3735_cast_fp16))[name = tensor("op_3976_cast_fp16")]; + tensor var_3977_to_fp16 = const()[name = tensor("op_3977_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_269_cast_fp16 = mul(x = var_3976_cast_fp16, y = var_3977_to_fp16)[name = tensor("aw_269_cast_fp16")]; + tensor var_3980_equation_0 = const()[name = tensor("op_3980_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3980_cast_fp16 = einsum(equation = var_3980_equation_0, values = (var_3822_cast_fp16, var_3739_cast_fp16))[name = tensor("op_3980_cast_fp16")]; + tensor var_3981_to_fp16 = const()[name = tensor("op_3981_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_271_cast_fp16 = mul(x = var_3980_cast_fp16, y = var_3981_to_fp16)[name = tensor("aw_271_cast_fp16")]; + tensor var_3984_equation_0 = const()[name = tensor("op_3984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3984_cast_fp16 = einsum(equation = var_3984_equation_0, values = (var_3826_cast_fp16, var_3743_cast_fp16))[name = tensor("op_3984_cast_fp16")]; + tensor var_3985_to_fp16 = const()[name = tensor("op_3985_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_273_cast_fp16 = mul(x = var_3984_cast_fp16, y = var_3985_to_fp16)[name = tensor("aw_273_cast_fp16")]; + tensor var_3988_equation_0 = const()[name = tensor("op_3988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3988_cast_fp16 = einsum(equation = var_3988_equation_0, values = (var_3830_cast_fp16, var_3747_cast_fp16))[name = tensor("op_3988_cast_fp16")]; + tensor var_3989_to_fp16 = const()[name = tensor("op_3989_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_275_cast_fp16 = mul(x = var_3988_cast_fp16, y = var_3989_to_fp16)[name = tensor("aw_275_cast_fp16")]; + tensor var_3992_equation_0 = const()[name = tensor("op_3992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3992_cast_fp16 = einsum(equation = var_3992_equation_0, values = (var_3834_cast_fp16, var_3751_cast_fp16))[name = tensor("op_3992_cast_fp16")]; + tensor var_3993_to_fp16 = const()[name = tensor("op_3993_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_277_cast_fp16 = mul(x = var_3992_cast_fp16, y = var_3993_to_fp16)[name = tensor("aw_277_cast_fp16")]; + tensor var_3996_equation_0 = const()[name = tensor("op_3996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3996_cast_fp16 = einsum(equation = var_3996_equation_0, values = (var_3838_cast_fp16, var_3755_cast_fp16))[name = tensor("op_3996_cast_fp16")]; + tensor var_3997_to_fp16 = const()[name = tensor("op_3997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_279_cast_fp16 = mul(x = var_3996_cast_fp16, y = var_3997_to_fp16)[name = tensor("aw_279_cast_fp16")]; + tensor var_3999_cast_fp16 = softmax(axis = var_2624, x = aw_241_cast_fp16)[name = tensor("op_3999_cast_fp16")]; + tensor var_4000_cast_fp16 = softmax(axis = var_2624, x = aw_243_cast_fp16)[name = tensor("op_4000_cast_fp16")]; + tensor var_4001_cast_fp16 = softmax(axis = var_2624, x = aw_245_cast_fp16)[name = tensor("op_4001_cast_fp16")]; + tensor var_4002_cast_fp16 = softmax(axis = var_2624, x = aw_247_cast_fp16)[name = tensor("op_4002_cast_fp16")]; + tensor var_4003_cast_fp16 = softmax(axis = var_2624, x = aw_249_cast_fp16)[name = tensor("op_4003_cast_fp16")]; + tensor var_4004_cast_fp16 = softmax(axis = var_2624, x = aw_251_cast_fp16)[name = tensor("op_4004_cast_fp16")]; + tensor var_4005_cast_fp16 = softmax(axis = var_2624, x = aw_253_cast_fp16)[name = tensor("op_4005_cast_fp16")]; + tensor var_4006_cast_fp16 = softmax(axis = var_2624, x = aw_255_cast_fp16)[name = tensor("op_4006_cast_fp16")]; + tensor var_4007_cast_fp16 = softmax(axis = var_2624, x = aw_257_cast_fp16)[name = tensor("op_4007_cast_fp16")]; + tensor var_4008_cast_fp16 = softmax(axis = var_2624, x = aw_259_cast_fp16)[name = tensor("op_4008_cast_fp16")]; + tensor var_4009_cast_fp16 = softmax(axis = var_2624, x = aw_261_cast_fp16)[name = tensor("op_4009_cast_fp16")]; + tensor var_4010_cast_fp16 = softmax(axis = var_2624, x = aw_263_cast_fp16)[name = tensor("op_4010_cast_fp16")]; + tensor var_4011_cast_fp16 = softmax(axis = var_2624, x = aw_265_cast_fp16)[name = tensor("op_4011_cast_fp16")]; + tensor var_4012_cast_fp16 = softmax(axis = var_2624, x = aw_267_cast_fp16)[name = tensor("op_4012_cast_fp16")]; + tensor var_4013_cast_fp16 = softmax(axis = var_2624, x = aw_269_cast_fp16)[name = tensor("op_4013_cast_fp16")]; + tensor var_4014_cast_fp16 = softmax(axis = var_2624, x = aw_271_cast_fp16)[name = tensor("op_4014_cast_fp16")]; + tensor var_4015_cast_fp16 = softmax(axis = var_2624, x = aw_273_cast_fp16)[name = tensor("op_4015_cast_fp16")]; + tensor var_4016_cast_fp16 = softmax(axis = var_2624, x = aw_275_cast_fp16)[name = tensor("op_4016_cast_fp16")]; + tensor var_4017_cast_fp16 = softmax(axis = var_2624, x = aw_277_cast_fp16)[name = tensor("op_4017_cast_fp16")]; + tensor var_4018_cast_fp16 = softmax(axis = var_2624, x = aw_279_cast_fp16)[name = tensor("op_4018_cast_fp16")]; + tensor var_4020_equation_0 = const()[name = tensor("op_4020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4020_cast_fp16 = einsum(equation = var_4020_equation_0, values = (var_3840_cast_fp16, var_3999_cast_fp16))[name = tensor("op_4020_cast_fp16")]; + tensor var_4022_equation_0 = const()[name = tensor("op_4022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4022_cast_fp16 = einsum(equation = var_4022_equation_0, values = (var_3844_cast_fp16, var_4000_cast_fp16))[name = tensor("op_4022_cast_fp16")]; + tensor var_4024_equation_0 = const()[name = tensor("op_4024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4024_cast_fp16 = einsum(equation = var_4024_equation_0, values = (var_3848_cast_fp16, var_4001_cast_fp16))[name = tensor("op_4024_cast_fp16")]; + tensor var_4026_equation_0 = const()[name = tensor("op_4026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4026_cast_fp16 = einsum(equation = var_4026_equation_0, values = (var_3852_cast_fp16, var_4002_cast_fp16))[name = tensor("op_4026_cast_fp16")]; + tensor var_4028_equation_0 = const()[name = tensor("op_4028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4028_cast_fp16 = einsum(equation = var_4028_equation_0, values = (var_3856_cast_fp16, var_4003_cast_fp16))[name = tensor("op_4028_cast_fp16")]; + tensor var_4030_equation_0 = const()[name = tensor("op_4030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4030_cast_fp16 = einsum(equation = var_4030_equation_0, values = (var_3860_cast_fp16, var_4004_cast_fp16))[name = tensor("op_4030_cast_fp16")]; + tensor var_4032_equation_0 = const()[name = tensor("op_4032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4032_cast_fp16 = einsum(equation = var_4032_equation_0, values = (var_3864_cast_fp16, var_4005_cast_fp16))[name = tensor("op_4032_cast_fp16")]; + tensor var_4034_equation_0 = const()[name = tensor("op_4034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4034_cast_fp16 = einsum(equation = var_4034_equation_0, values = (var_3868_cast_fp16, var_4006_cast_fp16))[name = tensor("op_4034_cast_fp16")]; + tensor var_4036_equation_0 = const()[name = tensor("op_4036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4036_cast_fp16 = einsum(equation = var_4036_equation_0, values = (var_3872_cast_fp16, var_4007_cast_fp16))[name = tensor("op_4036_cast_fp16")]; + tensor var_4038_equation_0 = const()[name = tensor("op_4038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4038_cast_fp16 = einsum(equation = var_4038_equation_0, values = (var_3876_cast_fp16, var_4008_cast_fp16))[name = tensor("op_4038_cast_fp16")]; + tensor var_4040_equation_0 = const()[name = tensor("op_4040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4040_cast_fp16 = einsum(equation = var_4040_equation_0, values = (var_3880_cast_fp16, var_4009_cast_fp16))[name = tensor("op_4040_cast_fp16")]; + tensor var_4042_equation_0 = const()[name = tensor("op_4042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4042_cast_fp16 = einsum(equation = var_4042_equation_0, values = (var_3884_cast_fp16, var_4010_cast_fp16))[name = tensor("op_4042_cast_fp16")]; + tensor var_4044_equation_0 = const()[name = tensor("op_4044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4044_cast_fp16 = einsum(equation = var_4044_equation_0, values = (var_3888_cast_fp16, var_4011_cast_fp16))[name = tensor("op_4044_cast_fp16")]; + tensor var_4046_equation_0 = const()[name = tensor("op_4046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4046_cast_fp16 = einsum(equation = var_4046_equation_0, values = (var_3892_cast_fp16, var_4012_cast_fp16))[name = tensor("op_4046_cast_fp16")]; + tensor var_4048_equation_0 = const()[name = tensor("op_4048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4048_cast_fp16 = einsum(equation = var_4048_equation_0, values = (var_3896_cast_fp16, var_4013_cast_fp16))[name = tensor("op_4048_cast_fp16")]; + tensor var_4050_equation_0 = const()[name = tensor("op_4050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4050_cast_fp16 = einsum(equation = var_4050_equation_0, values = (var_3900_cast_fp16, var_4014_cast_fp16))[name = tensor("op_4050_cast_fp16")]; + tensor var_4052_equation_0 = const()[name = tensor("op_4052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4052_cast_fp16 = einsum(equation = var_4052_equation_0, values = (var_3904_cast_fp16, var_4015_cast_fp16))[name = tensor("op_4052_cast_fp16")]; + tensor var_4054_equation_0 = const()[name = tensor("op_4054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4054_cast_fp16 = einsum(equation = var_4054_equation_0, values = (var_3908_cast_fp16, var_4016_cast_fp16))[name = tensor("op_4054_cast_fp16")]; + tensor var_4056_equation_0 = const()[name = tensor("op_4056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4056_cast_fp16 = einsum(equation = var_4056_equation_0, values = (var_3912_cast_fp16, var_4017_cast_fp16))[name = tensor("op_4056_cast_fp16")]; + tensor var_4058_equation_0 = const()[name = tensor("op_4058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4058_cast_fp16 = einsum(equation = var_4058_equation_0, values = (var_3916_cast_fp16, var_4018_cast_fp16))[name = tensor("op_4058_cast_fp16")]; + tensor input_139_interleave_0 = const()[name = tensor("input_139_interleave_0"), val = tensor(false)]; + tensor input_139_cast_fp16 = concat(axis = var_2624, interleave = input_139_interleave_0, values = (var_4020_cast_fp16, var_4022_cast_fp16, var_4024_cast_fp16, var_4026_cast_fp16, var_4028_cast_fp16, var_4030_cast_fp16, var_4032_cast_fp16, var_4034_cast_fp16, var_4036_cast_fp16, var_4038_cast_fp16, var_4040_cast_fp16, var_4042_cast_fp16, var_4044_cast_fp16, var_4046_cast_fp16, var_4048_cast_fp16, var_4050_cast_fp16, var_4052_cast_fp16, var_4054_cast_fp16, var_4056_cast_fp16, var_4058_cast_fp16))[name = tensor("input_139_cast_fp16")]; + tensor var_4068_pad_type_0 = const()[name = tensor("op_4068_pad_type_0"), val = tensor("valid")]; + tensor var_4068_strides_0 = const()[name = tensor("op_4068_strides_0"), val = tensor([1, 1])]; + tensor var_4068_pad_0 = const()[name = tensor("op_4068_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4068_dilations_0 = const()[name = tensor("op_4068_dilations_0"), val = tensor([1, 1])]; + tensor var_4068_groups_0 = const()[name = tensor("op_4068_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104229632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105458496))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105458688)))]; + tensor var_4068_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_4068_dilations_0, groups = var_4068_groups_0, pad = var_4068_pad_0, pad_type = var_4068_pad_type_0, strides = var_4068_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = tensor("op_4068_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = var_4068_cast_fp16, y = inputs_31_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; + tensor hidden_states_73_axes_0 = const()[name = tensor("hidden_states_73_axes_0"), val = tensor([1])]; + tensor hidden_states_73_gamma_0_to_fp16 = const()[name = tensor("hidden_states_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105461312)))]; + tensor hidden_states_73_beta_0_to_fp16 = const()[name = tensor("hidden_states_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105463936)))]; + tensor var_4078_to_fp16 = const()[name = tensor("op_4078_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_73_cast_fp16 = layer_norm(axes = hidden_states_73_axes_0, beta = hidden_states_73_beta_0_to_fp16, epsilon = var_4078_to_fp16, gamma = hidden_states_73_gamma_0_to_fp16, x = inputs_33_cast_fp16)[name = tensor("hidden_states_73_cast_fp16")]; + tensor q_23_pad_type_0 = const()[name = tensor("q_23_pad_type_0"), val = tensor("valid")]; + tensor q_23_strides_0 = const()[name = tensor("q_23_strides_0"), val = tensor([1, 1])]; + tensor q_23_pad_0 = const()[name = tensor("q_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_23_dilations_0 = const()[name = tensor("q_23_dilations_0"), val = tensor([1, 1])]; + tensor q_23_groups_0 = const()[name = tensor("q_23_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105466560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106695424))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_23_cast_fp16 = conv(dilations = q_23_dilations_0, groups = q_23_groups_0, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = q_23_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_73_cast_fp16)[name = tensor("q_23_cast_fp16")]; + tensor k_45_pad_type_0 = const()[name = tensor("k_45_pad_type_0"), val = tensor("valid")]; + tensor k_45_strides_0 = const()[name = tensor("k_45_strides_0"), val = tensor([1, 1])]; + tensor k_45_pad_0 = const()[name = tensor("k_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_45_dilations_0 = const()[name = tensor("k_45_dilations_0"), val = tensor([1, 1])]; + tensor k_45_groups_0 = const()[name = tensor("k_45_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106695616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108661760))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_45_cast_fp16 = conv(dilations = k_45_dilations_0, groups = k_45_groups_0, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = k_45_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_45_cast_fp16")]; + tensor v_23_pad_type_0 = const()[name = tensor("v_23_pad_type_0"), val = tensor("valid")]; + tensor v_23_strides_0 = const()[name = tensor("v_23_strides_0"), val = tensor([1, 1])]; + tensor v_23_pad_0 = const()[name = tensor("v_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_23_dilations_0 = const()[name = tensor("v_23_dilations_0"), val = tensor([1, 1])]; + tensor v_23_groups_0 = const()[name = tensor("v_23_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108661952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110628096))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_23_cast_fp16 = conv(dilations = v_23_dilations_0, groups = v_23_groups_0, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = v_23_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_23_cast_fp16")]; + tensor var_4111_begin_0 = const()[name = tensor("op_4111_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4111_end_0 = const()[name = tensor("op_4111_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_4111_end_mask_0 = const()[name = tensor("op_4111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4111_cast_fp16 = slice_by_index(begin = var_4111_begin_0, end = var_4111_end_0, end_mask = var_4111_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4111_cast_fp16")]; + tensor var_4115_begin_0 = const()[name = tensor("op_4115_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4115_end_0 = const()[name = tensor("op_4115_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_4115_end_mask_0 = const()[name = tensor("op_4115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4115_cast_fp16 = slice_by_index(begin = var_4115_begin_0, end = var_4115_end_0, end_mask = var_4115_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4115_cast_fp16")]; + tensor var_4119_begin_0 = const()[name = tensor("op_4119_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4119_end_0 = const()[name = tensor("op_4119_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_4119_end_mask_0 = const()[name = tensor("op_4119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4119_cast_fp16 = slice_by_index(begin = var_4119_begin_0, end = var_4119_end_0, end_mask = var_4119_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4119_cast_fp16")]; + tensor var_4123_begin_0 = const()[name = tensor("op_4123_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4123_end_0 = const()[name = tensor("op_4123_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_4123_end_mask_0 = const()[name = tensor("op_4123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4123_cast_fp16 = slice_by_index(begin = var_4123_begin_0, end = var_4123_end_0, end_mask = var_4123_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4123_cast_fp16")]; + tensor var_4127_begin_0 = const()[name = tensor("op_4127_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4127_end_0 = const()[name = tensor("op_4127_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4127_end_mask_0 = const()[name = tensor("op_4127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4127_cast_fp16 = slice_by_index(begin = var_4127_begin_0, end = var_4127_end_0, end_mask = var_4127_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4127_cast_fp16")]; + tensor var_4131_begin_0 = const()[name = tensor("op_4131_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4131_end_0 = const()[name = tensor("op_4131_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_4131_end_mask_0 = const()[name = tensor("op_4131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4131_cast_fp16 = slice_by_index(begin = var_4131_begin_0, end = var_4131_end_0, end_mask = var_4131_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4131_cast_fp16")]; + tensor var_4135_begin_0 = const()[name = tensor("op_4135_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4135_end_0 = const()[name = tensor("op_4135_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_4135_end_mask_0 = const()[name = tensor("op_4135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4135_cast_fp16 = slice_by_index(begin = var_4135_begin_0, end = var_4135_end_0, end_mask = var_4135_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4135_cast_fp16")]; + tensor var_4139_begin_0 = const()[name = tensor("op_4139_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4139_end_0 = const()[name = tensor("op_4139_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_4139_end_mask_0 = const()[name = tensor("op_4139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4139_cast_fp16 = slice_by_index(begin = var_4139_begin_0, end = var_4139_end_0, end_mask = var_4139_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4139_cast_fp16")]; + tensor var_4143_begin_0 = const()[name = tensor("op_4143_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4143_end_0 = const()[name = tensor("op_4143_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_4143_end_mask_0 = const()[name = tensor("op_4143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4143_cast_fp16 = slice_by_index(begin = var_4143_begin_0, end = var_4143_end_0, end_mask = var_4143_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4143_cast_fp16")]; + tensor var_4147_begin_0 = const()[name = tensor("op_4147_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4147_end_0 = const()[name = tensor("op_4147_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_4147_end_mask_0 = const()[name = tensor("op_4147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4147_cast_fp16 = slice_by_index(begin = var_4147_begin_0, end = var_4147_end_0, end_mask = var_4147_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4147_cast_fp16")]; + tensor var_4151_begin_0 = const()[name = tensor("op_4151_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4151_end_0 = const()[name = tensor("op_4151_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_4151_end_mask_0 = const()[name = tensor("op_4151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4151_cast_fp16 = slice_by_index(begin = var_4151_begin_0, end = var_4151_end_0, end_mask = var_4151_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4151_cast_fp16")]; + tensor var_4155_begin_0 = const()[name = tensor("op_4155_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4155_end_0 = const()[name = tensor("op_4155_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_4155_end_mask_0 = const()[name = tensor("op_4155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4155_cast_fp16 = slice_by_index(begin = var_4155_begin_0, end = var_4155_end_0, end_mask = var_4155_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4155_cast_fp16")]; + tensor var_4159_begin_0 = const()[name = tensor("op_4159_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4159_end_0 = const()[name = tensor("op_4159_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_4159_end_mask_0 = const()[name = tensor("op_4159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4159_cast_fp16 = slice_by_index(begin = var_4159_begin_0, end = var_4159_end_0, end_mask = var_4159_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4159_cast_fp16")]; + tensor var_4163_begin_0 = const()[name = tensor("op_4163_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4163_end_0 = const()[name = tensor("op_4163_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_4163_end_mask_0 = const()[name = tensor("op_4163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4163_cast_fp16 = slice_by_index(begin = var_4163_begin_0, end = var_4163_end_0, end_mask = var_4163_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4163_cast_fp16")]; + tensor var_4167_begin_0 = const()[name = tensor("op_4167_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4167_end_0 = const()[name = tensor("op_4167_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_4167_end_mask_0 = const()[name = tensor("op_4167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4167_cast_fp16 = slice_by_index(begin = var_4167_begin_0, end = var_4167_end_0, end_mask = var_4167_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4167_cast_fp16")]; + tensor var_4171_begin_0 = const()[name = tensor("op_4171_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4171_end_0 = const()[name = tensor("op_4171_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4171_end_mask_0 = const()[name = tensor("op_4171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4171_cast_fp16 = slice_by_index(begin = var_4171_begin_0, end = var_4171_end_0, end_mask = var_4171_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4171_cast_fp16")]; + tensor var_4175_begin_0 = const()[name = tensor("op_4175_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4175_end_0 = const()[name = tensor("op_4175_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_4175_end_mask_0 = const()[name = tensor("op_4175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4175_cast_fp16 = slice_by_index(begin = var_4175_begin_0, end = var_4175_end_0, end_mask = var_4175_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4175_cast_fp16")]; + tensor var_4179_begin_0 = const()[name = tensor("op_4179_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4179_end_0 = const()[name = tensor("op_4179_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_4179_end_mask_0 = const()[name = tensor("op_4179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4179_cast_fp16 = slice_by_index(begin = var_4179_begin_0, end = var_4179_end_0, end_mask = var_4179_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4179_cast_fp16")]; + tensor var_4183_begin_0 = const()[name = tensor("op_4183_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4183_end_0 = const()[name = tensor("op_4183_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_4183_end_mask_0 = const()[name = tensor("op_4183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4183_cast_fp16 = slice_by_index(begin = var_4183_begin_0, end = var_4183_end_0, end_mask = var_4183_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4183_cast_fp16")]; + tensor var_4187_begin_0 = const()[name = tensor("op_4187_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4187_end_0 = const()[name = tensor("op_4187_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_4187_end_mask_0 = const()[name = tensor("op_4187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4187_cast_fp16 = slice_by_index(begin = var_4187_begin_0, end = var_4187_end_0, end_mask = var_4187_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4187_cast_fp16")]; + tensor k_47_perm_0 = const()[name = tensor("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4194_begin_0 = const()[name = tensor("op_4194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4194_end_0 = const()[name = tensor("op_4194_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_4194_end_mask_0 = const()[name = tensor("op_4194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_47_cast_fp16 = transpose(perm = k_47_perm_0, x = k_45_cast_fp16)[name = tensor("transpose_56")]; + tensor var_4194_cast_fp16 = slice_by_index(begin = var_4194_begin_0, end = var_4194_end_0, end_mask = var_4194_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4194_cast_fp16")]; + tensor var_4198_begin_0 = const()[name = tensor("op_4198_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4198_end_0 = const()[name = tensor("op_4198_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_4198_end_mask_0 = const()[name = tensor("op_4198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4198_cast_fp16 = slice_by_index(begin = var_4198_begin_0, end = var_4198_end_0, end_mask = var_4198_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4198_cast_fp16")]; + tensor var_4202_begin_0 = const()[name = tensor("op_4202_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4202_end_0 = const()[name = tensor("op_4202_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_4202_end_mask_0 = const()[name = tensor("op_4202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4202_cast_fp16 = slice_by_index(begin = var_4202_begin_0, end = var_4202_end_0, end_mask = var_4202_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4202_cast_fp16")]; + tensor var_4206_begin_0 = const()[name = tensor("op_4206_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4206_end_0 = const()[name = tensor("op_4206_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_4206_end_mask_0 = const()[name = tensor("op_4206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4206_cast_fp16 = slice_by_index(begin = var_4206_begin_0, end = var_4206_end_0, end_mask = var_4206_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4206_cast_fp16")]; + tensor var_4210_begin_0 = const()[name = tensor("op_4210_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4210_end_0 = const()[name = tensor("op_4210_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_4210_end_mask_0 = const()[name = tensor("op_4210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4210_cast_fp16 = slice_by_index(begin = var_4210_begin_0, end = var_4210_end_0, end_mask = var_4210_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4210_cast_fp16")]; + tensor var_4214_begin_0 = const()[name = tensor("op_4214_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4214_end_0 = const()[name = tensor("op_4214_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_4214_end_mask_0 = const()[name = tensor("op_4214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4214_cast_fp16 = slice_by_index(begin = var_4214_begin_0, end = var_4214_end_0, end_mask = var_4214_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4214_cast_fp16")]; + tensor var_4218_begin_0 = const()[name = tensor("op_4218_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4218_end_0 = const()[name = tensor("op_4218_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_4218_end_mask_0 = const()[name = tensor("op_4218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4218_cast_fp16 = slice_by_index(begin = var_4218_begin_0, end = var_4218_end_0, end_mask = var_4218_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4218_cast_fp16")]; + tensor var_4222_begin_0 = const()[name = tensor("op_4222_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4222_end_0 = const()[name = tensor("op_4222_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_4222_end_mask_0 = const()[name = tensor("op_4222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4222_cast_fp16 = slice_by_index(begin = var_4222_begin_0, end = var_4222_end_0, end_mask = var_4222_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4222_cast_fp16")]; + tensor var_4226_begin_0 = const()[name = tensor("op_4226_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4226_end_0 = const()[name = tensor("op_4226_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_4226_end_mask_0 = const()[name = tensor("op_4226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4226_cast_fp16 = slice_by_index(begin = var_4226_begin_0, end = var_4226_end_0, end_mask = var_4226_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4226_cast_fp16")]; + tensor var_4230_begin_0 = const()[name = tensor("op_4230_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4230_end_0 = const()[name = tensor("op_4230_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_4230_end_mask_0 = const()[name = tensor("op_4230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4230_cast_fp16 = slice_by_index(begin = var_4230_begin_0, end = var_4230_end_0, end_mask = var_4230_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4230_cast_fp16")]; + tensor var_4234_begin_0 = const()[name = tensor("op_4234_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4234_end_0 = const()[name = tensor("op_4234_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_4234_end_mask_0 = const()[name = tensor("op_4234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4234_cast_fp16 = slice_by_index(begin = var_4234_begin_0, end = var_4234_end_0, end_mask = var_4234_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4234_cast_fp16")]; + tensor var_4238_begin_0 = const()[name = tensor("op_4238_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4238_end_0 = const()[name = tensor("op_4238_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_4238_end_mask_0 = const()[name = tensor("op_4238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4238_cast_fp16 = slice_by_index(begin = var_4238_begin_0, end = var_4238_end_0, end_mask = var_4238_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4238_cast_fp16")]; + tensor var_4242_begin_0 = const()[name = tensor("op_4242_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4242_end_0 = const()[name = tensor("op_4242_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_4242_end_mask_0 = const()[name = tensor("op_4242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4242_cast_fp16 = slice_by_index(begin = var_4242_begin_0, end = var_4242_end_0, end_mask = var_4242_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4242_cast_fp16")]; + tensor var_4246_begin_0 = const()[name = tensor("op_4246_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4246_end_0 = const()[name = tensor("op_4246_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_4246_end_mask_0 = const()[name = tensor("op_4246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4246_cast_fp16 = slice_by_index(begin = var_4246_begin_0, end = var_4246_end_0, end_mask = var_4246_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4246_cast_fp16")]; + tensor var_4250_begin_0 = const()[name = tensor("op_4250_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4250_end_0 = const()[name = tensor("op_4250_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_4250_end_mask_0 = const()[name = tensor("op_4250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4250_cast_fp16 = slice_by_index(begin = var_4250_begin_0, end = var_4250_end_0, end_mask = var_4250_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4250_cast_fp16")]; + tensor var_4254_begin_0 = const()[name = tensor("op_4254_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4254_end_0 = const()[name = tensor("op_4254_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_4254_end_mask_0 = const()[name = tensor("op_4254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4254_cast_fp16 = slice_by_index(begin = var_4254_begin_0, end = var_4254_end_0, end_mask = var_4254_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4254_cast_fp16")]; + tensor var_4258_begin_0 = const()[name = tensor("op_4258_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4258_end_0 = const()[name = tensor("op_4258_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_4258_end_mask_0 = const()[name = tensor("op_4258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4258_cast_fp16 = slice_by_index(begin = var_4258_begin_0, end = var_4258_end_0, end_mask = var_4258_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4258_cast_fp16")]; + tensor var_4262_begin_0 = const()[name = tensor("op_4262_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4262_end_0 = const()[name = tensor("op_4262_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_4262_end_mask_0 = const()[name = tensor("op_4262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4262_cast_fp16 = slice_by_index(begin = var_4262_begin_0, end = var_4262_end_0, end_mask = var_4262_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4262_cast_fp16")]; + tensor var_4266_begin_0 = const()[name = tensor("op_4266_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4266_end_0 = const()[name = tensor("op_4266_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_4266_end_mask_0 = const()[name = tensor("op_4266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4266_cast_fp16")]; + tensor var_4270_begin_0 = const()[name = tensor("op_4270_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4270_end_0 = const()[name = tensor("op_4270_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_4270_end_mask_0 = const()[name = tensor("op_4270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4270_cast_fp16 = slice_by_index(begin = var_4270_begin_0, end = var_4270_end_0, end_mask = var_4270_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4270_cast_fp16")]; + tensor var_4272_begin_0 = const()[name = tensor("op_4272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4272_end_0 = const()[name = tensor("op_4272_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_4272_end_mask_0 = const()[name = tensor("op_4272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4272_cast_fp16 = slice_by_index(begin = var_4272_begin_0, end = var_4272_end_0, end_mask = var_4272_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4272_cast_fp16")]; + tensor var_4276_begin_0 = const()[name = tensor("op_4276_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4276_end_0 = const()[name = tensor("op_4276_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_4276_end_mask_0 = const()[name = tensor("op_4276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4276_cast_fp16 = slice_by_index(begin = var_4276_begin_0, end = var_4276_end_0, end_mask = var_4276_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4276_cast_fp16")]; + tensor var_4280_begin_0 = const()[name = tensor("op_4280_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4280_end_0 = const()[name = tensor("op_4280_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_4280_end_mask_0 = const()[name = tensor("op_4280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4280_cast_fp16")]; + tensor var_4284_begin_0 = const()[name = tensor("op_4284_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4284_end_0 = const()[name = tensor("op_4284_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_4284_end_mask_0 = const()[name = tensor("op_4284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4284_cast_fp16 = slice_by_index(begin = var_4284_begin_0, end = var_4284_end_0, end_mask = var_4284_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4284_cast_fp16")]; + tensor var_4288_begin_0 = const()[name = tensor("op_4288_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4288_end_0 = const()[name = tensor("op_4288_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_4288_end_mask_0 = const()[name = tensor("op_4288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4288_cast_fp16 = slice_by_index(begin = var_4288_begin_0, end = var_4288_end_0, end_mask = var_4288_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4288_cast_fp16")]; + tensor var_4292_begin_0 = const()[name = tensor("op_4292_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4292_end_0 = const()[name = tensor("op_4292_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_4292_end_mask_0 = const()[name = tensor("op_4292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4292_cast_fp16 = slice_by_index(begin = var_4292_begin_0, end = var_4292_end_0, end_mask = var_4292_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4292_cast_fp16")]; + tensor var_4296_begin_0 = const()[name = tensor("op_4296_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4296_end_0 = const()[name = tensor("op_4296_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_4296_end_mask_0 = const()[name = tensor("op_4296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4296_cast_fp16 = slice_by_index(begin = var_4296_begin_0, end = var_4296_end_0, end_mask = var_4296_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4296_cast_fp16")]; + tensor var_4300_begin_0 = const()[name = tensor("op_4300_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4300_end_0 = const()[name = tensor("op_4300_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_4300_end_mask_0 = const()[name = tensor("op_4300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4300_cast_fp16 = slice_by_index(begin = var_4300_begin_0, end = var_4300_end_0, end_mask = var_4300_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4300_cast_fp16")]; + tensor var_4304_begin_0 = const()[name = tensor("op_4304_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4304_end_0 = const()[name = tensor("op_4304_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_4304_end_mask_0 = const()[name = tensor("op_4304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4304_cast_fp16 = slice_by_index(begin = var_4304_begin_0, end = var_4304_end_0, end_mask = var_4304_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4304_cast_fp16")]; + tensor var_4308_begin_0 = const()[name = tensor("op_4308_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4308_end_0 = const()[name = tensor("op_4308_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_4308_end_mask_0 = const()[name = tensor("op_4308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4308_cast_fp16 = slice_by_index(begin = var_4308_begin_0, end = var_4308_end_0, end_mask = var_4308_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4308_cast_fp16")]; + tensor var_4312_begin_0 = const()[name = tensor("op_4312_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4312_end_0 = const()[name = tensor("op_4312_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_4312_end_mask_0 = const()[name = tensor("op_4312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4312_cast_fp16 = slice_by_index(begin = var_4312_begin_0, end = var_4312_end_0, end_mask = var_4312_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4312_cast_fp16")]; + tensor var_4316_begin_0 = const()[name = tensor("op_4316_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4316_end_0 = const()[name = tensor("op_4316_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_4316_end_mask_0 = const()[name = tensor("op_4316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4316_cast_fp16 = slice_by_index(begin = var_4316_begin_0, end = var_4316_end_0, end_mask = var_4316_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4316_cast_fp16")]; + tensor var_4320_begin_0 = const()[name = tensor("op_4320_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4320_end_0 = const()[name = tensor("op_4320_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_4320_end_mask_0 = const()[name = tensor("op_4320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4320_cast_fp16 = slice_by_index(begin = var_4320_begin_0, end = var_4320_end_0, end_mask = var_4320_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4320_cast_fp16")]; + tensor var_4324_begin_0 = const()[name = tensor("op_4324_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4324_end_0 = const()[name = tensor("op_4324_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_4324_end_mask_0 = const()[name = tensor("op_4324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4324_cast_fp16 = slice_by_index(begin = var_4324_begin_0, end = var_4324_end_0, end_mask = var_4324_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4324_cast_fp16")]; + tensor var_4328_begin_0 = const()[name = tensor("op_4328_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4328_end_0 = const()[name = tensor("op_4328_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_4328_end_mask_0 = const()[name = tensor("op_4328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4328_cast_fp16 = slice_by_index(begin = var_4328_begin_0, end = var_4328_end_0, end_mask = var_4328_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4328_cast_fp16")]; + tensor var_4332_begin_0 = const()[name = tensor("op_4332_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4332_end_0 = const()[name = tensor("op_4332_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_4332_end_mask_0 = const()[name = tensor("op_4332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4332_cast_fp16 = slice_by_index(begin = var_4332_begin_0, end = var_4332_end_0, end_mask = var_4332_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4332_cast_fp16")]; + tensor var_4336_begin_0 = const()[name = tensor("op_4336_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4336_end_0 = const()[name = tensor("op_4336_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_4336_end_mask_0 = const()[name = tensor("op_4336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4336_cast_fp16 = slice_by_index(begin = var_4336_begin_0, end = var_4336_end_0, end_mask = var_4336_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4336_cast_fp16")]; + tensor var_4340_begin_0 = const()[name = tensor("op_4340_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4340_end_0 = const()[name = tensor("op_4340_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_4340_end_mask_0 = const()[name = tensor("op_4340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4340_cast_fp16 = slice_by_index(begin = var_4340_begin_0, end = var_4340_end_0, end_mask = var_4340_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4340_cast_fp16")]; + tensor var_4344_begin_0 = const()[name = tensor("op_4344_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4344_end_0 = const()[name = tensor("op_4344_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_4344_end_mask_0 = const()[name = tensor("op_4344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4344_cast_fp16 = slice_by_index(begin = var_4344_begin_0, end = var_4344_end_0, end_mask = var_4344_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4344_cast_fp16")]; + tensor var_4348_begin_0 = const()[name = tensor("op_4348_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4348_end_0 = const()[name = tensor("op_4348_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_4348_end_mask_0 = const()[name = tensor("op_4348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4348_cast_fp16 = slice_by_index(begin = var_4348_begin_0, end = var_4348_end_0, end_mask = var_4348_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4348_cast_fp16")]; + tensor var_4352_equation_0 = const()[name = tensor("op_4352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4352_cast_fp16 = einsum(equation = var_4352_equation_0, values = (var_4194_cast_fp16, var_4111_cast_fp16))[name = tensor("op_4352_cast_fp16")]; + tensor var_4353_to_fp16 = const()[name = tensor("op_4353_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_281_cast_fp16 = mul(x = var_4352_cast_fp16, y = var_4353_to_fp16)[name = tensor("aw_281_cast_fp16")]; + tensor var_4356_equation_0 = const()[name = tensor("op_4356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4356_cast_fp16 = einsum(equation = var_4356_equation_0, values = (var_4198_cast_fp16, var_4115_cast_fp16))[name = tensor("op_4356_cast_fp16")]; + tensor var_4357_to_fp16 = const()[name = tensor("op_4357_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_283_cast_fp16 = mul(x = var_4356_cast_fp16, y = var_4357_to_fp16)[name = tensor("aw_283_cast_fp16")]; + tensor var_4360_equation_0 = const()[name = tensor("op_4360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4360_cast_fp16 = einsum(equation = var_4360_equation_0, values = (var_4202_cast_fp16, var_4119_cast_fp16))[name = tensor("op_4360_cast_fp16")]; + tensor var_4361_to_fp16 = const()[name = tensor("op_4361_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_285_cast_fp16 = mul(x = var_4360_cast_fp16, y = var_4361_to_fp16)[name = tensor("aw_285_cast_fp16")]; + tensor var_4364_equation_0 = const()[name = tensor("op_4364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4364_cast_fp16 = einsum(equation = var_4364_equation_0, values = (var_4206_cast_fp16, var_4123_cast_fp16))[name = tensor("op_4364_cast_fp16")]; + tensor var_4365_to_fp16 = const()[name = tensor("op_4365_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_287_cast_fp16 = mul(x = var_4364_cast_fp16, y = var_4365_to_fp16)[name = tensor("aw_287_cast_fp16")]; + tensor var_4368_equation_0 = const()[name = tensor("op_4368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4368_cast_fp16 = einsum(equation = var_4368_equation_0, values = (var_4210_cast_fp16, var_4127_cast_fp16))[name = tensor("op_4368_cast_fp16")]; + tensor var_4369_to_fp16 = const()[name = tensor("op_4369_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_289_cast_fp16 = mul(x = var_4368_cast_fp16, y = var_4369_to_fp16)[name = tensor("aw_289_cast_fp16")]; + tensor var_4372_equation_0 = const()[name = tensor("op_4372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4372_cast_fp16 = einsum(equation = var_4372_equation_0, values = (var_4214_cast_fp16, var_4131_cast_fp16))[name = tensor("op_4372_cast_fp16")]; + tensor var_4373_to_fp16 = const()[name = tensor("op_4373_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_291_cast_fp16 = mul(x = var_4372_cast_fp16, y = var_4373_to_fp16)[name = tensor("aw_291_cast_fp16")]; + tensor var_4376_equation_0 = const()[name = tensor("op_4376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4376_cast_fp16 = einsum(equation = var_4376_equation_0, values = (var_4218_cast_fp16, var_4135_cast_fp16))[name = tensor("op_4376_cast_fp16")]; + tensor var_4377_to_fp16 = const()[name = tensor("op_4377_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_293_cast_fp16 = mul(x = var_4376_cast_fp16, y = var_4377_to_fp16)[name = tensor("aw_293_cast_fp16")]; + tensor var_4380_equation_0 = const()[name = tensor("op_4380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4380_cast_fp16 = einsum(equation = var_4380_equation_0, values = (var_4222_cast_fp16, var_4139_cast_fp16))[name = tensor("op_4380_cast_fp16")]; + tensor var_4381_to_fp16 = const()[name = tensor("op_4381_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_295_cast_fp16 = mul(x = var_4380_cast_fp16, y = var_4381_to_fp16)[name = tensor("aw_295_cast_fp16")]; + tensor var_4384_equation_0 = const()[name = tensor("op_4384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4384_cast_fp16 = einsum(equation = var_4384_equation_0, values = (var_4226_cast_fp16, var_4143_cast_fp16))[name = tensor("op_4384_cast_fp16")]; + tensor var_4385_to_fp16 = const()[name = tensor("op_4385_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_297_cast_fp16 = mul(x = var_4384_cast_fp16, y = var_4385_to_fp16)[name = tensor("aw_297_cast_fp16")]; + tensor var_4388_equation_0 = const()[name = tensor("op_4388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4388_cast_fp16 = einsum(equation = var_4388_equation_0, values = (var_4230_cast_fp16, var_4147_cast_fp16))[name = tensor("op_4388_cast_fp16")]; + tensor var_4389_to_fp16 = const()[name = tensor("op_4389_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_299_cast_fp16 = mul(x = var_4388_cast_fp16, y = var_4389_to_fp16)[name = tensor("aw_299_cast_fp16")]; + tensor var_4392_equation_0 = const()[name = tensor("op_4392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4392_cast_fp16 = einsum(equation = var_4392_equation_0, values = (var_4234_cast_fp16, var_4151_cast_fp16))[name = tensor("op_4392_cast_fp16")]; + tensor var_4393_to_fp16 = const()[name = tensor("op_4393_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_301_cast_fp16 = mul(x = var_4392_cast_fp16, y = var_4393_to_fp16)[name = tensor("aw_301_cast_fp16")]; + tensor var_4396_equation_0 = const()[name = tensor("op_4396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4396_cast_fp16 = einsum(equation = var_4396_equation_0, values = (var_4238_cast_fp16, var_4155_cast_fp16))[name = tensor("op_4396_cast_fp16")]; + tensor var_4397_to_fp16 = const()[name = tensor("op_4397_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_303_cast_fp16 = mul(x = var_4396_cast_fp16, y = var_4397_to_fp16)[name = tensor("aw_303_cast_fp16")]; + tensor var_4400_equation_0 = const()[name = tensor("op_4400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4400_cast_fp16 = einsum(equation = var_4400_equation_0, values = (var_4242_cast_fp16, var_4159_cast_fp16))[name = tensor("op_4400_cast_fp16")]; + tensor var_4401_to_fp16 = const()[name = tensor("op_4401_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_305_cast_fp16 = mul(x = var_4400_cast_fp16, y = var_4401_to_fp16)[name = tensor("aw_305_cast_fp16")]; + tensor var_4404_equation_0 = const()[name = tensor("op_4404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4404_cast_fp16 = einsum(equation = var_4404_equation_0, values = (var_4246_cast_fp16, var_4163_cast_fp16))[name = tensor("op_4404_cast_fp16")]; + tensor var_4405_to_fp16 = const()[name = tensor("op_4405_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_307_cast_fp16 = mul(x = var_4404_cast_fp16, y = var_4405_to_fp16)[name = tensor("aw_307_cast_fp16")]; + tensor var_4408_equation_0 = const()[name = tensor("op_4408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4408_cast_fp16 = einsum(equation = var_4408_equation_0, values = (var_4250_cast_fp16, var_4167_cast_fp16))[name = tensor("op_4408_cast_fp16")]; + tensor var_4409_to_fp16 = const()[name = tensor("op_4409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_309_cast_fp16 = mul(x = var_4408_cast_fp16, y = var_4409_to_fp16)[name = tensor("aw_309_cast_fp16")]; + tensor var_4412_equation_0 = const()[name = tensor("op_4412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4412_cast_fp16 = einsum(equation = var_4412_equation_0, values = (var_4254_cast_fp16, var_4171_cast_fp16))[name = tensor("op_4412_cast_fp16")]; + tensor var_4413_to_fp16 = const()[name = tensor("op_4413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_311_cast_fp16 = mul(x = var_4412_cast_fp16, y = var_4413_to_fp16)[name = tensor("aw_311_cast_fp16")]; + tensor var_4416_equation_0 = const()[name = tensor("op_4416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4416_cast_fp16 = einsum(equation = var_4416_equation_0, values = (var_4258_cast_fp16, var_4175_cast_fp16))[name = tensor("op_4416_cast_fp16")]; + tensor var_4417_to_fp16 = const()[name = tensor("op_4417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_313_cast_fp16 = mul(x = var_4416_cast_fp16, y = var_4417_to_fp16)[name = tensor("aw_313_cast_fp16")]; + tensor var_4420_equation_0 = const()[name = tensor("op_4420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4420_cast_fp16 = einsum(equation = var_4420_equation_0, values = (var_4262_cast_fp16, var_4179_cast_fp16))[name = tensor("op_4420_cast_fp16")]; + tensor var_4421_to_fp16 = const()[name = tensor("op_4421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_315_cast_fp16 = mul(x = var_4420_cast_fp16, y = var_4421_to_fp16)[name = tensor("aw_315_cast_fp16")]; + tensor var_4424_equation_0 = const()[name = tensor("op_4424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4424_cast_fp16 = einsum(equation = var_4424_equation_0, values = (var_4266_cast_fp16, var_4183_cast_fp16))[name = tensor("op_4424_cast_fp16")]; + tensor var_4425_to_fp16 = const()[name = tensor("op_4425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_317_cast_fp16 = mul(x = var_4424_cast_fp16, y = var_4425_to_fp16)[name = tensor("aw_317_cast_fp16")]; + tensor var_4428_equation_0 = const()[name = tensor("op_4428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4428_cast_fp16 = einsum(equation = var_4428_equation_0, values = (var_4270_cast_fp16, var_4187_cast_fp16))[name = tensor("op_4428_cast_fp16")]; + tensor var_4429_to_fp16 = const()[name = tensor("op_4429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_319_cast_fp16 = mul(x = var_4428_cast_fp16, y = var_4429_to_fp16)[name = tensor("aw_319_cast_fp16")]; + tensor var_4431_cast_fp16 = softmax(axis = var_2624, x = aw_281_cast_fp16)[name = tensor("op_4431_cast_fp16")]; + tensor var_4432_cast_fp16 = softmax(axis = var_2624, x = aw_283_cast_fp16)[name = tensor("op_4432_cast_fp16")]; + tensor var_4433_cast_fp16 = softmax(axis = var_2624, x = aw_285_cast_fp16)[name = tensor("op_4433_cast_fp16")]; + tensor var_4434_cast_fp16 = softmax(axis = var_2624, x = aw_287_cast_fp16)[name = tensor("op_4434_cast_fp16")]; + tensor var_4435_cast_fp16 = softmax(axis = var_2624, x = aw_289_cast_fp16)[name = tensor("op_4435_cast_fp16")]; + tensor var_4436_cast_fp16 = softmax(axis = var_2624, x = aw_291_cast_fp16)[name = tensor("op_4436_cast_fp16")]; + tensor var_4437_cast_fp16 = softmax(axis = var_2624, x = aw_293_cast_fp16)[name = tensor("op_4437_cast_fp16")]; + tensor var_4438_cast_fp16 = softmax(axis = var_2624, x = aw_295_cast_fp16)[name = tensor("op_4438_cast_fp16")]; + tensor var_4439_cast_fp16 = softmax(axis = var_2624, x = aw_297_cast_fp16)[name = tensor("op_4439_cast_fp16")]; + tensor var_4440_cast_fp16 = softmax(axis = var_2624, x = aw_299_cast_fp16)[name = tensor("op_4440_cast_fp16")]; + tensor var_4441_cast_fp16 = softmax(axis = var_2624, x = aw_301_cast_fp16)[name = tensor("op_4441_cast_fp16")]; + tensor var_4442_cast_fp16 = softmax(axis = var_2624, x = aw_303_cast_fp16)[name = tensor("op_4442_cast_fp16")]; + tensor var_4443_cast_fp16 = softmax(axis = var_2624, x = aw_305_cast_fp16)[name = tensor("op_4443_cast_fp16")]; + tensor var_4444_cast_fp16 = softmax(axis = var_2624, x = aw_307_cast_fp16)[name = tensor("op_4444_cast_fp16")]; + tensor var_4445_cast_fp16 = softmax(axis = var_2624, x = aw_309_cast_fp16)[name = tensor("op_4445_cast_fp16")]; + tensor var_4446_cast_fp16 = softmax(axis = var_2624, x = aw_311_cast_fp16)[name = tensor("op_4446_cast_fp16")]; + tensor var_4447_cast_fp16 = softmax(axis = var_2624, x = aw_313_cast_fp16)[name = tensor("op_4447_cast_fp16")]; + tensor var_4448_cast_fp16 = softmax(axis = var_2624, x = aw_315_cast_fp16)[name = tensor("op_4448_cast_fp16")]; + tensor var_4449_cast_fp16 = softmax(axis = var_2624, x = aw_317_cast_fp16)[name = tensor("op_4449_cast_fp16")]; + tensor var_4450_cast_fp16 = softmax(axis = var_2624, x = aw_319_cast_fp16)[name = tensor("op_4450_cast_fp16")]; + tensor var_4452_equation_0 = const()[name = tensor("op_4452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4452_cast_fp16 = einsum(equation = var_4452_equation_0, values = (var_4272_cast_fp16, var_4431_cast_fp16))[name = tensor("op_4452_cast_fp16")]; + tensor var_4454_equation_0 = const()[name = tensor("op_4454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4454_cast_fp16 = einsum(equation = var_4454_equation_0, values = (var_4276_cast_fp16, var_4432_cast_fp16))[name = tensor("op_4454_cast_fp16")]; + tensor var_4456_equation_0 = const()[name = tensor("op_4456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4456_cast_fp16 = einsum(equation = var_4456_equation_0, values = (var_4280_cast_fp16, var_4433_cast_fp16))[name = tensor("op_4456_cast_fp16")]; + tensor var_4458_equation_0 = const()[name = tensor("op_4458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4458_cast_fp16 = einsum(equation = var_4458_equation_0, values = (var_4284_cast_fp16, var_4434_cast_fp16))[name = tensor("op_4458_cast_fp16")]; + tensor var_4460_equation_0 = const()[name = tensor("op_4460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4460_cast_fp16 = einsum(equation = var_4460_equation_0, values = (var_4288_cast_fp16, var_4435_cast_fp16))[name = tensor("op_4460_cast_fp16")]; + tensor var_4462_equation_0 = const()[name = tensor("op_4462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4462_cast_fp16 = einsum(equation = var_4462_equation_0, values = (var_4292_cast_fp16, var_4436_cast_fp16))[name = tensor("op_4462_cast_fp16")]; + tensor var_4464_equation_0 = const()[name = tensor("op_4464_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4464_cast_fp16 = einsum(equation = var_4464_equation_0, values = (var_4296_cast_fp16, var_4437_cast_fp16))[name = tensor("op_4464_cast_fp16")]; + tensor var_4466_equation_0 = const()[name = tensor("op_4466_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4466_cast_fp16 = einsum(equation = var_4466_equation_0, values = (var_4300_cast_fp16, var_4438_cast_fp16))[name = tensor("op_4466_cast_fp16")]; + tensor var_4468_equation_0 = const()[name = tensor("op_4468_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4468_cast_fp16 = einsum(equation = var_4468_equation_0, values = (var_4304_cast_fp16, var_4439_cast_fp16))[name = tensor("op_4468_cast_fp16")]; + tensor var_4470_equation_0 = const()[name = tensor("op_4470_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4470_cast_fp16 = einsum(equation = var_4470_equation_0, values = (var_4308_cast_fp16, var_4440_cast_fp16))[name = tensor("op_4470_cast_fp16")]; + tensor var_4472_equation_0 = const()[name = tensor("op_4472_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4472_cast_fp16 = einsum(equation = var_4472_equation_0, values = (var_4312_cast_fp16, var_4441_cast_fp16))[name = tensor("op_4472_cast_fp16")]; + tensor var_4474_equation_0 = const()[name = tensor("op_4474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4474_cast_fp16 = einsum(equation = var_4474_equation_0, values = (var_4316_cast_fp16, var_4442_cast_fp16))[name = tensor("op_4474_cast_fp16")]; + tensor var_4476_equation_0 = const()[name = tensor("op_4476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4476_cast_fp16 = einsum(equation = var_4476_equation_0, values = (var_4320_cast_fp16, var_4443_cast_fp16))[name = tensor("op_4476_cast_fp16")]; + tensor var_4478_equation_0 = const()[name = tensor("op_4478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4478_cast_fp16 = einsum(equation = var_4478_equation_0, values = (var_4324_cast_fp16, var_4444_cast_fp16))[name = tensor("op_4478_cast_fp16")]; + tensor var_4480_equation_0 = const()[name = tensor("op_4480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4480_cast_fp16 = einsum(equation = var_4480_equation_0, values = (var_4328_cast_fp16, var_4445_cast_fp16))[name = tensor("op_4480_cast_fp16")]; + tensor var_4482_equation_0 = const()[name = tensor("op_4482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4482_cast_fp16 = einsum(equation = var_4482_equation_0, values = (var_4332_cast_fp16, var_4446_cast_fp16))[name = tensor("op_4482_cast_fp16")]; + tensor var_4484_equation_0 = const()[name = tensor("op_4484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4484_cast_fp16 = einsum(equation = var_4484_equation_0, values = (var_4336_cast_fp16, var_4447_cast_fp16))[name = tensor("op_4484_cast_fp16")]; + tensor var_4486_equation_0 = const()[name = tensor("op_4486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4486_cast_fp16 = einsum(equation = var_4486_equation_0, values = (var_4340_cast_fp16, var_4448_cast_fp16))[name = tensor("op_4486_cast_fp16")]; + tensor var_4488_equation_0 = const()[name = tensor("op_4488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4488_cast_fp16 = einsum(equation = var_4488_equation_0, values = (var_4344_cast_fp16, var_4449_cast_fp16))[name = tensor("op_4488_cast_fp16")]; + tensor var_4490_equation_0 = const()[name = tensor("op_4490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4490_cast_fp16 = einsum(equation = var_4490_equation_0, values = (var_4348_cast_fp16, var_4450_cast_fp16))[name = tensor("op_4490_cast_fp16")]; + tensor input_141_interleave_0 = const()[name = tensor("input_141_interleave_0"), val = tensor(false)]; + tensor input_141_cast_fp16 = concat(axis = var_2624, interleave = input_141_interleave_0, values = (var_4452_cast_fp16, var_4454_cast_fp16, var_4456_cast_fp16, var_4458_cast_fp16, var_4460_cast_fp16, var_4462_cast_fp16, var_4464_cast_fp16, var_4466_cast_fp16, var_4468_cast_fp16, var_4470_cast_fp16, var_4472_cast_fp16, var_4474_cast_fp16, var_4476_cast_fp16, var_4478_cast_fp16, var_4480_cast_fp16, var_4482_cast_fp16, var_4484_cast_fp16, var_4486_cast_fp16, var_4488_cast_fp16, var_4490_cast_fp16))[name = tensor("input_141_cast_fp16")]; + tensor var_4500_pad_type_0 = const()[name = tensor("op_4500_pad_type_0"), val = tensor("valid")]; + tensor var_4500_strides_0 = const()[name = tensor("op_4500_strides_0"), val = tensor([1, 1])]; + tensor var_4500_pad_0 = const()[name = tensor("op_4500_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4500_dilations_0 = const()[name = tensor("op_4500_dilations_0"), val = tensor([1, 1])]; + tensor var_4500_groups_0 = const()[name = tensor("op_4500_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110628288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111857152))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111857344)))]; + tensor var_4500_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_4500_dilations_0, groups = var_4500_groups_0, pad = var_4500_pad_0, pad_type = var_4500_pad_type_0, strides = var_4500_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor("op_4500_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = var_4500_cast_fp16, y = inputs_33_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; + tensor input_143_axes_0 = const()[name = tensor("input_143_axes_0"), val = tensor([1])]; + tensor input_143_gamma_0_to_fp16 = const()[name = tensor("input_143_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111859968)))]; + tensor input_143_beta_0_to_fp16 = const()[name = tensor("input_143_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111862592)))]; + tensor var_4510_to_fp16 = const()[name = tensor("op_4510_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_143_cast_fp16 = layer_norm(axes = input_143_axes_0, beta = input_143_beta_0_to_fp16, epsilon = var_4510_to_fp16, gamma = input_143_gamma_0_to_fp16, x = inputs_35_cast_fp16)[name = tensor("input_143_cast_fp16")]; + tensor var_4530_pad_type_0 = const()[name = tensor("op_4530_pad_type_0"), val = tensor("valid")]; + tensor var_4530_strides_0 = const()[name = tensor("op_4530_strides_0"), val = tensor([1, 1])]; + tensor var_4530_pad_0 = const()[name = tensor("op_4530_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4530_dilations_0 = const()[name = tensor("op_4530_dilations_0"), val = tensor([1, 1])]; + tensor var_4530_groups_0 = const()[name = tensor("op_4530_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111865216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121695680))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121695872)))]; + tensor var_4530_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_4530_dilations_0, groups = var_4530_groups_0, pad = var_4530_pad_0, pad_type = var_4530_pad_type_0, strides = var_4530_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = tensor("op_4530_cast_fp16")]; + tensor var_4531_split_sizes_0 = const()[name = tensor("op_4531_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_4531_axis_0 = const()[name = tensor("op_4531_axis_0"), val = tensor(1)]; + tensor var_4531_cast_fp16_0, tensor var_4531_cast_fp16_1 = split(axis = var_4531_axis_0, split_sizes = var_4531_split_sizes_0, x = var_4530_cast_fp16)[name = tensor("op_4531_cast_fp16")]; + tensor var_4533_mode_0 = const()[name = tensor("op_4533_mode_0"), val = tensor("EXACT")]; + tensor var_4533_cast_fp16 = gelu(mode = var_4533_mode_0, x = var_4531_cast_fp16_1)[name = tensor("op_4533_cast_fp16")]; + tensor input_145_cast_fp16 = mul(x = var_4531_cast_fp16_0, y = var_4533_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor var_4541_pad_type_0 = const()[name = tensor("op_4541_pad_type_0"), val = tensor("valid")]; + tensor var_4541_strides_0 = const()[name = tensor("op_4541_strides_0"), val = tensor([1, 1])]; + tensor var_4541_pad_0 = const()[name = tensor("op_4541_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4541_dilations_0 = const()[name = tensor("op_4541_dilations_0"), val = tensor([1, 1])]; + tensor var_4541_groups_0 = const()[name = tensor("op_4541_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121716416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126631680))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126631872)))]; + tensor var_4541_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_4541_dilations_0, groups = var_4541_groups_0, pad = var_4541_pad_0, pad_type = var_4541_pad_type_0, strides = var_4541_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor("op_4541_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = var_4541_cast_fp16, y = inputs_35_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; + tensor hidden_states_77_axes_0 = const()[name = tensor("hidden_states_77_axes_0"), val = tensor([1])]; + tensor hidden_states_77_gamma_0_to_fp16 = const()[name = tensor("hidden_states_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126634496)))]; + tensor hidden_states_77_beta_0_to_fp16 = const()[name = tensor("hidden_states_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126637120)))]; + tensor var_4557_to_fp16 = const()[name = tensor("op_4557_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_77_cast_fp16 = layer_norm(axes = hidden_states_77_axes_0, beta = hidden_states_77_beta_0_to_fp16, epsilon = var_4557_to_fp16, gamma = hidden_states_77_gamma_0_to_fp16, x = inputs_37_cast_fp16)[name = tensor("hidden_states_77_cast_fp16")]; + tensor q_25_pad_type_0 = const()[name = tensor("q_25_pad_type_0"), val = tensor("valid")]; + tensor q_25_strides_0 = const()[name = tensor("q_25_strides_0"), val = tensor([1, 1])]; + tensor q_25_pad_0 = const()[name = tensor("q_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_25_dilations_0 = const()[name = tensor("q_25_dilations_0"), val = tensor([1, 1])]; + tensor q_25_groups_0 = const()[name = tensor("q_25_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126639744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127868608))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_25_cast_fp16 = conv(dilations = q_25_dilations_0, groups = q_25_groups_0, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = q_25_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_77_cast_fp16)[name = tensor("q_25_cast_fp16")]; + tensor k_49_pad_type_0 = const()[name = tensor("k_49_pad_type_0"), val = tensor("valid")]; + tensor k_49_strides_0 = const()[name = tensor("k_49_strides_0"), val = tensor([1, 1])]; + tensor k_49_pad_0 = const()[name = tensor("k_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_49_dilations_0 = const()[name = tensor("k_49_dilations_0"), val = tensor([1, 1])]; + tensor k_49_groups_0 = const()[name = tensor("k_49_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127868800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129097664))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_49_cast_fp16 = conv(dilations = k_49_dilations_0, groups = k_49_groups_0, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = k_49_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_77_cast_fp16)[name = tensor("k_49_cast_fp16")]; + tensor v_25_pad_type_0 = const()[name = tensor("v_25_pad_type_0"), val = tensor("valid")]; + tensor v_25_strides_0 = const()[name = tensor("v_25_strides_0"), val = tensor([1, 1])]; + tensor v_25_pad_0 = const()[name = tensor("v_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_25_dilations_0 = const()[name = tensor("v_25_dilations_0"), val = tensor([1, 1])]; + tensor v_25_groups_0 = const()[name = tensor("v_25_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129097856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130326720))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_25_cast_fp16 = conv(dilations = v_25_dilations_0, groups = v_25_groups_0, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = v_25_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_77_cast_fp16)[name = tensor("v_25_cast_fp16")]; + tensor var_4590_begin_0 = const()[name = tensor("op_4590_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4590_end_0 = const()[name = tensor("op_4590_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_4590_end_mask_0 = const()[name = tensor("op_4590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4590_cast_fp16 = slice_by_index(begin = var_4590_begin_0, end = var_4590_end_0, end_mask = var_4590_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4590_cast_fp16")]; + tensor var_4594_begin_0 = const()[name = tensor("op_4594_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4594_end_0 = const()[name = tensor("op_4594_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_4594_end_mask_0 = const()[name = tensor("op_4594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4594_cast_fp16 = slice_by_index(begin = var_4594_begin_0, end = var_4594_end_0, end_mask = var_4594_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4594_cast_fp16")]; + tensor var_4598_begin_0 = const()[name = tensor("op_4598_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4598_end_0 = const()[name = tensor("op_4598_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_4598_end_mask_0 = const()[name = tensor("op_4598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4598_cast_fp16 = slice_by_index(begin = var_4598_begin_0, end = var_4598_end_0, end_mask = var_4598_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4598_cast_fp16")]; + tensor var_4602_begin_0 = const()[name = tensor("op_4602_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4602_end_0 = const()[name = tensor("op_4602_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_4602_end_mask_0 = const()[name = tensor("op_4602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4602_cast_fp16 = slice_by_index(begin = var_4602_begin_0, end = var_4602_end_0, end_mask = var_4602_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4602_cast_fp16")]; + tensor var_4606_begin_0 = const()[name = tensor("op_4606_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4606_end_0 = const()[name = tensor("op_4606_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4606_end_mask_0 = const()[name = tensor("op_4606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4606_cast_fp16 = slice_by_index(begin = var_4606_begin_0, end = var_4606_end_0, end_mask = var_4606_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4606_cast_fp16")]; + tensor var_4610_begin_0 = const()[name = tensor("op_4610_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4610_end_0 = const()[name = tensor("op_4610_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_4610_end_mask_0 = const()[name = tensor("op_4610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4610_cast_fp16 = slice_by_index(begin = var_4610_begin_0, end = var_4610_end_0, end_mask = var_4610_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4610_cast_fp16")]; + tensor var_4614_begin_0 = const()[name = tensor("op_4614_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4614_end_0 = const()[name = tensor("op_4614_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_4614_end_mask_0 = const()[name = tensor("op_4614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4614_cast_fp16 = slice_by_index(begin = var_4614_begin_0, end = var_4614_end_0, end_mask = var_4614_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4614_cast_fp16")]; + tensor var_4618_begin_0 = const()[name = tensor("op_4618_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4618_end_0 = const()[name = tensor("op_4618_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_4618_end_mask_0 = const()[name = tensor("op_4618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4618_cast_fp16 = slice_by_index(begin = var_4618_begin_0, end = var_4618_end_0, end_mask = var_4618_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4618_cast_fp16")]; + tensor var_4622_begin_0 = const()[name = tensor("op_4622_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4622_end_0 = const()[name = tensor("op_4622_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_4622_end_mask_0 = const()[name = tensor("op_4622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4622_cast_fp16 = slice_by_index(begin = var_4622_begin_0, end = var_4622_end_0, end_mask = var_4622_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4622_cast_fp16")]; + tensor var_4626_begin_0 = const()[name = tensor("op_4626_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4626_end_0 = const()[name = tensor("op_4626_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_4626_end_mask_0 = const()[name = tensor("op_4626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4626_cast_fp16 = slice_by_index(begin = var_4626_begin_0, end = var_4626_end_0, end_mask = var_4626_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4626_cast_fp16")]; + tensor var_4630_begin_0 = const()[name = tensor("op_4630_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4630_end_0 = const()[name = tensor("op_4630_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_4630_end_mask_0 = const()[name = tensor("op_4630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4630_cast_fp16 = slice_by_index(begin = var_4630_begin_0, end = var_4630_end_0, end_mask = var_4630_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4630_cast_fp16")]; + tensor var_4634_begin_0 = const()[name = tensor("op_4634_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4634_end_0 = const()[name = tensor("op_4634_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_4634_end_mask_0 = const()[name = tensor("op_4634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4634_cast_fp16 = slice_by_index(begin = var_4634_begin_0, end = var_4634_end_0, end_mask = var_4634_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4634_cast_fp16")]; + tensor var_4638_begin_0 = const()[name = tensor("op_4638_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4638_end_0 = const()[name = tensor("op_4638_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_4638_end_mask_0 = const()[name = tensor("op_4638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4638_cast_fp16 = slice_by_index(begin = var_4638_begin_0, end = var_4638_end_0, end_mask = var_4638_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4638_cast_fp16")]; + tensor var_4642_begin_0 = const()[name = tensor("op_4642_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4642_end_0 = const()[name = tensor("op_4642_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_4642_end_mask_0 = const()[name = tensor("op_4642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4642_cast_fp16 = slice_by_index(begin = var_4642_begin_0, end = var_4642_end_0, end_mask = var_4642_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4642_cast_fp16")]; + tensor var_4646_begin_0 = const()[name = tensor("op_4646_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4646_end_0 = const()[name = tensor("op_4646_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_4646_end_mask_0 = const()[name = tensor("op_4646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4646_cast_fp16 = slice_by_index(begin = var_4646_begin_0, end = var_4646_end_0, end_mask = var_4646_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4646_cast_fp16")]; + tensor var_4650_begin_0 = const()[name = tensor("op_4650_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4650_end_0 = const()[name = tensor("op_4650_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4650_end_mask_0 = const()[name = tensor("op_4650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4650_cast_fp16 = slice_by_index(begin = var_4650_begin_0, end = var_4650_end_0, end_mask = var_4650_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4650_cast_fp16")]; + tensor var_4654_begin_0 = const()[name = tensor("op_4654_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4654_end_0 = const()[name = tensor("op_4654_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_4654_end_mask_0 = const()[name = tensor("op_4654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4654_cast_fp16 = slice_by_index(begin = var_4654_begin_0, end = var_4654_end_0, end_mask = var_4654_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4654_cast_fp16")]; + tensor var_4658_begin_0 = const()[name = tensor("op_4658_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4658_end_0 = const()[name = tensor("op_4658_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_4658_end_mask_0 = const()[name = tensor("op_4658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4658_cast_fp16 = slice_by_index(begin = var_4658_begin_0, end = var_4658_end_0, end_mask = var_4658_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4658_cast_fp16")]; + tensor var_4662_begin_0 = const()[name = tensor("op_4662_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4662_end_0 = const()[name = tensor("op_4662_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_4662_end_mask_0 = const()[name = tensor("op_4662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4662_cast_fp16 = slice_by_index(begin = var_4662_begin_0, end = var_4662_end_0, end_mask = var_4662_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4662_cast_fp16")]; + tensor var_4666_begin_0 = const()[name = tensor("op_4666_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4666_end_0 = const()[name = tensor("op_4666_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_4666_end_mask_0 = const()[name = tensor("op_4666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4666_cast_fp16 = slice_by_index(begin = var_4666_begin_0, end = var_4666_end_0, end_mask = var_4666_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4666_cast_fp16")]; + tensor k_51_perm_0 = const()[name = tensor("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4673_begin_0 = const()[name = tensor("op_4673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4673_end_0 = const()[name = tensor("op_4673_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_4673_end_mask_0 = const()[name = tensor("op_4673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_51_cast_fp16 = transpose(perm = k_51_perm_0, x = k_49_cast_fp16)[name = tensor("transpose_55")]; + tensor var_4673_cast_fp16 = slice_by_index(begin = var_4673_begin_0, end = var_4673_end_0, end_mask = var_4673_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4673_cast_fp16")]; + tensor var_4677_begin_0 = const()[name = tensor("op_4677_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4677_end_0 = const()[name = tensor("op_4677_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_4677_end_mask_0 = const()[name = tensor("op_4677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4677_cast_fp16 = slice_by_index(begin = var_4677_begin_0, end = var_4677_end_0, end_mask = var_4677_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4677_cast_fp16")]; + tensor var_4681_begin_0 = const()[name = tensor("op_4681_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4681_end_0 = const()[name = tensor("op_4681_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_4681_end_mask_0 = const()[name = tensor("op_4681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4681_cast_fp16 = slice_by_index(begin = var_4681_begin_0, end = var_4681_end_0, end_mask = var_4681_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4681_cast_fp16")]; + tensor var_4685_begin_0 = const()[name = tensor("op_4685_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4685_end_0 = const()[name = tensor("op_4685_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_4685_end_mask_0 = const()[name = tensor("op_4685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4685_cast_fp16 = slice_by_index(begin = var_4685_begin_0, end = var_4685_end_0, end_mask = var_4685_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4685_cast_fp16")]; + tensor var_4689_begin_0 = const()[name = tensor("op_4689_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4689_end_0 = const()[name = tensor("op_4689_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_4689_end_mask_0 = const()[name = tensor("op_4689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4689_cast_fp16 = slice_by_index(begin = var_4689_begin_0, end = var_4689_end_0, end_mask = var_4689_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4689_cast_fp16")]; + tensor var_4693_begin_0 = const()[name = tensor("op_4693_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4693_end_0 = const()[name = tensor("op_4693_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_4693_end_mask_0 = const()[name = tensor("op_4693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4693_cast_fp16 = slice_by_index(begin = var_4693_begin_0, end = var_4693_end_0, end_mask = var_4693_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4693_cast_fp16")]; + tensor var_4697_begin_0 = const()[name = tensor("op_4697_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4697_end_0 = const()[name = tensor("op_4697_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_4697_end_mask_0 = const()[name = tensor("op_4697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4697_cast_fp16 = slice_by_index(begin = var_4697_begin_0, end = var_4697_end_0, end_mask = var_4697_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4697_cast_fp16")]; + tensor var_4701_begin_0 = const()[name = tensor("op_4701_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4701_end_0 = const()[name = tensor("op_4701_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_4701_end_mask_0 = const()[name = tensor("op_4701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4701_cast_fp16 = slice_by_index(begin = var_4701_begin_0, end = var_4701_end_0, end_mask = var_4701_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4701_cast_fp16")]; + tensor var_4705_begin_0 = const()[name = tensor("op_4705_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4705_end_0 = const()[name = tensor("op_4705_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_4705_end_mask_0 = const()[name = tensor("op_4705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4705_cast_fp16 = slice_by_index(begin = var_4705_begin_0, end = var_4705_end_0, end_mask = var_4705_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4705_cast_fp16")]; + tensor var_4709_begin_0 = const()[name = tensor("op_4709_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4709_end_0 = const()[name = tensor("op_4709_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_4709_end_mask_0 = const()[name = tensor("op_4709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4709_cast_fp16 = slice_by_index(begin = var_4709_begin_0, end = var_4709_end_0, end_mask = var_4709_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4709_cast_fp16")]; + tensor var_4713_begin_0 = const()[name = tensor("op_4713_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4713_end_0 = const()[name = tensor("op_4713_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_4713_end_mask_0 = const()[name = tensor("op_4713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4713_cast_fp16 = slice_by_index(begin = var_4713_begin_0, end = var_4713_end_0, end_mask = var_4713_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4713_cast_fp16")]; + tensor var_4717_begin_0 = const()[name = tensor("op_4717_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4717_end_0 = const()[name = tensor("op_4717_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_4717_end_mask_0 = const()[name = tensor("op_4717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4717_cast_fp16 = slice_by_index(begin = var_4717_begin_0, end = var_4717_end_0, end_mask = var_4717_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4717_cast_fp16")]; + tensor var_4721_begin_0 = const()[name = tensor("op_4721_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4721_end_0 = const()[name = tensor("op_4721_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_4721_end_mask_0 = const()[name = tensor("op_4721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4721_cast_fp16 = slice_by_index(begin = var_4721_begin_0, end = var_4721_end_0, end_mask = var_4721_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4721_cast_fp16")]; + tensor var_4725_begin_0 = const()[name = tensor("op_4725_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4725_end_0 = const()[name = tensor("op_4725_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_4725_end_mask_0 = const()[name = tensor("op_4725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4725_cast_fp16 = slice_by_index(begin = var_4725_begin_0, end = var_4725_end_0, end_mask = var_4725_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4725_cast_fp16")]; + tensor var_4729_begin_0 = const()[name = tensor("op_4729_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4729_end_0 = const()[name = tensor("op_4729_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_4729_end_mask_0 = const()[name = tensor("op_4729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4729_cast_fp16 = slice_by_index(begin = var_4729_begin_0, end = var_4729_end_0, end_mask = var_4729_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4729_cast_fp16")]; + tensor var_4733_begin_0 = const()[name = tensor("op_4733_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4733_end_0 = const()[name = tensor("op_4733_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4733_end_mask_0 = const()[name = tensor("op_4733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4733_cast_fp16 = slice_by_index(begin = var_4733_begin_0, end = var_4733_end_0, end_mask = var_4733_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4733_cast_fp16")]; + tensor var_4737_begin_0 = const()[name = tensor("op_4737_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4737_end_0 = const()[name = tensor("op_4737_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_4737_end_mask_0 = const()[name = tensor("op_4737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4737_cast_fp16 = slice_by_index(begin = var_4737_begin_0, end = var_4737_end_0, end_mask = var_4737_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4737_cast_fp16")]; + tensor var_4741_begin_0 = const()[name = tensor("op_4741_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4741_end_0 = const()[name = tensor("op_4741_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_4741_end_mask_0 = const()[name = tensor("op_4741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4741_cast_fp16 = slice_by_index(begin = var_4741_begin_0, end = var_4741_end_0, end_mask = var_4741_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4741_cast_fp16")]; + tensor var_4745_begin_0 = const()[name = tensor("op_4745_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4745_end_0 = const()[name = tensor("op_4745_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_4745_end_mask_0 = const()[name = tensor("op_4745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4745_cast_fp16 = slice_by_index(begin = var_4745_begin_0, end = var_4745_end_0, end_mask = var_4745_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4745_cast_fp16")]; + tensor var_4749_begin_0 = const()[name = tensor("op_4749_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4749_end_0 = const()[name = tensor("op_4749_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_4749_end_mask_0 = const()[name = tensor("op_4749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4749_cast_fp16 = slice_by_index(begin = var_4749_begin_0, end = var_4749_end_0, end_mask = var_4749_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4749_cast_fp16")]; + tensor var_4751_begin_0 = const()[name = tensor("op_4751_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4751_end_0 = const()[name = tensor("op_4751_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_4751_end_mask_0 = const()[name = tensor("op_4751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4751_cast_fp16 = slice_by_index(begin = var_4751_begin_0, end = var_4751_end_0, end_mask = var_4751_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4751_cast_fp16")]; + tensor var_4755_begin_0 = const()[name = tensor("op_4755_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4755_end_0 = const()[name = tensor("op_4755_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_4755_end_mask_0 = const()[name = tensor("op_4755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4755_cast_fp16 = slice_by_index(begin = var_4755_begin_0, end = var_4755_end_0, end_mask = var_4755_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4755_cast_fp16")]; + tensor var_4759_begin_0 = const()[name = tensor("op_4759_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4759_end_0 = const()[name = tensor("op_4759_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_4759_end_mask_0 = const()[name = tensor("op_4759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4759_cast_fp16 = slice_by_index(begin = var_4759_begin_0, end = var_4759_end_0, end_mask = var_4759_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4759_cast_fp16")]; + tensor var_4763_begin_0 = const()[name = tensor("op_4763_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4763_end_0 = const()[name = tensor("op_4763_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_4763_end_mask_0 = const()[name = tensor("op_4763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4763_cast_fp16 = slice_by_index(begin = var_4763_begin_0, end = var_4763_end_0, end_mask = var_4763_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4763_cast_fp16")]; + tensor var_4767_begin_0 = const()[name = tensor("op_4767_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4767_end_0 = const()[name = tensor("op_4767_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4767_end_mask_0 = const()[name = tensor("op_4767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4767_cast_fp16 = slice_by_index(begin = var_4767_begin_0, end = var_4767_end_0, end_mask = var_4767_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4767_cast_fp16")]; + tensor var_4771_begin_0 = const()[name = tensor("op_4771_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4771_end_0 = const()[name = tensor("op_4771_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_4771_end_mask_0 = const()[name = tensor("op_4771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4771_cast_fp16 = slice_by_index(begin = var_4771_begin_0, end = var_4771_end_0, end_mask = var_4771_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4771_cast_fp16")]; + tensor var_4775_begin_0 = const()[name = tensor("op_4775_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4775_end_0 = const()[name = tensor("op_4775_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_4775_end_mask_0 = const()[name = tensor("op_4775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4775_cast_fp16 = slice_by_index(begin = var_4775_begin_0, end = var_4775_end_0, end_mask = var_4775_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4775_cast_fp16")]; + tensor var_4779_begin_0 = const()[name = tensor("op_4779_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4779_end_0 = const()[name = tensor("op_4779_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_4779_end_mask_0 = const()[name = tensor("op_4779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4779_cast_fp16 = slice_by_index(begin = var_4779_begin_0, end = var_4779_end_0, end_mask = var_4779_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4779_cast_fp16")]; + tensor var_4783_begin_0 = const()[name = tensor("op_4783_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4783_end_0 = const()[name = tensor("op_4783_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_4783_end_mask_0 = const()[name = tensor("op_4783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4783_cast_fp16 = slice_by_index(begin = var_4783_begin_0, end = var_4783_end_0, end_mask = var_4783_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4783_cast_fp16")]; + tensor var_4787_begin_0 = const()[name = tensor("op_4787_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4787_end_0 = const()[name = tensor("op_4787_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_4787_end_mask_0 = const()[name = tensor("op_4787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4787_cast_fp16 = slice_by_index(begin = var_4787_begin_0, end = var_4787_end_0, end_mask = var_4787_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4787_cast_fp16")]; + tensor var_4791_begin_0 = const()[name = tensor("op_4791_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4791_end_0 = const()[name = tensor("op_4791_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_4791_end_mask_0 = const()[name = tensor("op_4791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4791_cast_fp16 = slice_by_index(begin = var_4791_begin_0, end = var_4791_end_0, end_mask = var_4791_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4791_cast_fp16")]; + tensor var_4795_begin_0 = const()[name = tensor("op_4795_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4795_end_0 = const()[name = tensor("op_4795_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_4795_end_mask_0 = const()[name = tensor("op_4795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4795_cast_fp16 = slice_by_index(begin = var_4795_begin_0, end = var_4795_end_0, end_mask = var_4795_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4795_cast_fp16")]; + tensor var_4799_begin_0 = const()[name = tensor("op_4799_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4799_end_0 = const()[name = tensor("op_4799_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_4799_end_mask_0 = const()[name = tensor("op_4799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4799_cast_fp16 = slice_by_index(begin = var_4799_begin_0, end = var_4799_end_0, end_mask = var_4799_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4799_cast_fp16")]; + tensor var_4803_begin_0 = const()[name = tensor("op_4803_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4803_end_0 = const()[name = tensor("op_4803_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_4803_end_mask_0 = const()[name = tensor("op_4803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4803_cast_fp16 = slice_by_index(begin = var_4803_begin_0, end = var_4803_end_0, end_mask = var_4803_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4803_cast_fp16")]; + tensor var_4807_begin_0 = const()[name = tensor("op_4807_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4807_end_0 = const()[name = tensor("op_4807_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_4807_end_mask_0 = const()[name = tensor("op_4807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4807_cast_fp16 = slice_by_index(begin = var_4807_begin_0, end = var_4807_end_0, end_mask = var_4807_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4807_cast_fp16")]; + tensor var_4811_begin_0 = const()[name = tensor("op_4811_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4811_end_0 = const()[name = tensor("op_4811_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4811_end_mask_0 = const()[name = tensor("op_4811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4811_cast_fp16 = slice_by_index(begin = var_4811_begin_0, end = var_4811_end_0, end_mask = var_4811_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4811_cast_fp16")]; + tensor var_4815_begin_0 = const()[name = tensor("op_4815_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4815_end_0 = const()[name = tensor("op_4815_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_4815_end_mask_0 = const()[name = tensor("op_4815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4815_cast_fp16 = slice_by_index(begin = var_4815_begin_0, end = var_4815_end_0, end_mask = var_4815_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4815_cast_fp16")]; + tensor var_4819_begin_0 = const()[name = tensor("op_4819_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4819_end_0 = const()[name = tensor("op_4819_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_4819_end_mask_0 = const()[name = tensor("op_4819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4819_cast_fp16 = slice_by_index(begin = var_4819_begin_0, end = var_4819_end_0, end_mask = var_4819_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4819_cast_fp16")]; + tensor var_4823_begin_0 = const()[name = tensor("op_4823_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4823_end_0 = const()[name = tensor("op_4823_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_4823_end_mask_0 = const()[name = tensor("op_4823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4823_cast_fp16 = slice_by_index(begin = var_4823_begin_0, end = var_4823_end_0, end_mask = var_4823_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4823_cast_fp16")]; + tensor var_4827_begin_0 = const()[name = tensor("op_4827_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4827_end_0 = const()[name = tensor("op_4827_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_4827_end_mask_0 = const()[name = tensor("op_4827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4827_cast_fp16 = slice_by_index(begin = var_4827_begin_0, end = var_4827_end_0, end_mask = var_4827_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4827_cast_fp16")]; + tensor var_4831_equation_0 = const()[name = tensor("op_4831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4831_cast_fp16 = einsum(equation = var_4831_equation_0, values = (var_4673_cast_fp16, var_4590_cast_fp16))[name = tensor("op_4831_cast_fp16")]; + tensor var_4832_to_fp16 = const()[name = tensor("op_4832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_321_cast_fp16 = mul(x = var_4831_cast_fp16, y = var_4832_to_fp16)[name = tensor("aw_321_cast_fp16")]; + tensor var_4835_equation_0 = const()[name = tensor("op_4835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4835_cast_fp16 = einsum(equation = var_4835_equation_0, values = (var_4677_cast_fp16, var_4594_cast_fp16))[name = tensor("op_4835_cast_fp16")]; + tensor var_4836_to_fp16 = const()[name = tensor("op_4836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_323_cast_fp16 = mul(x = var_4835_cast_fp16, y = var_4836_to_fp16)[name = tensor("aw_323_cast_fp16")]; + tensor var_4839_equation_0 = const()[name = tensor("op_4839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4839_cast_fp16 = einsum(equation = var_4839_equation_0, values = (var_4681_cast_fp16, var_4598_cast_fp16))[name = tensor("op_4839_cast_fp16")]; + tensor var_4840_to_fp16 = const()[name = tensor("op_4840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_325_cast_fp16 = mul(x = var_4839_cast_fp16, y = var_4840_to_fp16)[name = tensor("aw_325_cast_fp16")]; + tensor var_4843_equation_0 = const()[name = tensor("op_4843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4843_cast_fp16 = einsum(equation = var_4843_equation_0, values = (var_4685_cast_fp16, var_4602_cast_fp16))[name = tensor("op_4843_cast_fp16")]; + tensor var_4844_to_fp16 = const()[name = tensor("op_4844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_327_cast_fp16 = mul(x = var_4843_cast_fp16, y = var_4844_to_fp16)[name = tensor("aw_327_cast_fp16")]; + tensor var_4847_equation_0 = const()[name = tensor("op_4847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4847_cast_fp16 = einsum(equation = var_4847_equation_0, values = (var_4689_cast_fp16, var_4606_cast_fp16))[name = tensor("op_4847_cast_fp16")]; + tensor var_4848_to_fp16 = const()[name = tensor("op_4848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_329_cast_fp16 = mul(x = var_4847_cast_fp16, y = var_4848_to_fp16)[name = tensor("aw_329_cast_fp16")]; + tensor var_4851_equation_0 = const()[name = tensor("op_4851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4851_cast_fp16 = einsum(equation = var_4851_equation_0, values = (var_4693_cast_fp16, var_4610_cast_fp16))[name = tensor("op_4851_cast_fp16")]; + tensor var_4852_to_fp16 = const()[name = tensor("op_4852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_331_cast_fp16 = mul(x = var_4851_cast_fp16, y = var_4852_to_fp16)[name = tensor("aw_331_cast_fp16")]; + tensor var_4855_equation_0 = const()[name = tensor("op_4855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4855_cast_fp16 = einsum(equation = var_4855_equation_0, values = (var_4697_cast_fp16, var_4614_cast_fp16))[name = tensor("op_4855_cast_fp16")]; + tensor var_4856_to_fp16 = const()[name = tensor("op_4856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_333_cast_fp16 = mul(x = var_4855_cast_fp16, y = var_4856_to_fp16)[name = tensor("aw_333_cast_fp16")]; + tensor var_4859_equation_0 = const()[name = tensor("op_4859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4859_cast_fp16 = einsum(equation = var_4859_equation_0, values = (var_4701_cast_fp16, var_4618_cast_fp16))[name = tensor("op_4859_cast_fp16")]; + tensor var_4860_to_fp16 = const()[name = tensor("op_4860_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_335_cast_fp16 = mul(x = var_4859_cast_fp16, y = var_4860_to_fp16)[name = tensor("aw_335_cast_fp16")]; + tensor var_4863_equation_0 = const()[name = tensor("op_4863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4863_cast_fp16 = einsum(equation = var_4863_equation_0, values = (var_4705_cast_fp16, var_4622_cast_fp16))[name = tensor("op_4863_cast_fp16")]; + tensor var_4864_to_fp16 = const()[name = tensor("op_4864_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_337_cast_fp16 = mul(x = var_4863_cast_fp16, y = var_4864_to_fp16)[name = tensor("aw_337_cast_fp16")]; + tensor var_4867_equation_0 = const()[name = tensor("op_4867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4867_cast_fp16 = einsum(equation = var_4867_equation_0, values = (var_4709_cast_fp16, var_4626_cast_fp16))[name = tensor("op_4867_cast_fp16")]; + tensor var_4868_to_fp16 = const()[name = tensor("op_4868_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_339_cast_fp16 = mul(x = var_4867_cast_fp16, y = var_4868_to_fp16)[name = tensor("aw_339_cast_fp16")]; + tensor var_4871_equation_0 = const()[name = tensor("op_4871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4871_cast_fp16 = einsum(equation = var_4871_equation_0, values = (var_4713_cast_fp16, var_4630_cast_fp16))[name = tensor("op_4871_cast_fp16")]; + tensor var_4872_to_fp16 = const()[name = tensor("op_4872_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_341_cast_fp16 = mul(x = var_4871_cast_fp16, y = var_4872_to_fp16)[name = tensor("aw_341_cast_fp16")]; + tensor var_4875_equation_0 = const()[name = tensor("op_4875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4875_cast_fp16 = einsum(equation = var_4875_equation_0, values = (var_4717_cast_fp16, var_4634_cast_fp16))[name = tensor("op_4875_cast_fp16")]; + tensor var_4876_to_fp16 = const()[name = tensor("op_4876_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_343_cast_fp16 = mul(x = var_4875_cast_fp16, y = var_4876_to_fp16)[name = tensor("aw_343_cast_fp16")]; + tensor var_4879_equation_0 = const()[name = tensor("op_4879_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4879_cast_fp16 = einsum(equation = var_4879_equation_0, values = (var_4721_cast_fp16, var_4638_cast_fp16))[name = tensor("op_4879_cast_fp16")]; + tensor var_4880_to_fp16 = const()[name = tensor("op_4880_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_345_cast_fp16 = mul(x = var_4879_cast_fp16, y = var_4880_to_fp16)[name = tensor("aw_345_cast_fp16")]; + tensor var_4883_equation_0 = const()[name = tensor("op_4883_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4883_cast_fp16 = einsum(equation = var_4883_equation_0, values = (var_4725_cast_fp16, var_4642_cast_fp16))[name = tensor("op_4883_cast_fp16")]; + tensor var_4884_to_fp16 = const()[name = tensor("op_4884_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_347_cast_fp16 = mul(x = var_4883_cast_fp16, y = var_4884_to_fp16)[name = tensor("aw_347_cast_fp16")]; + tensor var_4887_equation_0 = const()[name = tensor("op_4887_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4887_cast_fp16 = einsum(equation = var_4887_equation_0, values = (var_4729_cast_fp16, var_4646_cast_fp16))[name = tensor("op_4887_cast_fp16")]; + tensor var_4888_to_fp16 = const()[name = tensor("op_4888_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_349_cast_fp16 = mul(x = var_4887_cast_fp16, y = var_4888_to_fp16)[name = tensor("aw_349_cast_fp16")]; + tensor var_4891_equation_0 = const()[name = tensor("op_4891_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4891_cast_fp16 = einsum(equation = var_4891_equation_0, values = (var_4733_cast_fp16, var_4650_cast_fp16))[name = tensor("op_4891_cast_fp16")]; + tensor var_4892_to_fp16 = const()[name = tensor("op_4892_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_351_cast_fp16 = mul(x = var_4891_cast_fp16, y = var_4892_to_fp16)[name = tensor("aw_351_cast_fp16")]; + tensor var_4895_equation_0 = const()[name = tensor("op_4895_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4895_cast_fp16 = einsum(equation = var_4895_equation_0, values = (var_4737_cast_fp16, var_4654_cast_fp16))[name = tensor("op_4895_cast_fp16")]; + tensor var_4896_to_fp16 = const()[name = tensor("op_4896_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_353_cast_fp16 = mul(x = var_4895_cast_fp16, y = var_4896_to_fp16)[name = tensor("aw_353_cast_fp16")]; + tensor var_4899_equation_0 = const()[name = tensor("op_4899_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4899_cast_fp16 = einsum(equation = var_4899_equation_0, values = (var_4741_cast_fp16, var_4658_cast_fp16))[name = tensor("op_4899_cast_fp16")]; + tensor var_4900_to_fp16 = const()[name = tensor("op_4900_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_355_cast_fp16 = mul(x = var_4899_cast_fp16, y = var_4900_to_fp16)[name = tensor("aw_355_cast_fp16")]; + tensor var_4903_equation_0 = const()[name = tensor("op_4903_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4903_cast_fp16 = einsum(equation = var_4903_equation_0, values = (var_4745_cast_fp16, var_4662_cast_fp16))[name = tensor("op_4903_cast_fp16")]; + tensor var_4904_to_fp16 = const()[name = tensor("op_4904_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_357_cast_fp16 = mul(x = var_4903_cast_fp16, y = var_4904_to_fp16)[name = tensor("aw_357_cast_fp16")]; + tensor var_4907_equation_0 = const()[name = tensor("op_4907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4907_cast_fp16 = einsum(equation = var_4907_equation_0, values = (var_4749_cast_fp16, var_4666_cast_fp16))[name = tensor("op_4907_cast_fp16")]; + tensor var_4908_to_fp16 = const()[name = tensor("op_4908_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_359_cast_fp16 = mul(x = var_4907_cast_fp16, y = var_4908_to_fp16)[name = tensor("aw_359_cast_fp16")]; + tensor var_4910_cast_fp16 = softmax(axis = var_2624, x = aw_321_cast_fp16)[name = tensor("op_4910_cast_fp16")]; + tensor var_4911_cast_fp16 = softmax(axis = var_2624, x = aw_323_cast_fp16)[name = tensor("op_4911_cast_fp16")]; + tensor var_4912_cast_fp16 = softmax(axis = var_2624, x = aw_325_cast_fp16)[name = tensor("op_4912_cast_fp16")]; + tensor var_4913_cast_fp16 = softmax(axis = var_2624, x = aw_327_cast_fp16)[name = tensor("op_4913_cast_fp16")]; + tensor var_4914_cast_fp16 = softmax(axis = var_2624, x = aw_329_cast_fp16)[name = tensor("op_4914_cast_fp16")]; + tensor var_4915_cast_fp16 = softmax(axis = var_2624, x = aw_331_cast_fp16)[name = tensor("op_4915_cast_fp16")]; + tensor var_4916_cast_fp16 = softmax(axis = var_2624, x = aw_333_cast_fp16)[name = tensor("op_4916_cast_fp16")]; + tensor var_4917_cast_fp16 = softmax(axis = var_2624, x = aw_335_cast_fp16)[name = tensor("op_4917_cast_fp16")]; + tensor var_4918_cast_fp16 = softmax(axis = var_2624, x = aw_337_cast_fp16)[name = tensor("op_4918_cast_fp16")]; + tensor var_4919_cast_fp16 = softmax(axis = var_2624, x = aw_339_cast_fp16)[name = tensor("op_4919_cast_fp16")]; + tensor var_4920_cast_fp16 = softmax(axis = var_2624, x = aw_341_cast_fp16)[name = tensor("op_4920_cast_fp16")]; + tensor var_4921_cast_fp16 = softmax(axis = var_2624, x = aw_343_cast_fp16)[name = tensor("op_4921_cast_fp16")]; + tensor var_4922_cast_fp16 = softmax(axis = var_2624, x = aw_345_cast_fp16)[name = tensor("op_4922_cast_fp16")]; + tensor var_4923_cast_fp16 = softmax(axis = var_2624, x = aw_347_cast_fp16)[name = tensor("op_4923_cast_fp16")]; + tensor var_4924_cast_fp16 = softmax(axis = var_2624, x = aw_349_cast_fp16)[name = tensor("op_4924_cast_fp16")]; + tensor var_4925_cast_fp16 = softmax(axis = var_2624, x = aw_351_cast_fp16)[name = tensor("op_4925_cast_fp16")]; + tensor var_4926_cast_fp16 = softmax(axis = var_2624, x = aw_353_cast_fp16)[name = tensor("op_4926_cast_fp16")]; + tensor var_4927_cast_fp16 = softmax(axis = var_2624, x = aw_355_cast_fp16)[name = tensor("op_4927_cast_fp16")]; + tensor var_4928_cast_fp16 = softmax(axis = var_2624, x = aw_357_cast_fp16)[name = tensor("op_4928_cast_fp16")]; + tensor var_4929_cast_fp16 = softmax(axis = var_2624, x = aw_359_cast_fp16)[name = tensor("op_4929_cast_fp16")]; + tensor var_4931_equation_0 = const()[name = tensor("op_4931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4931_cast_fp16 = einsum(equation = var_4931_equation_0, values = (var_4751_cast_fp16, var_4910_cast_fp16))[name = tensor("op_4931_cast_fp16")]; + tensor var_4933_equation_0 = const()[name = tensor("op_4933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4933_cast_fp16 = einsum(equation = var_4933_equation_0, values = (var_4755_cast_fp16, var_4911_cast_fp16))[name = tensor("op_4933_cast_fp16")]; + tensor var_4935_equation_0 = const()[name = tensor("op_4935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4935_cast_fp16 = einsum(equation = var_4935_equation_0, values = (var_4759_cast_fp16, var_4912_cast_fp16))[name = tensor("op_4935_cast_fp16")]; + tensor var_4937_equation_0 = const()[name = tensor("op_4937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4937_cast_fp16 = einsum(equation = var_4937_equation_0, values = (var_4763_cast_fp16, var_4913_cast_fp16))[name = tensor("op_4937_cast_fp16")]; + tensor var_4939_equation_0 = const()[name = tensor("op_4939_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4939_cast_fp16 = einsum(equation = var_4939_equation_0, values = (var_4767_cast_fp16, var_4914_cast_fp16))[name = tensor("op_4939_cast_fp16")]; + tensor var_4941_equation_0 = const()[name = tensor("op_4941_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4941_cast_fp16 = einsum(equation = var_4941_equation_0, values = (var_4771_cast_fp16, var_4915_cast_fp16))[name = tensor("op_4941_cast_fp16")]; + tensor var_4943_equation_0 = const()[name = tensor("op_4943_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4943_cast_fp16 = einsum(equation = var_4943_equation_0, values = (var_4775_cast_fp16, var_4916_cast_fp16))[name = tensor("op_4943_cast_fp16")]; + tensor var_4945_equation_0 = const()[name = tensor("op_4945_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4945_cast_fp16 = einsum(equation = var_4945_equation_0, values = (var_4779_cast_fp16, var_4917_cast_fp16))[name = tensor("op_4945_cast_fp16")]; + tensor var_4947_equation_0 = const()[name = tensor("op_4947_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4947_cast_fp16 = einsum(equation = var_4947_equation_0, values = (var_4783_cast_fp16, var_4918_cast_fp16))[name = tensor("op_4947_cast_fp16")]; + tensor var_4949_equation_0 = const()[name = tensor("op_4949_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4949_cast_fp16 = einsum(equation = var_4949_equation_0, values = (var_4787_cast_fp16, var_4919_cast_fp16))[name = tensor("op_4949_cast_fp16")]; + tensor var_4951_equation_0 = const()[name = tensor("op_4951_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4951_cast_fp16 = einsum(equation = var_4951_equation_0, values = (var_4791_cast_fp16, var_4920_cast_fp16))[name = tensor("op_4951_cast_fp16")]; + tensor var_4953_equation_0 = const()[name = tensor("op_4953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4953_cast_fp16 = einsum(equation = var_4953_equation_0, values = (var_4795_cast_fp16, var_4921_cast_fp16))[name = tensor("op_4953_cast_fp16")]; + tensor var_4955_equation_0 = const()[name = tensor("op_4955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4955_cast_fp16 = einsum(equation = var_4955_equation_0, values = (var_4799_cast_fp16, var_4922_cast_fp16))[name = tensor("op_4955_cast_fp16")]; + tensor var_4957_equation_0 = const()[name = tensor("op_4957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4957_cast_fp16 = einsum(equation = var_4957_equation_0, values = (var_4803_cast_fp16, var_4923_cast_fp16))[name = tensor("op_4957_cast_fp16")]; + tensor var_4959_equation_0 = const()[name = tensor("op_4959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4959_cast_fp16 = einsum(equation = var_4959_equation_0, values = (var_4807_cast_fp16, var_4924_cast_fp16))[name = tensor("op_4959_cast_fp16")]; + tensor var_4961_equation_0 = const()[name = tensor("op_4961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4961_cast_fp16 = einsum(equation = var_4961_equation_0, values = (var_4811_cast_fp16, var_4925_cast_fp16))[name = tensor("op_4961_cast_fp16")]; + tensor var_4963_equation_0 = const()[name = tensor("op_4963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4963_cast_fp16 = einsum(equation = var_4963_equation_0, values = (var_4815_cast_fp16, var_4926_cast_fp16))[name = tensor("op_4963_cast_fp16")]; + tensor var_4965_equation_0 = const()[name = tensor("op_4965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4965_cast_fp16 = einsum(equation = var_4965_equation_0, values = (var_4819_cast_fp16, var_4927_cast_fp16))[name = tensor("op_4965_cast_fp16")]; + tensor var_4967_equation_0 = const()[name = tensor("op_4967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4967_cast_fp16 = einsum(equation = var_4967_equation_0, values = (var_4823_cast_fp16, var_4928_cast_fp16))[name = tensor("op_4967_cast_fp16")]; + tensor var_4969_equation_0 = const()[name = tensor("op_4969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4969_cast_fp16 = einsum(equation = var_4969_equation_0, values = (var_4827_cast_fp16, var_4929_cast_fp16))[name = tensor("op_4969_cast_fp16")]; + tensor input_147_interleave_0 = const()[name = tensor("input_147_interleave_0"), val = tensor(false)]; + tensor input_147_cast_fp16 = concat(axis = var_2624, interleave = input_147_interleave_0, values = (var_4931_cast_fp16, var_4933_cast_fp16, var_4935_cast_fp16, var_4937_cast_fp16, var_4939_cast_fp16, var_4941_cast_fp16, var_4943_cast_fp16, var_4945_cast_fp16, var_4947_cast_fp16, var_4949_cast_fp16, var_4951_cast_fp16, var_4953_cast_fp16, var_4955_cast_fp16, var_4957_cast_fp16, var_4959_cast_fp16, var_4961_cast_fp16, var_4963_cast_fp16, var_4965_cast_fp16, var_4967_cast_fp16, var_4969_cast_fp16))[name = tensor("input_147_cast_fp16")]; + tensor var_4979_pad_type_0 = const()[name = tensor("op_4979_pad_type_0"), val = tensor("valid")]; + tensor var_4979_strides_0 = const()[name = tensor("op_4979_strides_0"), val = tensor([1, 1])]; + tensor var_4979_pad_0 = const()[name = tensor("op_4979_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4979_dilations_0 = const()[name = tensor("op_4979_dilations_0"), val = tensor([1, 1])]; + tensor var_4979_groups_0 = const()[name = tensor("op_4979_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130326912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131555776))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131555968)))]; + tensor var_4979_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_4979_dilations_0, groups = var_4979_groups_0, pad = var_4979_pad_0, pad_type = var_4979_pad_type_0, strides = var_4979_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = tensor("op_4979_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = var_4979_cast_fp16, y = inputs_37_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; + tensor hidden_states_79_axes_0 = const()[name = tensor("hidden_states_79_axes_0"), val = tensor([1])]; + tensor hidden_states_79_gamma_0_to_fp16 = const()[name = tensor("hidden_states_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131558592)))]; + tensor hidden_states_79_beta_0_to_fp16 = const()[name = tensor("hidden_states_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131561216)))]; + tensor var_4989_to_fp16 = const()[name = tensor("op_4989_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_79_cast_fp16 = layer_norm(axes = hidden_states_79_axes_0, beta = hidden_states_79_beta_0_to_fp16, epsilon = var_4989_to_fp16, gamma = hidden_states_79_gamma_0_to_fp16, x = inputs_39_cast_fp16)[name = tensor("hidden_states_79_cast_fp16")]; + tensor q_27_pad_type_0 = const()[name = tensor("q_27_pad_type_0"), val = tensor("valid")]; + tensor q_27_strides_0 = const()[name = tensor("q_27_strides_0"), val = tensor([1, 1])]; + tensor q_27_pad_0 = const()[name = tensor("q_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_27_dilations_0 = const()[name = tensor("q_27_dilations_0"), val = tensor([1, 1])]; + tensor q_27_groups_0 = const()[name = tensor("q_27_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131563840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132792704))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_27_cast_fp16 = conv(dilations = q_27_dilations_0, groups = q_27_groups_0, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = q_27_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_79_cast_fp16)[name = tensor("q_27_cast_fp16")]; + tensor k_53_pad_type_0 = const()[name = tensor("k_53_pad_type_0"), val = tensor("valid")]; + tensor k_53_strides_0 = const()[name = tensor("k_53_strides_0"), val = tensor([1, 1])]; + tensor k_53_pad_0 = const()[name = tensor("k_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_53_dilations_0 = const()[name = tensor("k_53_dilations_0"), val = tensor([1, 1])]; + tensor k_53_groups_0 = const()[name = tensor("k_53_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132792896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134759040))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_53_cast_fp16 = conv(dilations = k_53_dilations_0, groups = k_53_groups_0, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = k_53_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_53_cast_fp16")]; + tensor v_27_pad_type_0 = const()[name = tensor("v_27_pad_type_0"), val = tensor("valid")]; + tensor v_27_strides_0 = const()[name = tensor("v_27_strides_0"), val = tensor([1, 1])]; + tensor v_27_pad_0 = const()[name = tensor("v_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_27_dilations_0 = const()[name = tensor("v_27_dilations_0"), val = tensor([1, 1])]; + tensor v_27_groups_0 = const()[name = tensor("v_27_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134759232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136725376))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_27_cast_fp16 = conv(dilations = v_27_dilations_0, groups = v_27_groups_0, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = v_27_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_27_cast_fp16")]; + tensor var_5022_begin_0 = const()[name = tensor("op_5022_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5022_end_0 = const()[name = tensor("op_5022_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5022_end_mask_0 = const()[name = tensor("op_5022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5022_cast_fp16 = slice_by_index(begin = var_5022_begin_0, end = var_5022_end_0, end_mask = var_5022_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5022_cast_fp16")]; + tensor var_5026_begin_0 = const()[name = tensor("op_5026_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5026_end_0 = const()[name = tensor("op_5026_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5026_end_mask_0 = const()[name = tensor("op_5026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5026_cast_fp16 = slice_by_index(begin = var_5026_begin_0, end = var_5026_end_0, end_mask = var_5026_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5026_cast_fp16")]; + tensor var_5030_begin_0 = const()[name = tensor("op_5030_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5030_end_0 = const()[name = tensor("op_5030_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5030_end_mask_0 = const()[name = tensor("op_5030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5030_cast_fp16 = slice_by_index(begin = var_5030_begin_0, end = var_5030_end_0, end_mask = var_5030_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5030_cast_fp16")]; + tensor var_5034_begin_0 = const()[name = tensor("op_5034_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5034_end_0 = const()[name = tensor("op_5034_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5034_end_mask_0 = const()[name = tensor("op_5034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5034_cast_fp16 = slice_by_index(begin = var_5034_begin_0, end = var_5034_end_0, end_mask = var_5034_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5034_cast_fp16")]; + tensor var_5038_begin_0 = const()[name = tensor("op_5038_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5038_end_0 = const()[name = tensor("op_5038_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5038_end_mask_0 = const()[name = tensor("op_5038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5038_cast_fp16 = slice_by_index(begin = var_5038_begin_0, end = var_5038_end_0, end_mask = var_5038_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5038_cast_fp16")]; + tensor var_5042_begin_0 = const()[name = tensor("op_5042_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5042_end_0 = const()[name = tensor("op_5042_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5042_end_mask_0 = const()[name = tensor("op_5042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5042_cast_fp16 = slice_by_index(begin = var_5042_begin_0, end = var_5042_end_0, end_mask = var_5042_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5042_cast_fp16")]; + tensor var_5046_begin_0 = const()[name = tensor("op_5046_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5046_end_0 = const()[name = tensor("op_5046_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5046_end_mask_0 = const()[name = tensor("op_5046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5046_cast_fp16 = slice_by_index(begin = var_5046_begin_0, end = var_5046_end_0, end_mask = var_5046_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5046_cast_fp16")]; + tensor var_5050_begin_0 = const()[name = tensor("op_5050_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5050_end_0 = const()[name = tensor("op_5050_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5050_end_mask_0 = const()[name = tensor("op_5050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5050_cast_fp16 = slice_by_index(begin = var_5050_begin_0, end = var_5050_end_0, end_mask = var_5050_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5050_cast_fp16")]; + tensor var_5054_begin_0 = const()[name = tensor("op_5054_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5054_end_0 = const()[name = tensor("op_5054_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5054_end_mask_0 = const()[name = tensor("op_5054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5054_cast_fp16 = slice_by_index(begin = var_5054_begin_0, end = var_5054_end_0, end_mask = var_5054_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5054_cast_fp16")]; + tensor var_5058_begin_0 = const()[name = tensor("op_5058_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5058_end_0 = const()[name = tensor("op_5058_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5058_end_mask_0 = const()[name = tensor("op_5058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5058_cast_fp16 = slice_by_index(begin = var_5058_begin_0, end = var_5058_end_0, end_mask = var_5058_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5058_cast_fp16")]; + tensor var_5062_begin_0 = const()[name = tensor("op_5062_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5062_end_0 = const()[name = tensor("op_5062_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5062_end_mask_0 = const()[name = tensor("op_5062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5062_cast_fp16 = slice_by_index(begin = var_5062_begin_0, end = var_5062_end_0, end_mask = var_5062_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5062_cast_fp16")]; + tensor var_5066_begin_0 = const()[name = tensor("op_5066_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5066_end_0 = const()[name = tensor("op_5066_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5066_end_mask_0 = const()[name = tensor("op_5066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5066_cast_fp16 = slice_by_index(begin = var_5066_begin_0, end = var_5066_end_0, end_mask = var_5066_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5066_cast_fp16")]; + tensor var_5070_begin_0 = const()[name = tensor("op_5070_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5070_end_0 = const()[name = tensor("op_5070_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5070_end_mask_0 = const()[name = tensor("op_5070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5070_cast_fp16 = slice_by_index(begin = var_5070_begin_0, end = var_5070_end_0, end_mask = var_5070_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5070_cast_fp16")]; + tensor var_5074_begin_0 = const()[name = tensor("op_5074_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5074_end_0 = const()[name = tensor("op_5074_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5074_end_mask_0 = const()[name = tensor("op_5074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5074_cast_fp16 = slice_by_index(begin = var_5074_begin_0, end = var_5074_end_0, end_mask = var_5074_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5074_cast_fp16")]; + tensor var_5078_begin_0 = const()[name = tensor("op_5078_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5078_end_0 = const()[name = tensor("op_5078_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5078_end_mask_0 = const()[name = tensor("op_5078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5078_cast_fp16 = slice_by_index(begin = var_5078_begin_0, end = var_5078_end_0, end_mask = var_5078_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5078_cast_fp16")]; + tensor var_5082_begin_0 = const()[name = tensor("op_5082_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5082_end_0 = const()[name = tensor("op_5082_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5082_end_mask_0 = const()[name = tensor("op_5082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5082_cast_fp16 = slice_by_index(begin = var_5082_begin_0, end = var_5082_end_0, end_mask = var_5082_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5082_cast_fp16")]; + tensor var_5086_begin_0 = const()[name = tensor("op_5086_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5086_end_0 = const()[name = tensor("op_5086_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5086_end_mask_0 = const()[name = tensor("op_5086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5086_cast_fp16 = slice_by_index(begin = var_5086_begin_0, end = var_5086_end_0, end_mask = var_5086_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5086_cast_fp16")]; + tensor var_5090_begin_0 = const()[name = tensor("op_5090_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5090_end_0 = const()[name = tensor("op_5090_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_5090_end_mask_0 = const()[name = tensor("op_5090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5090_cast_fp16 = slice_by_index(begin = var_5090_begin_0, end = var_5090_end_0, end_mask = var_5090_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5090_cast_fp16")]; + tensor var_5094_begin_0 = const()[name = tensor("op_5094_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5094_end_0 = const()[name = tensor("op_5094_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_5094_end_mask_0 = const()[name = tensor("op_5094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5094_cast_fp16 = slice_by_index(begin = var_5094_begin_0, end = var_5094_end_0, end_mask = var_5094_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5094_cast_fp16")]; + tensor var_5098_begin_0 = const()[name = tensor("op_5098_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5098_end_0 = const()[name = tensor("op_5098_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_5098_end_mask_0 = const()[name = tensor("op_5098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5098_cast_fp16 = slice_by_index(begin = var_5098_begin_0, end = var_5098_end_0, end_mask = var_5098_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5098_cast_fp16")]; + tensor k_55_perm_0 = const()[name = tensor("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5105_begin_0 = const()[name = tensor("op_5105_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5105_end_0 = const()[name = tensor("op_5105_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_5105_end_mask_0 = const()[name = tensor("op_5105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_55_cast_fp16 = transpose(perm = k_55_perm_0, x = k_53_cast_fp16)[name = tensor("transpose_54")]; + tensor var_5105_cast_fp16 = slice_by_index(begin = var_5105_begin_0, end = var_5105_end_0, end_mask = var_5105_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5105_cast_fp16")]; + tensor var_5109_begin_0 = const()[name = tensor("op_5109_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5109_end_0 = const()[name = tensor("op_5109_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_5109_end_mask_0 = const()[name = tensor("op_5109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5109_cast_fp16 = slice_by_index(begin = var_5109_begin_0, end = var_5109_end_0, end_mask = var_5109_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5109_cast_fp16")]; + tensor var_5113_begin_0 = const()[name = tensor("op_5113_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5113_end_0 = const()[name = tensor("op_5113_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_5113_end_mask_0 = const()[name = tensor("op_5113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5113_cast_fp16 = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5113_cast_fp16")]; + tensor var_5117_begin_0 = const()[name = tensor("op_5117_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5117_end_0 = const()[name = tensor("op_5117_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_5117_end_mask_0 = const()[name = tensor("op_5117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5117_cast_fp16 = slice_by_index(begin = var_5117_begin_0, end = var_5117_end_0, end_mask = var_5117_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5117_cast_fp16")]; + tensor var_5121_begin_0 = const()[name = tensor("op_5121_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5121_end_0 = const()[name = tensor("op_5121_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_5121_end_mask_0 = const()[name = tensor("op_5121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5121_cast_fp16 = slice_by_index(begin = var_5121_begin_0, end = var_5121_end_0, end_mask = var_5121_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5121_cast_fp16")]; + tensor var_5125_begin_0 = const()[name = tensor("op_5125_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5125_end_0 = const()[name = tensor("op_5125_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_5125_end_mask_0 = const()[name = tensor("op_5125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5125_cast_fp16 = slice_by_index(begin = var_5125_begin_0, end = var_5125_end_0, end_mask = var_5125_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5125_cast_fp16")]; + tensor var_5129_begin_0 = const()[name = tensor("op_5129_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5129_end_0 = const()[name = tensor("op_5129_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_5129_end_mask_0 = const()[name = tensor("op_5129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5129_cast_fp16 = slice_by_index(begin = var_5129_begin_0, end = var_5129_end_0, end_mask = var_5129_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5129_cast_fp16")]; + tensor var_5133_begin_0 = const()[name = tensor("op_5133_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5133_end_0 = const()[name = tensor("op_5133_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_5133_end_mask_0 = const()[name = tensor("op_5133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5133_cast_fp16 = slice_by_index(begin = var_5133_begin_0, end = var_5133_end_0, end_mask = var_5133_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5133_cast_fp16")]; + tensor var_5137_begin_0 = const()[name = tensor("op_5137_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5137_end_0 = const()[name = tensor("op_5137_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_5137_end_mask_0 = const()[name = tensor("op_5137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5137_cast_fp16 = slice_by_index(begin = var_5137_begin_0, end = var_5137_end_0, end_mask = var_5137_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5137_cast_fp16")]; + tensor var_5141_begin_0 = const()[name = tensor("op_5141_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5141_end_0 = const()[name = tensor("op_5141_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_5141_end_mask_0 = const()[name = tensor("op_5141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5141_cast_fp16 = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5141_cast_fp16")]; + tensor var_5145_begin_0 = const()[name = tensor("op_5145_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5145_end_0 = const()[name = tensor("op_5145_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_5145_end_mask_0 = const()[name = tensor("op_5145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5145_cast_fp16 = slice_by_index(begin = var_5145_begin_0, end = var_5145_end_0, end_mask = var_5145_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5145_cast_fp16")]; + tensor var_5149_begin_0 = const()[name = tensor("op_5149_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5149_end_0 = const()[name = tensor("op_5149_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_5149_end_mask_0 = const()[name = tensor("op_5149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5149_cast_fp16 = slice_by_index(begin = var_5149_begin_0, end = var_5149_end_0, end_mask = var_5149_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5149_cast_fp16")]; + tensor var_5153_begin_0 = const()[name = tensor("op_5153_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5153_end_0 = const()[name = tensor("op_5153_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_5153_end_mask_0 = const()[name = tensor("op_5153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5153_cast_fp16 = slice_by_index(begin = var_5153_begin_0, end = var_5153_end_0, end_mask = var_5153_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5153_cast_fp16")]; + tensor var_5157_begin_0 = const()[name = tensor("op_5157_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5157_end_0 = const()[name = tensor("op_5157_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_5157_end_mask_0 = const()[name = tensor("op_5157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5157_cast_fp16 = slice_by_index(begin = var_5157_begin_0, end = var_5157_end_0, end_mask = var_5157_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5157_cast_fp16")]; + tensor var_5161_begin_0 = const()[name = tensor("op_5161_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5161_end_0 = const()[name = tensor("op_5161_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_5161_end_mask_0 = const()[name = tensor("op_5161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5161_cast_fp16 = slice_by_index(begin = var_5161_begin_0, end = var_5161_end_0, end_mask = var_5161_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5161_cast_fp16")]; + tensor var_5165_begin_0 = const()[name = tensor("op_5165_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5165_end_0 = const()[name = tensor("op_5165_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_5165_end_mask_0 = const()[name = tensor("op_5165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5165_cast_fp16 = slice_by_index(begin = var_5165_begin_0, end = var_5165_end_0, end_mask = var_5165_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5165_cast_fp16")]; + tensor var_5169_begin_0 = const()[name = tensor("op_5169_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5169_end_0 = const()[name = tensor("op_5169_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_5169_end_mask_0 = const()[name = tensor("op_5169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5169_cast_fp16 = slice_by_index(begin = var_5169_begin_0, end = var_5169_end_0, end_mask = var_5169_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5169_cast_fp16")]; + tensor var_5173_begin_0 = const()[name = tensor("op_5173_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5173_end_0 = const()[name = tensor("op_5173_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_5173_end_mask_0 = const()[name = tensor("op_5173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5173_cast_fp16 = slice_by_index(begin = var_5173_begin_0, end = var_5173_end_0, end_mask = var_5173_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5173_cast_fp16")]; + tensor var_5177_begin_0 = const()[name = tensor("op_5177_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5177_end_0 = const()[name = tensor("op_5177_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_5177_end_mask_0 = const()[name = tensor("op_5177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5177_cast_fp16 = slice_by_index(begin = var_5177_begin_0, end = var_5177_end_0, end_mask = var_5177_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5177_cast_fp16")]; + tensor var_5181_begin_0 = const()[name = tensor("op_5181_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5181_end_0 = const()[name = tensor("op_5181_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_5181_end_mask_0 = const()[name = tensor("op_5181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5181_cast_fp16 = slice_by_index(begin = var_5181_begin_0, end = var_5181_end_0, end_mask = var_5181_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5181_cast_fp16")]; + tensor var_5183_begin_0 = const()[name = tensor("op_5183_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5183_end_0 = const()[name = tensor("op_5183_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_5183_end_mask_0 = const()[name = tensor("op_5183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5183_cast_fp16 = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5183_cast_fp16")]; + tensor var_5187_begin_0 = const()[name = tensor("op_5187_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5187_end_0 = const()[name = tensor("op_5187_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_5187_end_mask_0 = const()[name = tensor("op_5187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5187_cast_fp16 = slice_by_index(begin = var_5187_begin_0, end = var_5187_end_0, end_mask = var_5187_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5187_cast_fp16")]; + tensor var_5191_begin_0 = const()[name = tensor("op_5191_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5191_end_0 = const()[name = tensor("op_5191_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_5191_end_mask_0 = const()[name = tensor("op_5191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5191_cast_fp16 = slice_by_index(begin = var_5191_begin_0, end = var_5191_end_0, end_mask = var_5191_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5191_cast_fp16")]; + tensor var_5195_begin_0 = const()[name = tensor("op_5195_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5195_end_0 = const()[name = tensor("op_5195_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_5195_end_mask_0 = const()[name = tensor("op_5195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5195_cast_fp16 = slice_by_index(begin = var_5195_begin_0, end = var_5195_end_0, end_mask = var_5195_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5195_cast_fp16")]; + tensor var_5199_begin_0 = const()[name = tensor("op_5199_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5199_end_0 = const()[name = tensor("op_5199_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_5199_end_mask_0 = const()[name = tensor("op_5199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5199_cast_fp16 = slice_by_index(begin = var_5199_begin_0, end = var_5199_end_0, end_mask = var_5199_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5199_cast_fp16")]; + tensor var_5203_begin_0 = const()[name = tensor("op_5203_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5203_end_0 = const()[name = tensor("op_5203_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_5203_end_mask_0 = const()[name = tensor("op_5203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5203_cast_fp16 = slice_by_index(begin = var_5203_begin_0, end = var_5203_end_0, end_mask = var_5203_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5203_cast_fp16")]; + tensor var_5207_begin_0 = const()[name = tensor("op_5207_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5207_end_0 = const()[name = tensor("op_5207_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_5207_end_mask_0 = const()[name = tensor("op_5207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5207_cast_fp16 = slice_by_index(begin = var_5207_begin_0, end = var_5207_end_0, end_mask = var_5207_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5207_cast_fp16")]; + tensor var_5211_begin_0 = const()[name = tensor("op_5211_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5211_end_0 = const()[name = tensor("op_5211_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_5211_end_mask_0 = const()[name = tensor("op_5211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5211_cast_fp16 = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5211_cast_fp16")]; + tensor var_5215_begin_0 = const()[name = tensor("op_5215_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5215_end_0 = const()[name = tensor("op_5215_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_5215_end_mask_0 = const()[name = tensor("op_5215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5215_cast_fp16 = slice_by_index(begin = var_5215_begin_0, end = var_5215_end_0, end_mask = var_5215_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5215_cast_fp16")]; + tensor var_5219_begin_0 = const()[name = tensor("op_5219_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5219_end_0 = const()[name = tensor("op_5219_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_5219_end_mask_0 = const()[name = tensor("op_5219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5219_cast_fp16 = slice_by_index(begin = var_5219_begin_0, end = var_5219_end_0, end_mask = var_5219_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5219_cast_fp16")]; + tensor var_5223_begin_0 = const()[name = tensor("op_5223_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5223_end_0 = const()[name = tensor("op_5223_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_5223_end_mask_0 = const()[name = tensor("op_5223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5223_cast_fp16 = slice_by_index(begin = var_5223_begin_0, end = var_5223_end_0, end_mask = var_5223_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5223_cast_fp16")]; + tensor var_5227_begin_0 = const()[name = tensor("op_5227_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5227_end_0 = const()[name = tensor("op_5227_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_5227_end_mask_0 = const()[name = tensor("op_5227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5227_cast_fp16 = slice_by_index(begin = var_5227_begin_0, end = var_5227_end_0, end_mask = var_5227_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5227_cast_fp16")]; + tensor var_5231_begin_0 = const()[name = tensor("op_5231_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5231_end_0 = const()[name = tensor("op_5231_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_5231_end_mask_0 = const()[name = tensor("op_5231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5231_cast_fp16 = slice_by_index(begin = var_5231_begin_0, end = var_5231_end_0, end_mask = var_5231_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5231_cast_fp16")]; + tensor var_5235_begin_0 = const()[name = tensor("op_5235_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5235_end_0 = const()[name = tensor("op_5235_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_5235_end_mask_0 = const()[name = tensor("op_5235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5235_cast_fp16 = slice_by_index(begin = var_5235_begin_0, end = var_5235_end_0, end_mask = var_5235_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5235_cast_fp16")]; + tensor var_5239_begin_0 = const()[name = tensor("op_5239_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5239_end_0 = const()[name = tensor("op_5239_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_5239_end_mask_0 = const()[name = tensor("op_5239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5239_cast_fp16 = slice_by_index(begin = var_5239_begin_0, end = var_5239_end_0, end_mask = var_5239_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5239_cast_fp16")]; + tensor var_5243_begin_0 = const()[name = tensor("op_5243_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5243_end_0 = const()[name = tensor("op_5243_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_5243_end_mask_0 = const()[name = tensor("op_5243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5243_cast_fp16 = slice_by_index(begin = var_5243_begin_0, end = var_5243_end_0, end_mask = var_5243_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5243_cast_fp16")]; + tensor var_5247_begin_0 = const()[name = tensor("op_5247_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5247_end_0 = const()[name = tensor("op_5247_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_5247_end_mask_0 = const()[name = tensor("op_5247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5247_cast_fp16 = slice_by_index(begin = var_5247_begin_0, end = var_5247_end_0, end_mask = var_5247_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5247_cast_fp16")]; + tensor var_5251_begin_0 = const()[name = tensor("op_5251_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5251_end_0 = const()[name = tensor("op_5251_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_5251_end_mask_0 = const()[name = tensor("op_5251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5251_cast_fp16 = slice_by_index(begin = var_5251_begin_0, end = var_5251_end_0, end_mask = var_5251_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5251_cast_fp16")]; + tensor var_5255_begin_0 = const()[name = tensor("op_5255_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5255_end_0 = const()[name = tensor("op_5255_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_5255_end_mask_0 = const()[name = tensor("op_5255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5255_cast_fp16 = slice_by_index(begin = var_5255_begin_0, end = var_5255_end_0, end_mask = var_5255_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5255_cast_fp16")]; + tensor var_5259_begin_0 = const()[name = tensor("op_5259_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5259_end_0 = const()[name = tensor("op_5259_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_5259_end_mask_0 = const()[name = tensor("op_5259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5259_cast_fp16 = slice_by_index(begin = var_5259_begin_0, end = var_5259_end_0, end_mask = var_5259_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5259_cast_fp16")]; + tensor var_5263_equation_0 = const()[name = tensor("op_5263_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5263_cast_fp16 = einsum(equation = var_5263_equation_0, values = (var_5105_cast_fp16, var_5022_cast_fp16))[name = tensor("op_5263_cast_fp16")]; + tensor var_5264_to_fp16 = const()[name = tensor("op_5264_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_361_cast_fp16 = mul(x = var_5263_cast_fp16, y = var_5264_to_fp16)[name = tensor("aw_361_cast_fp16")]; + tensor var_5267_equation_0 = const()[name = tensor("op_5267_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5267_cast_fp16 = einsum(equation = var_5267_equation_0, values = (var_5109_cast_fp16, var_5026_cast_fp16))[name = tensor("op_5267_cast_fp16")]; + tensor var_5268_to_fp16 = const()[name = tensor("op_5268_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_363_cast_fp16 = mul(x = var_5267_cast_fp16, y = var_5268_to_fp16)[name = tensor("aw_363_cast_fp16")]; + tensor var_5271_equation_0 = const()[name = tensor("op_5271_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5271_cast_fp16 = einsum(equation = var_5271_equation_0, values = (var_5113_cast_fp16, var_5030_cast_fp16))[name = tensor("op_5271_cast_fp16")]; + tensor var_5272_to_fp16 = const()[name = tensor("op_5272_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_365_cast_fp16 = mul(x = var_5271_cast_fp16, y = var_5272_to_fp16)[name = tensor("aw_365_cast_fp16")]; + tensor var_5275_equation_0 = const()[name = tensor("op_5275_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5275_cast_fp16 = einsum(equation = var_5275_equation_0, values = (var_5117_cast_fp16, var_5034_cast_fp16))[name = tensor("op_5275_cast_fp16")]; + tensor var_5276_to_fp16 = const()[name = tensor("op_5276_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_367_cast_fp16 = mul(x = var_5275_cast_fp16, y = var_5276_to_fp16)[name = tensor("aw_367_cast_fp16")]; + tensor var_5279_equation_0 = const()[name = tensor("op_5279_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5279_cast_fp16 = einsum(equation = var_5279_equation_0, values = (var_5121_cast_fp16, var_5038_cast_fp16))[name = tensor("op_5279_cast_fp16")]; + tensor var_5280_to_fp16 = const()[name = tensor("op_5280_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_369_cast_fp16 = mul(x = var_5279_cast_fp16, y = var_5280_to_fp16)[name = tensor("aw_369_cast_fp16")]; + tensor var_5283_equation_0 = const()[name = tensor("op_5283_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5283_cast_fp16 = einsum(equation = var_5283_equation_0, values = (var_5125_cast_fp16, var_5042_cast_fp16))[name = tensor("op_5283_cast_fp16")]; + tensor var_5284_to_fp16 = const()[name = tensor("op_5284_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_371_cast_fp16 = mul(x = var_5283_cast_fp16, y = var_5284_to_fp16)[name = tensor("aw_371_cast_fp16")]; + tensor var_5287_equation_0 = const()[name = tensor("op_5287_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5287_cast_fp16 = einsum(equation = var_5287_equation_0, values = (var_5129_cast_fp16, var_5046_cast_fp16))[name = tensor("op_5287_cast_fp16")]; + tensor var_5288_to_fp16 = const()[name = tensor("op_5288_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_373_cast_fp16 = mul(x = var_5287_cast_fp16, y = var_5288_to_fp16)[name = tensor("aw_373_cast_fp16")]; + tensor var_5291_equation_0 = const()[name = tensor("op_5291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5291_cast_fp16 = einsum(equation = var_5291_equation_0, values = (var_5133_cast_fp16, var_5050_cast_fp16))[name = tensor("op_5291_cast_fp16")]; + tensor var_5292_to_fp16 = const()[name = tensor("op_5292_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_375_cast_fp16 = mul(x = var_5291_cast_fp16, y = var_5292_to_fp16)[name = tensor("aw_375_cast_fp16")]; + tensor var_5295_equation_0 = const()[name = tensor("op_5295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5295_cast_fp16 = einsum(equation = var_5295_equation_0, values = (var_5137_cast_fp16, var_5054_cast_fp16))[name = tensor("op_5295_cast_fp16")]; + tensor var_5296_to_fp16 = const()[name = tensor("op_5296_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_377_cast_fp16 = mul(x = var_5295_cast_fp16, y = var_5296_to_fp16)[name = tensor("aw_377_cast_fp16")]; + tensor var_5299_equation_0 = const()[name = tensor("op_5299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5299_cast_fp16 = einsum(equation = var_5299_equation_0, values = (var_5141_cast_fp16, var_5058_cast_fp16))[name = tensor("op_5299_cast_fp16")]; + tensor var_5300_to_fp16 = const()[name = tensor("op_5300_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_379_cast_fp16 = mul(x = var_5299_cast_fp16, y = var_5300_to_fp16)[name = tensor("aw_379_cast_fp16")]; + tensor var_5303_equation_0 = const()[name = tensor("op_5303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5303_cast_fp16 = einsum(equation = var_5303_equation_0, values = (var_5145_cast_fp16, var_5062_cast_fp16))[name = tensor("op_5303_cast_fp16")]; + tensor var_5304_to_fp16 = const()[name = tensor("op_5304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_381_cast_fp16 = mul(x = var_5303_cast_fp16, y = var_5304_to_fp16)[name = tensor("aw_381_cast_fp16")]; + tensor var_5307_equation_0 = const()[name = tensor("op_5307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5307_cast_fp16 = einsum(equation = var_5307_equation_0, values = (var_5149_cast_fp16, var_5066_cast_fp16))[name = tensor("op_5307_cast_fp16")]; + tensor var_5308_to_fp16 = const()[name = tensor("op_5308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_383_cast_fp16 = mul(x = var_5307_cast_fp16, y = var_5308_to_fp16)[name = tensor("aw_383_cast_fp16")]; + tensor var_5311_equation_0 = const()[name = tensor("op_5311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5311_cast_fp16 = einsum(equation = var_5311_equation_0, values = (var_5153_cast_fp16, var_5070_cast_fp16))[name = tensor("op_5311_cast_fp16")]; + tensor var_5312_to_fp16 = const()[name = tensor("op_5312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_385_cast_fp16 = mul(x = var_5311_cast_fp16, y = var_5312_to_fp16)[name = tensor("aw_385_cast_fp16")]; + tensor var_5315_equation_0 = const()[name = tensor("op_5315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5315_cast_fp16 = einsum(equation = var_5315_equation_0, values = (var_5157_cast_fp16, var_5074_cast_fp16))[name = tensor("op_5315_cast_fp16")]; + tensor var_5316_to_fp16 = const()[name = tensor("op_5316_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_387_cast_fp16 = mul(x = var_5315_cast_fp16, y = var_5316_to_fp16)[name = tensor("aw_387_cast_fp16")]; + tensor var_5319_equation_0 = const()[name = tensor("op_5319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5319_cast_fp16 = einsum(equation = var_5319_equation_0, values = (var_5161_cast_fp16, var_5078_cast_fp16))[name = tensor("op_5319_cast_fp16")]; + tensor var_5320_to_fp16 = const()[name = tensor("op_5320_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_389_cast_fp16 = mul(x = var_5319_cast_fp16, y = var_5320_to_fp16)[name = tensor("aw_389_cast_fp16")]; + tensor var_5323_equation_0 = const()[name = tensor("op_5323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5323_cast_fp16 = einsum(equation = var_5323_equation_0, values = (var_5165_cast_fp16, var_5082_cast_fp16))[name = tensor("op_5323_cast_fp16")]; + tensor var_5324_to_fp16 = const()[name = tensor("op_5324_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_391_cast_fp16 = mul(x = var_5323_cast_fp16, y = var_5324_to_fp16)[name = tensor("aw_391_cast_fp16")]; + tensor var_5327_equation_0 = const()[name = tensor("op_5327_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5327_cast_fp16 = einsum(equation = var_5327_equation_0, values = (var_5169_cast_fp16, var_5086_cast_fp16))[name = tensor("op_5327_cast_fp16")]; + tensor var_5328_to_fp16 = const()[name = tensor("op_5328_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_393_cast_fp16 = mul(x = var_5327_cast_fp16, y = var_5328_to_fp16)[name = tensor("aw_393_cast_fp16")]; + tensor var_5331_equation_0 = const()[name = tensor("op_5331_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5331_cast_fp16 = einsum(equation = var_5331_equation_0, values = (var_5173_cast_fp16, var_5090_cast_fp16))[name = tensor("op_5331_cast_fp16")]; + tensor var_5332_to_fp16 = const()[name = tensor("op_5332_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_395_cast_fp16 = mul(x = var_5331_cast_fp16, y = var_5332_to_fp16)[name = tensor("aw_395_cast_fp16")]; + tensor var_5335_equation_0 = const()[name = tensor("op_5335_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5335_cast_fp16 = einsum(equation = var_5335_equation_0, values = (var_5177_cast_fp16, var_5094_cast_fp16))[name = tensor("op_5335_cast_fp16")]; + tensor var_5336_to_fp16 = const()[name = tensor("op_5336_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_397_cast_fp16 = mul(x = var_5335_cast_fp16, y = var_5336_to_fp16)[name = tensor("aw_397_cast_fp16")]; + tensor var_5339_equation_0 = const()[name = tensor("op_5339_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5339_cast_fp16 = einsum(equation = var_5339_equation_0, values = (var_5181_cast_fp16, var_5098_cast_fp16))[name = tensor("op_5339_cast_fp16")]; + tensor var_5340_to_fp16 = const()[name = tensor("op_5340_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_399_cast_fp16 = mul(x = var_5339_cast_fp16, y = var_5340_to_fp16)[name = tensor("aw_399_cast_fp16")]; + tensor var_5342_cast_fp16 = softmax(axis = var_2624, x = aw_361_cast_fp16)[name = tensor("op_5342_cast_fp16")]; + tensor var_5343_cast_fp16 = softmax(axis = var_2624, x = aw_363_cast_fp16)[name = tensor("op_5343_cast_fp16")]; + tensor var_5344_cast_fp16 = softmax(axis = var_2624, x = aw_365_cast_fp16)[name = tensor("op_5344_cast_fp16")]; + tensor var_5345_cast_fp16 = softmax(axis = var_2624, x = aw_367_cast_fp16)[name = tensor("op_5345_cast_fp16")]; + tensor var_5346_cast_fp16 = softmax(axis = var_2624, x = aw_369_cast_fp16)[name = tensor("op_5346_cast_fp16")]; + tensor var_5347_cast_fp16 = softmax(axis = var_2624, x = aw_371_cast_fp16)[name = tensor("op_5347_cast_fp16")]; + tensor var_5348_cast_fp16 = softmax(axis = var_2624, x = aw_373_cast_fp16)[name = tensor("op_5348_cast_fp16")]; + tensor var_5349_cast_fp16 = softmax(axis = var_2624, x = aw_375_cast_fp16)[name = tensor("op_5349_cast_fp16")]; + tensor var_5350_cast_fp16 = softmax(axis = var_2624, x = aw_377_cast_fp16)[name = tensor("op_5350_cast_fp16")]; + tensor var_5351_cast_fp16 = softmax(axis = var_2624, x = aw_379_cast_fp16)[name = tensor("op_5351_cast_fp16")]; + tensor var_5352_cast_fp16 = softmax(axis = var_2624, x = aw_381_cast_fp16)[name = tensor("op_5352_cast_fp16")]; + tensor var_5353_cast_fp16 = softmax(axis = var_2624, x = aw_383_cast_fp16)[name = tensor("op_5353_cast_fp16")]; + tensor var_5354_cast_fp16 = softmax(axis = var_2624, x = aw_385_cast_fp16)[name = tensor("op_5354_cast_fp16")]; + tensor var_5355_cast_fp16 = softmax(axis = var_2624, x = aw_387_cast_fp16)[name = tensor("op_5355_cast_fp16")]; + tensor var_5356_cast_fp16 = softmax(axis = var_2624, x = aw_389_cast_fp16)[name = tensor("op_5356_cast_fp16")]; + tensor var_5357_cast_fp16 = softmax(axis = var_2624, x = aw_391_cast_fp16)[name = tensor("op_5357_cast_fp16")]; + tensor var_5358_cast_fp16 = softmax(axis = var_2624, x = aw_393_cast_fp16)[name = tensor("op_5358_cast_fp16")]; + tensor var_5359_cast_fp16 = softmax(axis = var_2624, x = aw_395_cast_fp16)[name = tensor("op_5359_cast_fp16")]; + tensor var_5360_cast_fp16 = softmax(axis = var_2624, x = aw_397_cast_fp16)[name = tensor("op_5360_cast_fp16")]; + tensor var_5361_cast_fp16 = softmax(axis = var_2624, x = aw_399_cast_fp16)[name = tensor("op_5361_cast_fp16")]; + tensor var_5363_equation_0 = const()[name = tensor("op_5363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5363_cast_fp16 = einsum(equation = var_5363_equation_0, values = (var_5183_cast_fp16, var_5342_cast_fp16))[name = tensor("op_5363_cast_fp16")]; + tensor var_5365_equation_0 = const()[name = tensor("op_5365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5365_cast_fp16 = einsum(equation = var_5365_equation_0, values = (var_5187_cast_fp16, var_5343_cast_fp16))[name = tensor("op_5365_cast_fp16")]; + tensor var_5367_equation_0 = const()[name = tensor("op_5367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5367_cast_fp16 = einsum(equation = var_5367_equation_0, values = (var_5191_cast_fp16, var_5344_cast_fp16))[name = tensor("op_5367_cast_fp16")]; + tensor var_5369_equation_0 = const()[name = tensor("op_5369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5369_cast_fp16 = einsum(equation = var_5369_equation_0, values = (var_5195_cast_fp16, var_5345_cast_fp16))[name = tensor("op_5369_cast_fp16")]; + tensor var_5371_equation_0 = const()[name = tensor("op_5371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5371_cast_fp16 = einsum(equation = var_5371_equation_0, values = (var_5199_cast_fp16, var_5346_cast_fp16))[name = tensor("op_5371_cast_fp16")]; + tensor var_5373_equation_0 = const()[name = tensor("op_5373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5373_cast_fp16 = einsum(equation = var_5373_equation_0, values = (var_5203_cast_fp16, var_5347_cast_fp16))[name = tensor("op_5373_cast_fp16")]; + tensor var_5375_equation_0 = const()[name = tensor("op_5375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5375_cast_fp16 = einsum(equation = var_5375_equation_0, values = (var_5207_cast_fp16, var_5348_cast_fp16))[name = tensor("op_5375_cast_fp16")]; + tensor var_5377_equation_0 = const()[name = tensor("op_5377_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5377_cast_fp16 = einsum(equation = var_5377_equation_0, values = (var_5211_cast_fp16, var_5349_cast_fp16))[name = tensor("op_5377_cast_fp16")]; + tensor var_5379_equation_0 = const()[name = tensor("op_5379_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5379_cast_fp16 = einsum(equation = var_5379_equation_0, values = (var_5215_cast_fp16, var_5350_cast_fp16))[name = tensor("op_5379_cast_fp16")]; + tensor var_5381_equation_0 = const()[name = tensor("op_5381_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5381_cast_fp16 = einsum(equation = var_5381_equation_0, values = (var_5219_cast_fp16, var_5351_cast_fp16))[name = tensor("op_5381_cast_fp16")]; + tensor var_5383_equation_0 = const()[name = tensor("op_5383_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5383_cast_fp16 = einsum(equation = var_5383_equation_0, values = (var_5223_cast_fp16, var_5352_cast_fp16))[name = tensor("op_5383_cast_fp16")]; + tensor var_5385_equation_0 = const()[name = tensor("op_5385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5385_cast_fp16 = einsum(equation = var_5385_equation_0, values = (var_5227_cast_fp16, var_5353_cast_fp16))[name = tensor("op_5385_cast_fp16")]; + tensor var_5387_equation_0 = const()[name = tensor("op_5387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5387_cast_fp16 = einsum(equation = var_5387_equation_0, values = (var_5231_cast_fp16, var_5354_cast_fp16))[name = tensor("op_5387_cast_fp16")]; + tensor var_5389_equation_0 = const()[name = tensor("op_5389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5389_cast_fp16 = einsum(equation = var_5389_equation_0, values = (var_5235_cast_fp16, var_5355_cast_fp16))[name = tensor("op_5389_cast_fp16")]; + tensor var_5391_equation_0 = const()[name = tensor("op_5391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5391_cast_fp16 = einsum(equation = var_5391_equation_0, values = (var_5239_cast_fp16, var_5356_cast_fp16))[name = tensor("op_5391_cast_fp16")]; + tensor var_5393_equation_0 = const()[name = tensor("op_5393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5393_cast_fp16 = einsum(equation = var_5393_equation_0, values = (var_5243_cast_fp16, var_5357_cast_fp16))[name = tensor("op_5393_cast_fp16")]; + tensor var_5395_equation_0 = const()[name = tensor("op_5395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5395_cast_fp16 = einsum(equation = var_5395_equation_0, values = (var_5247_cast_fp16, var_5358_cast_fp16))[name = tensor("op_5395_cast_fp16")]; + tensor var_5397_equation_0 = const()[name = tensor("op_5397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5397_cast_fp16 = einsum(equation = var_5397_equation_0, values = (var_5251_cast_fp16, var_5359_cast_fp16))[name = tensor("op_5397_cast_fp16")]; + tensor var_5399_equation_0 = const()[name = tensor("op_5399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5399_cast_fp16 = einsum(equation = var_5399_equation_0, values = (var_5255_cast_fp16, var_5360_cast_fp16))[name = tensor("op_5399_cast_fp16")]; + tensor var_5401_equation_0 = const()[name = tensor("op_5401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5401_cast_fp16 = einsum(equation = var_5401_equation_0, values = (var_5259_cast_fp16, var_5361_cast_fp16))[name = tensor("op_5401_cast_fp16")]; + tensor input_149_interleave_0 = const()[name = tensor("input_149_interleave_0"), val = tensor(false)]; + tensor input_149_cast_fp16 = concat(axis = var_2624, interleave = input_149_interleave_0, values = (var_5363_cast_fp16, var_5365_cast_fp16, var_5367_cast_fp16, var_5369_cast_fp16, var_5371_cast_fp16, var_5373_cast_fp16, var_5375_cast_fp16, var_5377_cast_fp16, var_5379_cast_fp16, var_5381_cast_fp16, var_5383_cast_fp16, var_5385_cast_fp16, var_5387_cast_fp16, var_5389_cast_fp16, var_5391_cast_fp16, var_5393_cast_fp16, var_5395_cast_fp16, var_5397_cast_fp16, var_5399_cast_fp16, var_5401_cast_fp16))[name = tensor("input_149_cast_fp16")]; + tensor var_5411_pad_type_0 = const()[name = tensor("op_5411_pad_type_0"), val = tensor("valid")]; + tensor var_5411_strides_0 = const()[name = tensor("op_5411_strides_0"), val = tensor([1, 1])]; + tensor var_5411_pad_0 = const()[name = tensor("op_5411_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5411_dilations_0 = const()[name = tensor("op_5411_dilations_0"), val = tensor([1, 1])]; + tensor var_5411_groups_0 = const()[name = tensor("op_5411_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136725568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137954432))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137954624)))]; + tensor var_5411_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_5411_dilations_0, groups = var_5411_groups_0, pad = var_5411_pad_0, pad_type = var_5411_pad_type_0, strides = var_5411_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = tensor("op_5411_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = var_5411_cast_fp16, y = inputs_39_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; + tensor input_151_axes_0 = const()[name = tensor("input_151_axes_0"), val = tensor([1])]; + tensor input_151_gamma_0_to_fp16 = const()[name = tensor("input_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137957248)))]; + tensor input_151_beta_0_to_fp16 = const()[name = tensor("input_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137959872)))]; + tensor var_5421_to_fp16 = const()[name = tensor("op_5421_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_151_cast_fp16 = layer_norm(axes = input_151_axes_0, beta = input_151_beta_0_to_fp16, epsilon = var_5421_to_fp16, gamma = input_151_gamma_0_to_fp16, x = inputs_41_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor var_5441_pad_type_0 = const()[name = tensor("op_5441_pad_type_0"), val = tensor("valid")]; + tensor var_5441_strides_0 = const()[name = tensor("op_5441_strides_0"), val = tensor([1, 1])]; + tensor var_5441_pad_0 = const()[name = tensor("op_5441_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5441_dilations_0 = const()[name = tensor("op_5441_dilations_0"), val = tensor([1, 1])]; + tensor var_5441_groups_0 = const()[name = tensor("op_5441_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137962496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147792960))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147793152)))]; + tensor var_5441_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_5441_dilations_0, groups = var_5441_groups_0, pad = var_5441_pad_0, pad_type = var_5441_pad_type_0, strides = var_5441_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = tensor("op_5441_cast_fp16")]; + tensor var_5442_split_sizes_0 = const()[name = tensor("op_5442_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_5442_axis_0 = const()[name = tensor("op_5442_axis_0"), val = tensor(1)]; + tensor var_5442_cast_fp16_0, tensor var_5442_cast_fp16_1 = split(axis = var_5442_axis_0, split_sizes = var_5442_split_sizes_0, x = var_5441_cast_fp16)[name = tensor("op_5442_cast_fp16")]; + tensor var_5444_mode_0 = const()[name = tensor("op_5444_mode_0"), val = tensor("EXACT")]; + tensor var_5444_cast_fp16 = gelu(mode = var_5444_mode_0, x = var_5442_cast_fp16_1)[name = tensor("op_5444_cast_fp16")]; + tensor input_153_cast_fp16 = mul(x = var_5442_cast_fp16_0, y = var_5444_cast_fp16)[name = tensor("input_153_cast_fp16")]; + tensor var_5452_pad_type_0 = const()[name = tensor("op_5452_pad_type_0"), val = tensor("valid")]; + tensor var_5452_strides_0 = const()[name = tensor("op_5452_strides_0"), val = tensor([1, 1])]; + tensor var_5452_pad_0 = const()[name = tensor("op_5452_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5452_dilations_0 = const()[name = tensor("op_5452_dilations_0"), val = tensor([1, 1])]; + tensor var_5452_groups_0 = const()[name = tensor("op_5452_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147813696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152728960))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152729152)))]; + tensor var_5452_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_5452_dilations_0, groups = var_5452_groups_0, pad = var_5452_pad_0, pad_type = var_5452_pad_type_0, strides = var_5452_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor("op_5452_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = var_5452_cast_fp16, y = inputs_41_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; + tensor hidden_states_83_axes_0 = const()[name = tensor("hidden_states_83_axes_0"), val = tensor([1])]; + tensor hidden_states_83_gamma_0_to_fp16 = const()[name = tensor("hidden_states_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152731776)))]; + tensor hidden_states_83_beta_0_to_fp16 = const()[name = tensor("hidden_states_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152734400)))]; + tensor var_5468_to_fp16 = const()[name = tensor("op_5468_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_83_cast_fp16 = layer_norm(axes = hidden_states_83_axes_0, beta = hidden_states_83_beta_0_to_fp16, epsilon = var_5468_to_fp16, gamma = hidden_states_83_gamma_0_to_fp16, x = inputs_43_cast_fp16)[name = tensor("hidden_states_83_cast_fp16")]; + tensor q_29_pad_type_0 = const()[name = tensor("q_29_pad_type_0"), val = tensor("valid")]; + tensor q_29_strides_0 = const()[name = tensor("q_29_strides_0"), val = tensor([1, 1])]; + tensor q_29_pad_0 = const()[name = tensor("q_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_29_dilations_0 = const()[name = tensor("q_29_dilations_0"), val = tensor([1, 1])]; + tensor q_29_groups_0 = const()[name = tensor("q_29_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152737024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153965888))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_29_cast_fp16 = conv(dilations = q_29_dilations_0, groups = q_29_groups_0, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = q_29_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_83_cast_fp16)[name = tensor("q_29_cast_fp16")]; + tensor k_57_pad_type_0 = const()[name = tensor("k_57_pad_type_0"), val = tensor("valid")]; + tensor k_57_strides_0 = const()[name = tensor("k_57_strides_0"), val = tensor([1, 1])]; + tensor k_57_pad_0 = const()[name = tensor("k_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_57_dilations_0 = const()[name = tensor("k_57_dilations_0"), val = tensor([1, 1])]; + tensor k_57_groups_0 = const()[name = tensor("k_57_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153966080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155194944))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_57_cast_fp16 = conv(dilations = k_57_dilations_0, groups = k_57_groups_0, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = k_57_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_83_cast_fp16)[name = tensor("k_57_cast_fp16")]; + tensor v_29_pad_type_0 = const()[name = tensor("v_29_pad_type_0"), val = tensor("valid")]; + tensor v_29_strides_0 = const()[name = tensor("v_29_strides_0"), val = tensor([1, 1])]; + tensor v_29_pad_0 = const()[name = tensor("v_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_29_dilations_0 = const()[name = tensor("v_29_dilations_0"), val = tensor([1, 1])]; + tensor v_29_groups_0 = const()[name = tensor("v_29_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155195136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156424000))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_29_cast_fp16 = conv(dilations = v_29_dilations_0, groups = v_29_groups_0, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = v_29_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_83_cast_fp16)[name = tensor("v_29_cast_fp16")]; + tensor var_5501_begin_0 = const()[name = tensor("op_5501_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5501_end_0 = const()[name = tensor("op_5501_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5501_end_mask_0 = const()[name = tensor("op_5501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5501_cast_fp16 = slice_by_index(begin = var_5501_begin_0, end = var_5501_end_0, end_mask = var_5501_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5501_cast_fp16")]; + tensor var_5505_begin_0 = const()[name = tensor("op_5505_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5505_end_0 = const()[name = tensor("op_5505_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5505_end_mask_0 = const()[name = tensor("op_5505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5505_cast_fp16 = slice_by_index(begin = var_5505_begin_0, end = var_5505_end_0, end_mask = var_5505_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5505_cast_fp16")]; + tensor var_5509_begin_0 = const()[name = tensor("op_5509_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5509_end_0 = const()[name = tensor("op_5509_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5509_end_mask_0 = const()[name = tensor("op_5509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5509_cast_fp16 = slice_by_index(begin = var_5509_begin_0, end = var_5509_end_0, end_mask = var_5509_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5509_cast_fp16")]; + tensor var_5513_begin_0 = const()[name = tensor("op_5513_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5513_end_0 = const()[name = tensor("op_5513_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5513_end_mask_0 = const()[name = tensor("op_5513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5513_cast_fp16 = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5513_cast_fp16")]; + tensor var_5517_begin_0 = const()[name = tensor("op_5517_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5517_end_0 = const()[name = tensor("op_5517_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5517_end_mask_0 = const()[name = tensor("op_5517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5517_cast_fp16 = slice_by_index(begin = var_5517_begin_0, end = var_5517_end_0, end_mask = var_5517_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5517_cast_fp16")]; + tensor var_5521_begin_0 = const()[name = tensor("op_5521_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5521_end_0 = const()[name = tensor("op_5521_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5521_end_mask_0 = const()[name = tensor("op_5521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5521_cast_fp16 = slice_by_index(begin = var_5521_begin_0, end = var_5521_end_0, end_mask = var_5521_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5521_cast_fp16")]; + tensor var_5525_begin_0 = const()[name = tensor("op_5525_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5525_end_0 = const()[name = tensor("op_5525_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5525_end_mask_0 = const()[name = tensor("op_5525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5525_cast_fp16 = slice_by_index(begin = var_5525_begin_0, end = var_5525_end_0, end_mask = var_5525_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5525_cast_fp16")]; + tensor var_5529_begin_0 = const()[name = tensor("op_5529_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5529_end_0 = const()[name = tensor("op_5529_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5529_end_mask_0 = const()[name = tensor("op_5529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5529_cast_fp16 = slice_by_index(begin = var_5529_begin_0, end = var_5529_end_0, end_mask = var_5529_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5529_cast_fp16")]; + tensor var_5533_begin_0 = const()[name = tensor("op_5533_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5533_end_0 = const()[name = tensor("op_5533_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5533_end_mask_0 = const()[name = tensor("op_5533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5533_cast_fp16 = slice_by_index(begin = var_5533_begin_0, end = var_5533_end_0, end_mask = var_5533_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5533_cast_fp16")]; + tensor var_5537_begin_0 = const()[name = tensor("op_5537_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5537_end_0 = const()[name = tensor("op_5537_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5537_end_mask_0 = const()[name = tensor("op_5537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5537_cast_fp16 = slice_by_index(begin = var_5537_begin_0, end = var_5537_end_0, end_mask = var_5537_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5537_cast_fp16")]; + tensor var_5541_begin_0 = const()[name = tensor("op_5541_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5541_end_0 = const()[name = tensor("op_5541_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5541_end_mask_0 = const()[name = tensor("op_5541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5541_cast_fp16 = slice_by_index(begin = var_5541_begin_0, end = var_5541_end_0, end_mask = var_5541_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5541_cast_fp16")]; + tensor var_5545_begin_0 = const()[name = tensor("op_5545_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5545_end_0 = const()[name = tensor("op_5545_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5545_end_mask_0 = const()[name = tensor("op_5545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5545_cast_fp16 = slice_by_index(begin = var_5545_begin_0, end = var_5545_end_0, end_mask = var_5545_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5545_cast_fp16")]; + tensor var_5549_begin_0 = const()[name = tensor("op_5549_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5549_end_0 = const()[name = tensor("op_5549_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5549_end_mask_0 = const()[name = tensor("op_5549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5549_cast_fp16 = slice_by_index(begin = var_5549_begin_0, end = var_5549_end_0, end_mask = var_5549_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5549_cast_fp16")]; + tensor var_5553_begin_0 = const()[name = tensor("op_5553_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5553_end_0 = const()[name = tensor("op_5553_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5553_end_mask_0 = const()[name = tensor("op_5553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5553_cast_fp16 = slice_by_index(begin = var_5553_begin_0, end = var_5553_end_0, end_mask = var_5553_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5553_cast_fp16")]; + tensor var_5557_begin_0 = const()[name = tensor("op_5557_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5557_end_0 = const()[name = tensor("op_5557_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5557_end_mask_0 = const()[name = tensor("op_5557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5557_cast_fp16 = slice_by_index(begin = var_5557_begin_0, end = var_5557_end_0, end_mask = var_5557_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5557_cast_fp16")]; + tensor var_5561_begin_0 = const()[name = tensor("op_5561_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5561_end_0 = const()[name = tensor("op_5561_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5561_end_mask_0 = const()[name = tensor("op_5561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5561_cast_fp16 = slice_by_index(begin = var_5561_begin_0, end = var_5561_end_0, end_mask = var_5561_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5561_cast_fp16")]; + tensor var_5565_begin_0 = const()[name = tensor("op_5565_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5565_end_0 = const()[name = tensor("op_5565_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5565_end_mask_0 = const()[name = tensor("op_5565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5565_cast_fp16 = slice_by_index(begin = var_5565_begin_0, end = var_5565_end_0, end_mask = var_5565_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5565_cast_fp16")]; + tensor var_5569_begin_0 = const()[name = tensor("op_5569_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5569_end_0 = const()[name = tensor("op_5569_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_5569_end_mask_0 = const()[name = tensor("op_5569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5569_cast_fp16 = slice_by_index(begin = var_5569_begin_0, end = var_5569_end_0, end_mask = var_5569_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5569_cast_fp16")]; + tensor var_5573_begin_0 = const()[name = tensor("op_5573_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5573_end_0 = const()[name = tensor("op_5573_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_5573_end_mask_0 = const()[name = tensor("op_5573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5573_cast_fp16 = slice_by_index(begin = var_5573_begin_0, end = var_5573_end_0, end_mask = var_5573_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5573_cast_fp16")]; + tensor var_5577_begin_0 = const()[name = tensor("op_5577_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5577_end_0 = const()[name = tensor("op_5577_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_5577_end_mask_0 = const()[name = tensor("op_5577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5577_cast_fp16 = slice_by_index(begin = var_5577_begin_0, end = var_5577_end_0, end_mask = var_5577_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5577_cast_fp16")]; + tensor k_59_perm_0 = const()[name = tensor("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5584_begin_0 = const()[name = tensor("op_5584_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5584_end_0 = const()[name = tensor("op_5584_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_5584_end_mask_0 = const()[name = tensor("op_5584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_59_cast_fp16 = transpose(perm = k_59_perm_0, x = k_57_cast_fp16)[name = tensor("transpose_53")]; + tensor var_5584_cast_fp16 = slice_by_index(begin = var_5584_begin_0, end = var_5584_end_0, end_mask = var_5584_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5584_cast_fp16")]; + tensor var_5588_begin_0 = const()[name = tensor("op_5588_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5588_end_0 = const()[name = tensor("op_5588_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_5588_end_mask_0 = const()[name = tensor("op_5588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5588_cast_fp16 = slice_by_index(begin = var_5588_begin_0, end = var_5588_end_0, end_mask = var_5588_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5588_cast_fp16")]; + tensor var_5592_begin_0 = const()[name = tensor("op_5592_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5592_end_0 = const()[name = tensor("op_5592_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_5592_end_mask_0 = const()[name = tensor("op_5592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5592_cast_fp16 = slice_by_index(begin = var_5592_begin_0, end = var_5592_end_0, end_mask = var_5592_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5592_cast_fp16")]; + tensor var_5596_begin_0 = const()[name = tensor("op_5596_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5596_end_0 = const()[name = tensor("op_5596_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_5596_end_mask_0 = const()[name = tensor("op_5596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5596_cast_fp16 = slice_by_index(begin = var_5596_begin_0, end = var_5596_end_0, end_mask = var_5596_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5596_cast_fp16")]; + tensor var_5600_begin_0 = const()[name = tensor("op_5600_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5600_end_0 = const()[name = tensor("op_5600_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_5600_end_mask_0 = const()[name = tensor("op_5600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5600_cast_fp16 = slice_by_index(begin = var_5600_begin_0, end = var_5600_end_0, end_mask = var_5600_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5600_cast_fp16")]; + tensor var_5604_begin_0 = const()[name = tensor("op_5604_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5604_end_0 = const()[name = tensor("op_5604_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_5604_end_mask_0 = const()[name = tensor("op_5604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5604_cast_fp16 = slice_by_index(begin = var_5604_begin_0, end = var_5604_end_0, end_mask = var_5604_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5604_cast_fp16")]; + tensor var_5608_begin_0 = const()[name = tensor("op_5608_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5608_end_0 = const()[name = tensor("op_5608_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_5608_end_mask_0 = const()[name = tensor("op_5608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5608_cast_fp16 = slice_by_index(begin = var_5608_begin_0, end = var_5608_end_0, end_mask = var_5608_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5608_cast_fp16")]; + tensor var_5612_begin_0 = const()[name = tensor("op_5612_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5612_end_0 = const()[name = tensor("op_5612_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_5612_end_mask_0 = const()[name = tensor("op_5612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5612_cast_fp16 = slice_by_index(begin = var_5612_begin_0, end = var_5612_end_0, end_mask = var_5612_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5612_cast_fp16")]; + tensor var_5616_begin_0 = const()[name = tensor("op_5616_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5616_end_0 = const()[name = tensor("op_5616_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_5616_end_mask_0 = const()[name = tensor("op_5616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5616_cast_fp16 = slice_by_index(begin = var_5616_begin_0, end = var_5616_end_0, end_mask = var_5616_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5616_cast_fp16")]; + tensor var_5620_begin_0 = const()[name = tensor("op_5620_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5620_end_0 = const()[name = tensor("op_5620_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_5620_end_mask_0 = const()[name = tensor("op_5620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5620_cast_fp16 = slice_by_index(begin = var_5620_begin_0, end = var_5620_end_0, end_mask = var_5620_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5620_cast_fp16")]; + tensor var_5624_begin_0 = const()[name = tensor("op_5624_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5624_end_0 = const()[name = tensor("op_5624_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_5624_end_mask_0 = const()[name = tensor("op_5624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5624_cast_fp16 = slice_by_index(begin = var_5624_begin_0, end = var_5624_end_0, end_mask = var_5624_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5624_cast_fp16")]; + tensor var_5628_begin_0 = const()[name = tensor("op_5628_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5628_end_0 = const()[name = tensor("op_5628_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_5628_end_mask_0 = const()[name = tensor("op_5628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5628_cast_fp16 = slice_by_index(begin = var_5628_begin_0, end = var_5628_end_0, end_mask = var_5628_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5628_cast_fp16")]; + tensor var_5632_begin_0 = const()[name = tensor("op_5632_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5632_end_0 = const()[name = tensor("op_5632_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_5632_end_mask_0 = const()[name = tensor("op_5632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5632_cast_fp16 = slice_by_index(begin = var_5632_begin_0, end = var_5632_end_0, end_mask = var_5632_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5632_cast_fp16")]; + tensor var_5636_begin_0 = const()[name = tensor("op_5636_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5636_end_0 = const()[name = tensor("op_5636_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_5636_end_mask_0 = const()[name = tensor("op_5636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5636_cast_fp16 = slice_by_index(begin = var_5636_begin_0, end = var_5636_end_0, end_mask = var_5636_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5636_cast_fp16")]; + tensor var_5640_begin_0 = const()[name = tensor("op_5640_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5640_end_0 = const()[name = tensor("op_5640_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_5640_end_mask_0 = const()[name = tensor("op_5640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5640_cast_fp16 = slice_by_index(begin = var_5640_begin_0, end = var_5640_end_0, end_mask = var_5640_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5640_cast_fp16")]; + tensor var_5644_begin_0 = const()[name = tensor("op_5644_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5644_end_0 = const()[name = tensor("op_5644_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5644_end_mask_0 = const()[name = tensor("op_5644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5644_cast_fp16 = slice_by_index(begin = var_5644_begin_0, end = var_5644_end_0, end_mask = var_5644_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5644_cast_fp16")]; + tensor var_5648_begin_0 = const()[name = tensor("op_5648_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5648_end_0 = const()[name = tensor("op_5648_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_5648_end_mask_0 = const()[name = tensor("op_5648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5648_cast_fp16 = slice_by_index(begin = var_5648_begin_0, end = var_5648_end_0, end_mask = var_5648_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5648_cast_fp16")]; + tensor var_5652_begin_0 = const()[name = tensor("op_5652_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5652_end_0 = const()[name = tensor("op_5652_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_5652_end_mask_0 = const()[name = tensor("op_5652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5652_cast_fp16 = slice_by_index(begin = var_5652_begin_0, end = var_5652_end_0, end_mask = var_5652_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5652_cast_fp16")]; + tensor var_5656_begin_0 = const()[name = tensor("op_5656_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5656_end_0 = const()[name = tensor("op_5656_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_5656_end_mask_0 = const()[name = tensor("op_5656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5656_cast_fp16 = slice_by_index(begin = var_5656_begin_0, end = var_5656_end_0, end_mask = var_5656_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5656_cast_fp16")]; + tensor var_5660_begin_0 = const()[name = tensor("op_5660_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5660_end_0 = const()[name = tensor("op_5660_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_5660_end_mask_0 = const()[name = tensor("op_5660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5660_cast_fp16 = slice_by_index(begin = var_5660_begin_0, end = var_5660_end_0, end_mask = var_5660_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5660_cast_fp16")]; + tensor var_5662_begin_0 = const()[name = tensor("op_5662_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5662_end_0 = const()[name = tensor("op_5662_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5662_end_mask_0 = const()[name = tensor("op_5662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5662_cast_fp16 = slice_by_index(begin = var_5662_begin_0, end = var_5662_end_0, end_mask = var_5662_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5662_cast_fp16")]; + tensor var_5666_begin_0 = const()[name = tensor("op_5666_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5666_end_0 = const()[name = tensor("op_5666_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5666_end_mask_0 = const()[name = tensor("op_5666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5666_cast_fp16 = slice_by_index(begin = var_5666_begin_0, end = var_5666_end_0, end_mask = var_5666_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5666_cast_fp16")]; + tensor var_5670_begin_0 = const()[name = tensor("op_5670_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5670_end_0 = const()[name = tensor("op_5670_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5670_end_mask_0 = const()[name = tensor("op_5670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5670_cast_fp16 = slice_by_index(begin = var_5670_begin_0, end = var_5670_end_0, end_mask = var_5670_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5670_cast_fp16")]; + tensor var_5674_begin_0 = const()[name = tensor("op_5674_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5674_end_0 = const()[name = tensor("op_5674_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5674_end_mask_0 = const()[name = tensor("op_5674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5674_cast_fp16 = slice_by_index(begin = var_5674_begin_0, end = var_5674_end_0, end_mask = var_5674_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5674_cast_fp16")]; + tensor var_5678_begin_0 = const()[name = tensor("op_5678_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5678_end_0 = const()[name = tensor("op_5678_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5678_end_mask_0 = const()[name = tensor("op_5678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5678_cast_fp16 = slice_by_index(begin = var_5678_begin_0, end = var_5678_end_0, end_mask = var_5678_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5678_cast_fp16")]; + tensor var_5682_begin_0 = const()[name = tensor("op_5682_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5682_end_0 = const()[name = tensor("op_5682_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5682_end_mask_0 = const()[name = tensor("op_5682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5682_cast_fp16 = slice_by_index(begin = var_5682_begin_0, end = var_5682_end_0, end_mask = var_5682_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5682_cast_fp16")]; + tensor var_5686_begin_0 = const()[name = tensor("op_5686_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5686_end_0 = const()[name = tensor("op_5686_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5686_end_mask_0 = const()[name = tensor("op_5686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5686_cast_fp16 = slice_by_index(begin = var_5686_begin_0, end = var_5686_end_0, end_mask = var_5686_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5686_cast_fp16")]; + tensor var_5690_begin_0 = const()[name = tensor("op_5690_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5690_end_0 = const()[name = tensor("op_5690_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5690_end_mask_0 = const()[name = tensor("op_5690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5690_cast_fp16 = slice_by_index(begin = var_5690_begin_0, end = var_5690_end_0, end_mask = var_5690_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5690_cast_fp16")]; + tensor var_5694_begin_0 = const()[name = tensor("op_5694_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5694_end_0 = const()[name = tensor("op_5694_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5694_end_mask_0 = const()[name = tensor("op_5694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5694_cast_fp16 = slice_by_index(begin = var_5694_begin_0, end = var_5694_end_0, end_mask = var_5694_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5694_cast_fp16")]; + tensor var_5698_begin_0 = const()[name = tensor("op_5698_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5698_end_0 = const()[name = tensor("op_5698_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5698_end_mask_0 = const()[name = tensor("op_5698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5698_cast_fp16 = slice_by_index(begin = var_5698_begin_0, end = var_5698_end_0, end_mask = var_5698_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5698_cast_fp16")]; + tensor var_5702_begin_0 = const()[name = tensor("op_5702_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5702_end_0 = const()[name = tensor("op_5702_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5702_end_mask_0 = const()[name = tensor("op_5702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5702_cast_fp16 = slice_by_index(begin = var_5702_begin_0, end = var_5702_end_0, end_mask = var_5702_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5702_cast_fp16")]; + tensor var_5706_begin_0 = const()[name = tensor("op_5706_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5706_end_0 = const()[name = tensor("op_5706_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5706_end_mask_0 = const()[name = tensor("op_5706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5706_cast_fp16 = slice_by_index(begin = var_5706_begin_0, end = var_5706_end_0, end_mask = var_5706_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5706_cast_fp16")]; + tensor var_5710_begin_0 = const()[name = tensor("op_5710_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5710_end_0 = const()[name = tensor("op_5710_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5710_end_mask_0 = const()[name = tensor("op_5710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5710_cast_fp16 = slice_by_index(begin = var_5710_begin_0, end = var_5710_end_0, end_mask = var_5710_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5710_cast_fp16")]; + tensor var_5714_begin_0 = const()[name = tensor("op_5714_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5714_end_0 = const()[name = tensor("op_5714_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5714_end_mask_0 = const()[name = tensor("op_5714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5714_cast_fp16 = slice_by_index(begin = var_5714_begin_0, end = var_5714_end_0, end_mask = var_5714_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5714_cast_fp16")]; + tensor var_5718_begin_0 = const()[name = tensor("op_5718_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5718_end_0 = const()[name = tensor("op_5718_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5718_end_mask_0 = const()[name = tensor("op_5718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5718_cast_fp16 = slice_by_index(begin = var_5718_begin_0, end = var_5718_end_0, end_mask = var_5718_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5718_cast_fp16")]; + tensor var_5722_begin_0 = const()[name = tensor("op_5722_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5722_end_0 = const()[name = tensor("op_5722_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5722_end_mask_0 = const()[name = tensor("op_5722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5722_cast_fp16 = slice_by_index(begin = var_5722_begin_0, end = var_5722_end_0, end_mask = var_5722_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5722_cast_fp16")]; + tensor var_5726_begin_0 = const()[name = tensor("op_5726_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5726_end_0 = const()[name = tensor("op_5726_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5726_end_mask_0 = const()[name = tensor("op_5726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5726_cast_fp16 = slice_by_index(begin = var_5726_begin_0, end = var_5726_end_0, end_mask = var_5726_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5726_cast_fp16")]; + tensor var_5730_begin_0 = const()[name = tensor("op_5730_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5730_end_0 = const()[name = tensor("op_5730_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_5730_end_mask_0 = const()[name = tensor("op_5730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5730_cast_fp16 = slice_by_index(begin = var_5730_begin_0, end = var_5730_end_0, end_mask = var_5730_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5730_cast_fp16")]; + tensor var_5734_begin_0 = const()[name = tensor("op_5734_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5734_end_0 = const()[name = tensor("op_5734_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_5734_end_mask_0 = const()[name = tensor("op_5734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5734_cast_fp16 = slice_by_index(begin = var_5734_begin_0, end = var_5734_end_0, end_mask = var_5734_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5734_cast_fp16")]; + tensor var_5738_begin_0 = const()[name = tensor("op_5738_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5738_end_0 = const()[name = tensor("op_5738_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_5738_end_mask_0 = const()[name = tensor("op_5738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5738_cast_fp16 = slice_by_index(begin = var_5738_begin_0, end = var_5738_end_0, end_mask = var_5738_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5738_cast_fp16")]; + tensor var_5742_equation_0 = const()[name = tensor("op_5742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5742_cast_fp16 = einsum(equation = var_5742_equation_0, values = (var_5584_cast_fp16, var_5501_cast_fp16))[name = tensor("op_5742_cast_fp16")]; + tensor var_5743_to_fp16 = const()[name = tensor("op_5743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_401_cast_fp16 = mul(x = var_5742_cast_fp16, y = var_5743_to_fp16)[name = tensor("aw_401_cast_fp16")]; + tensor var_5746_equation_0 = const()[name = tensor("op_5746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5746_cast_fp16 = einsum(equation = var_5746_equation_0, values = (var_5588_cast_fp16, var_5505_cast_fp16))[name = tensor("op_5746_cast_fp16")]; + tensor var_5747_to_fp16 = const()[name = tensor("op_5747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_403_cast_fp16 = mul(x = var_5746_cast_fp16, y = var_5747_to_fp16)[name = tensor("aw_403_cast_fp16")]; + tensor var_5750_equation_0 = const()[name = tensor("op_5750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5750_cast_fp16 = einsum(equation = var_5750_equation_0, values = (var_5592_cast_fp16, var_5509_cast_fp16))[name = tensor("op_5750_cast_fp16")]; + tensor var_5751_to_fp16 = const()[name = tensor("op_5751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_405_cast_fp16 = mul(x = var_5750_cast_fp16, y = var_5751_to_fp16)[name = tensor("aw_405_cast_fp16")]; + tensor var_5754_equation_0 = const()[name = tensor("op_5754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5754_cast_fp16 = einsum(equation = var_5754_equation_0, values = (var_5596_cast_fp16, var_5513_cast_fp16))[name = tensor("op_5754_cast_fp16")]; + tensor var_5755_to_fp16 = const()[name = tensor("op_5755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_407_cast_fp16 = mul(x = var_5754_cast_fp16, y = var_5755_to_fp16)[name = tensor("aw_407_cast_fp16")]; + tensor var_5758_equation_0 = const()[name = tensor("op_5758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5758_cast_fp16 = einsum(equation = var_5758_equation_0, values = (var_5600_cast_fp16, var_5517_cast_fp16))[name = tensor("op_5758_cast_fp16")]; + tensor var_5759_to_fp16 = const()[name = tensor("op_5759_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_409_cast_fp16 = mul(x = var_5758_cast_fp16, y = var_5759_to_fp16)[name = tensor("aw_409_cast_fp16")]; + tensor var_5762_equation_0 = const()[name = tensor("op_5762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5762_cast_fp16 = einsum(equation = var_5762_equation_0, values = (var_5604_cast_fp16, var_5521_cast_fp16))[name = tensor("op_5762_cast_fp16")]; + tensor var_5763_to_fp16 = const()[name = tensor("op_5763_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_411_cast_fp16 = mul(x = var_5762_cast_fp16, y = var_5763_to_fp16)[name = tensor("aw_411_cast_fp16")]; + tensor var_5766_equation_0 = const()[name = tensor("op_5766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5766_cast_fp16 = einsum(equation = var_5766_equation_0, values = (var_5608_cast_fp16, var_5525_cast_fp16))[name = tensor("op_5766_cast_fp16")]; + tensor var_5767_to_fp16 = const()[name = tensor("op_5767_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_413_cast_fp16 = mul(x = var_5766_cast_fp16, y = var_5767_to_fp16)[name = tensor("aw_413_cast_fp16")]; + tensor var_5770_equation_0 = const()[name = tensor("op_5770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5770_cast_fp16 = einsum(equation = var_5770_equation_0, values = (var_5612_cast_fp16, var_5529_cast_fp16))[name = tensor("op_5770_cast_fp16")]; + tensor var_5771_to_fp16 = const()[name = tensor("op_5771_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_415_cast_fp16 = mul(x = var_5770_cast_fp16, y = var_5771_to_fp16)[name = tensor("aw_415_cast_fp16")]; + tensor var_5774_equation_0 = const()[name = tensor("op_5774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5774_cast_fp16 = einsum(equation = var_5774_equation_0, values = (var_5616_cast_fp16, var_5533_cast_fp16))[name = tensor("op_5774_cast_fp16")]; + tensor var_5775_to_fp16 = const()[name = tensor("op_5775_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_417_cast_fp16 = mul(x = var_5774_cast_fp16, y = var_5775_to_fp16)[name = tensor("aw_417_cast_fp16")]; + tensor var_5778_equation_0 = const()[name = tensor("op_5778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5778_cast_fp16 = einsum(equation = var_5778_equation_0, values = (var_5620_cast_fp16, var_5537_cast_fp16))[name = tensor("op_5778_cast_fp16")]; + tensor var_5779_to_fp16 = const()[name = tensor("op_5779_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_419_cast_fp16 = mul(x = var_5778_cast_fp16, y = var_5779_to_fp16)[name = tensor("aw_419_cast_fp16")]; + tensor var_5782_equation_0 = const()[name = tensor("op_5782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5782_cast_fp16 = einsum(equation = var_5782_equation_0, values = (var_5624_cast_fp16, var_5541_cast_fp16))[name = tensor("op_5782_cast_fp16")]; + tensor var_5783_to_fp16 = const()[name = tensor("op_5783_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_421_cast_fp16 = mul(x = var_5782_cast_fp16, y = var_5783_to_fp16)[name = tensor("aw_421_cast_fp16")]; + tensor var_5786_equation_0 = const()[name = tensor("op_5786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5786_cast_fp16 = einsum(equation = var_5786_equation_0, values = (var_5628_cast_fp16, var_5545_cast_fp16))[name = tensor("op_5786_cast_fp16")]; + tensor var_5787_to_fp16 = const()[name = tensor("op_5787_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_423_cast_fp16 = mul(x = var_5786_cast_fp16, y = var_5787_to_fp16)[name = tensor("aw_423_cast_fp16")]; + tensor var_5790_equation_0 = const()[name = tensor("op_5790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5790_cast_fp16 = einsum(equation = var_5790_equation_0, values = (var_5632_cast_fp16, var_5549_cast_fp16))[name = tensor("op_5790_cast_fp16")]; + tensor var_5791_to_fp16 = const()[name = tensor("op_5791_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_425_cast_fp16 = mul(x = var_5790_cast_fp16, y = var_5791_to_fp16)[name = tensor("aw_425_cast_fp16")]; + tensor var_5794_equation_0 = const()[name = tensor("op_5794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5794_cast_fp16 = einsum(equation = var_5794_equation_0, values = (var_5636_cast_fp16, var_5553_cast_fp16))[name = tensor("op_5794_cast_fp16")]; + tensor var_5795_to_fp16 = const()[name = tensor("op_5795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_427_cast_fp16 = mul(x = var_5794_cast_fp16, y = var_5795_to_fp16)[name = tensor("aw_427_cast_fp16")]; + tensor var_5798_equation_0 = const()[name = tensor("op_5798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5798_cast_fp16 = einsum(equation = var_5798_equation_0, values = (var_5640_cast_fp16, var_5557_cast_fp16))[name = tensor("op_5798_cast_fp16")]; + tensor var_5799_to_fp16 = const()[name = tensor("op_5799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_429_cast_fp16 = mul(x = var_5798_cast_fp16, y = var_5799_to_fp16)[name = tensor("aw_429_cast_fp16")]; + tensor var_5802_equation_0 = const()[name = tensor("op_5802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5802_cast_fp16 = einsum(equation = var_5802_equation_0, values = (var_5644_cast_fp16, var_5561_cast_fp16))[name = tensor("op_5802_cast_fp16")]; + tensor var_5803_to_fp16 = const()[name = tensor("op_5803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_431_cast_fp16 = mul(x = var_5802_cast_fp16, y = var_5803_to_fp16)[name = tensor("aw_431_cast_fp16")]; + tensor var_5806_equation_0 = const()[name = tensor("op_5806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5806_cast_fp16 = einsum(equation = var_5806_equation_0, values = (var_5648_cast_fp16, var_5565_cast_fp16))[name = tensor("op_5806_cast_fp16")]; + tensor var_5807_to_fp16 = const()[name = tensor("op_5807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_433_cast_fp16 = mul(x = var_5806_cast_fp16, y = var_5807_to_fp16)[name = tensor("aw_433_cast_fp16")]; + tensor var_5810_equation_0 = const()[name = tensor("op_5810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5810_cast_fp16 = einsum(equation = var_5810_equation_0, values = (var_5652_cast_fp16, var_5569_cast_fp16))[name = tensor("op_5810_cast_fp16")]; + tensor var_5811_to_fp16 = const()[name = tensor("op_5811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_435_cast_fp16 = mul(x = var_5810_cast_fp16, y = var_5811_to_fp16)[name = tensor("aw_435_cast_fp16")]; + tensor var_5814_equation_0 = const()[name = tensor("op_5814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5814_cast_fp16 = einsum(equation = var_5814_equation_0, values = (var_5656_cast_fp16, var_5573_cast_fp16))[name = tensor("op_5814_cast_fp16")]; + tensor var_5815_to_fp16 = const()[name = tensor("op_5815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_437_cast_fp16 = mul(x = var_5814_cast_fp16, y = var_5815_to_fp16)[name = tensor("aw_437_cast_fp16")]; + tensor var_5818_equation_0 = const()[name = tensor("op_5818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5818_cast_fp16 = einsum(equation = var_5818_equation_0, values = (var_5660_cast_fp16, var_5577_cast_fp16))[name = tensor("op_5818_cast_fp16")]; + tensor var_5819_to_fp16 = const()[name = tensor("op_5819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_439_cast_fp16 = mul(x = var_5818_cast_fp16, y = var_5819_to_fp16)[name = tensor("aw_439_cast_fp16")]; + tensor var_5821_cast_fp16 = softmax(axis = var_2624, x = aw_401_cast_fp16)[name = tensor("op_5821_cast_fp16")]; + tensor var_5822_cast_fp16 = softmax(axis = var_2624, x = aw_403_cast_fp16)[name = tensor("op_5822_cast_fp16")]; + tensor var_5823_cast_fp16 = softmax(axis = var_2624, x = aw_405_cast_fp16)[name = tensor("op_5823_cast_fp16")]; + tensor var_5824_cast_fp16 = softmax(axis = var_2624, x = aw_407_cast_fp16)[name = tensor("op_5824_cast_fp16")]; + tensor var_5825_cast_fp16 = softmax(axis = var_2624, x = aw_409_cast_fp16)[name = tensor("op_5825_cast_fp16")]; + tensor var_5826_cast_fp16 = softmax(axis = var_2624, x = aw_411_cast_fp16)[name = tensor("op_5826_cast_fp16")]; + tensor var_5827_cast_fp16 = softmax(axis = var_2624, x = aw_413_cast_fp16)[name = tensor("op_5827_cast_fp16")]; + tensor var_5828_cast_fp16 = softmax(axis = var_2624, x = aw_415_cast_fp16)[name = tensor("op_5828_cast_fp16")]; + tensor var_5829_cast_fp16 = softmax(axis = var_2624, x = aw_417_cast_fp16)[name = tensor("op_5829_cast_fp16")]; + tensor var_5830_cast_fp16 = softmax(axis = var_2624, x = aw_419_cast_fp16)[name = tensor("op_5830_cast_fp16")]; + tensor var_5831_cast_fp16 = softmax(axis = var_2624, x = aw_421_cast_fp16)[name = tensor("op_5831_cast_fp16")]; + tensor var_5832_cast_fp16 = softmax(axis = var_2624, x = aw_423_cast_fp16)[name = tensor("op_5832_cast_fp16")]; + tensor var_5833_cast_fp16 = softmax(axis = var_2624, x = aw_425_cast_fp16)[name = tensor("op_5833_cast_fp16")]; + tensor var_5834_cast_fp16 = softmax(axis = var_2624, x = aw_427_cast_fp16)[name = tensor("op_5834_cast_fp16")]; + tensor var_5835_cast_fp16 = softmax(axis = var_2624, x = aw_429_cast_fp16)[name = tensor("op_5835_cast_fp16")]; + tensor var_5836_cast_fp16 = softmax(axis = var_2624, x = aw_431_cast_fp16)[name = tensor("op_5836_cast_fp16")]; + tensor var_5837_cast_fp16 = softmax(axis = var_2624, x = aw_433_cast_fp16)[name = tensor("op_5837_cast_fp16")]; + tensor var_5838_cast_fp16 = softmax(axis = var_2624, x = aw_435_cast_fp16)[name = tensor("op_5838_cast_fp16")]; + tensor var_5839_cast_fp16 = softmax(axis = var_2624, x = aw_437_cast_fp16)[name = tensor("op_5839_cast_fp16")]; + tensor var_5840_cast_fp16 = softmax(axis = var_2624, x = aw_439_cast_fp16)[name = tensor("op_5840_cast_fp16")]; + tensor var_5842_equation_0 = const()[name = tensor("op_5842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5842_cast_fp16 = einsum(equation = var_5842_equation_0, values = (var_5662_cast_fp16, var_5821_cast_fp16))[name = tensor("op_5842_cast_fp16")]; + tensor var_5844_equation_0 = const()[name = tensor("op_5844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5844_cast_fp16 = einsum(equation = var_5844_equation_0, values = (var_5666_cast_fp16, var_5822_cast_fp16))[name = tensor("op_5844_cast_fp16")]; + tensor var_5846_equation_0 = const()[name = tensor("op_5846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5846_cast_fp16 = einsum(equation = var_5846_equation_0, values = (var_5670_cast_fp16, var_5823_cast_fp16))[name = tensor("op_5846_cast_fp16")]; + tensor var_5848_equation_0 = const()[name = tensor("op_5848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5848_cast_fp16 = einsum(equation = var_5848_equation_0, values = (var_5674_cast_fp16, var_5824_cast_fp16))[name = tensor("op_5848_cast_fp16")]; + tensor var_5850_equation_0 = const()[name = tensor("op_5850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5850_cast_fp16 = einsum(equation = var_5850_equation_0, values = (var_5678_cast_fp16, var_5825_cast_fp16))[name = tensor("op_5850_cast_fp16")]; + tensor var_5852_equation_0 = const()[name = tensor("op_5852_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5852_cast_fp16 = einsum(equation = var_5852_equation_0, values = (var_5682_cast_fp16, var_5826_cast_fp16))[name = tensor("op_5852_cast_fp16")]; + tensor var_5854_equation_0 = const()[name = tensor("op_5854_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5854_cast_fp16 = einsum(equation = var_5854_equation_0, values = (var_5686_cast_fp16, var_5827_cast_fp16))[name = tensor("op_5854_cast_fp16")]; + tensor var_5856_equation_0 = const()[name = tensor("op_5856_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5856_cast_fp16 = einsum(equation = var_5856_equation_0, values = (var_5690_cast_fp16, var_5828_cast_fp16))[name = tensor("op_5856_cast_fp16")]; + tensor var_5858_equation_0 = const()[name = tensor("op_5858_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5858_cast_fp16 = einsum(equation = var_5858_equation_0, values = (var_5694_cast_fp16, var_5829_cast_fp16))[name = tensor("op_5858_cast_fp16")]; + tensor var_5860_equation_0 = const()[name = tensor("op_5860_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5860_cast_fp16 = einsum(equation = var_5860_equation_0, values = (var_5698_cast_fp16, var_5830_cast_fp16))[name = tensor("op_5860_cast_fp16")]; + tensor var_5862_equation_0 = const()[name = tensor("op_5862_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5862_cast_fp16 = einsum(equation = var_5862_equation_0, values = (var_5702_cast_fp16, var_5831_cast_fp16))[name = tensor("op_5862_cast_fp16")]; + tensor var_5864_equation_0 = const()[name = tensor("op_5864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5864_cast_fp16 = einsum(equation = var_5864_equation_0, values = (var_5706_cast_fp16, var_5832_cast_fp16))[name = tensor("op_5864_cast_fp16")]; + tensor var_5866_equation_0 = const()[name = tensor("op_5866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5866_cast_fp16 = einsum(equation = var_5866_equation_0, values = (var_5710_cast_fp16, var_5833_cast_fp16))[name = tensor("op_5866_cast_fp16")]; + tensor var_5868_equation_0 = const()[name = tensor("op_5868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5868_cast_fp16 = einsum(equation = var_5868_equation_0, values = (var_5714_cast_fp16, var_5834_cast_fp16))[name = tensor("op_5868_cast_fp16")]; + tensor var_5870_equation_0 = const()[name = tensor("op_5870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5870_cast_fp16 = einsum(equation = var_5870_equation_0, values = (var_5718_cast_fp16, var_5835_cast_fp16))[name = tensor("op_5870_cast_fp16")]; + tensor var_5872_equation_0 = const()[name = tensor("op_5872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5872_cast_fp16 = einsum(equation = var_5872_equation_0, values = (var_5722_cast_fp16, var_5836_cast_fp16))[name = tensor("op_5872_cast_fp16")]; + tensor var_5874_equation_0 = const()[name = tensor("op_5874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5874_cast_fp16 = einsum(equation = var_5874_equation_0, values = (var_5726_cast_fp16, var_5837_cast_fp16))[name = tensor("op_5874_cast_fp16")]; + tensor var_5876_equation_0 = const()[name = tensor("op_5876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5876_cast_fp16 = einsum(equation = var_5876_equation_0, values = (var_5730_cast_fp16, var_5838_cast_fp16))[name = tensor("op_5876_cast_fp16")]; + tensor var_5878_equation_0 = const()[name = tensor("op_5878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5878_cast_fp16 = einsum(equation = var_5878_equation_0, values = (var_5734_cast_fp16, var_5839_cast_fp16))[name = tensor("op_5878_cast_fp16")]; + tensor var_5880_equation_0 = const()[name = tensor("op_5880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5880_cast_fp16 = einsum(equation = var_5880_equation_0, values = (var_5738_cast_fp16, var_5840_cast_fp16))[name = tensor("op_5880_cast_fp16")]; + tensor input_155_interleave_0 = const()[name = tensor("input_155_interleave_0"), val = tensor(false)]; + tensor input_155_cast_fp16 = concat(axis = var_2624, interleave = input_155_interleave_0, values = (var_5842_cast_fp16, var_5844_cast_fp16, var_5846_cast_fp16, var_5848_cast_fp16, var_5850_cast_fp16, var_5852_cast_fp16, var_5854_cast_fp16, var_5856_cast_fp16, var_5858_cast_fp16, var_5860_cast_fp16, var_5862_cast_fp16, var_5864_cast_fp16, var_5866_cast_fp16, var_5868_cast_fp16, var_5870_cast_fp16, var_5872_cast_fp16, var_5874_cast_fp16, var_5876_cast_fp16, var_5878_cast_fp16, var_5880_cast_fp16))[name = tensor("input_155_cast_fp16")]; + tensor var_5890_pad_type_0 = const()[name = tensor("op_5890_pad_type_0"), val = tensor("valid")]; + tensor var_5890_strides_0 = const()[name = tensor("op_5890_strides_0"), val = tensor([1, 1])]; + tensor var_5890_pad_0 = const()[name = tensor("op_5890_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5890_dilations_0 = const()[name = tensor("op_5890_dilations_0"), val = tensor([1, 1])]; + tensor var_5890_groups_0 = const()[name = tensor("op_5890_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156424192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157653056))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157653248)))]; + tensor var_5890_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_5890_dilations_0, groups = var_5890_groups_0, pad = var_5890_pad_0, pad_type = var_5890_pad_type_0, strides = var_5890_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor("op_5890_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = var_5890_cast_fp16, y = inputs_43_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; + tensor hidden_states_85_axes_0 = const()[name = tensor("hidden_states_85_axes_0"), val = tensor([1])]; + tensor hidden_states_85_gamma_0_to_fp16 = const()[name = tensor("hidden_states_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157655872)))]; + tensor hidden_states_85_beta_0_to_fp16 = const()[name = tensor("hidden_states_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157658496)))]; + tensor var_5900_to_fp16 = const()[name = tensor("op_5900_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = hidden_states_85_beta_0_to_fp16, epsilon = var_5900_to_fp16, gamma = hidden_states_85_gamma_0_to_fp16, x = inputs_45_cast_fp16)[name = tensor("hidden_states_85_cast_fp16")]; + tensor q_31_pad_type_0 = const()[name = tensor("q_31_pad_type_0"), val = tensor("valid")]; + tensor q_31_strides_0 = const()[name = tensor("q_31_strides_0"), val = tensor([1, 1])]; + tensor q_31_pad_0 = const()[name = tensor("q_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_31_dilations_0 = const()[name = tensor("q_31_dilations_0"), val = tensor([1, 1])]; + tensor q_31_groups_0 = const()[name = tensor("q_31_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157661120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158889984))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_31_cast_fp16 = conv(dilations = q_31_dilations_0, groups = q_31_groups_0, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = q_31_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_85_cast_fp16)[name = tensor("q_31_cast_fp16")]; + tensor k_61_pad_type_0 = const()[name = tensor("k_61_pad_type_0"), val = tensor("valid")]; + tensor k_61_strides_0 = const()[name = tensor("k_61_strides_0"), val = tensor([1, 1])]; + tensor k_61_pad_0 = const()[name = tensor("k_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_61_dilations_0 = const()[name = tensor("k_61_dilations_0"), val = tensor([1, 1])]; + tensor k_61_groups_0 = const()[name = tensor("k_61_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158890176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160856320))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_61_cast_fp16 = conv(dilations = k_61_dilations_0, groups = k_61_groups_0, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = k_61_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_61_cast_fp16")]; + tensor v_31_pad_type_0 = const()[name = tensor("v_31_pad_type_0"), val = tensor("valid")]; + tensor v_31_strides_0 = const()[name = tensor("v_31_strides_0"), val = tensor([1, 1])]; + tensor v_31_pad_0 = const()[name = tensor("v_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_31_dilations_0 = const()[name = tensor("v_31_dilations_0"), val = tensor([1, 1])]; + tensor v_31_groups_0 = const()[name = tensor("v_31_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160856512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162822656))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_31_cast_fp16 = conv(dilations = v_31_dilations_0, groups = v_31_groups_0, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = v_31_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_31_cast_fp16")]; + tensor var_5933_begin_0 = const()[name = tensor("op_5933_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5933_end_0 = const()[name = tensor("op_5933_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5933_end_mask_0 = const()[name = tensor("op_5933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5933_cast_fp16 = slice_by_index(begin = var_5933_begin_0, end = var_5933_end_0, end_mask = var_5933_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5933_cast_fp16")]; + tensor var_5937_begin_0 = const()[name = tensor("op_5937_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5937_end_0 = const()[name = tensor("op_5937_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5937_end_mask_0 = const()[name = tensor("op_5937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5937_cast_fp16 = slice_by_index(begin = var_5937_begin_0, end = var_5937_end_0, end_mask = var_5937_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5937_cast_fp16")]; + tensor var_5941_begin_0 = const()[name = tensor("op_5941_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5941_end_0 = const()[name = tensor("op_5941_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5941_end_mask_0 = const()[name = tensor("op_5941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5941_cast_fp16 = slice_by_index(begin = var_5941_begin_0, end = var_5941_end_0, end_mask = var_5941_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5941_cast_fp16")]; + tensor var_5945_begin_0 = const()[name = tensor("op_5945_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5945_end_0 = const()[name = tensor("op_5945_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5945_end_mask_0 = const()[name = tensor("op_5945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5945_cast_fp16 = slice_by_index(begin = var_5945_begin_0, end = var_5945_end_0, end_mask = var_5945_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5945_cast_fp16")]; + tensor var_5949_begin_0 = const()[name = tensor("op_5949_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5949_end_0 = const()[name = tensor("op_5949_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5949_end_mask_0 = const()[name = tensor("op_5949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5949_cast_fp16 = slice_by_index(begin = var_5949_begin_0, end = var_5949_end_0, end_mask = var_5949_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5949_cast_fp16")]; + tensor var_5953_begin_0 = const()[name = tensor("op_5953_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5953_end_0 = const()[name = tensor("op_5953_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5953_end_mask_0 = const()[name = tensor("op_5953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5953_cast_fp16 = slice_by_index(begin = var_5953_begin_0, end = var_5953_end_0, end_mask = var_5953_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5953_cast_fp16")]; + tensor var_5957_begin_0 = const()[name = tensor("op_5957_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5957_end_0 = const()[name = tensor("op_5957_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5957_end_mask_0 = const()[name = tensor("op_5957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5957_cast_fp16 = slice_by_index(begin = var_5957_begin_0, end = var_5957_end_0, end_mask = var_5957_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5957_cast_fp16")]; + tensor var_5961_begin_0 = const()[name = tensor("op_5961_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5961_end_0 = const()[name = tensor("op_5961_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5961_end_mask_0 = const()[name = tensor("op_5961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5961_cast_fp16 = slice_by_index(begin = var_5961_begin_0, end = var_5961_end_0, end_mask = var_5961_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5961_cast_fp16")]; + tensor var_5965_begin_0 = const()[name = tensor("op_5965_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5965_end_0 = const()[name = tensor("op_5965_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5965_end_mask_0 = const()[name = tensor("op_5965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5965_cast_fp16 = slice_by_index(begin = var_5965_begin_0, end = var_5965_end_0, end_mask = var_5965_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5965_cast_fp16")]; + tensor var_5969_begin_0 = const()[name = tensor("op_5969_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5969_end_0 = const()[name = tensor("op_5969_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5969_end_mask_0 = const()[name = tensor("op_5969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5969_cast_fp16 = slice_by_index(begin = var_5969_begin_0, end = var_5969_end_0, end_mask = var_5969_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5969_cast_fp16")]; + tensor var_5973_begin_0 = const()[name = tensor("op_5973_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5973_end_0 = const()[name = tensor("op_5973_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5973_end_mask_0 = const()[name = tensor("op_5973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5973_cast_fp16 = slice_by_index(begin = var_5973_begin_0, end = var_5973_end_0, end_mask = var_5973_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5973_cast_fp16")]; + tensor var_5977_begin_0 = const()[name = tensor("op_5977_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5977_end_0 = const()[name = tensor("op_5977_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5977_end_mask_0 = const()[name = tensor("op_5977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5977_cast_fp16 = slice_by_index(begin = var_5977_begin_0, end = var_5977_end_0, end_mask = var_5977_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5977_cast_fp16")]; + tensor var_5981_begin_0 = const()[name = tensor("op_5981_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5981_end_0 = const()[name = tensor("op_5981_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5981_end_mask_0 = const()[name = tensor("op_5981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5981_cast_fp16 = slice_by_index(begin = var_5981_begin_0, end = var_5981_end_0, end_mask = var_5981_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5981_cast_fp16")]; + tensor var_5985_begin_0 = const()[name = tensor("op_5985_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5985_end_0 = const()[name = tensor("op_5985_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5985_end_mask_0 = const()[name = tensor("op_5985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5985_cast_fp16 = slice_by_index(begin = var_5985_begin_0, end = var_5985_end_0, end_mask = var_5985_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5985_cast_fp16")]; + tensor var_5989_begin_0 = const()[name = tensor("op_5989_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5989_end_0 = const()[name = tensor("op_5989_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5989_end_mask_0 = const()[name = tensor("op_5989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5989_cast_fp16 = slice_by_index(begin = var_5989_begin_0, end = var_5989_end_0, end_mask = var_5989_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5989_cast_fp16")]; + tensor var_5993_begin_0 = const()[name = tensor("op_5993_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5993_end_0 = const()[name = tensor("op_5993_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5993_end_mask_0 = const()[name = tensor("op_5993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5993_cast_fp16 = slice_by_index(begin = var_5993_begin_0, end = var_5993_end_0, end_mask = var_5993_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5993_cast_fp16")]; + tensor var_5997_begin_0 = const()[name = tensor("op_5997_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5997_end_0 = const()[name = tensor("op_5997_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5997_end_mask_0 = const()[name = tensor("op_5997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5997_cast_fp16 = slice_by_index(begin = var_5997_begin_0, end = var_5997_end_0, end_mask = var_5997_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5997_cast_fp16")]; + tensor var_6001_begin_0 = const()[name = tensor("op_6001_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6001_end_0 = const()[name = tensor("op_6001_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6001_end_mask_0 = const()[name = tensor("op_6001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6001_cast_fp16 = slice_by_index(begin = var_6001_begin_0, end = var_6001_end_0, end_mask = var_6001_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6001_cast_fp16")]; + tensor var_6005_begin_0 = const()[name = tensor("op_6005_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6005_end_0 = const()[name = tensor("op_6005_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6005_end_mask_0 = const()[name = tensor("op_6005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6005_cast_fp16 = slice_by_index(begin = var_6005_begin_0, end = var_6005_end_0, end_mask = var_6005_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6005_cast_fp16")]; + tensor var_6009_begin_0 = const()[name = tensor("op_6009_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6009_end_0 = const()[name = tensor("op_6009_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6009_end_mask_0 = const()[name = tensor("op_6009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6009_cast_fp16 = slice_by_index(begin = var_6009_begin_0, end = var_6009_end_0, end_mask = var_6009_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6009_cast_fp16")]; + tensor k_63_perm_0 = const()[name = tensor("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6016_begin_0 = const()[name = tensor("op_6016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6016_end_0 = const()[name = tensor("op_6016_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_6016_end_mask_0 = const()[name = tensor("op_6016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_63_cast_fp16 = transpose(perm = k_63_perm_0, x = k_61_cast_fp16)[name = tensor("transpose_52")]; + tensor var_6016_cast_fp16 = slice_by_index(begin = var_6016_begin_0, end = var_6016_end_0, end_mask = var_6016_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6016_cast_fp16")]; + tensor var_6020_begin_0 = const()[name = tensor("op_6020_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6020_end_0 = const()[name = tensor("op_6020_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_6020_end_mask_0 = const()[name = tensor("op_6020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6020_cast_fp16 = slice_by_index(begin = var_6020_begin_0, end = var_6020_end_0, end_mask = var_6020_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6020_cast_fp16")]; + tensor var_6024_begin_0 = const()[name = tensor("op_6024_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6024_end_0 = const()[name = tensor("op_6024_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_6024_end_mask_0 = const()[name = tensor("op_6024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6024_cast_fp16 = slice_by_index(begin = var_6024_begin_0, end = var_6024_end_0, end_mask = var_6024_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6024_cast_fp16")]; + tensor var_6028_begin_0 = const()[name = tensor("op_6028_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6028_end_0 = const()[name = tensor("op_6028_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_6028_end_mask_0 = const()[name = tensor("op_6028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6028_cast_fp16 = slice_by_index(begin = var_6028_begin_0, end = var_6028_end_0, end_mask = var_6028_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6028_cast_fp16")]; + tensor var_6032_begin_0 = const()[name = tensor("op_6032_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6032_end_0 = const()[name = tensor("op_6032_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_6032_end_mask_0 = const()[name = tensor("op_6032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6032_cast_fp16 = slice_by_index(begin = var_6032_begin_0, end = var_6032_end_0, end_mask = var_6032_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6032_cast_fp16")]; + tensor var_6036_begin_0 = const()[name = tensor("op_6036_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6036_end_0 = const()[name = tensor("op_6036_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_6036_end_mask_0 = const()[name = tensor("op_6036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6036_cast_fp16 = slice_by_index(begin = var_6036_begin_0, end = var_6036_end_0, end_mask = var_6036_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6036_cast_fp16")]; + tensor var_6040_begin_0 = const()[name = tensor("op_6040_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6040_end_0 = const()[name = tensor("op_6040_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_6040_end_mask_0 = const()[name = tensor("op_6040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6040_cast_fp16 = slice_by_index(begin = var_6040_begin_0, end = var_6040_end_0, end_mask = var_6040_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6040_cast_fp16")]; + tensor var_6044_begin_0 = const()[name = tensor("op_6044_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6044_end_0 = const()[name = tensor("op_6044_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_6044_end_mask_0 = const()[name = tensor("op_6044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6044_cast_fp16 = slice_by_index(begin = var_6044_begin_0, end = var_6044_end_0, end_mask = var_6044_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6044_cast_fp16")]; + tensor var_6048_begin_0 = const()[name = tensor("op_6048_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6048_end_0 = const()[name = tensor("op_6048_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_6048_end_mask_0 = const()[name = tensor("op_6048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6048_cast_fp16 = slice_by_index(begin = var_6048_begin_0, end = var_6048_end_0, end_mask = var_6048_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6048_cast_fp16")]; + tensor var_6052_begin_0 = const()[name = tensor("op_6052_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6052_end_0 = const()[name = tensor("op_6052_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_6052_end_mask_0 = const()[name = tensor("op_6052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6052_cast_fp16 = slice_by_index(begin = var_6052_begin_0, end = var_6052_end_0, end_mask = var_6052_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6052_cast_fp16")]; + tensor var_6056_begin_0 = const()[name = tensor("op_6056_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6056_end_0 = const()[name = tensor("op_6056_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_6056_end_mask_0 = const()[name = tensor("op_6056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6056_cast_fp16 = slice_by_index(begin = var_6056_begin_0, end = var_6056_end_0, end_mask = var_6056_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6056_cast_fp16")]; + tensor var_6060_begin_0 = const()[name = tensor("op_6060_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6060_end_0 = const()[name = tensor("op_6060_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_6060_end_mask_0 = const()[name = tensor("op_6060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6060_cast_fp16 = slice_by_index(begin = var_6060_begin_0, end = var_6060_end_0, end_mask = var_6060_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6060_cast_fp16")]; + tensor var_6064_begin_0 = const()[name = tensor("op_6064_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6064_end_0 = const()[name = tensor("op_6064_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_6064_end_mask_0 = const()[name = tensor("op_6064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6064_cast_fp16 = slice_by_index(begin = var_6064_begin_0, end = var_6064_end_0, end_mask = var_6064_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6064_cast_fp16")]; + tensor var_6068_begin_0 = const()[name = tensor("op_6068_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6068_end_0 = const()[name = tensor("op_6068_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_6068_end_mask_0 = const()[name = tensor("op_6068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6068_cast_fp16 = slice_by_index(begin = var_6068_begin_0, end = var_6068_end_0, end_mask = var_6068_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6068_cast_fp16")]; + tensor var_6072_begin_0 = const()[name = tensor("op_6072_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6072_end_0 = const()[name = tensor("op_6072_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_6072_end_mask_0 = const()[name = tensor("op_6072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6072_cast_fp16 = slice_by_index(begin = var_6072_begin_0, end = var_6072_end_0, end_mask = var_6072_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6072_cast_fp16")]; + tensor var_6076_begin_0 = const()[name = tensor("op_6076_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6076_end_0 = const()[name = tensor("op_6076_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_6076_end_mask_0 = const()[name = tensor("op_6076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6076_cast_fp16 = slice_by_index(begin = var_6076_begin_0, end = var_6076_end_0, end_mask = var_6076_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6076_cast_fp16")]; + tensor var_6080_begin_0 = const()[name = tensor("op_6080_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6080_end_0 = const()[name = tensor("op_6080_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_6080_end_mask_0 = const()[name = tensor("op_6080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6080_cast_fp16 = slice_by_index(begin = var_6080_begin_0, end = var_6080_end_0, end_mask = var_6080_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6080_cast_fp16")]; + tensor var_6084_begin_0 = const()[name = tensor("op_6084_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6084_end_0 = const()[name = tensor("op_6084_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_6084_end_mask_0 = const()[name = tensor("op_6084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6084_cast_fp16 = slice_by_index(begin = var_6084_begin_0, end = var_6084_end_0, end_mask = var_6084_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6084_cast_fp16")]; + tensor var_6088_begin_0 = const()[name = tensor("op_6088_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6088_end_0 = const()[name = tensor("op_6088_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_6088_end_mask_0 = const()[name = tensor("op_6088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6088_cast_fp16 = slice_by_index(begin = var_6088_begin_0, end = var_6088_end_0, end_mask = var_6088_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6088_cast_fp16")]; + tensor var_6092_begin_0 = const()[name = tensor("op_6092_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_6092_end_0 = const()[name = tensor("op_6092_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_6092_end_mask_0 = const()[name = tensor("op_6092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6092_cast_fp16 = slice_by_index(begin = var_6092_begin_0, end = var_6092_end_0, end_mask = var_6092_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6092_cast_fp16")]; + tensor var_6094_begin_0 = const()[name = tensor("op_6094_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6094_end_0 = const()[name = tensor("op_6094_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_6094_end_mask_0 = const()[name = tensor("op_6094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6094_cast_fp16 = slice_by_index(begin = var_6094_begin_0, end = var_6094_end_0, end_mask = var_6094_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6094_cast_fp16")]; + tensor var_6098_begin_0 = const()[name = tensor("op_6098_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6098_end_0 = const()[name = tensor("op_6098_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_6098_end_mask_0 = const()[name = tensor("op_6098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6098_cast_fp16 = slice_by_index(begin = var_6098_begin_0, end = var_6098_end_0, end_mask = var_6098_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6098_cast_fp16")]; + tensor var_6102_begin_0 = const()[name = tensor("op_6102_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6102_end_0 = const()[name = tensor("op_6102_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_6102_end_mask_0 = const()[name = tensor("op_6102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6102_cast_fp16 = slice_by_index(begin = var_6102_begin_0, end = var_6102_end_0, end_mask = var_6102_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6102_cast_fp16")]; + tensor var_6106_begin_0 = const()[name = tensor("op_6106_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6106_end_0 = const()[name = tensor("op_6106_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_6106_end_mask_0 = const()[name = tensor("op_6106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6106_cast_fp16 = slice_by_index(begin = var_6106_begin_0, end = var_6106_end_0, end_mask = var_6106_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6106_cast_fp16")]; + tensor var_6110_begin_0 = const()[name = tensor("op_6110_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6110_end_0 = const()[name = tensor("op_6110_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_6110_end_mask_0 = const()[name = tensor("op_6110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6110_cast_fp16 = slice_by_index(begin = var_6110_begin_0, end = var_6110_end_0, end_mask = var_6110_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6110_cast_fp16")]; + tensor var_6114_begin_0 = const()[name = tensor("op_6114_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6114_end_0 = const()[name = tensor("op_6114_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_6114_end_mask_0 = const()[name = tensor("op_6114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6114_cast_fp16 = slice_by_index(begin = var_6114_begin_0, end = var_6114_end_0, end_mask = var_6114_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6114_cast_fp16")]; + tensor var_6118_begin_0 = const()[name = tensor("op_6118_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6118_end_0 = const()[name = tensor("op_6118_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_6118_end_mask_0 = const()[name = tensor("op_6118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6118_cast_fp16 = slice_by_index(begin = var_6118_begin_0, end = var_6118_end_0, end_mask = var_6118_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6118_cast_fp16")]; + tensor var_6122_begin_0 = const()[name = tensor("op_6122_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6122_end_0 = const()[name = tensor("op_6122_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_6122_end_mask_0 = const()[name = tensor("op_6122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6122_cast_fp16 = slice_by_index(begin = var_6122_begin_0, end = var_6122_end_0, end_mask = var_6122_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6122_cast_fp16")]; + tensor var_6126_begin_0 = const()[name = tensor("op_6126_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6126_end_0 = const()[name = tensor("op_6126_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_6126_end_mask_0 = const()[name = tensor("op_6126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6126_cast_fp16 = slice_by_index(begin = var_6126_begin_0, end = var_6126_end_0, end_mask = var_6126_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6126_cast_fp16")]; + tensor var_6130_begin_0 = const()[name = tensor("op_6130_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6130_end_0 = const()[name = tensor("op_6130_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_6130_end_mask_0 = const()[name = tensor("op_6130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6130_cast_fp16 = slice_by_index(begin = var_6130_begin_0, end = var_6130_end_0, end_mask = var_6130_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6130_cast_fp16")]; + tensor var_6134_begin_0 = const()[name = tensor("op_6134_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6134_end_0 = const()[name = tensor("op_6134_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_6134_end_mask_0 = const()[name = tensor("op_6134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6134_cast_fp16 = slice_by_index(begin = var_6134_begin_0, end = var_6134_end_0, end_mask = var_6134_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6134_cast_fp16")]; + tensor var_6138_begin_0 = const()[name = tensor("op_6138_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6138_end_0 = const()[name = tensor("op_6138_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_6138_end_mask_0 = const()[name = tensor("op_6138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6138_cast_fp16 = slice_by_index(begin = var_6138_begin_0, end = var_6138_end_0, end_mask = var_6138_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6138_cast_fp16")]; + tensor var_6142_begin_0 = const()[name = tensor("op_6142_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6142_end_0 = const()[name = tensor("op_6142_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_6142_end_mask_0 = const()[name = tensor("op_6142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6142_cast_fp16 = slice_by_index(begin = var_6142_begin_0, end = var_6142_end_0, end_mask = var_6142_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6142_cast_fp16")]; + tensor var_6146_begin_0 = const()[name = tensor("op_6146_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6146_end_0 = const()[name = tensor("op_6146_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_6146_end_mask_0 = const()[name = tensor("op_6146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6146_cast_fp16 = slice_by_index(begin = var_6146_begin_0, end = var_6146_end_0, end_mask = var_6146_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6146_cast_fp16")]; + tensor var_6150_begin_0 = const()[name = tensor("op_6150_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6150_end_0 = const()[name = tensor("op_6150_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_6150_end_mask_0 = const()[name = tensor("op_6150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6150_cast_fp16 = slice_by_index(begin = var_6150_begin_0, end = var_6150_end_0, end_mask = var_6150_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6150_cast_fp16")]; + tensor var_6154_begin_0 = const()[name = tensor("op_6154_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6154_end_0 = const()[name = tensor("op_6154_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_6154_end_mask_0 = const()[name = tensor("op_6154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6154_cast_fp16 = slice_by_index(begin = var_6154_begin_0, end = var_6154_end_0, end_mask = var_6154_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6154_cast_fp16")]; + tensor var_6158_begin_0 = const()[name = tensor("op_6158_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6158_end_0 = const()[name = tensor("op_6158_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_6158_end_mask_0 = const()[name = tensor("op_6158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6158_cast_fp16 = slice_by_index(begin = var_6158_begin_0, end = var_6158_end_0, end_mask = var_6158_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6158_cast_fp16")]; + tensor var_6162_begin_0 = const()[name = tensor("op_6162_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6162_end_0 = const()[name = tensor("op_6162_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_6162_end_mask_0 = const()[name = tensor("op_6162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6162_cast_fp16 = slice_by_index(begin = var_6162_begin_0, end = var_6162_end_0, end_mask = var_6162_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6162_cast_fp16")]; + tensor var_6166_begin_0 = const()[name = tensor("op_6166_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6166_end_0 = const()[name = tensor("op_6166_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_6166_end_mask_0 = const()[name = tensor("op_6166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6166_cast_fp16 = slice_by_index(begin = var_6166_begin_0, end = var_6166_end_0, end_mask = var_6166_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6166_cast_fp16")]; + tensor var_6170_begin_0 = const()[name = tensor("op_6170_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6170_end_0 = const()[name = tensor("op_6170_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_6170_end_mask_0 = const()[name = tensor("op_6170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6170_cast_fp16 = slice_by_index(begin = var_6170_begin_0, end = var_6170_end_0, end_mask = var_6170_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6170_cast_fp16")]; + tensor var_6174_equation_0 = const()[name = tensor("op_6174_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6174_cast_fp16 = einsum(equation = var_6174_equation_0, values = (var_6016_cast_fp16, var_5933_cast_fp16))[name = tensor("op_6174_cast_fp16")]; + tensor var_6175_to_fp16 = const()[name = tensor("op_6175_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_441_cast_fp16 = mul(x = var_6174_cast_fp16, y = var_6175_to_fp16)[name = tensor("aw_441_cast_fp16")]; + tensor var_6178_equation_0 = const()[name = tensor("op_6178_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6178_cast_fp16 = einsum(equation = var_6178_equation_0, values = (var_6020_cast_fp16, var_5937_cast_fp16))[name = tensor("op_6178_cast_fp16")]; + tensor var_6179_to_fp16 = const()[name = tensor("op_6179_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_443_cast_fp16 = mul(x = var_6178_cast_fp16, y = var_6179_to_fp16)[name = tensor("aw_443_cast_fp16")]; + tensor var_6182_equation_0 = const()[name = tensor("op_6182_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6182_cast_fp16 = einsum(equation = var_6182_equation_0, values = (var_6024_cast_fp16, var_5941_cast_fp16))[name = tensor("op_6182_cast_fp16")]; + tensor var_6183_to_fp16 = const()[name = tensor("op_6183_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_445_cast_fp16 = mul(x = var_6182_cast_fp16, y = var_6183_to_fp16)[name = tensor("aw_445_cast_fp16")]; + tensor var_6186_equation_0 = const()[name = tensor("op_6186_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6186_cast_fp16 = einsum(equation = var_6186_equation_0, values = (var_6028_cast_fp16, var_5945_cast_fp16))[name = tensor("op_6186_cast_fp16")]; + tensor var_6187_to_fp16 = const()[name = tensor("op_6187_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_447_cast_fp16 = mul(x = var_6186_cast_fp16, y = var_6187_to_fp16)[name = tensor("aw_447_cast_fp16")]; + tensor var_6190_equation_0 = const()[name = tensor("op_6190_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6190_cast_fp16 = einsum(equation = var_6190_equation_0, values = (var_6032_cast_fp16, var_5949_cast_fp16))[name = tensor("op_6190_cast_fp16")]; + tensor var_6191_to_fp16 = const()[name = tensor("op_6191_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_449_cast_fp16 = mul(x = var_6190_cast_fp16, y = var_6191_to_fp16)[name = tensor("aw_449_cast_fp16")]; + tensor var_6194_equation_0 = const()[name = tensor("op_6194_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6194_cast_fp16 = einsum(equation = var_6194_equation_0, values = (var_6036_cast_fp16, var_5953_cast_fp16))[name = tensor("op_6194_cast_fp16")]; + tensor var_6195_to_fp16 = const()[name = tensor("op_6195_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_451_cast_fp16 = mul(x = var_6194_cast_fp16, y = var_6195_to_fp16)[name = tensor("aw_451_cast_fp16")]; + tensor var_6198_equation_0 = const()[name = tensor("op_6198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6198_cast_fp16 = einsum(equation = var_6198_equation_0, values = (var_6040_cast_fp16, var_5957_cast_fp16))[name = tensor("op_6198_cast_fp16")]; + tensor var_6199_to_fp16 = const()[name = tensor("op_6199_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_453_cast_fp16 = mul(x = var_6198_cast_fp16, y = var_6199_to_fp16)[name = tensor("aw_453_cast_fp16")]; + tensor var_6202_equation_0 = const()[name = tensor("op_6202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6202_cast_fp16 = einsum(equation = var_6202_equation_0, values = (var_6044_cast_fp16, var_5961_cast_fp16))[name = tensor("op_6202_cast_fp16")]; + tensor var_6203_to_fp16 = const()[name = tensor("op_6203_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_455_cast_fp16 = mul(x = var_6202_cast_fp16, y = var_6203_to_fp16)[name = tensor("aw_455_cast_fp16")]; + tensor var_6206_equation_0 = const()[name = tensor("op_6206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6206_cast_fp16 = einsum(equation = var_6206_equation_0, values = (var_6048_cast_fp16, var_5965_cast_fp16))[name = tensor("op_6206_cast_fp16")]; + tensor var_6207_to_fp16 = const()[name = tensor("op_6207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_457_cast_fp16 = mul(x = var_6206_cast_fp16, y = var_6207_to_fp16)[name = tensor("aw_457_cast_fp16")]; + tensor var_6210_equation_0 = const()[name = tensor("op_6210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6210_cast_fp16 = einsum(equation = var_6210_equation_0, values = (var_6052_cast_fp16, var_5969_cast_fp16))[name = tensor("op_6210_cast_fp16")]; + tensor var_6211_to_fp16 = const()[name = tensor("op_6211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_459_cast_fp16 = mul(x = var_6210_cast_fp16, y = var_6211_to_fp16)[name = tensor("aw_459_cast_fp16")]; + tensor var_6214_equation_0 = const()[name = tensor("op_6214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6214_cast_fp16 = einsum(equation = var_6214_equation_0, values = (var_6056_cast_fp16, var_5973_cast_fp16))[name = tensor("op_6214_cast_fp16")]; + tensor var_6215_to_fp16 = const()[name = tensor("op_6215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_461_cast_fp16 = mul(x = var_6214_cast_fp16, y = var_6215_to_fp16)[name = tensor("aw_461_cast_fp16")]; + tensor var_6218_equation_0 = const()[name = tensor("op_6218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6218_cast_fp16 = einsum(equation = var_6218_equation_0, values = (var_6060_cast_fp16, var_5977_cast_fp16))[name = tensor("op_6218_cast_fp16")]; + tensor var_6219_to_fp16 = const()[name = tensor("op_6219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_463_cast_fp16 = mul(x = var_6218_cast_fp16, y = var_6219_to_fp16)[name = tensor("aw_463_cast_fp16")]; + tensor var_6222_equation_0 = const()[name = tensor("op_6222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6222_cast_fp16 = einsum(equation = var_6222_equation_0, values = (var_6064_cast_fp16, var_5981_cast_fp16))[name = tensor("op_6222_cast_fp16")]; + tensor var_6223_to_fp16 = const()[name = tensor("op_6223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_465_cast_fp16 = mul(x = var_6222_cast_fp16, y = var_6223_to_fp16)[name = tensor("aw_465_cast_fp16")]; + tensor var_6226_equation_0 = const()[name = tensor("op_6226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6226_cast_fp16 = einsum(equation = var_6226_equation_0, values = (var_6068_cast_fp16, var_5985_cast_fp16))[name = tensor("op_6226_cast_fp16")]; + tensor var_6227_to_fp16 = const()[name = tensor("op_6227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_467_cast_fp16 = mul(x = var_6226_cast_fp16, y = var_6227_to_fp16)[name = tensor("aw_467_cast_fp16")]; + tensor var_6230_equation_0 = const()[name = tensor("op_6230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6230_cast_fp16 = einsum(equation = var_6230_equation_0, values = (var_6072_cast_fp16, var_5989_cast_fp16))[name = tensor("op_6230_cast_fp16")]; + tensor var_6231_to_fp16 = const()[name = tensor("op_6231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_469_cast_fp16 = mul(x = var_6230_cast_fp16, y = var_6231_to_fp16)[name = tensor("aw_469_cast_fp16")]; + tensor var_6234_equation_0 = const()[name = tensor("op_6234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6234_cast_fp16 = einsum(equation = var_6234_equation_0, values = (var_6076_cast_fp16, var_5993_cast_fp16))[name = tensor("op_6234_cast_fp16")]; + tensor var_6235_to_fp16 = const()[name = tensor("op_6235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_471_cast_fp16 = mul(x = var_6234_cast_fp16, y = var_6235_to_fp16)[name = tensor("aw_471_cast_fp16")]; + tensor var_6238_equation_0 = const()[name = tensor("op_6238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6238_cast_fp16 = einsum(equation = var_6238_equation_0, values = (var_6080_cast_fp16, var_5997_cast_fp16))[name = tensor("op_6238_cast_fp16")]; + tensor var_6239_to_fp16 = const()[name = tensor("op_6239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_473_cast_fp16 = mul(x = var_6238_cast_fp16, y = var_6239_to_fp16)[name = tensor("aw_473_cast_fp16")]; + tensor var_6242_equation_0 = const()[name = tensor("op_6242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6242_cast_fp16 = einsum(equation = var_6242_equation_0, values = (var_6084_cast_fp16, var_6001_cast_fp16))[name = tensor("op_6242_cast_fp16")]; + tensor var_6243_to_fp16 = const()[name = tensor("op_6243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_475_cast_fp16 = mul(x = var_6242_cast_fp16, y = var_6243_to_fp16)[name = tensor("aw_475_cast_fp16")]; + tensor var_6246_equation_0 = const()[name = tensor("op_6246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6246_cast_fp16 = einsum(equation = var_6246_equation_0, values = (var_6088_cast_fp16, var_6005_cast_fp16))[name = tensor("op_6246_cast_fp16")]; + tensor var_6247_to_fp16 = const()[name = tensor("op_6247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_477_cast_fp16 = mul(x = var_6246_cast_fp16, y = var_6247_to_fp16)[name = tensor("aw_477_cast_fp16")]; + tensor var_6250_equation_0 = const()[name = tensor("op_6250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6250_cast_fp16 = einsum(equation = var_6250_equation_0, values = (var_6092_cast_fp16, var_6009_cast_fp16))[name = tensor("op_6250_cast_fp16")]; + tensor var_6251_to_fp16 = const()[name = tensor("op_6251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_479_cast_fp16 = mul(x = var_6250_cast_fp16, y = var_6251_to_fp16)[name = tensor("aw_479_cast_fp16")]; + tensor var_6253_cast_fp16 = softmax(axis = var_2624, x = aw_441_cast_fp16)[name = tensor("op_6253_cast_fp16")]; + tensor var_6254_cast_fp16 = softmax(axis = var_2624, x = aw_443_cast_fp16)[name = tensor("op_6254_cast_fp16")]; + tensor var_6255_cast_fp16 = softmax(axis = var_2624, x = aw_445_cast_fp16)[name = tensor("op_6255_cast_fp16")]; + tensor var_6256_cast_fp16 = softmax(axis = var_2624, x = aw_447_cast_fp16)[name = tensor("op_6256_cast_fp16")]; + tensor var_6257_cast_fp16 = softmax(axis = var_2624, x = aw_449_cast_fp16)[name = tensor("op_6257_cast_fp16")]; + tensor var_6258_cast_fp16 = softmax(axis = var_2624, x = aw_451_cast_fp16)[name = tensor("op_6258_cast_fp16")]; + tensor var_6259_cast_fp16 = softmax(axis = var_2624, x = aw_453_cast_fp16)[name = tensor("op_6259_cast_fp16")]; + tensor var_6260_cast_fp16 = softmax(axis = var_2624, x = aw_455_cast_fp16)[name = tensor("op_6260_cast_fp16")]; + tensor var_6261_cast_fp16 = softmax(axis = var_2624, x = aw_457_cast_fp16)[name = tensor("op_6261_cast_fp16")]; + tensor var_6262_cast_fp16 = softmax(axis = var_2624, x = aw_459_cast_fp16)[name = tensor("op_6262_cast_fp16")]; + tensor var_6263_cast_fp16 = softmax(axis = var_2624, x = aw_461_cast_fp16)[name = tensor("op_6263_cast_fp16")]; + tensor var_6264_cast_fp16 = softmax(axis = var_2624, x = aw_463_cast_fp16)[name = tensor("op_6264_cast_fp16")]; + tensor var_6265_cast_fp16 = softmax(axis = var_2624, x = aw_465_cast_fp16)[name = tensor("op_6265_cast_fp16")]; + tensor var_6266_cast_fp16 = softmax(axis = var_2624, x = aw_467_cast_fp16)[name = tensor("op_6266_cast_fp16")]; + tensor var_6267_cast_fp16 = softmax(axis = var_2624, x = aw_469_cast_fp16)[name = tensor("op_6267_cast_fp16")]; + tensor var_6268_cast_fp16 = softmax(axis = var_2624, x = aw_471_cast_fp16)[name = tensor("op_6268_cast_fp16")]; + tensor var_6269_cast_fp16 = softmax(axis = var_2624, x = aw_473_cast_fp16)[name = tensor("op_6269_cast_fp16")]; + tensor var_6270_cast_fp16 = softmax(axis = var_2624, x = aw_475_cast_fp16)[name = tensor("op_6270_cast_fp16")]; + tensor var_6271_cast_fp16 = softmax(axis = var_2624, x = aw_477_cast_fp16)[name = tensor("op_6271_cast_fp16")]; + tensor var_6272_cast_fp16 = softmax(axis = var_2624, x = aw_479_cast_fp16)[name = tensor("op_6272_cast_fp16")]; + tensor var_6274_equation_0 = const()[name = tensor("op_6274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6274_cast_fp16 = einsum(equation = var_6274_equation_0, values = (var_6094_cast_fp16, var_6253_cast_fp16))[name = tensor("op_6274_cast_fp16")]; + tensor var_6276_equation_0 = const()[name = tensor("op_6276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6276_cast_fp16 = einsum(equation = var_6276_equation_0, values = (var_6098_cast_fp16, var_6254_cast_fp16))[name = tensor("op_6276_cast_fp16")]; + tensor var_6278_equation_0 = const()[name = tensor("op_6278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6278_cast_fp16 = einsum(equation = var_6278_equation_0, values = (var_6102_cast_fp16, var_6255_cast_fp16))[name = tensor("op_6278_cast_fp16")]; + tensor var_6280_equation_0 = const()[name = tensor("op_6280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6280_cast_fp16 = einsum(equation = var_6280_equation_0, values = (var_6106_cast_fp16, var_6256_cast_fp16))[name = tensor("op_6280_cast_fp16")]; + tensor var_6282_equation_0 = const()[name = tensor("op_6282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6282_cast_fp16 = einsum(equation = var_6282_equation_0, values = (var_6110_cast_fp16, var_6257_cast_fp16))[name = tensor("op_6282_cast_fp16")]; + tensor var_6284_equation_0 = const()[name = tensor("op_6284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6284_cast_fp16 = einsum(equation = var_6284_equation_0, values = (var_6114_cast_fp16, var_6258_cast_fp16))[name = tensor("op_6284_cast_fp16")]; + tensor var_6286_equation_0 = const()[name = tensor("op_6286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6286_cast_fp16 = einsum(equation = var_6286_equation_0, values = (var_6118_cast_fp16, var_6259_cast_fp16))[name = tensor("op_6286_cast_fp16")]; + tensor var_6288_equation_0 = const()[name = tensor("op_6288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6288_cast_fp16 = einsum(equation = var_6288_equation_0, values = (var_6122_cast_fp16, var_6260_cast_fp16))[name = tensor("op_6288_cast_fp16")]; + tensor var_6290_equation_0 = const()[name = tensor("op_6290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6290_cast_fp16 = einsum(equation = var_6290_equation_0, values = (var_6126_cast_fp16, var_6261_cast_fp16))[name = tensor("op_6290_cast_fp16")]; + tensor var_6292_equation_0 = const()[name = tensor("op_6292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6292_cast_fp16 = einsum(equation = var_6292_equation_0, values = (var_6130_cast_fp16, var_6262_cast_fp16))[name = tensor("op_6292_cast_fp16")]; + tensor var_6294_equation_0 = const()[name = tensor("op_6294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6294_cast_fp16 = einsum(equation = var_6294_equation_0, values = (var_6134_cast_fp16, var_6263_cast_fp16))[name = tensor("op_6294_cast_fp16")]; + tensor var_6296_equation_0 = const()[name = tensor("op_6296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6296_cast_fp16 = einsum(equation = var_6296_equation_0, values = (var_6138_cast_fp16, var_6264_cast_fp16))[name = tensor("op_6296_cast_fp16")]; + tensor var_6298_equation_0 = const()[name = tensor("op_6298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6298_cast_fp16 = einsum(equation = var_6298_equation_0, values = (var_6142_cast_fp16, var_6265_cast_fp16))[name = tensor("op_6298_cast_fp16")]; + tensor var_6300_equation_0 = const()[name = tensor("op_6300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6300_cast_fp16 = einsum(equation = var_6300_equation_0, values = (var_6146_cast_fp16, var_6266_cast_fp16))[name = tensor("op_6300_cast_fp16")]; + tensor var_6302_equation_0 = const()[name = tensor("op_6302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6302_cast_fp16 = einsum(equation = var_6302_equation_0, values = (var_6150_cast_fp16, var_6267_cast_fp16))[name = tensor("op_6302_cast_fp16")]; + tensor var_6304_equation_0 = const()[name = tensor("op_6304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6304_cast_fp16 = einsum(equation = var_6304_equation_0, values = (var_6154_cast_fp16, var_6268_cast_fp16))[name = tensor("op_6304_cast_fp16")]; + tensor var_6306_equation_0 = const()[name = tensor("op_6306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6306_cast_fp16 = einsum(equation = var_6306_equation_0, values = (var_6158_cast_fp16, var_6269_cast_fp16))[name = tensor("op_6306_cast_fp16")]; + tensor var_6308_equation_0 = const()[name = tensor("op_6308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6308_cast_fp16 = einsum(equation = var_6308_equation_0, values = (var_6162_cast_fp16, var_6270_cast_fp16))[name = tensor("op_6308_cast_fp16")]; + tensor var_6310_equation_0 = const()[name = tensor("op_6310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6310_cast_fp16 = einsum(equation = var_6310_equation_0, values = (var_6166_cast_fp16, var_6271_cast_fp16))[name = tensor("op_6310_cast_fp16")]; + tensor var_6312_equation_0 = const()[name = tensor("op_6312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6312_cast_fp16 = einsum(equation = var_6312_equation_0, values = (var_6170_cast_fp16, var_6272_cast_fp16))[name = tensor("op_6312_cast_fp16")]; + tensor input_157_interleave_0 = const()[name = tensor("input_157_interleave_0"), val = tensor(false)]; + tensor input_157_cast_fp16 = concat(axis = var_2624, interleave = input_157_interleave_0, values = (var_6274_cast_fp16, var_6276_cast_fp16, var_6278_cast_fp16, var_6280_cast_fp16, var_6282_cast_fp16, var_6284_cast_fp16, var_6286_cast_fp16, var_6288_cast_fp16, var_6290_cast_fp16, var_6292_cast_fp16, var_6294_cast_fp16, var_6296_cast_fp16, var_6298_cast_fp16, var_6300_cast_fp16, var_6302_cast_fp16, var_6304_cast_fp16, var_6306_cast_fp16, var_6308_cast_fp16, var_6310_cast_fp16, var_6312_cast_fp16))[name = tensor("input_157_cast_fp16")]; + tensor var_6322_pad_type_0 = const()[name = tensor("op_6322_pad_type_0"), val = tensor("valid")]; + tensor var_6322_strides_0 = const()[name = tensor("op_6322_strides_0"), val = tensor([1, 1])]; + tensor var_6322_pad_0 = const()[name = tensor("op_6322_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6322_dilations_0 = const()[name = tensor("op_6322_dilations_0"), val = tensor([1, 1])]; + tensor var_6322_groups_0 = const()[name = tensor("op_6322_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162822848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164051712))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164051904)))]; + tensor var_6322_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_6322_dilations_0, groups = var_6322_groups_0, pad = var_6322_pad_0, pad_type = var_6322_pad_type_0, strides = var_6322_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor("op_6322_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = var_6322_cast_fp16, y = inputs_45_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; + tensor input_159_axes_0 = const()[name = tensor("input_159_axes_0"), val = tensor([1])]; + tensor input_159_gamma_0_to_fp16 = const()[name = tensor("input_159_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164054528)))]; + tensor input_159_beta_0_to_fp16 = const()[name = tensor("input_159_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164057152)))]; + tensor var_6332_to_fp16 = const()[name = tensor("op_6332_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_159_cast_fp16 = layer_norm(axes = input_159_axes_0, beta = input_159_beta_0_to_fp16, epsilon = var_6332_to_fp16, gamma = input_159_gamma_0_to_fp16, x = inputs_47_cast_fp16)[name = tensor("input_159_cast_fp16")]; + tensor var_6352_pad_type_0 = const()[name = tensor("op_6352_pad_type_0"), val = tensor("valid")]; + tensor var_6352_strides_0 = const()[name = tensor("op_6352_strides_0"), val = tensor([1, 1])]; + tensor var_6352_pad_0 = const()[name = tensor("op_6352_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6352_dilations_0 = const()[name = tensor("op_6352_dilations_0"), val = tensor([1, 1])]; + tensor var_6352_groups_0 = const()[name = tensor("op_6352_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164059776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173890240))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173890432)))]; + tensor var_6352_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_6352_dilations_0, groups = var_6352_groups_0, pad = var_6352_pad_0, pad_type = var_6352_pad_type_0, strides = var_6352_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized, x = input_159_cast_fp16)[name = tensor("op_6352_cast_fp16")]; + tensor var_6353_split_sizes_0 = const()[name = tensor("op_6353_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_6353_axis_0 = const()[name = tensor("op_6353_axis_0"), val = tensor(1)]; + tensor var_6353_cast_fp16_0, tensor var_6353_cast_fp16_1 = split(axis = var_6353_axis_0, split_sizes = var_6353_split_sizes_0, x = var_6352_cast_fp16)[name = tensor("op_6353_cast_fp16")]; + tensor var_6355_mode_0 = const()[name = tensor("op_6355_mode_0"), val = tensor("EXACT")]; + tensor var_6355_cast_fp16 = gelu(mode = var_6355_mode_0, x = var_6353_cast_fp16_1)[name = tensor("op_6355_cast_fp16")]; + tensor input_161_cast_fp16 = mul(x = var_6353_cast_fp16_0, y = var_6355_cast_fp16)[name = tensor("input_161_cast_fp16")]; + tensor var_6363_pad_type_0 = const()[name = tensor("op_6363_pad_type_0"), val = tensor("valid")]; + tensor var_6363_strides_0 = const()[name = tensor("op_6363_strides_0"), val = tensor([1, 1])]; + tensor var_6363_pad_0 = const()[name = tensor("op_6363_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6363_dilations_0 = const()[name = tensor("op_6363_dilations_0"), val = tensor([1, 1])]; + tensor var_6363_groups_0 = const()[name = tensor("op_6363_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173910976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178826240))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178826432)))]; + tensor var_6363_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_6363_dilations_0, groups = var_6363_groups_0, pad = var_6363_pad_0, pad_type = var_6363_pad_type_0, strides = var_6363_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor("op_6363_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = var_6363_cast_fp16, y = inputs_47_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; + tensor hidden_states_89_axes_0 = const()[name = tensor("hidden_states_89_axes_0"), val = tensor([1])]; + tensor hidden_states_89_gamma_0_to_fp16 = const()[name = tensor("hidden_states_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178829056)))]; + tensor hidden_states_89_beta_0_to_fp16 = const()[name = tensor("hidden_states_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178831680)))]; + tensor var_6379_to_fp16 = const()[name = tensor("op_6379_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_89_cast_fp16 = layer_norm(axes = hidden_states_89_axes_0, beta = hidden_states_89_beta_0_to_fp16, epsilon = var_6379_to_fp16, gamma = hidden_states_89_gamma_0_to_fp16, x = inputs_49_cast_fp16)[name = tensor("hidden_states_89_cast_fp16")]; + tensor q_33_pad_type_0 = const()[name = tensor("q_33_pad_type_0"), val = tensor("valid")]; + tensor q_33_strides_0 = const()[name = tensor("q_33_strides_0"), val = tensor([1, 1])]; + tensor q_33_pad_0 = const()[name = tensor("q_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_33_dilations_0 = const()[name = tensor("q_33_dilations_0"), val = tensor([1, 1])]; + tensor q_33_groups_0 = const()[name = tensor("q_33_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178834304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180063168))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_33_cast_fp16 = conv(dilations = q_33_dilations_0, groups = q_33_groups_0, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = q_33_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_89_cast_fp16)[name = tensor("q_33_cast_fp16")]; + tensor k_65_pad_type_0 = const()[name = tensor("k_65_pad_type_0"), val = tensor("valid")]; + tensor k_65_strides_0 = const()[name = tensor("k_65_strides_0"), val = tensor([1, 1])]; + tensor k_65_pad_0 = const()[name = tensor("k_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_65_dilations_0 = const()[name = tensor("k_65_dilations_0"), val = tensor([1, 1])]; + tensor k_65_groups_0 = const()[name = tensor("k_65_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180063360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181292224))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_65_cast_fp16 = conv(dilations = k_65_dilations_0, groups = k_65_groups_0, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = k_65_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_89_cast_fp16)[name = tensor("k_65_cast_fp16")]; + tensor v_33_pad_type_0 = const()[name = tensor("v_33_pad_type_0"), val = tensor("valid")]; + tensor v_33_strides_0 = const()[name = tensor("v_33_strides_0"), val = tensor([1, 1])]; + tensor v_33_pad_0 = const()[name = tensor("v_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_33_dilations_0 = const()[name = tensor("v_33_dilations_0"), val = tensor([1, 1])]; + tensor v_33_groups_0 = const()[name = tensor("v_33_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181292416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(182521280))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_33_cast_fp16 = conv(dilations = v_33_dilations_0, groups = v_33_groups_0, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = v_33_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_89_cast_fp16)[name = tensor("v_33_cast_fp16")]; + tensor var_6412_begin_0 = const()[name = tensor("op_6412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6412_end_0 = const()[name = tensor("op_6412_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_6412_end_mask_0 = const()[name = tensor("op_6412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6412_cast_fp16 = slice_by_index(begin = var_6412_begin_0, end = var_6412_end_0, end_mask = var_6412_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6412_cast_fp16")]; + tensor var_6416_begin_0 = const()[name = tensor("op_6416_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6416_end_0 = const()[name = tensor("op_6416_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_6416_end_mask_0 = const()[name = tensor("op_6416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6416_cast_fp16 = slice_by_index(begin = var_6416_begin_0, end = var_6416_end_0, end_mask = var_6416_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6416_cast_fp16")]; + tensor var_6420_begin_0 = const()[name = tensor("op_6420_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6420_end_0 = const()[name = tensor("op_6420_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_6420_end_mask_0 = const()[name = tensor("op_6420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6420_cast_fp16 = slice_by_index(begin = var_6420_begin_0, end = var_6420_end_0, end_mask = var_6420_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6420_cast_fp16")]; + tensor var_6424_begin_0 = const()[name = tensor("op_6424_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6424_end_0 = const()[name = tensor("op_6424_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_6424_end_mask_0 = const()[name = tensor("op_6424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6424_cast_fp16 = slice_by_index(begin = var_6424_begin_0, end = var_6424_end_0, end_mask = var_6424_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6424_cast_fp16")]; + tensor var_6428_begin_0 = const()[name = tensor("op_6428_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6428_end_0 = const()[name = tensor("op_6428_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_6428_end_mask_0 = const()[name = tensor("op_6428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6428_cast_fp16 = slice_by_index(begin = var_6428_begin_0, end = var_6428_end_0, end_mask = var_6428_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6428_cast_fp16")]; + tensor var_6432_begin_0 = const()[name = tensor("op_6432_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6432_end_0 = const()[name = tensor("op_6432_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_6432_end_mask_0 = const()[name = tensor("op_6432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6432_cast_fp16 = slice_by_index(begin = var_6432_begin_0, end = var_6432_end_0, end_mask = var_6432_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6432_cast_fp16")]; + tensor var_6436_begin_0 = const()[name = tensor("op_6436_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6436_end_0 = const()[name = tensor("op_6436_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_6436_end_mask_0 = const()[name = tensor("op_6436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6436_cast_fp16 = slice_by_index(begin = var_6436_begin_0, end = var_6436_end_0, end_mask = var_6436_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6436_cast_fp16")]; + tensor var_6440_begin_0 = const()[name = tensor("op_6440_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6440_end_0 = const()[name = tensor("op_6440_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_6440_end_mask_0 = const()[name = tensor("op_6440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6440_cast_fp16 = slice_by_index(begin = var_6440_begin_0, end = var_6440_end_0, end_mask = var_6440_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6440_cast_fp16")]; + tensor var_6444_begin_0 = const()[name = tensor("op_6444_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6444_end_0 = const()[name = tensor("op_6444_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_6444_end_mask_0 = const()[name = tensor("op_6444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6444_cast_fp16 = slice_by_index(begin = var_6444_begin_0, end = var_6444_end_0, end_mask = var_6444_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6444_cast_fp16")]; + tensor var_6448_begin_0 = const()[name = tensor("op_6448_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6448_end_0 = const()[name = tensor("op_6448_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_6448_end_mask_0 = const()[name = tensor("op_6448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6448_cast_fp16 = slice_by_index(begin = var_6448_begin_0, end = var_6448_end_0, end_mask = var_6448_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6448_cast_fp16")]; + tensor var_6452_begin_0 = const()[name = tensor("op_6452_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6452_end_0 = const()[name = tensor("op_6452_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_6452_end_mask_0 = const()[name = tensor("op_6452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6452_cast_fp16 = slice_by_index(begin = var_6452_begin_0, end = var_6452_end_0, end_mask = var_6452_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6452_cast_fp16")]; + tensor var_6456_begin_0 = const()[name = tensor("op_6456_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6456_end_0 = const()[name = tensor("op_6456_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_6456_end_mask_0 = const()[name = tensor("op_6456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6456_cast_fp16 = slice_by_index(begin = var_6456_begin_0, end = var_6456_end_0, end_mask = var_6456_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6456_cast_fp16")]; + tensor var_6460_begin_0 = const()[name = tensor("op_6460_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6460_end_0 = const()[name = tensor("op_6460_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_6460_end_mask_0 = const()[name = tensor("op_6460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6460_cast_fp16 = slice_by_index(begin = var_6460_begin_0, end = var_6460_end_0, end_mask = var_6460_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6460_cast_fp16")]; + tensor var_6464_begin_0 = const()[name = tensor("op_6464_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6464_end_0 = const()[name = tensor("op_6464_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_6464_end_mask_0 = const()[name = tensor("op_6464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6464_cast_fp16 = slice_by_index(begin = var_6464_begin_0, end = var_6464_end_0, end_mask = var_6464_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6464_cast_fp16")]; + tensor var_6468_begin_0 = const()[name = tensor("op_6468_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6468_end_0 = const()[name = tensor("op_6468_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_6468_end_mask_0 = const()[name = tensor("op_6468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6468_cast_fp16 = slice_by_index(begin = var_6468_begin_0, end = var_6468_end_0, end_mask = var_6468_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6468_cast_fp16")]; + tensor var_6472_begin_0 = const()[name = tensor("op_6472_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6472_end_0 = const()[name = tensor("op_6472_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6472_end_mask_0 = const()[name = tensor("op_6472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6472_cast_fp16 = slice_by_index(begin = var_6472_begin_0, end = var_6472_end_0, end_mask = var_6472_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6472_cast_fp16")]; + tensor var_6476_begin_0 = const()[name = tensor("op_6476_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6476_end_0 = const()[name = tensor("op_6476_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_6476_end_mask_0 = const()[name = tensor("op_6476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6476_cast_fp16 = slice_by_index(begin = var_6476_begin_0, end = var_6476_end_0, end_mask = var_6476_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6476_cast_fp16")]; + tensor var_6480_begin_0 = const()[name = tensor("op_6480_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6480_end_0 = const()[name = tensor("op_6480_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6480_end_mask_0 = const()[name = tensor("op_6480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6480_cast_fp16 = slice_by_index(begin = var_6480_begin_0, end = var_6480_end_0, end_mask = var_6480_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6480_cast_fp16")]; + tensor var_6484_begin_0 = const()[name = tensor("op_6484_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6484_end_0 = const()[name = tensor("op_6484_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6484_end_mask_0 = const()[name = tensor("op_6484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6484_cast_fp16 = slice_by_index(begin = var_6484_begin_0, end = var_6484_end_0, end_mask = var_6484_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6484_cast_fp16")]; + tensor var_6488_begin_0 = const()[name = tensor("op_6488_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6488_end_0 = const()[name = tensor("op_6488_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6488_end_mask_0 = const()[name = tensor("op_6488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6488_cast_fp16 = slice_by_index(begin = var_6488_begin_0, end = var_6488_end_0, end_mask = var_6488_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6488_cast_fp16")]; + tensor k_67_perm_0 = const()[name = tensor("k_67_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6495_begin_0 = const()[name = tensor("op_6495_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6495_end_0 = const()[name = tensor("op_6495_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_6495_end_mask_0 = const()[name = tensor("op_6495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_67_cast_fp16 = transpose(perm = k_67_perm_0, x = k_65_cast_fp16)[name = tensor("transpose_51")]; + tensor var_6495_cast_fp16 = slice_by_index(begin = var_6495_begin_0, end = var_6495_end_0, end_mask = var_6495_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6495_cast_fp16")]; + tensor var_6499_begin_0 = const()[name = tensor("op_6499_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6499_end_0 = const()[name = tensor("op_6499_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_6499_end_mask_0 = const()[name = tensor("op_6499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6499_cast_fp16 = slice_by_index(begin = var_6499_begin_0, end = var_6499_end_0, end_mask = var_6499_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6499_cast_fp16")]; + tensor var_6503_begin_0 = const()[name = tensor("op_6503_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6503_end_0 = const()[name = tensor("op_6503_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_6503_end_mask_0 = const()[name = tensor("op_6503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6503_cast_fp16 = slice_by_index(begin = var_6503_begin_0, end = var_6503_end_0, end_mask = var_6503_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6503_cast_fp16")]; + tensor var_6507_begin_0 = const()[name = tensor("op_6507_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6507_end_0 = const()[name = tensor("op_6507_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_6507_end_mask_0 = const()[name = tensor("op_6507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6507_cast_fp16 = slice_by_index(begin = var_6507_begin_0, end = var_6507_end_0, end_mask = var_6507_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6507_cast_fp16")]; + tensor var_6511_begin_0 = const()[name = tensor("op_6511_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6511_end_0 = const()[name = tensor("op_6511_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_6511_end_mask_0 = const()[name = tensor("op_6511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6511_cast_fp16 = slice_by_index(begin = var_6511_begin_0, end = var_6511_end_0, end_mask = var_6511_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6511_cast_fp16")]; + tensor var_6515_begin_0 = const()[name = tensor("op_6515_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6515_end_0 = const()[name = tensor("op_6515_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_6515_end_mask_0 = const()[name = tensor("op_6515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6515_cast_fp16 = slice_by_index(begin = var_6515_begin_0, end = var_6515_end_0, end_mask = var_6515_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6515_cast_fp16")]; + tensor var_6519_begin_0 = const()[name = tensor("op_6519_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6519_end_0 = const()[name = tensor("op_6519_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_6519_end_mask_0 = const()[name = tensor("op_6519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6519_cast_fp16 = slice_by_index(begin = var_6519_begin_0, end = var_6519_end_0, end_mask = var_6519_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6519_cast_fp16")]; + tensor var_6523_begin_0 = const()[name = tensor("op_6523_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6523_end_0 = const()[name = tensor("op_6523_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_6523_end_mask_0 = const()[name = tensor("op_6523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6523_cast_fp16 = slice_by_index(begin = var_6523_begin_0, end = var_6523_end_0, end_mask = var_6523_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6523_cast_fp16")]; + tensor var_6527_begin_0 = const()[name = tensor("op_6527_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6527_end_0 = const()[name = tensor("op_6527_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_6527_end_mask_0 = const()[name = tensor("op_6527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6527_cast_fp16 = slice_by_index(begin = var_6527_begin_0, end = var_6527_end_0, end_mask = var_6527_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6527_cast_fp16")]; + tensor var_6531_begin_0 = const()[name = tensor("op_6531_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6531_end_0 = const()[name = tensor("op_6531_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_6531_end_mask_0 = const()[name = tensor("op_6531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6531_cast_fp16 = slice_by_index(begin = var_6531_begin_0, end = var_6531_end_0, end_mask = var_6531_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6531_cast_fp16")]; + tensor var_6535_begin_0 = const()[name = tensor("op_6535_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6535_end_0 = const()[name = tensor("op_6535_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_6535_end_mask_0 = const()[name = tensor("op_6535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6535_cast_fp16 = slice_by_index(begin = var_6535_begin_0, end = var_6535_end_0, end_mask = var_6535_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6535_cast_fp16")]; + tensor var_6539_begin_0 = const()[name = tensor("op_6539_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6539_end_0 = const()[name = tensor("op_6539_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_6539_end_mask_0 = const()[name = tensor("op_6539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6539_cast_fp16 = slice_by_index(begin = var_6539_begin_0, end = var_6539_end_0, end_mask = var_6539_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6539_cast_fp16")]; + tensor var_6543_begin_0 = const()[name = tensor("op_6543_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6543_end_0 = const()[name = tensor("op_6543_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_6543_end_mask_0 = const()[name = tensor("op_6543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6543_cast_fp16 = slice_by_index(begin = var_6543_begin_0, end = var_6543_end_0, end_mask = var_6543_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6543_cast_fp16")]; + tensor var_6547_begin_0 = const()[name = tensor("op_6547_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6547_end_0 = const()[name = tensor("op_6547_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_6547_end_mask_0 = const()[name = tensor("op_6547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6547_cast_fp16 = slice_by_index(begin = var_6547_begin_0, end = var_6547_end_0, end_mask = var_6547_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6547_cast_fp16")]; + tensor var_6551_begin_0 = const()[name = tensor("op_6551_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6551_end_0 = const()[name = tensor("op_6551_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_6551_end_mask_0 = const()[name = tensor("op_6551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6551_cast_fp16 = slice_by_index(begin = var_6551_begin_0, end = var_6551_end_0, end_mask = var_6551_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6551_cast_fp16")]; + tensor var_6555_begin_0 = const()[name = tensor("op_6555_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6555_end_0 = const()[name = tensor("op_6555_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6555_end_mask_0 = const()[name = tensor("op_6555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6555_cast_fp16 = slice_by_index(begin = var_6555_begin_0, end = var_6555_end_0, end_mask = var_6555_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6555_cast_fp16")]; + tensor var_6559_begin_0 = const()[name = tensor("op_6559_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6559_end_0 = const()[name = tensor("op_6559_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_6559_end_mask_0 = const()[name = tensor("op_6559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6559_cast_fp16 = slice_by_index(begin = var_6559_begin_0, end = var_6559_end_0, end_mask = var_6559_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6559_cast_fp16")]; + tensor var_6563_begin_0 = const()[name = tensor("op_6563_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6563_end_0 = const()[name = tensor("op_6563_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_6563_end_mask_0 = const()[name = tensor("op_6563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6563_cast_fp16 = slice_by_index(begin = var_6563_begin_0, end = var_6563_end_0, end_mask = var_6563_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6563_cast_fp16")]; + tensor var_6567_begin_0 = const()[name = tensor("op_6567_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6567_end_0 = const()[name = tensor("op_6567_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_6567_end_mask_0 = const()[name = tensor("op_6567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6567_cast_fp16 = slice_by_index(begin = var_6567_begin_0, end = var_6567_end_0, end_mask = var_6567_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6567_cast_fp16")]; + tensor var_6571_begin_0 = const()[name = tensor("op_6571_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_6571_end_0 = const()[name = tensor("op_6571_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_6571_end_mask_0 = const()[name = tensor("op_6571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6571_cast_fp16 = slice_by_index(begin = var_6571_begin_0, end = var_6571_end_0, end_mask = var_6571_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6571_cast_fp16")]; + tensor var_6573_begin_0 = const()[name = tensor("op_6573_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6573_end_0 = const()[name = tensor("op_6573_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_6573_end_mask_0 = const()[name = tensor("op_6573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6573_cast_fp16 = slice_by_index(begin = var_6573_begin_0, end = var_6573_end_0, end_mask = var_6573_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6573_cast_fp16")]; + tensor var_6577_begin_0 = const()[name = tensor("op_6577_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6577_end_0 = const()[name = tensor("op_6577_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_6577_end_mask_0 = const()[name = tensor("op_6577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6577_cast_fp16 = slice_by_index(begin = var_6577_begin_0, end = var_6577_end_0, end_mask = var_6577_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6577_cast_fp16")]; + tensor var_6581_begin_0 = const()[name = tensor("op_6581_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6581_end_0 = const()[name = tensor("op_6581_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_6581_end_mask_0 = const()[name = tensor("op_6581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6581_cast_fp16 = slice_by_index(begin = var_6581_begin_0, end = var_6581_end_0, end_mask = var_6581_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6581_cast_fp16")]; + tensor var_6585_begin_0 = const()[name = tensor("op_6585_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6585_end_0 = const()[name = tensor("op_6585_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_6585_end_mask_0 = const()[name = tensor("op_6585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6585_cast_fp16 = slice_by_index(begin = var_6585_begin_0, end = var_6585_end_0, end_mask = var_6585_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6585_cast_fp16")]; + tensor var_6589_begin_0 = const()[name = tensor("op_6589_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6589_end_0 = const()[name = tensor("op_6589_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_6589_end_mask_0 = const()[name = tensor("op_6589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6589_cast_fp16 = slice_by_index(begin = var_6589_begin_0, end = var_6589_end_0, end_mask = var_6589_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6589_cast_fp16")]; + tensor var_6593_begin_0 = const()[name = tensor("op_6593_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6593_end_0 = const()[name = tensor("op_6593_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_6593_end_mask_0 = const()[name = tensor("op_6593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6593_cast_fp16 = slice_by_index(begin = var_6593_begin_0, end = var_6593_end_0, end_mask = var_6593_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6593_cast_fp16")]; + tensor var_6597_begin_0 = const()[name = tensor("op_6597_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6597_end_0 = const()[name = tensor("op_6597_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_6597_end_mask_0 = const()[name = tensor("op_6597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6597_cast_fp16 = slice_by_index(begin = var_6597_begin_0, end = var_6597_end_0, end_mask = var_6597_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6597_cast_fp16")]; + tensor var_6601_begin_0 = const()[name = tensor("op_6601_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6601_end_0 = const()[name = tensor("op_6601_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_6601_end_mask_0 = const()[name = tensor("op_6601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6601_cast_fp16 = slice_by_index(begin = var_6601_begin_0, end = var_6601_end_0, end_mask = var_6601_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6601_cast_fp16")]; + tensor var_6605_begin_0 = const()[name = tensor("op_6605_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6605_end_0 = const()[name = tensor("op_6605_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_6605_end_mask_0 = const()[name = tensor("op_6605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6605_cast_fp16 = slice_by_index(begin = var_6605_begin_0, end = var_6605_end_0, end_mask = var_6605_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6605_cast_fp16")]; + tensor var_6609_begin_0 = const()[name = tensor("op_6609_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6609_end_0 = const()[name = tensor("op_6609_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_6609_end_mask_0 = const()[name = tensor("op_6609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6609_cast_fp16 = slice_by_index(begin = var_6609_begin_0, end = var_6609_end_0, end_mask = var_6609_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6609_cast_fp16")]; + tensor var_6613_begin_0 = const()[name = tensor("op_6613_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6613_end_0 = const()[name = tensor("op_6613_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_6613_end_mask_0 = const()[name = tensor("op_6613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6613_cast_fp16 = slice_by_index(begin = var_6613_begin_0, end = var_6613_end_0, end_mask = var_6613_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6613_cast_fp16")]; + tensor var_6617_begin_0 = const()[name = tensor("op_6617_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6617_end_0 = const()[name = tensor("op_6617_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_6617_end_mask_0 = const()[name = tensor("op_6617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6617_cast_fp16 = slice_by_index(begin = var_6617_begin_0, end = var_6617_end_0, end_mask = var_6617_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6617_cast_fp16")]; + tensor var_6621_begin_0 = const()[name = tensor("op_6621_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6621_end_0 = const()[name = tensor("op_6621_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_6621_end_mask_0 = const()[name = tensor("op_6621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6621_cast_fp16 = slice_by_index(begin = var_6621_begin_0, end = var_6621_end_0, end_mask = var_6621_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6621_cast_fp16")]; + tensor var_6625_begin_0 = const()[name = tensor("op_6625_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6625_end_0 = const()[name = tensor("op_6625_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_6625_end_mask_0 = const()[name = tensor("op_6625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6625_cast_fp16 = slice_by_index(begin = var_6625_begin_0, end = var_6625_end_0, end_mask = var_6625_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6625_cast_fp16")]; + tensor var_6629_begin_0 = const()[name = tensor("op_6629_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6629_end_0 = const()[name = tensor("op_6629_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_6629_end_mask_0 = const()[name = tensor("op_6629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6629_cast_fp16 = slice_by_index(begin = var_6629_begin_0, end = var_6629_end_0, end_mask = var_6629_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6629_cast_fp16")]; + tensor var_6633_begin_0 = const()[name = tensor("op_6633_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6633_end_0 = const()[name = tensor("op_6633_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6633_end_mask_0 = const()[name = tensor("op_6633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6633_cast_fp16 = slice_by_index(begin = var_6633_begin_0, end = var_6633_end_0, end_mask = var_6633_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6633_cast_fp16")]; + tensor var_6637_begin_0 = const()[name = tensor("op_6637_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6637_end_0 = const()[name = tensor("op_6637_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_6637_end_mask_0 = const()[name = tensor("op_6637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6637_cast_fp16 = slice_by_index(begin = var_6637_begin_0, end = var_6637_end_0, end_mask = var_6637_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6637_cast_fp16")]; + tensor var_6641_begin_0 = const()[name = tensor("op_6641_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6641_end_0 = const()[name = tensor("op_6641_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6641_end_mask_0 = const()[name = tensor("op_6641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6641_cast_fp16 = slice_by_index(begin = var_6641_begin_0, end = var_6641_end_0, end_mask = var_6641_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6641_cast_fp16")]; + tensor var_6645_begin_0 = const()[name = tensor("op_6645_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6645_end_0 = const()[name = tensor("op_6645_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6645_end_mask_0 = const()[name = tensor("op_6645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6645_cast_fp16 = slice_by_index(begin = var_6645_begin_0, end = var_6645_end_0, end_mask = var_6645_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6645_cast_fp16")]; + tensor var_6649_begin_0 = const()[name = tensor("op_6649_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6649_end_0 = const()[name = tensor("op_6649_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6649_end_mask_0 = const()[name = tensor("op_6649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6649_cast_fp16 = slice_by_index(begin = var_6649_begin_0, end = var_6649_end_0, end_mask = var_6649_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6649_cast_fp16")]; + tensor var_6653_equation_0 = const()[name = tensor("op_6653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6653_cast_fp16 = einsum(equation = var_6653_equation_0, values = (var_6495_cast_fp16, var_6412_cast_fp16))[name = tensor("op_6653_cast_fp16")]; + tensor var_6654_to_fp16 = const()[name = tensor("op_6654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_481_cast_fp16 = mul(x = var_6653_cast_fp16, y = var_6654_to_fp16)[name = tensor("aw_481_cast_fp16")]; + tensor var_6657_equation_0 = const()[name = tensor("op_6657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6657_cast_fp16 = einsum(equation = var_6657_equation_0, values = (var_6499_cast_fp16, var_6416_cast_fp16))[name = tensor("op_6657_cast_fp16")]; + tensor var_6658_to_fp16 = const()[name = tensor("op_6658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_483_cast_fp16 = mul(x = var_6657_cast_fp16, y = var_6658_to_fp16)[name = tensor("aw_483_cast_fp16")]; + tensor var_6661_equation_0 = const()[name = tensor("op_6661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6661_cast_fp16 = einsum(equation = var_6661_equation_0, values = (var_6503_cast_fp16, var_6420_cast_fp16))[name = tensor("op_6661_cast_fp16")]; + tensor var_6662_to_fp16 = const()[name = tensor("op_6662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_485_cast_fp16 = mul(x = var_6661_cast_fp16, y = var_6662_to_fp16)[name = tensor("aw_485_cast_fp16")]; + tensor var_6665_equation_0 = const()[name = tensor("op_6665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6665_cast_fp16 = einsum(equation = var_6665_equation_0, values = (var_6507_cast_fp16, var_6424_cast_fp16))[name = tensor("op_6665_cast_fp16")]; + tensor var_6666_to_fp16 = const()[name = tensor("op_6666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_487_cast_fp16 = mul(x = var_6665_cast_fp16, y = var_6666_to_fp16)[name = tensor("aw_487_cast_fp16")]; + tensor var_6669_equation_0 = const()[name = tensor("op_6669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6669_cast_fp16 = einsum(equation = var_6669_equation_0, values = (var_6511_cast_fp16, var_6428_cast_fp16))[name = tensor("op_6669_cast_fp16")]; + tensor var_6670_to_fp16 = const()[name = tensor("op_6670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_489_cast_fp16 = mul(x = var_6669_cast_fp16, y = var_6670_to_fp16)[name = tensor("aw_489_cast_fp16")]; + tensor var_6673_equation_0 = const()[name = tensor("op_6673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6673_cast_fp16 = einsum(equation = var_6673_equation_0, values = (var_6515_cast_fp16, var_6432_cast_fp16))[name = tensor("op_6673_cast_fp16")]; + tensor var_6674_to_fp16 = const()[name = tensor("op_6674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_491_cast_fp16 = mul(x = var_6673_cast_fp16, y = var_6674_to_fp16)[name = tensor("aw_491_cast_fp16")]; + tensor var_6677_equation_0 = const()[name = tensor("op_6677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6677_cast_fp16 = einsum(equation = var_6677_equation_0, values = (var_6519_cast_fp16, var_6436_cast_fp16))[name = tensor("op_6677_cast_fp16")]; + tensor var_6678_to_fp16 = const()[name = tensor("op_6678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_493_cast_fp16 = mul(x = var_6677_cast_fp16, y = var_6678_to_fp16)[name = tensor("aw_493_cast_fp16")]; + tensor var_6681_equation_0 = const()[name = tensor("op_6681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6681_cast_fp16 = einsum(equation = var_6681_equation_0, values = (var_6523_cast_fp16, var_6440_cast_fp16))[name = tensor("op_6681_cast_fp16")]; + tensor var_6682_to_fp16 = const()[name = tensor("op_6682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_495_cast_fp16 = mul(x = var_6681_cast_fp16, y = var_6682_to_fp16)[name = tensor("aw_495_cast_fp16")]; + tensor var_6685_equation_0 = const()[name = tensor("op_6685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6685_cast_fp16 = einsum(equation = var_6685_equation_0, values = (var_6527_cast_fp16, var_6444_cast_fp16))[name = tensor("op_6685_cast_fp16")]; + tensor var_6686_to_fp16 = const()[name = tensor("op_6686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_497_cast_fp16 = mul(x = var_6685_cast_fp16, y = var_6686_to_fp16)[name = tensor("aw_497_cast_fp16")]; + tensor var_6689_equation_0 = const()[name = tensor("op_6689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6689_cast_fp16 = einsum(equation = var_6689_equation_0, values = (var_6531_cast_fp16, var_6448_cast_fp16))[name = tensor("op_6689_cast_fp16")]; + tensor var_6690_to_fp16 = const()[name = tensor("op_6690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_499_cast_fp16 = mul(x = var_6689_cast_fp16, y = var_6690_to_fp16)[name = tensor("aw_499_cast_fp16")]; + tensor var_6693_equation_0 = const()[name = tensor("op_6693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6693_cast_fp16 = einsum(equation = var_6693_equation_0, values = (var_6535_cast_fp16, var_6452_cast_fp16))[name = tensor("op_6693_cast_fp16")]; + tensor var_6694_to_fp16 = const()[name = tensor("op_6694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_501_cast_fp16 = mul(x = var_6693_cast_fp16, y = var_6694_to_fp16)[name = tensor("aw_501_cast_fp16")]; + tensor var_6697_equation_0 = const()[name = tensor("op_6697_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6697_cast_fp16 = einsum(equation = var_6697_equation_0, values = (var_6539_cast_fp16, var_6456_cast_fp16))[name = tensor("op_6697_cast_fp16")]; + tensor var_6698_to_fp16 = const()[name = tensor("op_6698_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_503_cast_fp16 = mul(x = var_6697_cast_fp16, y = var_6698_to_fp16)[name = tensor("aw_503_cast_fp16")]; + tensor var_6701_equation_0 = const()[name = tensor("op_6701_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6701_cast_fp16 = einsum(equation = var_6701_equation_0, values = (var_6543_cast_fp16, var_6460_cast_fp16))[name = tensor("op_6701_cast_fp16")]; + tensor var_6702_to_fp16 = const()[name = tensor("op_6702_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_505_cast_fp16 = mul(x = var_6701_cast_fp16, y = var_6702_to_fp16)[name = tensor("aw_505_cast_fp16")]; + tensor var_6705_equation_0 = const()[name = tensor("op_6705_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6705_cast_fp16 = einsum(equation = var_6705_equation_0, values = (var_6547_cast_fp16, var_6464_cast_fp16))[name = tensor("op_6705_cast_fp16")]; + tensor var_6706_to_fp16 = const()[name = tensor("op_6706_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_507_cast_fp16 = mul(x = var_6705_cast_fp16, y = var_6706_to_fp16)[name = tensor("aw_507_cast_fp16")]; + tensor var_6709_equation_0 = const()[name = tensor("op_6709_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6709_cast_fp16 = einsum(equation = var_6709_equation_0, values = (var_6551_cast_fp16, var_6468_cast_fp16))[name = tensor("op_6709_cast_fp16")]; + tensor var_6710_to_fp16 = const()[name = tensor("op_6710_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_509_cast_fp16 = mul(x = var_6709_cast_fp16, y = var_6710_to_fp16)[name = tensor("aw_509_cast_fp16")]; + tensor var_6713_equation_0 = const()[name = tensor("op_6713_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6713_cast_fp16 = einsum(equation = var_6713_equation_0, values = (var_6555_cast_fp16, var_6472_cast_fp16))[name = tensor("op_6713_cast_fp16")]; + tensor var_6714_to_fp16 = const()[name = tensor("op_6714_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_511_cast_fp16 = mul(x = var_6713_cast_fp16, y = var_6714_to_fp16)[name = tensor("aw_511_cast_fp16")]; + tensor var_6717_equation_0 = const()[name = tensor("op_6717_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6717_cast_fp16 = einsum(equation = var_6717_equation_0, values = (var_6559_cast_fp16, var_6476_cast_fp16))[name = tensor("op_6717_cast_fp16")]; + tensor var_6718_to_fp16 = const()[name = tensor("op_6718_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_513_cast_fp16 = mul(x = var_6717_cast_fp16, y = var_6718_to_fp16)[name = tensor("aw_513_cast_fp16")]; + tensor var_6721_equation_0 = const()[name = tensor("op_6721_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6721_cast_fp16 = einsum(equation = var_6721_equation_0, values = (var_6563_cast_fp16, var_6480_cast_fp16))[name = tensor("op_6721_cast_fp16")]; + tensor var_6722_to_fp16 = const()[name = tensor("op_6722_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_515_cast_fp16 = mul(x = var_6721_cast_fp16, y = var_6722_to_fp16)[name = tensor("aw_515_cast_fp16")]; + tensor var_6725_equation_0 = const()[name = tensor("op_6725_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6725_cast_fp16 = einsum(equation = var_6725_equation_0, values = (var_6567_cast_fp16, var_6484_cast_fp16))[name = tensor("op_6725_cast_fp16")]; + tensor var_6726_to_fp16 = const()[name = tensor("op_6726_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_517_cast_fp16 = mul(x = var_6725_cast_fp16, y = var_6726_to_fp16)[name = tensor("aw_517_cast_fp16")]; + tensor var_6729_equation_0 = const()[name = tensor("op_6729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6729_cast_fp16 = einsum(equation = var_6729_equation_0, values = (var_6571_cast_fp16, var_6488_cast_fp16))[name = tensor("op_6729_cast_fp16")]; + tensor var_6730_to_fp16 = const()[name = tensor("op_6730_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_519_cast_fp16 = mul(x = var_6729_cast_fp16, y = var_6730_to_fp16)[name = tensor("aw_519_cast_fp16")]; + tensor var_6732_cast_fp16 = softmax(axis = var_2624, x = aw_481_cast_fp16)[name = tensor("op_6732_cast_fp16")]; + tensor var_6733_cast_fp16 = softmax(axis = var_2624, x = aw_483_cast_fp16)[name = tensor("op_6733_cast_fp16")]; + tensor var_6734_cast_fp16 = softmax(axis = var_2624, x = aw_485_cast_fp16)[name = tensor("op_6734_cast_fp16")]; + tensor var_6735_cast_fp16 = softmax(axis = var_2624, x = aw_487_cast_fp16)[name = tensor("op_6735_cast_fp16")]; + tensor var_6736_cast_fp16 = softmax(axis = var_2624, x = aw_489_cast_fp16)[name = tensor("op_6736_cast_fp16")]; + tensor var_6737_cast_fp16 = softmax(axis = var_2624, x = aw_491_cast_fp16)[name = tensor("op_6737_cast_fp16")]; + tensor var_6738_cast_fp16 = softmax(axis = var_2624, x = aw_493_cast_fp16)[name = tensor("op_6738_cast_fp16")]; + tensor var_6739_cast_fp16 = softmax(axis = var_2624, x = aw_495_cast_fp16)[name = tensor("op_6739_cast_fp16")]; + tensor var_6740_cast_fp16 = softmax(axis = var_2624, x = aw_497_cast_fp16)[name = tensor("op_6740_cast_fp16")]; + tensor var_6741_cast_fp16 = softmax(axis = var_2624, x = aw_499_cast_fp16)[name = tensor("op_6741_cast_fp16")]; + tensor var_6742_cast_fp16 = softmax(axis = var_2624, x = aw_501_cast_fp16)[name = tensor("op_6742_cast_fp16")]; + tensor var_6743_cast_fp16 = softmax(axis = var_2624, x = aw_503_cast_fp16)[name = tensor("op_6743_cast_fp16")]; + tensor var_6744_cast_fp16 = softmax(axis = var_2624, x = aw_505_cast_fp16)[name = tensor("op_6744_cast_fp16")]; + tensor var_6745_cast_fp16 = softmax(axis = var_2624, x = aw_507_cast_fp16)[name = tensor("op_6745_cast_fp16")]; + tensor var_6746_cast_fp16 = softmax(axis = var_2624, x = aw_509_cast_fp16)[name = tensor("op_6746_cast_fp16")]; + tensor var_6747_cast_fp16 = softmax(axis = var_2624, x = aw_511_cast_fp16)[name = tensor("op_6747_cast_fp16")]; + tensor var_6748_cast_fp16 = softmax(axis = var_2624, x = aw_513_cast_fp16)[name = tensor("op_6748_cast_fp16")]; + tensor var_6749_cast_fp16 = softmax(axis = var_2624, x = aw_515_cast_fp16)[name = tensor("op_6749_cast_fp16")]; + tensor var_6750_cast_fp16 = softmax(axis = var_2624, x = aw_517_cast_fp16)[name = tensor("op_6750_cast_fp16")]; + tensor var_6751_cast_fp16 = softmax(axis = var_2624, x = aw_519_cast_fp16)[name = tensor("op_6751_cast_fp16")]; + tensor var_6753_equation_0 = const()[name = tensor("op_6753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6753_cast_fp16 = einsum(equation = var_6753_equation_0, values = (var_6573_cast_fp16, var_6732_cast_fp16))[name = tensor("op_6753_cast_fp16")]; + tensor var_6755_equation_0 = const()[name = tensor("op_6755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6755_cast_fp16 = einsum(equation = var_6755_equation_0, values = (var_6577_cast_fp16, var_6733_cast_fp16))[name = tensor("op_6755_cast_fp16")]; + tensor var_6757_equation_0 = const()[name = tensor("op_6757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6757_cast_fp16 = einsum(equation = var_6757_equation_0, values = (var_6581_cast_fp16, var_6734_cast_fp16))[name = tensor("op_6757_cast_fp16")]; + tensor var_6759_equation_0 = const()[name = tensor("op_6759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6759_cast_fp16 = einsum(equation = var_6759_equation_0, values = (var_6585_cast_fp16, var_6735_cast_fp16))[name = tensor("op_6759_cast_fp16")]; + tensor var_6761_equation_0 = const()[name = tensor("op_6761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6761_cast_fp16 = einsum(equation = var_6761_equation_0, values = (var_6589_cast_fp16, var_6736_cast_fp16))[name = tensor("op_6761_cast_fp16")]; + tensor var_6763_equation_0 = const()[name = tensor("op_6763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6763_cast_fp16 = einsum(equation = var_6763_equation_0, values = (var_6593_cast_fp16, var_6737_cast_fp16))[name = tensor("op_6763_cast_fp16")]; + tensor var_6765_equation_0 = const()[name = tensor("op_6765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6765_cast_fp16 = einsum(equation = var_6765_equation_0, values = (var_6597_cast_fp16, var_6738_cast_fp16))[name = tensor("op_6765_cast_fp16")]; + tensor var_6767_equation_0 = const()[name = tensor("op_6767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6767_cast_fp16 = einsum(equation = var_6767_equation_0, values = (var_6601_cast_fp16, var_6739_cast_fp16))[name = tensor("op_6767_cast_fp16")]; + tensor var_6769_equation_0 = const()[name = tensor("op_6769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6769_cast_fp16 = einsum(equation = var_6769_equation_0, values = (var_6605_cast_fp16, var_6740_cast_fp16))[name = tensor("op_6769_cast_fp16")]; + tensor var_6771_equation_0 = const()[name = tensor("op_6771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6771_cast_fp16 = einsum(equation = var_6771_equation_0, values = (var_6609_cast_fp16, var_6741_cast_fp16))[name = tensor("op_6771_cast_fp16")]; + tensor var_6773_equation_0 = const()[name = tensor("op_6773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6773_cast_fp16 = einsum(equation = var_6773_equation_0, values = (var_6613_cast_fp16, var_6742_cast_fp16))[name = tensor("op_6773_cast_fp16")]; + tensor var_6775_equation_0 = const()[name = tensor("op_6775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6775_cast_fp16 = einsum(equation = var_6775_equation_0, values = (var_6617_cast_fp16, var_6743_cast_fp16))[name = tensor("op_6775_cast_fp16")]; + tensor var_6777_equation_0 = const()[name = tensor("op_6777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6777_cast_fp16 = einsum(equation = var_6777_equation_0, values = (var_6621_cast_fp16, var_6744_cast_fp16))[name = tensor("op_6777_cast_fp16")]; + tensor var_6779_equation_0 = const()[name = tensor("op_6779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6779_cast_fp16 = einsum(equation = var_6779_equation_0, values = (var_6625_cast_fp16, var_6745_cast_fp16))[name = tensor("op_6779_cast_fp16")]; + tensor var_6781_equation_0 = const()[name = tensor("op_6781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6781_cast_fp16 = einsum(equation = var_6781_equation_0, values = (var_6629_cast_fp16, var_6746_cast_fp16))[name = tensor("op_6781_cast_fp16")]; + tensor var_6783_equation_0 = const()[name = tensor("op_6783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6783_cast_fp16 = einsum(equation = var_6783_equation_0, values = (var_6633_cast_fp16, var_6747_cast_fp16))[name = tensor("op_6783_cast_fp16")]; + tensor var_6785_equation_0 = const()[name = tensor("op_6785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6785_cast_fp16 = einsum(equation = var_6785_equation_0, values = (var_6637_cast_fp16, var_6748_cast_fp16))[name = tensor("op_6785_cast_fp16")]; + tensor var_6787_equation_0 = const()[name = tensor("op_6787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6787_cast_fp16 = einsum(equation = var_6787_equation_0, values = (var_6641_cast_fp16, var_6749_cast_fp16))[name = tensor("op_6787_cast_fp16")]; + tensor var_6789_equation_0 = const()[name = tensor("op_6789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6789_cast_fp16 = einsum(equation = var_6789_equation_0, values = (var_6645_cast_fp16, var_6750_cast_fp16))[name = tensor("op_6789_cast_fp16")]; + tensor var_6791_equation_0 = const()[name = tensor("op_6791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6791_cast_fp16 = einsum(equation = var_6791_equation_0, values = (var_6649_cast_fp16, var_6751_cast_fp16))[name = tensor("op_6791_cast_fp16")]; + tensor input_163_interleave_0 = const()[name = tensor("input_163_interleave_0"), val = tensor(false)]; + tensor input_163_cast_fp16 = concat(axis = var_2624, interleave = input_163_interleave_0, values = (var_6753_cast_fp16, var_6755_cast_fp16, var_6757_cast_fp16, var_6759_cast_fp16, var_6761_cast_fp16, var_6763_cast_fp16, var_6765_cast_fp16, var_6767_cast_fp16, var_6769_cast_fp16, var_6771_cast_fp16, var_6773_cast_fp16, var_6775_cast_fp16, var_6777_cast_fp16, var_6779_cast_fp16, var_6781_cast_fp16, var_6783_cast_fp16, var_6785_cast_fp16, var_6787_cast_fp16, var_6789_cast_fp16, var_6791_cast_fp16))[name = tensor("input_163_cast_fp16")]; + tensor var_6801_pad_type_0 = const()[name = tensor("op_6801_pad_type_0"), val = tensor("valid")]; + tensor var_6801_strides_0 = const()[name = tensor("op_6801_strides_0"), val = tensor([1, 1])]; + tensor var_6801_pad_0 = const()[name = tensor("op_6801_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6801_dilations_0 = const()[name = tensor("op_6801_dilations_0"), val = tensor([1, 1])]; + tensor var_6801_groups_0 = const()[name = tensor("op_6801_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(182521472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183750336))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183750528)))]; + tensor var_6801_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_6801_dilations_0, groups = var_6801_groups_0, pad = var_6801_pad_0, pad_type = var_6801_pad_type_0, strides = var_6801_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = tensor("op_6801_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = var_6801_cast_fp16, y = inputs_49_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; + tensor hidden_states_91_axes_0 = const()[name = tensor("hidden_states_91_axes_0"), val = tensor([1])]; + tensor hidden_states_91_gamma_0_to_fp16 = const()[name = tensor("hidden_states_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183753152)))]; + tensor hidden_states_91_beta_0_to_fp16 = const()[name = tensor("hidden_states_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183755776)))]; + tensor var_6811_to_fp16 = const()[name = tensor("op_6811_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_91_cast_fp16 = layer_norm(axes = hidden_states_91_axes_0, beta = hidden_states_91_beta_0_to_fp16, epsilon = var_6811_to_fp16, gamma = hidden_states_91_gamma_0_to_fp16, x = inputs_51_cast_fp16)[name = tensor("hidden_states_91_cast_fp16")]; + tensor q_35_pad_type_0 = const()[name = tensor("q_35_pad_type_0"), val = tensor("valid")]; + tensor q_35_strides_0 = const()[name = tensor("q_35_strides_0"), val = tensor([1, 1])]; + tensor q_35_pad_0 = const()[name = tensor("q_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_35_dilations_0 = const()[name = tensor("q_35_dilations_0"), val = tensor([1, 1])]; + tensor q_35_groups_0 = const()[name = tensor("q_35_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183758400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184987264))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_35_cast_fp16 = conv(dilations = q_35_dilations_0, groups = q_35_groups_0, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = q_35_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_91_cast_fp16)[name = tensor("q_35_cast_fp16")]; + tensor k_69_pad_type_0 = const()[name = tensor("k_69_pad_type_0"), val = tensor("valid")]; + tensor k_69_strides_0 = const()[name = tensor("k_69_strides_0"), val = tensor([1, 1])]; + tensor k_69_pad_0 = const()[name = tensor("k_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_69_dilations_0 = const()[name = tensor("k_69_dilations_0"), val = tensor([1, 1])]; + tensor k_69_groups_0 = const()[name = tensor("k_69_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184987456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186953600))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_69_cast_fp16 = conv(dilations = k_69_dilations_0, groups = k_69_groups_0, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = k_69_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_69_cast_fp16")]; + tensor v_35_pad_type_0 = const()[name = tensor("v_35_pad_type_0"), val = tensor("valid")]; + tensor v_35_strides_0 = const()[name = tensor("v_35_strides_0"), val = tensor([1, 1])]; + tensor v_35_pad_0 = const()[name = tensor("v_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_35_dilations_0 = const()[name = tensor("v_35_dilations_0"), val = tensor([1, 1])]; + tensor v_35_groups_0 = const()[name = tensor("v_35_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186953792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188919936))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_35_cast_fp16 = conv(dilations = v_35_dilations_0, groups = v_35_groups_0, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = v_35_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_35_cast_fp16")]; + tensor var_6844_begin_0 = const()[name = tensor("op_6844_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6844_end_0 = const()[name = tensor("op_6844_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_6844_end_mask_0 = const()[name = tensor("op_6844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6844_cast_fp16 = slice_by_index(begin = var_6844_begin_0, end = var_6844_end_0, end_mask = var_6844_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6844_cast_fp16")]; + tensor var_6848_begin_0 = const()[name = tensor("op_6848_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6848_end_0 = const()[name = tensor("op_6848_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_6848_end_mask_0 = const()[name = tensor("op_6848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6848_cast_fp16 = slice_by_index(begin = var_6848_begin_0, end = var_6848_end_0, end_mask = var_6848_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6848_cast_fp16")]; + tensor var_6852_begin_0 = const()[name = tensor("op_6852_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6852_end_0 = const()[name = tensor("op_6852_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_6852_end_mask_0 = const()[name = tensor("op_6852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6852_cast_fp16 = slice_by_index(begin = var_6852_begin_0, end = var_6852_end_0, end_mask = var_6852_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6852_cast_fp16")]; + tensor var_6856_begin_0 = const()[name = tensor("op_6856_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6856_end_0 = const()[name = tensor("op_6856_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_6856_end_mask_0 = const()[name = tensor("op_6856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6856_cast_fp16 = slice_by_index(begin = var_6856_begin_0, end = var_6856_end_0, end_mask = var_6856_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6856_cast_fp16")]; + tensor var_6860_begin_0 = const()[name = tensor("op_6860_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6860_end_0 = const()[name = tensor("op_6860_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_6860_end_mask_0 = const()[name = tensor("op_6860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6860_cast_fp16 = slice_by_index(begin = var_6860_begin_0, end = var_6860_end_0, end_mask = var_6860_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6860_cast_fp16")]; + tensor var_6864_begin_0 = const()[name = tensor("op_6864_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6864_end_0 = const()[name = tensor("op_6864_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_6864_end_mask_0 = const()[name = tensor("op_6864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6864_cast_fp16 = slice_by_index(begin = var_6864_begin_0, end = var_6864_end_0, end_mask = var_6864_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6864_cast_fp16")]; + tensor var_6868_begin_0 = const()[name = tensor("op_6868_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6868_end_0 = const()[name = tensor("op_6868_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_6868_end_mask_0 = const()[name = tensor("op_6868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6868_cast_fp16 = slice_by_index(begin = var_6868_begin_0, end = var_6868_end_0, end_mask = var_6868_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6868_cast_fp16")]; + tensor var_6872_begin_0 = const()[name = tensor("op_6872_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6872_end_0 = const()[name = tensor("op_6872_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_6872_end_mask_0 = const()[name = tensor("op_6872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6872_cast_fp16 = slice_by_index(begin = var_6872_begin_0, end = var_6872_end_0, end_mask = var_6872_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6872_cast_fp16")]; + tensor var_6876_begin_0 = const()[name = tensor("op_6876_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6876_end_0 = const()[name = tensor("op_6876_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_6876_end_mask_0 = const()[name = tensor("op_6876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6876_cast_fp16 = slice_by_index(begin = var_6876_begin_0, end = var_6876_end_0, end_mask = var_6876_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6876_cast_fp16")]; + tensor var_6880_begin_0 = const()[name = tensor("op_6880_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6880_end_0 = const()[name = tensor("op_6880_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_6880_end_mask_0 = const()[name = tensor("op_6880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6880_cast_fp16 = slice_by_index(begin = var_6880_begin_0, end = var_6880_end_0, end_mask = var_6880_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6880_cast_fp16")]; + tensor var_6884_begin_0 = const()[name = tensor("op_6884_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6884_end_0 = const()[name = tensor("op_6884_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_6884_end_mask_0 = const()[name = tensor("op_6884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6884_cast_fp16 = slice_by_index(begin = var_6884_begin_0, end = var_6884_end_0, end_mask = var_6884_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6884_cast_fp16")]; + tensor var_6888_begin_0 = const()[name = tensor("op_6888_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6888_end_0 = const()[name = tensor("op_6888_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_6888_end_mask_0 = const()[name = tensor("op_6888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6888_cast_fp16 = slice_by_index(begin = var_6888_begin_0, end = var_6888_end_0, end_mask = var_6888_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6888_cast_fp16")]; + tensor var_6892_begin_0 = const()[name = tensor("op_6892_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6892_end_0 = const()[name = tensor("op_6892_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_6892_end_mask_0 = const()[name = tensor("op_6892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6892_cast_fp16 = slice_by_index(begin = var_6892_begin_0, end = var_6892_end_0, end_mask = var_6892_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6892_cast_fp16")]; + tensor var_6896_begin_0 = const()[name = tensor("op_6896_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6896_end_0 = const()[name = tensor("op_6896_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_6896_end_mask_0 = const()[name = tensor("op_6896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6896_cast_fp16 = slice_by_index(begin = var_6896_begin_0, end = var_6896_end_0, end_mask = var_6896_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6896_cast_fp16")]; + tensor var_6900_begin_0 = const()[name = tensor("op_6900_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6900_end_0 = const()[name = tensor("op_6900_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_6900_end_mask_0 = const()[name = tensor("op_6900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6900_cast_fp16 = slice_by_index(begin = var_6900_begin_0, end = var_6900_end_0, end_mask = var_6900_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6900_cast_fp16")]; + tensor var_6904_begin_0 = const()[name = tensor("op_6904_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6904_end_0 = const()[name = tensor("op_6904_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6904_end_mask_0 = const()[name = tensor("op_6904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6904_cast_fp16 = slice_by_index(begin = var_6904_begin_0, end = var_6904_end_0, end_mask = var_6904_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6904_cast_fp16")]; + tensor var_6908_begin_0 = const()[name = tensor("op_6908_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6908_end_0 = const()[name = tensor("op_6908_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_6908_end_mask_0 = const()[name = tensor("op_6908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6908_cast_fp16 = slice_by_index(begin = var_6908_begin_0, end = var_6908_end_0, end_mask = var_6908_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6908_cast_fp16")]; + tensor var_6912_begin_0 = const()[name = tensor("op_6912_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6912_end_0 = const()[name = tensor("op_6912_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6912_end_mask_0 = const()[name = tensor("op_6912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6912_cast_fp16 = slice_by_index(begin = var_6912_begin_0, end = var_6912_end_0, end_mask = var_6912_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6912_cast_fp16")]; + tensor var_6916_begin_0 = const()[name = tensor("op_6916_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6916_end_0 = const()[name = tensor("op_6916_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6916_end_mask_0 = const()[name = tensor("op_6916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6916_cast_fp16 = slice_by_index(begin = var_6916_begin_0, end = var_6916_end_0, end_mask = var_6916_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6916_cast_fp16")]; + tensor var_6920_begin_0 = const()[name = tensor("op_6920_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6920_end_0 = const()[name = tensor("op_6920_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6920_end_mask_0 = const()[name = tensor("op_6920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6920_cast_fp16 = slice_by_index(begin = var_6920_begin_0, end = var_6920_end_0, end_mask = var_6920_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6920_cast_fp16")]; + tensor k_71_perm_0 = const()[name = tensor("k_71_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6927_begin_0 = const()[name = tensor("op_6927_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6927_end_0 = const()[name = tensor("op_6927_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_6927_end_mask_0 = const()[name = tensor("op_6927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_71_cast_fp16 = transpose(perm = k_71_perm_0, x = k_69_cast_fp16)[name = tensor("transpose_50")]; + tensor var_6927_cast_fp16 = slice_by_index(begin = var_6927_begin_0, end = var_6927_end_0, end_mask = var_6927_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6927_cast_fp16")]; + tensor var_6931_begin_0 = const()[name = tensor("op_6931_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6931_end_0 = const()[name = tensor("op_6931_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_6931_end_mask_0 = const()[name = tensor("op_6931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6931_cast_fp16 = slice_by_index(begin = var_6931_begin_0, end = var_6931_end_0, end_mask = var_6931_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6931_cast_fp16")]; + tensor var_6935_begin_0 = const()[name = tensor("op_6935_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6935_end_0 = const()[name = tensor("op_6935_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_6935_end_mask_0 = const()[name = tensor("op_6935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6935_cast_fp16 = slice_by_index(begin = var_6935_begin_0, end = var_6935_end_0, end_mask = var_6935_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6935_cast_fp16")]; + tensor var_6939_begin_0 = const()[name = tensor("op_6939_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6939_end_0 = const()[name = tensor("op_6939_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_6939_end_mask_0 = const()[name = tensor("op_6939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6939_cast_fp16 = slice_by_index(begin = var_6939_begin_0, end = var_6939_end_0, end_mask = var_6939_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6939_cast_fp16")]; + tensor var_6943_begin_0 = const()[name = tensor("op_6943_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6943_end_0 = const()[name = tensor("op_6943_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_6943_end_mask_0 = const()[name = tensor("op_6943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6943_cast_fp16 = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6943_cast_fp16")]; + tensor var_6947_begin_0 = const()[name = tensor("op_6947_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6947_end_0 = const()[name = tensor("op_6947_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_6947_end_mask_0 = const()[name = tensor("op_6947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6947_cast_fp16 = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6947_cast_fp16")]; + tensor var_6951_begin_0 = const()[name = tensor("op_6951_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6951_end_0 = const()[name = tensor("op_6951_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_6951_end_mask_0 = const()[name = tensor("op_6951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6951_cast_fp16 = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6951_cast_fp16")]; + tensor var_6955_begin_0 = const()[name = tensor("op_6955_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6955_end_0 = const()[name = tensor("op_6955_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_6955_end_mask_0 = const()[name = tensor("op_6955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6955_cast_fp16")]; + tensor var_6959_begin_0 = const()[name = tensor("op_6959_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6959_end_0 = const()[name = tensor("op_6959_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_6959_end_mask_0 = const()[name = tensor("op_6959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6959_cast_fp16")]; + tensor var_6963_begin_0 = const()[name = tensor("op_6963_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6963_end_0 = const()[name = tensor("op_6963_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_6963_end_mask_0 = const()[name = tensor("op_6963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6963_cast_fp16 = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6963_cast_fp16")]; + tensor var_6967_begin_0 = const()[name = tensor("op_6967_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6967_end_0 = const()[name = tensor("op_6967_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_6967_end_mask_0 = const()[name = tensor("op_6967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6967_cast_fp16 = slice_by_index(begin = var_6967_begin_0, end = var_6967_end_0, end_mask = var_6967_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6967_cast_fp16")]; + tensor var_6971_begin_0 = const()[name = tensor("op_6971_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6971_end_0 = const()[name = tensor("op_6971_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_6971_end_mask_0 = const()[name = tensor("op_6971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6971_cast_fp16 = slice_by_index(begin = var_6971_begin_0, end = var_6971_end_0, end_mask = var_6971_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6971_cast_fp16")]; + tensor var_6975_begin_0 = const()[name = tensor("op_6975_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6975_end_0 = const()[name = tensor("op_6975_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_6975_end_mask_0 = const()[name = tensor("op_6975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6975_cast_fp16 = slice_by_index(begin = var_6975_begin_0, end = var_6975_end_0, end_mask = var_6975_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6975_cast_fp16")]; + tensor var_6979_begin_0 = const()[name = tensor("op_6979_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6979_end_0 = const()[name = tensor("op_6979_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_6979_end_mask_0 = const()[name = tensor("op_6979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6979_cast_fp16 = slice_by_index(begin = var_6979_begin_0, end = var_6979_end_0, end_mask = var_6979_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6979_cast_fp16")]; + tensor var_6983_begin_0 = const()[name = tensor("op_6983_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6983_end_0 = const()[name = tensor("op_6983_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_6983_end_mask_0 = const()[name = tensor("op_6983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6983_cast_fp16 = slice_by_index(begin = var_6983_begin_0, end = var_6983_end_0, end_mask = var_6983_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6983_cast_fp16")]; + tensor var_6987_begin_0 = const()[name = tensor("op_6987_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6987_end_0 = const()[name = tensor("op_6987_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_6987_end_mask_0 = const()[name = tensor("op_6987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6987_cast_fp16 = slice_by_index(begin = var_6987_begin_0, end = var_6987_end_0, end_mask = var_6987_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6987_cast_fp16")]; + tensor var_6991_begin_0 = const()[name = tensor("op_6991_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6991_end_0 = const()[name = tensor("op_6991_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_6991_end_mask_0 = const()[name = tensor("op_6991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6991_cast_fp16 = slice_by_index(begin = var_6991_begin_0, end = var_6991_end_0, end_mask = var_6991_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6991_cast_fp16")]; + tensor var_6995_begin_0 = const()[name = tensor("op_6995_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6995_end_0 = const()[name = tensor("op_6995_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_6995_end_mask_0 = const()[name = tensor("op_6995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6995_cast_fp16 = slice_by_index(begin = var_6995_begin_0, end = var_6995_end_0, end_mask = var_6995_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6995_cast_fp16")]; + tensor var_6999_begin_0 = const()[name = tensor("op_6999_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6999_end_0 = const()[name = tensor("op_6999_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_6999_end_mask_0 = const()[name = tensor("op_6999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6999_cast_fp16 = slice_by_index(begin = var_6999_begin_0, end = var_6999_end_0, end_mask = var_6999_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6999_cast_fp16")]; + tensor var_7003_begin_0 = const()[name = tensor("op_7003_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7003_end_0 = const()[name = tensor("op_7003_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_7003_end_mask_0 = const()[name = tensor("op_7003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7003_cast_fp16 = slice_by_index(begin = var_7003_begin_0, end = var_7003_end_0, end_mask = var_7003_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7003_cast_fp16")]; + tensor var_7005_begin_0 = const()[name = tensor("op_7005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7005_end_0 = const()[name = tensor("op_7005_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_7005_end_mask_0 = const()[name = tensor("op_7005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7005_cast_fp16 = slice_by_index(begin = var_7005_begin_0, end = var_7005_end_0, end_mask = var_7005_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7005_cast_fp16")]; + tensor var_7009_begin_0 = const()[name = tensor("op_7009_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7009_end_0 = const()[name = tensor("op_7009_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_7009_end_mask_0 = const()[name = tensor("op_7009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7009_cast_fp16 = slice_by_index(begin = var_7009_begin_0, end = var_7009_end_0, end_mask = var_7009_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7009_cast_fp16")]; + tensor var_7013_begin_0 = const()[name = tensor("op_7013_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7013_end_0 = const()[name = tensor("op_7013_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_7013_end_mask_0 = const()[name = tensor("op_7013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7013_cast_fp16 = slice_by_index(begin = var_7013_begin_0, end = var_7013_end_0, end_mask = var_7013_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7013_cast_fp16")]; + tensor var_7017_begin_0 = const()[name = tensor("op_7017_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7017_end_0 = const()[name = tensor("op_7017_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_7017_end_mask_0 = const()[name = tensor("op_7017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7017_cast_fp16")]; + tensor var_7021_begin_0 = const()[name = tensor("op_7021_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7021_end_0 = const()[name = tensor("op_7021_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_7021_end_mask_0 = const()[name = tensor("op_7021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7021_cast_fp16 = slice_by_index(begin = var_7021_begin_0, end = var_7021_end_0, end_mask = var_7021_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7021_cast_fp16")]; + tensor var_7025_begin_0 = const()[name = tensor("op_7025_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7025_end_0 = const()[name = tensor("op_7025_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_7025_end_mask_0 = const()[name = tensor("op_7025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7025_cast_fp16 = slice_by_index(begin = var_7025_begin_0, end = var_7025_end_0, end_mask = var_7025_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7025_cast_fp16")]; + tensor var_7029_begin_0 = const()[name = tensor("op_7029_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7029_end_0 = const()[name = tensor("op_7029_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_7029_end_mask_0 = const()[name = tensor("op_7029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7029_cast_fp16 = slice_by_index(begin = var_7029_begin_0, end = var_7029_end_0, end_mask = var_7029_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7029_cast_fp16")]; + tensor var_7033_begin_0 = const()[name = tensor("op_7033_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7033_end_0 = const()[name = tensor("op_7033_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_7033_end_mask_0 = const()[name = tensor("op_7033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7033_cast_fp16 = slice_by_index(begin = var_7033_begin_0, end = var_7033_end_0, end_mask = var_7033_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7033_cast_fp16")]; + tensor var_7037_begin_0 = const()[name = tensor("op_7037_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7037_end_0 = const()[name = tensor("op_7037_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_7037_end_mask_0 = const()[name = tensor("op_7037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7037_cast_fp16 = slice_by_index(begin = var_7037_begin_0, end = var_7037_end_0, end_mask = var_7037_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7037_cast_fp16")]; + tensor var_7041_begin_0 = const()[name = tensor("op_7041_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7041_end_0 = const()[name = tensor("op_7041_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_7041_end_mask_0 = const()[name = tensor("op_7041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7041_cast_fp16 = slice_by_index(begin = var_7041_begin_0, end = var_7041_end_0, end_mask = var_7041_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7041_cast_fp16")]; + tensor var_7045_begin_0 = const()[name = tensor("op_7045_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7045_end_0 = const()[name = tensor("op_7045_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_7045_end_mask_0 = const()[name = tensor("op_7045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7045_cast_fp16")]; + tensor var_7049_begin_0 = const()[name = tensor("op_7049_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7049_end_0 = const()[name = tensor("op_7049_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_7049_end_mask_0 = const()[name = tensor("op_7049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7049_cast_fp16 = slice_by_index(begin = var_7049_begin_0, end = var_7049_end_0, end_mask = var_7049_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7049_cast_fp16")]; + tensor var_7053_begin_0 = const()[name = tensor("op_7053_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7053_end_0 = const()[name = tensor("op_7053_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_7053_end_mask_0 = const()[name = tensor("op_7053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7053_cast_fp16 = slice_by_index(begin = var_7053_begin_0, end = var_7053_end_0, end_mask = var_7053_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7053_cast_fp16")]; + tensor var_7057_begin_0 = const()[name = tensor("op_7057_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7057_end_0 = const()[name = tensor("op_7057_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_7057_end_mask_0 = const()[name = tensor("op_7057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7057_cast_fp16 = slice_by_index(begin = var_7057_begin_0, end = var_7057_end_0, end_mask = var_7057_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7057_cast_fp16")]; + tensor var_7061_begin_0 = const()[name = tensor("op_7061_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7061_end_0 = const()[name = tensor("op_7061_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_7061_end_mask_0 = const()[name = tensor("op_7061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7061_cast_fp16 = slice_by_index(begin = var_7061_begin_0, end = var_7061_end_0, end_mask = var_7061_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7061_cast_fp16")]; + tensor var_7065_begin_0 = const()[name = tensor("op_7065_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7065_end_0 = const()[name = tensor("op_7065_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_7065_end_mask_0 = const()[name = tensor("op_7065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7065_cast_fp16 = slice_by_index(begin = var_7065_begin_0, end = var_7065_end_0, end_mask = var_7065_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7065_cast_fp16")]; + tensor var_7069_begin_0 = const()[name = tensor("op_7069_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7069_end_0 = const()[name = tensor("op_7069_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_7069_end_mask_0 = const()[name = tensor("op_7069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7069_cast_fp16 = slice_by_index(begin = var_7069_begin_0, end = var_7069_end_0, end_mask = var_7069_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7069_cast_fp16")]; + tensor var_7073_begin_0 = const()[name = tensor("op_7073_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7073_end_0 = const()[name = tensor("op_7073_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_7073_end_mask_0 = const()[name = tensor("op_7073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7073_cast_fp16")]; + tensor var_7077_begin_0 = const()[name = tensor("op_7077_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7077_end_0 = const()[name = tensor("op_7077_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_7077_end_mask_0 = const()[name = tensor("op_7077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7077_cast_fp16 = slice_by_index(begin = var_7077_begin_0, end = var_7077_end_0, end_mask = var_7077_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7077_cast_fp16")]; + tensor var_7081_begin_0 = const()[name = tensor("op_7081_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7081_end_0 = const()[name = tensor("op_7081_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_7081_end_mask_0 = const()[name = tensor("op_7081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7081_cast_fp16 = slice_by_index(begin = var_7081_begin_0, end = var_7081_end_0, end_mask = var_7081_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7081_cast_fp16")]; + tensor var_7085_equation_0 = const()[name = tensor("op_7085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7085_cast_fp16 = einsum(equation = var_7085_equation_0, values = (var_6927_cast_fp16, var_6844_cast_fp16))[name = tensor("op_7085_cast_fp16")]; + tensor var_7086_to_fp16 = const()[name = tensor("op_7086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_521_cast_fp16 = mul(x = var_7085_cast_fp16, y = var_7086_to_fp16)[name = tensor("aw_521_cast_fp16")]; + tensor var_7089_equation_0 = const()[name = tensor("op_7089_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7089_cast_fp16 = einsum(equation = var_7089_equation_0, values = (var_6931_cast_fp16, var_6848_cast_fp16))[name = tensor("op_7089_cast_fp16")]; + tensor var_7090_to_fp16 = const()[name = tensor("op_7090_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_523_cast_fp16 = mul(x = var_7089_cast_fp16, y = var_7090_to_fp16)[name = tensor("aw_523_cast_fp16")]; + tensor var_7093_equation_0 = const()[name = tensor("op_7093_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7093_cast_fp16 = einsum(equation = var_7093_equation_0, values = (var_6935_cast_fp16, var_6852_cast_fp16))[name = tensor("op_7093_cast_fp16")]; + tensor var_7094_to_fp16 = const()[name = tensor("op_7094_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_525_cast_fp16 = mul(x = var_7093_cast_fp16, y = var_7094_to_fp16)[name = tensor("aw_525_cast_fp16")]; + tensor var_7097_equation_0 = const()[name = tensor("op_7097_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7097_cast_fp16 = einsum(equation = var_7097_equation_0, values = (var_6939_cast_fp16, var_6856_cast_fp16))[name = tensor("op_7097_cast_fp16")]; + tensor var_7098_to_fp16 = const()[name = tensor("op_7098_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_527_cast_fp16 = mul(x = var_7097_cast_fp16, y = var_7098_to_fp16)[name = tensor("aw_527_cast_fp16")]; + tensor var_7101_equation_0 = const()[name = tensor("op_7101_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7101_cast_fp16 = einsum(equation = var_7101_equation_0, values = (var_6943_cast_fp16, var_6860_cast_fp16))[name = tensor("op_7101_cast_fp16")]; + tensor var_7102_to_fp16 = const()[name = tensor("op_7102_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_529_cast_fp16 = mul(x = var_7101_cast_fp16, y = var_7102_to_fp16)[name = tensor("aw_529_cast_fp16")]; + tensor var_7105_equation_0 = const()[name = tensor("op_7105_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7105_cast_fp16 = einsum(equation = var_7105_equation_0, values = (var_6947_cast_fp16, var_6864_cast_fp16))[name = tensor("op_7105_cast_fp16")]; + tensor var_7106_to_fp16 = const()[name = tensor("op_7106_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_531_cast_fp16 = mul(x = var_7105_cast_fp16, y = var_7106_to_fp16)[name = tensor("aw_531_cast_fp16")]; + tensor var_7109_equation_0 = const()[name = tensor("op_7109_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7109_cast_fp16 = einsum(equation = var_7109_equation_0, values = (var_6951_cast_fp16, var_6868_cast_fp16))[name = tensor("op_7109_cast_fp16")]; + tensor var_7110_to_fp16 = const()[name = tensor("op_7110_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_533_cast_fp16 = mul(x = var_7109_cast_fp16, y = var_7110_to_fp16)[name = tensor("aw_533_cast_fp16")]; + tensor var_7113_equation_0 = const()[name = tensor("op_7113_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7113_cast_fp16 = einsum(equation = var_7113_equation_0, values = (var_6955_cast_fp16, var_6872_cast_fp16))[name = tensor("op_7113_cast_fp16")]; + tensor var_7114_to_fp16 = const()[name = tensor("op_7114_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_535_cast_fp16 = mul(x = var_7113_cast_fp16, y = var_7114_to_fp16)[name = tensor("aw_535_cast_fp16")]; + tensor var_7117_equation_0 = const()[name = tensor("op_7117_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7117_cast_fp16 = einsum(equation = var_7117_equation_0, values = (var_6959_cast_fp16, var_6876_cast_fp16))[name = tensor("op_7117_cast_fp16")]; + tensor var_7118_to_fp16 = const()[name = tensor("op_7118_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_537_cast_fp16 = mul(x = var_7117_cast_fp16, y = var_7118_to_fp16)[name = tensor("aw_537_cast_fp16")]; + tensor var_7121_equation_0 = const()[name = tensor("op_7121_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7121_cast_fp16 = einsum(equation = var_7121_equation_0, values = (var_6963_cast_fp16, var_6880_cast_fp16))[name = tensor("op_7121_cast_fp16")]; + tensor var_7122_to_fp16 = const()[name = tensor("op_7122_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_539_cast_fp16 = mul(x = var_7121_cast_fp16, y = var_7122_to_fp16)[name = tensor("aw_539_cast_fp16")]; + tensor var_7125_equation_0 = const()[name = tensor("op_7125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7125_cast_fp16 = einsum(equation = var_7125_equation_0, values = (var_6967_cast_fp16, var_6884_cast_fp16))[name = tensor("op_7125_cast_fp16")]; + tensor var_7126_to_fp16 = const()[name = tensor("op_7126_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_541_cast_fp16 = mul(x = var_7125_cast_fp16, y = var_7126_to_fp16)[name = tensor("aw_541_cast_fp16")]; + tensor var_7129_equation_0 = const()[name = tensor("op_7129_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7129_cast_fp16 = einsum(equation = var_7129_equation_0, values = (var_6971_cast_fp16, var_6888_cast_fp16))[name = tensor("op_7129_cast_fp16")]; + tensor var_7130_to_fp16 = const()[name = tensor("op_7130_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_543_cast_fp16 = mul(x = var_7129_cast_fp16, y = var_7130_to_fp16)[name = tensor("aw_543_cast_fp16")]; + tensor var_7133_equation_0 = const()[name = tensor("op_7133_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7133_cast_fp16 = einsum(equation = var_7133_equation_0, values = (var_6975_cast_fp16, var_6892_cast_fp16))[name = tensor("op_7133_cast_fp16")]; + tensor var_7134_to_fp16 = const()[name = tensor("op_7134_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_545_cast_fp16 = mul(x = var_7133_cast_fp16, y = var_7134_to_fp16)[name = tensor("aw_545_cast_fp16")]; + tensor var_7137_equation_0 = const()[name = tensor("op_7137_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7137_cast_fp16 = einsum(equation = var_7137_equation_0, values = (var_6979_cast_fp16, var_6896_cast_fp16))[name = tensor("op_7137_cast_fp16")]; + tensor var_7138_to_fp16 = const()[name = tensor("op_7138_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_547_cast_fp16 = mul(x = var_7137_cast_fp16, y = var_7138_to_fp16)[name = tensor("aw_547_cast_fp16")]; + tensor var_7141_equation_0 = const()[name = tensor("op_7141_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7141_cast_fp16 = einsum(equation = var_7141_equation_0, values = (var_6983_cast_fp16, var_6900_cast_fp16))[name = tensor("op_7141_cast_fp16")]; + tensor var_7142_to_fp16 = const()[name = tensor("op_7142_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_549_cast_fp16 = mul(x = var_7141_cast_fp16, y = var_7142_to_fp16)[name = tensor("aw_549_cast_fp16")]; + tensor var_7145_equation_0 = const()[name = tensor("op_7145_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7145_cast_fp16 = einsum(equation = var_7145_equation_0, values = (var_6987_cast_fp16, var_6904_cast_fp16))[name = tensor("op_7145_cast_fp16")]; + tensor var_7146_to_fp16 = const()[name = tensor("op_7146_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_551_cast_fp16 = mul(x = var_7145_cast_fp16, y = var_7146_to_fp16)[name = tensor("aw_551_cast_fp16")]; + tensor var_7149_equation_0 = const()[name = tensor("op_7149_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7149_cast_fp16 = einsum(equation = var_7149_equation_0, values = (var_6991_cast_fp16, var_6908_cast_fp16))[name = tensor("op_7149_cast_fp16")]; + tensor var_7150_to_fp16 = const()[name = tensor("op_7150_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_553_cast_fp16 = mul(x = var_7149_cast_fp16, y = var_7150_to_fp16)[name = tensor("aw_553_cast_fp16")]; + tensor var_7153_equation_0 = const()[name = tensor("op_7153_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7153_cast_fp16 = einsum(equation = var_7153_equation_0, values = (var_6995_cast_fp16, var_6912_cast_fp16))[name = tensor("op_7153_cast_fp16")]; + tensor var_7154_to_fp16 = const()[name = tensor("op_7154_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_555_cast_fp16 = mul(x = var_7153_cast_fp16, y = var_7154_to_fp16)[name = tensor("aw_555_cast_fp16")]; + tensor var_7157_equation_0 = const()[name = tensor("op_7157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7157_cast_fp16 = einsum(equation = var_7157_equation_0, values = (var_6999_cast_fp16, var_6916_cast_fp16))[name = tensor("op_7157_cast_fp16")]; + tensor var_7158_to_fp16 = const()[name = tensor("op_7158_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_557_cast_fp16 = mul(x = var_7157_cast_fp16, y = var_7158_to_fp16)[name = tensor("aw_557_cast_fp16")]; + tensor var_7161_equation_0 = const()[name = tensor("op_7161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7161_cast_fp16 = einsum(equation = var_7161_equation_0, values = (var_7003_cast_fp16, var_6920_cast_fp16))[name = tensor("op_7161_cast_fp16")]; + tensor var_7162_to_fp16 = const()[name = tensor("op_7162_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_559_cast_fp16 = mul(x = var_7161_cast_fp16, y = var_7162_to_fp16)[name = tensor("aw_559_cast_fp16")]; + tensor var_7164_cast_fp16 = softmax(axis = var_2624, x = aw_521_cast_fp16)[name = tensor("op_7164_cast_fp16")]; + tensor var_7165_cast_fp16 = softmax(axis = var_2624, x = aw_523_cast_fp16)[name = tensor("op_7165_cast_fp16")]; + tensor var_7166_cast_fp16 = softmax(axis = var_2624, x = aw_525_cast_fp16)[name = tensor("op_7166_cast_fp16")]; + tensor var_7167_cast_fp16 = softmax(axis = var_2624, x = aw_527_cast_fp16)[name = tensor("op_7167_cast_fp16")]; + tensor var_7168_cast_fp16 = softmax(axis = var_2624, x = aw_529_cast_fp16)[name = tensor("op_7168_cast_fp16")]; + tensor var_7169_cast_fp16 = softmax(axis = var_2624, x = aw_531_cast_fp16)[name = tensor("op_7169_cast_fp16")]; + tensor var_7170_cast_fp16 = softmax(axis = var_2624, x = aw_533_cast_fp16)[name = tensor("op_7170_cast_fp16")]; + tensor var_7171_cast_fp16 = softmax(axis = var_2624, x = aw_535_cast_fp16)[name = tensor("op_7171_cast_fp16")]; + tensor var_7172_cast_fp16 = softmax(axis = var_2624, x = aw_537_cast_fp16)[name = tensor("op_7172_cast_fp16")]; + tensor var_7173_cast_fp16 = softmax(axis = var_2624, x = aw_539_cast_fp16)[name = tensor("op_7173_cast_fp16")]; + tensor var_7174_cast_fp16 = softmax(axis = var_2624, x = aw_541_cast_fp16)[name = tensor("op_7174_cast_fp16")]; + tensor var_7175_cast_fp16 = softmax(axis = var_2624, x = aw_543_cast_fp16)[name = tensor("op_7175_cast_fp16")]; + tensor var_7176_cast_fp16 = softmax(axis = var_2624, x = aw_545_cast_fp16)[name = tensor("op_7176_cast_fp16")]; + tensor var_7177_cast_fp16 = softmax(axis = var_2624, x = aw_547_cast_fp16)[name = tensor("op_7177_cast_fp16")]; + tensor var_7178_cast_fp16 = softmax(axis = var_2624, x = aw_549_cast_fp16)[name = tensor("op_7178_cast_fp16")]; + tensor var_7179_cast_fp16 = softmax(axis = var_2624, x = aw_551_cast_fp16)[name = tensor("op_7179_cast_fp16")]; + tensor var_7180_cast_fp16 = softmax(axis = var_2624, x = aw_553_cast_fp16)[name = tensor("op_7180_cast_fp16")]; + tensor var_7181_cast_fp16 = softmax(axis = var_2624, x = aw_555_cast_fp16)[name = tensor("op_7181_cast_fp16")]; + tensor var_7182_cast_fp16 = softmax(axis = var_2624, x = aw_557_cast_fp16)[name = tensor("op_7182_cast_fp16")]; + tensor var_7183_cast_fp16 = softmax(axis = var_2624, x = aw_559_cast_fp16)[name = tensor("op_7183_cast_fp16")]; + tensor var_7185_equation_0 = const()[name = tensor("op_7185_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7185_cast_fp16 = einsum(equation = var_7185_equation_0, values = (var_7005_cast_fp16, var_7164_cast_fp16))[name = tensor("op_7185_cast_fp16")]; + tensor var_7187_equation_0 = const()[name = tensor("op_7187_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7187_cast_fp16 = einsum(equation = var_7187_equation_0, values = (var_7009_cast_fp16, var_7165_cast_fp16))[name = tensor("op_7187_cast_fp16")]; + tensor var_7189_equation_0 = const()[name = tensor("op_7189_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7189_cast_fp16 = einsum(equation = var_7189_equation_0, values = (var_7013_cast_fp16, var_7166_cast_fp16))[name = tensor("op_7189_cast_fp16")]; + tensor var_7191_equation_0 = const()[name = tensor("op_7191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7191_cast_fp16 = einsum(equation = var_7191_equation_0, values = (var_7017_cast_fp16, var_7167_cast_fp16))[name = tensor("op_7191_cast_fp16")]; + tensor var_7193_equation_0 = const()[name = tensor("op_7193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7193_cast_fp16 = einsum(equation = var_7193_equation_0, values = (var_7021_cast_fp16, var_7168_cast_fp16))[name = tensor("op_7193_cast_fp16")]; + tensor var_7195_equation_0 = const()[name = tensor("op_7195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7195_cast_fp16 = einsum(equation = var_7195_equation_0, values = (var_7025_cast_fp16, var_7169_cast_fp16))[name = tensor("op_7195_cast_fp16")]; + tensor var_7197_equation_0 = const()[name = tensor("op_7197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7197_cast_fp16 = einsum(equation = var_7197_equation_0, values = (var_7029_cast_fp16, var_7170_cast_fp16))[name = tensor("op_7197_cast_fp16")]; + tensor var_7199_equation_0 = const()[name = tensor("op_7199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7199_cast_fp16 = einsum(equation = var_7199_equation_0, values = (var_7033_cast_fp16, var_7171_cast_fp16))[name = tensor("op_7199_cast_fp16")]; + tensor var_7201_equation_0 = const()[name = tensor("op_7201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7201_cast_fp16 = einsum(equation = var_7201_equation_0, values = (var_7037_cast_fp16, var_7172_cast_fp16))[name = tensor("op_7201_cast_fp16")]; + tensor var_7203_equation_0 = const()[name = tensor("op_7203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7203_cast_fp16 = einsum(equation = var_7203_equation_0, values = (var_7041_cast_fp16, var_7173_cast_fp16))[name = tensor("op_7203_cast_fp16")]; + tensor var_7205_equation_0 = const()[name = tensor("op_7205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7205_cast_fp16 = einsum(equation = var_7205_equation_0, values = (var_7045_cast_fp16, var_7174_cast_fp16))[name = tensor("op_7205_cast_fp16")]; + tensor var_7207_equation_0 = const()[name = tensor("op_7207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7207_cast_fp16 = einsum(equation = var_7207_equation_0, values = (var_7049_cast_fp16, var_7175_cast_fp16))[name = tensor("op_7207_cast_fp16")]; + tensor var_7209_equation_0 = const()[name = tensor("op_7209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7209_cast_fp16 = einsum(equation = var_7209_equation_0, values = (var_7053_cast_fp16, var_7176_cast_fp16))[name = tensor("op_7209_cast_fp16")]; + tensor var_7211_equation_0 = const()[name = tensor("op_7211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7211_cast_fp16 = einsum(equation = var_7211_equation_0, values = (var_7057_cast_fp16, var_7177_cast_fp16))[name = tensor("op_7211_cast_fp16")]; + tensor var_7213_equation_0 = const()[name = tensor("op_7213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7213_cast_fp16 = einsum(equation = var_7213_equation_0, values = (var_7061_cast_fp16, var_7178_cast_fp16))[name = tensor("op_7213_cast_fp16")]; + tensor var_7215_equation_0 = const()[name = tensor("op_7215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7215_cast_fp16 = einsum(equation = var_7215_equation_0, values = (var_7065_cast_fp16, var_7179_cast_fp16))[name = tensor("op_7215_cast_fp16")]; + tensor var_7217_equation_0 = const()[name = tensor("op_7217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7217_cast_fp16 = einsum(equation = var_7217_equation_0, values = (var_7069_cast_fp16, var_7180_cast_fp16))[name = tensor("op_7217_cast_fp16")]; + tensor var_7219_equation_0 = const()[name = tensor("op_7219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7219_cast_fp16 = einsum(equation = var_7219_equation_0, values = (var_7073_cast_fp16, var_7181_cast_fp16))[name = tensor("op_7219_cast_fp16")]; + tensor var_7221_equation_0 = const()[name = tensor("op_7221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7221_cast_fp16 = einsum(equation = var_7221_equation_0, values = (var_7077_cast_fp16, var_7182_cast_fp16))[name = tensor("op_7221_cast_fp16")]; + tensor var_7223_equation_0 = const()[name = tensor("op_7223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7223_cast_fp16 = einsum(equation = var_7223_equation_0, values = (var_7081_cast_fp16, var_7183_cast_fp16))[name = tensor("op_7223_cast_fp16")]; + tensor input_165_interleave_0 = const()[name = tensor("input_165_interleave_0"), val = tensor(false)]; + tensor input_165_cast_fp16 = concat(axis = var_2624, interleave = input_165_interleave_0, values = (var_7185_cast_fp16, var_7187_cast_fp16, var_7189_cast_fp16, var_7191_cast_fp16, var_7193_cast_fp16, var_7195_cast_fp16, var_7197_cast_fp16, var_7199_cast_fp16, var_7201_cast_fp16, var_7203_cast_fp16, var_7205_cast_fp16, var_7207_cast_fp16, var_7209_cast_fp16, var_7211_cast_fp16, var_7213_cast_fp16, var_7215_cast_fp16, var_7217_cast_fp16, var_7219_cast_fp16, var_7221_cast_fp16, var_7223_cast_fp16))[name = tensor("input_165_cast_fp16")]; + tensor var_7233_pad_type_0 = const()[name = tensor("op_7233_pad_type_0"), val = tensor("valid")]; + tensor var_7233_strides_0 = const()[name = tensor("op_7233_strides_0"), val = tensor([1, 1])]; + tensor var_7233_pad_0 = const()[name = tensor("op_7233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7233_dilations_0 = const()[name = tensor("op_7233_dilations_0"), val = tensor([1, 1])]; + tensor var_7233_groups_0 = const()[name = tensor("op_7233_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188920128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190148992))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190149184)))]; + tensor var_7233_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_7233_dilations_0, groups = var_7233_groups_0, pad = var_7233_pad_0, pad_type = var_7233_pad_type_0, strides = var_7233_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor("op_7233_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = var_7233_cast_fp16, y = inputs_51_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; + tensor input_167_axes_0 = const()[name = tensor("input_167_axes_0"), val = tensor([1])]; + tensor input_167_gamma_0_to_fp16 = const()[name = tensor("input_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190151808)))]; + tensor input_167_beta_0_to_fp16 = const()[name = tensor("input_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190154432)))]; + tensor var_7243_to_fp16 = const()[name = tensor("op_7243_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_167_cast_fp16 = layer_norm(axes = input_167_axes_0, beta = input_167_beta_0_to_fp16, epsilon = var_7243_to_fp16, gamma = input_167_gamma_0_to_fp16, x = inputs_53_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor var_7263_pad_type_0 = const()[name = tensor("op_7263_pad_type_0"), val = tensor("valid")]; + tensor var_7263_strides_0 = const()[name = tensor("op_7263_strides_0"), val = tensor([1, 1])]; + tensor var_7263_pad_0 = const()[name = tensor("op_7263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7263_dilations_0 = const()[name = tensor("op_7263_dilations_0"), val = tensor([1, 1])]; + tensor var_7263_groups_0 = const()[name = tensor("op_7263_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190157056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199987520))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199987712)))]; + tensor var_7263_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_7263_dilations_0, groups = var_7263_groups_0, pad = var_7263_pad_0, pad_type = var_7263_pad_type_0, strides = var_7263_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("op_7263_cast_fp16")]; + tensor var_7264_split_sizes_0 = const()[name = tensor("op_7264_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_7264_axis_0 = const()[name = tensor("op_7264_axis_0"), val = tensor(1)]; + tensor var_7264_cast_fp16_0, tensor var_7264_cast_fp16_1 = split(axis = var_7264_axis_0, split_sizes = var_7264_split_sizes_0, x = var_7263_cast_fp16)[name = tensor("op_7264_cast_fp16")]; + tensor var_7266_mode_0 = const()[name = tensor("op_7266_mode_0"), val = tensor("EXACT")]; + tensor var_7266_cast_fp16 = gelu(mode = var_7266_mode_0, x = var_7264_cast_fp16_1)[name = tensor("op_7266_cast_fp16")]; + tensor input_169_cast_fp16 = mul(x = var_7264_cast_fp16_0, y = var_7266_cast_fp16)[name = tensor("input_169_cast_fp16")]; + tensor var_7274_pad_type_0 = const()[name = tensor("op_7274_pad_type_0"), val = tensor("valid")]; + tensor var_7274_strides_0 = const()[name = tensor("op_7274_strides_0"), val = tensor([1, 1])]; + tensor var_7274_pad_0 = const()[name = tensor("op_7274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7274_dilations_0 = const()[name = tensor("op_7274_dilations_0"), val = tensor([1, 1])]; + tensor var_7274_groups_0 = const()[name = tensor("op_7274_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200008256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204923520))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204923712)))]; + tensor var_7274_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_7274_dilations_0, groups = var_7274_groups_0, pad = var_7274_pad_0, pad_type = var_7274_pad_type_0, strides = var_7274_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor("op_7274_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = var_7274_cast_fp16, y = inputs_53_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; + tensor hidden_states_95_axes_0 = const()[name = tensor("hidden_states_95_axes_0"), val = tensor([1])]; + tensor hidden_states_95_gamma_0_to_fp16 = const()[name = tensor("hidden_states_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204926336)))]; + tensor hidden_states_95_beta_0_to_fp16 = const()[name = tensor("hidden_states_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204928960)))]; + tensor var_7290_to_fp16 = const()[name = tensor("op_7290_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_95_cast_fp16 = layer_norm(axes = hidden_states_95_axes_0, beta = hidden_states_95_beta_0_to_fp16, epsilon = var_7290_to_fp16, gamma = hidden_states_95_gamma_0_to_fp16, x = inputs_55_cast_fp16)[name = tensor("hidden_states_95_cast_fp16")]; + tensor q_37_pad_type_0 = const()[name = tensor("q_37_pad_type_0"), val = tensor("valid")]; + tensor q_37_strides_0 = const()[name = tensor("q_37_strides_0"), val = tensor([1, 1])]; + tensor q_37_pad_0 = const()[name = tensor("q_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_37_dilations_0 = const()[name = tensor("q_37_dilations_0"), val = tensor([1, 1])]; + tensor q_37_groups_0 = const()[name = tensor("q_37_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204931584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206160448))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_37_cast_fp16 = conv(dilations = q_37_dilations_0, groups = q_37_groups_0, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = q_37_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_95_cast_fp16)[name = tensor("q_37_cast_fp16")]; + tensor k_73_pad_type_0 = const()[name = tensor("k_73_pad_type_0"), val = tensor("valid")]; + tensor k_73_strides_0 = const()[name = tensor("k_73_strides_0"), val = tensor([1, 1])]; + tensor k_73_pad_0 = const()[name = tensor("k_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_73_dilations_0 = const()[name = tensor("k_73_dilations_0"), val = tensor([1, 1])]; + tensor k_73_groups_0 = const()[name = tensor("k_73_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206160640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207389504))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_73_cast_fp16 = conv(dilations = k_73_dilations_0, groups = k_73_groups_0, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = k_73_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_95_cast_fp16)[name = tensor("k_73_cast_fp16")]; + tensor v_37_pad_type_0 = const()[name = tensor("v_37_pad_type_0"), val = tensor("valid")]; + tensor v_37_strides_0 = const()[name = tensor("v_37_strides_0"), val = tensor([1, 1])]; + tensor v_37_pad_0 = const()[name = tensor("v_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_37_dilations_0 = const()[name = tensor("v_37_dilations_0"), val = tensor([1, 1])]; + tensor v_37_groups_0 = const()[name = tensor("v_37_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207389696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208618560))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_37_cast_fp16 = conv(dilations = v_37_dilations_0, groups = v_37_groups_0, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = v_37_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_95_cast_fp16)[name = tensor("v_37_cast_fp16")]; + tensor var_7323_begin_0 = const()[name = tensor("op_7323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7323_end_0 = const()[name = tensor("op_7323_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_7323_end_mask_0 = const()[name = tensor("op_7323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7323_cast_fp16 = slice_by_index(begin = var_7323_begin_0, end = var_7323_end_0, end_mask = var_7323_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7323_cast_fp16")]; + tensor var_7327_begin_0 = const()[name = tensor("op_7327_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7327_end_0 = const()[name = tensor("op_7327_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_7327_end_mask_0 = const()[name = tensor("op_7327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7327_cast_fp16 = slice_by_index(begin = var_7327_begin_0, end = var_7327_end_0, end_mask = var_7327_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7327_cast_fp16")]; + tensor var_7331_begin_0 = const()[name = tensor("op_7331_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7331_end_0 = const()[name = tensor("op_7331_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_7331_end_mask_0 = const()[name = tensor("op_7331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7331_cast_fp16 = slice_by_index(begin = var_7331_begin_0, end = var_7331_end_0, end_mask = var_7331_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7331_cast_fp16")]; + tensor var_7335_begin_0 = const()[name = tensor("op_7335_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7335_end_0 = const()[name = tensor("op_7335_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_7335_end_mask_0 = const()[name = tensor("op_7335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7335_cast_fp16 = slice_by_index(begin = var_7335_begin_0, end = var_7335_end_0, end_mask = var_7335_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7335_cast_fp16")]; + tensor var_7339_begin_0 = const()[name = tensor("op_7339_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7339_end_0 = const()[name = tensor("op_7339_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_7339_end_mask_0 = const()[name = tensor("op_7339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7339_cast_fp16 = slice_by_index(begin = var_7339_begin_0, end = var_7339_end_0, end_mask = var_7339_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7339_cast_fp16")]; + tensor var_7343_begin_0 = const()[name = tensor("op_7343_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7343_end_0 = const()[name = tensor("op_7343_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_7343_end_mask_0 = const()[name = tensor("op_7343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7343_cast_fp16 = slice_by_index(begin = var_7343_begin_0, end = var_7343_end_0, end_mask = var_7343_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7343_cast_fp16")]; + tensor var_7347_begin_0 = const()[name = tensor("op_7347_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7347_end_0 = const()[name = tensor("op_7347_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_7347_end_mask_0 = const()[name = tensor("op_7347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7347_cast_fp16 = slice_by_index(begin = var_7347_begin_0, end = var_7347_end_0, end_mask = var_7347_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7347_cast_fp16")]; + tensor var_7351_begin_0 = const()[name = tensor("op_7351_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7351_end_0 = const()[name = tensor("op_7351_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_7351_end_mask_0 = const()[name = tensor("op_7351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7351_cast_fp16 = slice_by_index(begin = var_7351_begin_0, end = var_7351_end_0, end_mask = var_7351_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7351_cast_fp16")]; + tensor var_7355_begin_0 = const()[name = tensor("op_7355_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7355_end_0 = const()[name = tensor("op_7355_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_7355_end_mask_0 = const()[name = tensor("op_7355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7355_cast_fp16 = slice_by_index(begin = var_7355_begin_0, end = var_7355_end_0, end_mask = var_7355_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7355_cast_fp16")]; + tensor var_7359_begin_0 = const()[name = tensor("op_7359_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7359_end_0 = const()[name = tensor("op_7359_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_7359_end_mask_0 = const()[name = tensor("op_7359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7359_cast_fp16 = slice_by_index(begin = var_7359_begin_0, end = var_7359_end_0, end_mask = var_7359_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7359_cast_fp16")]; + tensor var_7363_begin_0 = const()[name = tensor("op_7363_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7363_end_0 = const()[name = tensor("op_7363_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_7363_end_mask_0 = const()[name = tensor("op_7363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7363_cast_fp16 = slice_by_index(begin = var_7363_begin_0, end = var_7363_end_0, end_mask = var_7363_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7363_cast_fp16")]; + tensor var_7367_begin_0 = const()[name = tensor("op_7367_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7367_end_0 = const()[name = tensor("op_7367_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_7367_end_mask_0 = const()[name = tensor("op_7367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7367_cast_fp16 = slice_by_index(begin = var_7367_begin_0, end = var_7367_end_0, end_mask = var_7367_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7367_cast_fp16")]; + tensor var_7371_begin_0 = const()[name = tensor("op_7371_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7371_end_0 = const()[name = tensor("op_7371_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_7371_end_mask_0 = const()[name = tensor("op_7371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7371_cast_fp16 = slice_by_index(begin = var_7371_begin_0, end = var_7371_end_0, end_mask = var_7371_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7371_cast_fp16")]; + tensor var_7375_begin_0 = const()[name = tensor("op_7375_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7375_end_0 = const()[name = tensor("op_7375_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_7375_end_mask_0 = const()[name = tensor("op_7375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7375_cast_fp16 = slice_by_index(begin = var_7375_begin_0, end = var_7375_end_0, end_mask = var_7375_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7375_cast_fp16")]; + tensor var_7379_begin_0 = const()[name = tensor("op_7379_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7379_end_0 = const()[name = tensor("op_7379_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_7379_end_mask_0 = const()[name = tensor("op_7379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7379_cast_fp16 = slice_by_index(begin = var_7379_begin_0, end = var_7379_end_0, end_mask = var_7379_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7379_cast_fp16")]; + tensor var_7383_begin_0 = const()[name = tensor("op_7383_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7383_end_0 = const()[name = tensor("op_7383_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7383_end_mask_0 = const()[name = tensor("op_7383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7383_cast_fp16 = slice_by_index(begin = var_7383_begin_0, end = var_7383_end_0, end_mask = var_7383_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7383_cast_fp16")]; + tensor var_7387_begin_0 = const()[name = tensor("op_7387_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7387_end_0 = const()[name = tensor("op_7387_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_7387_end_mask_0 = const()[name = tensor("op_7387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7387_cast_fp16 = slice_by_index(begin = var_7387_begin_0, end = var_7387_end_0, end_mask = var_7387_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7387_cast_fp16")]; + tensor var_7391_begin_0 = const()[name = tensor("op_7391_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7391_end_0 = const()[name = tensor("op_7391_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_7391_end_mask_0 = const()[name = tensor("op_7391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7391_cast_fp16 = slice_by_index(begin = var_7391_begin_0, end = var_7391_end_0, end_mask = var_7391_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7391_cast_fp16")]; + tensor var_7395_begin_0 = const()[name = tensor("op_7395_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7395_end_0 = const()[name = tensor("op_7395_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_7395_end_mask_0 = const()[name = tensor("op_7395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7395_cast_fp16 = slice_by_index(begin = var_7395_begin_0, end = var_7395_end_0, end_mask = var_7395_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7395_cast_fp16")]; + tensor var_7399_begin_0 = const()[name = tensor("op_7399_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7399_end_0 = const()[name = tensor("op_7399_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_7399_end_mask_0 = const()[name = tensor("op_7399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7399_cast_fp16 = slice_by_index(begin = var_7399_begin_0, end = var_7399_end_0, end_mask = var_7399_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7399_cast_fp16")]; + tensor k_75_perm_0 = const()[name = tensor("k_75_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7406_begin_0 = const()[name = tensor("op_7406_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7406_end_0 = const()[name = tensor("op_7406_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_7406_end_mask_0 = const()[name = tensor("op_7406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_75_cast_fp16 = transpose(perm = k_75_perm_0, x = k_73_cast_fp16)[name = tensor("transpose_49")]; + tensor var_7406_cast_fp16 = slice_by_index(begin = var_7406_begin_0, end = var_7406_end_0, end_mask = var_7406_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7406_cast_fp16")]; + tensor var_7410_begin_0 = const()[name = tensor("op_7410_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7410_end_0 = const()[name = tensor("op_7410_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_7410_end_mask_0 = const()[name = tensor("op_7410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7410_cast_fp16 = slice_by_index(begin = var_7410_begin_0, end = var_7410_end_0, end_mask = var_7410_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7410_cast_fp16")]; + tensor var_7414_begin_0 = const()[name = tensor("op_7414_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7414_end_0 = const()[name = tensor("op_7414_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_7414_end_mask_0 = const()[name = tensor("op_7414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7414_cast_fp16 = slice_by_index(begin = var_7414_begin_0, end = var_7414_end_0, end_mask = var_7414_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7414_cast_fp16")]; + tensor var_7418_begin_0 = const()[name = tensor("op_7418_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7418_end_0 = const()[name = tensor("op_7418_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_7418_end_mask_0 = const()[name = tensor("op_7418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7418_cast_fp16 = slice_by_index(begin = var_7418_begin_0, end = var_7418_end_0, end_mask = var_7418_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7418_cast_fp16")]; + tensor var_7422_begin_0 = const()[name = tensor("op_7422_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7422_end_0 = const()[name = tensor("op_7422_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_7422_end_mask_0 = const()[name = tensor("op_7422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7422_cast_fp16 = slice_by_index(begin = var_7422_begin_0, end = var_7422_end_0, end_mask = var_7422_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7422_cast_fp16")]; + tensor var_7426_begin_0 = const()[name = tensor("op_7426_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7426_end_0 = const()[name = tensor("op_7426_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_7426_end_mask_0 = const()[name = tensor("op_7426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7426_cast_fp16 = slice_by_index(begin = var_7426_begin_0, end = var_7426_end_0, end_mask = var_7426_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7426_cast_fp16")]; + tensor var_7430_begin_0 = const()[name = tensor("op_7430_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7430_end_0 = const()[name = tensor("op_7430_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_7430_end_mask_0 = const()[name = tensor("op_7430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7430_cast_fp16 = slice_by_index(begin = var_7430_begin_0, end = var_7430_end_0, end_mask = var_7430_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7430_cast_fp16")]; + tensor var_7434_begin_0 = const()[name = tensor("op_7434_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7434_end_0 = const()[name = tensor("op_7434_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_7434_end_mask_0 = const()[name = tensor("op_7434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7434_cast_fp16 = slice_by_index(begin = var_7434_begin_0, end = var_7434_end_0, end_mask = var_7434_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7434_cast_fp16")]; + tensor var_7438_begin_0 = const()[name = tensor("op_7438_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7438_end_0 = const()[name = tensor("op_7438_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_7438_end_mask_0 = const()[name = tensor("op_7438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7438_cast_fp16 = slice_by_index(begin = var_7438_begin_0, end = var_7438_end_0, end_mask = var_7438_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7438_cast_fp16")]; + tensor var_7442_begin_0 = const()[name = tensor("op_7442_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7442_end_0 = const()[name = tensor("op_7442_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_7442_end_mask_0 = const()[name = tensor("op_7442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7442_cast_fp16 = slice_by_index(begin = var_7442_begin_0, end = var_7442_end_0, end_mask = var_7442_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7442_cast_fp16")]; + tensor var_7446_begin_0 = const()[name = tensor("op_7446_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7446_end_0 = const()[name = tensor("op_7446_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_7446_end_mask_0 = const()[name = tensor("op_7446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7446_cast_fp16 = slice_by_index(begin = var_7446_begin_0, end = var_7446_end_0, end_mask = var_7446_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7446_cast_fp16")]; + tensor var_7450_begin_0 = const()[name = tensor("op_7450_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7450_end_0 = const()[name = tensor("op_7450_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_7450_end_mask_0 = const()[name = tensor("op_7450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7450_cast_fp16 = slice_by_index(begin = var_7450_begin_0, end = var_7450_end_0, end_mask = var_7450_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7450_cast_fp16")]; + tensor var_7454_begin_0 = const()[name = tensor("op_7454_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7454_end_0 = const()[name = tensor("op_7454_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_7454_end_mask_0 = const()[name = tensor("op_7454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7454_cast_fp16 = slice_by_index(begin = var_7454_begin_0, end = var_7454_end_0, end_mask = var_7454_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7454_cast_fp16")]; + tensor var_7458_begin_0 = const()[name = tensor("op_7458_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7458_end_0 = const()[name = tensor("op_7458_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_7458_end_mask_0 = const()[name = tensor("op_7458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7458_cast_fp16 = slice_by_index(begin = var_7458_begin_0, end = var_7458_end_0, end_mask = var_7458_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7458_cast_fp16")]; + tensor var_7462_begin_0 = const()[name = tensor("op_7462_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7462_end_0 = const()[name = tensor("op_7462_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_7462_end_mask_0 = const()[name = tensor("op_7462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7462_cast_fp16 = slice_by_index(begin = var_7462_begin_0, end = var_7462_end_0, end_mask = var_7462_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7462_cast_fp16")]; + tensor var_7466_begin_0 = const()[name = tensor("op_7466_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7466_end_0 = const()[name = tensor("op_7466_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7466_end_mask_0 = const()[name = tensor("op_7466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7466_cast_fp16 = slice_by_index(begin = var_7466_begin_0, end = var_7466_end_0, end_mask = var_7466_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7466_cast_fp16")]; + tensor var_7470_begin_0 = const()[name = tensor("op_7470_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7470_end_0 = const()[name = tensor("op_7470_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_7470_end_mask_0 = const()[name = tensor("op_7470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7470_cast_fp16 = slice_by_index(begin = var_7470_begin_0, end = var_7470_end_0, end_mask = var_7470_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7470_cast_fp16")]; + tensor var_7474_begin_0 = const()[name = tensor("op_7474_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7474_end_0 = const()[name = tensor("op_7474_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_7474_end_mask_0 = const()[name = tensor("op_7474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7474_cast_fp16 = slice_by_index(begin = var_7474_begin_0, end = var_7474_end_0, end_mask = var_7474_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7474_cast_fp16")]; + tensor var_7478_begin_0 = const()[name = tensor("op_7478_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7478_end_0 = const()[name = tensor("op_7478_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_7478_end_mask_0 = const()[name = tensor("op_7478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7478_cast_fp16 = slice_by_index(begin = var_7478_begin_0, end = var_7478_end_0, end_mask = var_7478_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7478_cast_fp16")]; + tensor var_7482_begin_0 = const()[name = tensor("op_7482_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7482_end_0 = const()[name = tensor("op_7482_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_7482_end_mask_0 = const()[name = tensor("op_7482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7482_cast_fp16 = slice_by_index(begin = var_7482_begin_0, end = var_7482_end_0, end_mask = var_7482_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7482_cast_fp16")]; + tensor var_7484_begin_0 = const()[name = tensor("op_7484_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7484_end_0 = const()[name = tensor("op_7484_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_7484_end_mask_0 = const()[name = tensor("op_7484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7484_cast_fp16 = slice_by_index(begin = var_7484_begin_0, end = var_7484_end_0, end_mask = var_7484_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7484_cast_fp16")]; + tensor var_7488_begin_0 = const()[name = tensor("op_7488_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7488_end_0 = const()[name = tensor("op_7488_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_7488_end_mask_0 = const()[name = tensor("op_7488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7488_cast_fp16 = slice_by_index(begin = var_7488_begin_0, end = var_7488_end_0, end_mask = var_7488_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7488_cast_fp16")]; + tensor var_7492_begin_0 = const()[name = tensor("op_7492_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7492_end_0 = const()[name = tensor("op_7492_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_7492_end_mask_0 = const()[name = tensor("op_7492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7492_cast_fp16 = slice_by_index(begin = var_7492_begin_0, end = var_7492_end_0, end_mask = var_7492_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7492_cast_fp16")]; + tensor var_7496_begin_0 = const()[name = tensor("op_7496_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7496_end_0 = const()[name = tensor("op_7496_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_7496_end_mask_0 = const()[name = tensor("op_7496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7496_cast_fp16 = slice_by_index(begin = var_7496_begin_0, end = var_7496_end_0, end_mask = var_7496_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7496_cast_fp16")]; + tensor var_7500_begin_0 = const()[name = tensor("op_7500_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7500_end_0 = const()[name = tensor("op_7500_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_7500_end_mask_0 = const()[name = tensor("op_7500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7500_cast_fp16 = slice_by_index(begin = var_7500_begin_0, end = var_7500_end_0, end_mask = var_7500_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7500_cast_fp16")]; + tensor var_7504_begin_0 = const()[name = tensor("op_7504_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7504_end_0 = const()[name = tensor("op_7504_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_7504_end_mask_0 = const()[name = tensor("op_7504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7504_cast_fp16 = slice_by_index(begin = var_7504_begin_0, end = var_7504_end_0, end_mask = var_7504_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7504_cast_fp16")]; + tensor var_7508_begin_0 = const()[name = tensor("op_7508_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7508_end_0 = const()[name = tensor("op_7508_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_7508_end_mask_0 = const()[name = tensor("op_7508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7508_cast_fp16 = slice_by_index(begin = var_7508_begin_0, end = var_7508_end_0, end_mask = var_7508_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7508_cast_fp16")]; + tensor var_7512_begin_0 = const()[name = tensor("op_7512_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7512_end_0 = const()[name = tensor("op_7512_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_7512_end_mask_0 = const()[name = tensor("op_7512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7512_cast_fp16 = slice_by_index(begin = var_7512_begin_0, end = var_7512_end_0, end_mask = var_7512_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7512_cast_fp16")]; + tensor var_7516_begin_0 = const()[name = tensor("op_7516_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7516_end_0 = const()[name = tensor("op_7516_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_7516_end_mask_0 = const()[name = tensor("op_7516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7516_cast_fp16 = slice_by_index(begin = var_7516_begin_0, end = var_7516_end_0, end_mask = var_7516_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7516_cast_fp16")]; + tensor var_7520_begin_0 = const()[name = tensor("op_7520_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7520_end_0 = const()[name = tensor("op_7520_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_7520_end_mask_0 = const()[name = tensor("op_7520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7520_cast_fp16 = slice_by_index(begin = var_7520_begin_0, end = var_7520_end_0, end_mask = var_7520_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7520_cast_fp16")]; + tensor var_7524_begin_0 = const()[name = tensor("op_7524_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7524_end_0 = const()[name = tensor("op_7524_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_7524_end_mask_0 = const()[name = tensor("op_7524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7524_cast_fp16 = slice_by_index(begin = var_7524_begin_0, end = var_7524_end_0, end_mask = var_7524_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7524_cast_fp16")]; + tensor var_7528_begin_0 = const()[name = tensor("op_7528_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7528_end_0 = const()[name = tensor("op_7528_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_7528_end_mask_0 = const()[name = tensor("op_7528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7528_cast_fp16 = slice_by_index(begin = var_7528_begin_0, end = var_7528_end_0, end_mask = var_7528_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7528_cast_fp16")]; + tensor var_7532_begin_0 = const()[name = tensor("op_7532_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7532_end_0 = const()[name = tensor("op_7532_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_7532_end_mask_0 = const()[name = tensor("op_7532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7532_cast_fp16 = slice_by_index(begin = var_7532_begin_0, end = var_7532_end_0, end_mask = var_7532_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7532_cast_fp16")]; + tensor var_7536_begin_0 = const()[name = tensor("op_7536_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7536_end_0 = const()[name = tensor("op_7536_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_7536_end_mask_0 = const()[name = tensor("op_7536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7536_cast_fp16 = slice_by_index(begin = var_7536_begin_0, end = var_7536_end_0, end_mask = var_7536_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7536_cast_fp16")]; + tensor var_7540_begin_0 = const()[name = tensor("op_7540_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7540_end_0 = const()[name = tensor("op_7540_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_7540_end_mask_0 = const()[name = tensor("op_7540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7540_cast_fp16 = slice_by_index(begin = var_7540_begin_0, end = var_7540_end_0, end_mask = var_7540_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7540_cast_fp16")]; + tensor var_7544_begin_0 = const()[name = tensor("op_7544_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7544_end_0 = const()[name = tensor("op_7544_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7544_end_mask_0 = const()[name = tensor("op_7544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7544_cast_fp16 = slice_by_index(begin = var_7544_begin_0, end = var_7544_end_0, end_mask = var_7544_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7544_cast_fp16")]; + tensor var_7548_begin_0 = const()[name = tensor("op_7548_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7548_end_0 = const()[name = tensor("op_7548_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_7548_end_mask_0 = const()[name = tensor("op_7548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7548_cast_fp16 = slice_by_index(begin = var_7548_begin_0, end = var_7548_end_0, end_mask = var_7548_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7548_cast_fp16")]; + tensor var_7552_begin_0 = const()[name = tensor("op_7552_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7552_end_0 = const()[name = tensor("op_7552_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_7552_end_mask_0 = const()[name = tensor("op_7552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7552_cast_fp16 = slice_by_index(begin = var_7552_begin_0, end = var_7552_end_0, end_mask = var_7552_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7552_cast_fp16")]; + tensor var_7556_begin_0 = const()[name = tensor("op_7556_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7556_end_0 = const()[name = tensor("op_7556_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_7556_end_mask_0 = const()[name = tensor("op_7556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7556_cast_fp16 = slice_by_index(begin = var_7556_begin_0, end = var_7556_end_0, end_mask = var_7556_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7556_cast_fp16")]; + tensor var_7560_begin_0 = const()[name = tensor("op_7560_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7560_end_0 = const()[name = tensor("op_7560_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_7560_end_mask_0 = const()[name = tensor("op_7560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7560_cast_fp16 = slice_by_index(begin = var_7560_begin_0, end = var_7560_end_0, end_mask = var_7560_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7560_cast_fp16")]; + tensor var_7564_equation_0 = const()[name = tensor("op_7564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7564_cast_fp16 = einsum(equation = var_7564_equation_0, values = (var_7406_cast_fp16, var_7323_cast_fp16))[name = tensor("op_7564_cast_fp16")]; + tensor var_7565_to_fp16 = const()[name = tensor("op_7565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_561_cast_fp16 = mul(x = var_7564_cast_fp16, y = var_7565_to_fp16)[name = tensor("aw_561_cast_fp16")]; + tensor var_7568_equation_0 = const()[name = tensor("op_7568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7568_cast_fp16 = einsum(equation = var_7568_equation_0, values = (var_7410_cast_fp16, var_7327_cast_fp16))[name = tensor("op_7568_cast_fp16")]; + tensor var_7569_to_fp16 = const()[name = tensor("op_7569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_563_cast_fp16 = mul(x = var_7568_cast_fp16, y = var_7569_to_fp16)[name = tensor("aw_563_cast_fp16")]; + tensor var_7572_equation_0 = const()[name = tensor("op_7572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7572_cast_fp16 = einsum(equation = var_7572_equation_0, values = (var_7414_cast_fp16, var_7331_cast_fp16))[name = tensor("op_7572_cast_fp16")]; + tensor var_7573_to_fp16 = const()[name = tensor("op_7573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_565_cast_fp16 = mul(x = var_7572_cast_fp16, y = var_7573_to_fp16)[name = tensor("aw_565_cast_fp16")]; + tensor var_7576_equation_0 = const()[name = tensor("op_7576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7576_cast_fp16 = einsum(equation = var_7576_equation_0, values = (var_7418_cast_fp16, var_7335_cast_fp16))[name = tensor("op_7576_cast_fp16")]; + tensor var_7577_to_fp16 = const()[name = tensor("op_7577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_567_cast_fp16 = mul(x = var_7576_cast_fp16, y = var_7577_to_fp16)[name = tensor("aw_567_cast_fp16")]; + tensor var_7580_equation_0 = const()[name = tensor("op_7580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7580_cast_fp16 = einsum(equation = var_7580_equation_0, values = (var_7422_cast_fp16, var_7339_cast_fp16))[name = tensor("op_7580_cast_fp16")]; + tensor var_7581_to_fp16 = const()[name = tensor("op_7581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_569_cast_fp16 = mul(x = var_7580_cast_fp16, y = var_7581_to_fp16)[name = tensor("aw_569_cast_fp16")]; + tensor var_7584_equation_0 = const()[name = tensor("op_7584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7584_cast_fp16 = einsum(equation = var_7584_equation_0, values = (var_7426_cast_fp16, var_7343_cast_fp16))[name = tensor("op_7584_cast_fp16")]; + tensor var_7585_to_fp16 = const()[name = tensor("op_7585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_571_cast_fp16 = mul(x = var_7584_cast_fp16, y = var_7585_to_fp16)[name = tensor("aw_571_cast_fp16")]; + tensor var_7588_equation_0 = const()[name = tensor("op_7588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7588_cast_fp16 = einsum(equation = var_7588_equation_0, values = (var_7430_cast_fp16, var_7347_cast_fp16))[name = tensor("op_7588_cast_fp16")]; + tensor var_7589_to_fp16 = const()[name = tensor("op_7589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_573_cast_fp16 = mul(x = var_7588_cast_fp16, y = var_7589_to_fp16)[name = tensor("aw_573_cast_fp16")]; + tensor var_7592_equation_0 = const()[name = tensor("op_7592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7592_cast_fp16 = einsum(equation = var_7592_equation_0, values = (var_7434_cast_fp16, var_7351_cast_fp16))[name = tensor("op_7592_cast_fp16")]; + tensor var_7593_to_fp16 = const()[name = tensor("op_7593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_575_cast_fp16 = mul(x = var_7592_cast_fp16, y = var_7593_to_fp16)[name = tensor("aw_575_cast_fp16")]; + tensor var_7596_equation_0 = const()[name = tensor("op_7596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7596_cast_fp16 = einsum(equation = var_7596_equation_0, values = (var_7438_cast_fp16, var_7355_cast_fp16))[name = tensor("op_7596_cast_fp16")]; + tensor var_7597_to_fp16 = const()[name = tensor("op_7597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_577_cast_fp16 = mul(x = var_7596_cast_fp16, y = var_7597_to_fp16)[name = tensor("aw_577_cast_fp16")]; + tensor var_7600_equation_0 = const()[name = tensor("op_7600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7600_cast_fp16 = einsum(equation = var_7600_equation_0, values = (var_7442_cast_fp16, var_7359_cast_fp16))[name = tensor("op_7600_cast_fp16")]; + tensor var_7601_to_fp16 = const()[name = tensor("op_7601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_579_cast_fp16 = mul(x = var_7600_cast_fp16, y = var_7601_to_fp16)[name = tensor("aw_579_cast_fp16")]; + tensor var_7604_equation_0 = const()[name = tensor("op_7604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7604_cast_fp16 = einsum(equation = var_7604_equation_0, values = (var_7446_cast_fp16, var_7363_cast_fp16))[name = tensor("op_7604_cast_fp16")]; + tensor var_7605_to_fp16 = const()[name = tensor("op_7605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_581_cast_fp16 = mul(x = var_7604_cast_fp16, y = var_7605_to_fp16)[name = tensor("aw_581_cast_fp16")]; + tensor var_7608_equation_0 = const()[name = tensor("op_7608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7608_cast_fp16 = einsum(equation = var_7608_equation_0, values = (var_7450_cast_fp16, var_7367_cast_fp16))[name = tensor("op_7608_cast_fp16")]; + tensor var_7609_to_fp16 = const()[name = tensor("op_7609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_583_cast_fp16 = mul(x = var_7608_cast_fp16, y = var_7609_to_fp16)[name = tensor("aw_583_cast_fp16")]; + tensor var_7612_equation_0 = const()[name = tensor("op_7612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7612_cast_fp16 = einsum(equation = var_7612_equation_0, values = (var_7454_cast_fp16, var_7371_cast_fp16))[name = tensor("op_7612_cast_fp16")]; + tensor var_7613_to_fp16 = const()[name = tensor("op_7613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_585_cast_fp16 = mul(x = var_7612_cast_fp16, y = var_7613_to_fp16)[name = tensor("aw_585_cast_fp16")]; + tensor var_7616_equation_0 = const()[name = tensor("op_7616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7616_cast_fp16 = einsum(equation = var_7616_equation_0, values = (var_7458_cast_fp16, var_7375_cast_fp16))[name = tensor("op_7616_cast_fp16")]; + tensor var_7617_to_fp16 = const()[name = tensor("op_7617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_587_cast_fp16 = mul(x = var_7616_cast_fp16, y = var_7617_to_fp16)[name = tensor("aw_587_cast_fp16")]; + tensor var_7620_equation_0 = const()[name = tensor("op_7620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7620_cast_fp16 = einsum(equation = var_7620_equation_0, values = (var_7462_cast_fp16, var_7379_cast_fp16))[name = tensor("op_7620_cast_fp16")]; + tensor var_7621_to_fp16 = const()[name = tensor("op_7621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_589_cast_fp16 = mul(x = var_7620_cast_fp16, y = var_7621_to_fp16)[name = tensor("aw_589_cast_fp16")]; + tensor var_7624_equation_0 = const()[name = tensor("op_7624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7624_cast_fp16 = einsum(equation = var_7624_equation_0, values = (var_7466_cast_fp16, var_7383_cast_fp16))[name = tensor("op_7624_cast_fp16")]; + tensor var_7625_to_fp16 = const()[name = tensor("op_7625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_591_cast_fp16 = mul(x = var_7624_cast_fp16, y = var_7625_to_fp16)[name = tensor("aw_591_cast_fp16")]; + tensor var_7628_equation_0 = const()[name = tensor("op_7628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7628_cast_fp16 = einsum(equation = var_7628_equation_0, values = (var_7470_cast_fp16, var_7387_cast_fp16))[name = tensor("op_7628_cast_fp16")]; + tensor var_7629_to_fp16 = const()[name = tensor("op_7629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_593_cast_fp16 = mul(x = var_7628_cast_fp16, y = var_7629_to_fp16)[name = tensor("aw_593_cast_fp16")]; + tensor var_7632_equation_0 = const()[name = tensor("op_7632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7632_cast_fp16 = einsum(equation = var_7632_equation_0, values = (var_7474_cast_fp16, var_7391_cast_fp16))[name = tensor("op_7632_cast_fp16")]; + tensor var_7633_to_fp16 = const()[name = tensor("op_7633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_595_cast_fp16 = mul(x = var_7632_cast_fp16, y = var_7633_to_fp16)[name = tensor("aw_595_cast_fp16")]; + tensor var_7636_equation_0 = const()[name = tensor("op_7636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7636_cast_fp16 = einsum(equation = var_7636_equation_0, values = (var_7478_cast_fp16, var_7395_cast_fp16))[name = tensor("op_7636_cast_fp16")]; + tensor var_7637_to_fp16 = const()[name = tensor("op_7637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_597_cast_fp16 = mul(x = var_7636_cast_fp16, y = var_7637_to_fp16)[name = tensor("aw_597_cast_fp16")]; + tensor var_7640_equation_0 = const()[name = tensor("op_7640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7640_cast_fp16 = einsum(equation = var_7640_equation_0, values = (var_7482_cast_fp16, var_7399_cast_fp16))[name = tensor("op_7640_cast_fp16")]; + tensor var_7641_to_fp16 = const()[name = tensor("op_7641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_599_cast_fp16 = mul(x = var_7640_cast_fp16, y = var_7641_to_fp16)[name = tensor("aw_599_cast_fp16")]; + tensor var_7643_cast_fp16 = softmax(axis = var_2624, x = aw_561_cast_fp16)[name = tensor("op_7643_cast_fp16")]; + tensor var_7644_cast_fp16 = softmax(axis = var_2624, x = aw_563_cast_fp16)[name = tensor("op_7644_cast_fp16")]; + tensor var_7645_cast_fp16 = softmax(axis = var_2624, x = aw_565_cast_fp16)[name = tensor("op_7645_cast_fp16")]; + tensor var_7646_cast_fp16 = softmax(axis = var_2624, x = aw_567_cast_fp16)[name = tensor("op_7646_cast_fp16")]; + tensor var_7647_cast_fp16 = softmax(axis = var_2624, x = aw_569_cast_fp16)[name = tensor("op_7647_cast_fp16")]; + tensor var_7648_cast_fp16 = softmax(axis = var_2624, x = aw_571_cast_fp16)[name = tensor("op_7648_cast_fp16")]; + tensor var_7649_cast_fp16 = softmax(axis = var_2624, x = aw_573_cast_fp16)[name = tensor("op_7649_cast_fp16")]; + tensor var_7650_cast_fp16 = softmax(axis = var_2624, x = aw_575_cast_fp16)[name = tensor("op_7650_cast_fp16")]; + tensor var_7651_cast_fp16 = softmax(axis = var_2624, x = aw_577_cast_fp16)[name = tensor("op_7651_cast_fp16")]; + tensor var_7652_cast_fp16 = softmax(axis = var_2624, x = aw_579_cast_fp16)[name = tensor("op_7652_cast_fp16")]; + tensor var_7653_cast_fp16 = softmax(axis = var_2624, x = aw_581_cast_fp16)[name = tensor("op_7653_cast_fp16")]; + tensor var_7654_cast_fp16 = softmax(axis = var_2624, x = aw_583_cast_fp16)[name = tensor("op_7654_cast_fp16")]; + tensor var_7655_cast_fp16 = softmax(axis = var_2624, x = aw_585_cast_fp16)[name = tensor("op_7655_cast_fp16")]; + tensor var_7656_cast_fp16 = softmax(axis = var_2624, x = aw_587_cast_fp16)[name = tensor("op_7656_cast_fp16")]; + tensor var_7657_cast_fp16 = softmax(axis = var_2624, x = aw_589_cast_fp16)[name = tensor("op_7657_cast_fp16")]; + tensor var_7658_cast_fp16 = softmax(axis = var_2624, x = aw_591_cast_fp16)[name = tensor("op_7658_cast_fp16")]; + tensor var_7659_cast_fp16 = softmax(axis = var_2624, x = aw_593_cast_fp16)[name = tensor("op_7659_cast_fp16")]; + tensor var_7660_cast_fp16 = softmax(axis = var_2624, x = aw_595_cast_fp16)[name = tensor("op_7660_cast_fp16")]; + tensor var_7661_cast_fp16 = softmax(axis = var_2624, x = aw_597_cast_fp16)[name = tensor("op_7661_cast_fp16")]; + tensor var_7662_cast_fp16 = softmax(axis = var_2624, x = aw_599_cast_fp16)[name = tensor("op_7662_cast_fp16")]; + tensor var_7664_equation_0 = const()[name = tensor("op_7664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7664_cast_fp16 = einsum(equation = var_7664_equation_0, values = (var_7484_cast_fp16, var_7643_cast_fp16))[name = tensor("op_7664_cast_fp16")]; + tensor var_7666_equation_0 = const()[name = tensor("op_7666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7666_cast_fp16 = einsum(equation = var_7666_equation_0, values = (var_7488_cast_fp16, var_7644_cast_fp16))[name = tensor("op_7666_cast_fp16")]; + tensor var_7668_equation_0 = const()[name = tensor("op_7668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7668_cast_fp16 = einsum(equation = var_7668_equation_0, values = (var_7492_cast_fp16, var_7645_cast_fp16))[name = tensor("op_7668_cast_fp16")]; + tensor var_7670_equation_0 = const()[name = tensor("op_7670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7670_cast_fp16 = einsum(equation = var_7670_equation_0, values = (var_7496_cast_fp16, var_7646_cast_fp16))[name = tensor("op_7670_cast_fp16")]; + tensor var_7672_equation_0 = const()[name = tensor("op_7672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7672_cast_fp16 = einsum(equation = var_7672_equation_0, values = (var_7500_cast_fp16, var_7647_cast_fp16))[name = tensor("op_7672_cast_fp16")]; + tensor var_7674_equation_0 = const()[name = tensor("op_7674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7674_cast_fp16 = einsum(equation = var_7674_equation_0, values = (var_7504_cast_fp16, var_7648_cast_fp16))[name = tensor("op_7674_cast_fp16")]; + tensor var_7676_equation_0 = const()[name = tensor("op_7676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7676_cast_fp16 = einsum(equation = var_7676_equation_0, values = (var_7508_cast_fp16, var_7649_cast_fp16))[name = tensor("op_7676_cast_fp16")]; + tensor var_7678_equation_0 = const()[name = tensor("op_7678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7678_cast_fp16 = einsum(equation = var_7678_equation_0, values = (var_7512_cast_fp16, var_7650_cast_fp16))[name = tensor("op_7678_cast_fp16")]; + tensor var_7680_equation_0 = const()[name = tensor("op_7680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7680_cast_fp16 = einsum(equation = var_7680_equation_0, values = (var_7516_cast_fp16, var_7651_cast_fp16))[name = tensor("op_7680_cast_fp16")]; + tensor var_7682_equation_0 = const()[name = tensor("op_7682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7682_cast_fp16 = einsum(equation = var_7682_equation_0, values = (var_7520_cast_fp16, var_7652_cast_fp16))[name = tensor("op_7682_cast_fp16")]; + tensor var_7684_equation_0 = const()[name = tensor("op_7684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7684_cast_fp16 = einsum(equation = var_7684_equation_0, values = (var_7524_cast_fp16, var_7653_cast_fp16))[name = tensor("op_7684_cast_fp16")]; + tensor var_7686_equation_0 = const()[name = tensor("op_7686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7686_cast_fp16 = einsum(equation = var_7686_equation_0, values = (var_7528_cast_fp16, var_7654_cast_fp16))[name = tensor("op_7686_cast_fp16")]; + tensor var_7688_equation_0 = const()[name = tensor("op_7688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7688_cast_fp16 = einsum(equation = var_7688_equation_0, values = (var_7532_cast_fp16, var_7655_cast_fp16))[name = tensor("op_7688_cast_fp16")]; + tensor var_7690_equation_0 = const()[name = tensor("op_7690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7690_cast_fp16 = einsum(equation = var_7690_equation_0, values = (var_7536_cast_fp16, var_7656_cast_fp16))[name = tensor("op_7690_cast_fp16")]; + tensor var_7692_equation_0 = const()[name = tensor("op_7692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7692_cast_fp16 = einsum(equation = var_7692_equation_0, values = (var_7540_cast_fp16, var_7657_cast_fp16))[name = tensor("op_7692_cast_fp16")]; + tensor var_7694_equation_0 = const()[name = tensor("op_7694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7694_cast_fp16 = einsum(equation = var_7694_equation_0, values = (var_7544_cast_fp16, var_7658_cast_fp16))[name = tensor("op_7694_cast_fp16")]; + tensor var_7696_equation_0 = const()[name = tensor("op_7696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7696_cast_fp16 = einsum(equation = var_7696_equation_0, values = (var_7548_cast_fp16, var_7659_cast_fp16))[name = tensor("op_7696_cast_fp16")]; + tensor var_7698_equation_0 = const()[name = tensor("op_7698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7698_cast_fp16 = einsum(equation = var_7698_equation_0, values = (var_7552_cast_fp16, var_7660_cast_fp16))[name = tensor("op_7698_cast_fp16")]; + tensor var_7700_equation_0 = const()[name = tensor("op_7700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7700_cast_fp16 = einsum(equation = var_7700_equation_0, values = (var_7556_cast_fp16, var_7661_cast_fp16))[name = tensor("op_7700_cast_fp16")]; + tensor var_7702_equation_0 = const()[name = tensor("op_7702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7702_cast_fp16 = einsum(equation = var_7702_equation_0, values = (var_7560_cast_fp16, var_7662_cast_fp16))[name = tensor("op_7702_cast_fp16")]; + tensor input_171_interleave_0 = const()[name = tensor("input_171_interleave_0"), val = tensor(false)]; + tensor input_171_cast_fp16 = concat(axis = var_2624, interleave = input_171_interleave_0, values = (var_7664_cast_fp16, var_7666_cast_fp16, var_7668_cast_fp16, var_7670_cast_fp16, var_7672_cast_fp16, var_7674_cast_fp16, var_7676_cast_fp16, var_7678_cast_fp16, var_7680_cast_fp16, var_7682_cast_fp16, var_7684_cast_fp16, var_7686_cast_fp16, var_7688_cast_fp16, var_7690_cast_fp16, var_7692_cast_fp16, var_7694_cast_fp16, var_7696_cast_fp16, var_7698_cast_fp16, var_7700_cast_fp16, var_7702_cast_fp16))[name = tensor("input_171_cast_fp16")]; + tensor var_7712_pad_type_0 = const()[name = tensor("op_7712_pad_type_0"), val = tensor("valid")]; + tensor var_7712_strides_0 = const()[name = tensor("op_7712_strides_0"), val = tensor([1, 1])]; + tensor var_7712_pad_0 = const()[name = tensor("op_7712_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7712_dilations_0 = const()[name = tensor("op_7712_dilations_0"), val = tensor([1, 1])]; + tensor var_7712_groups_0 = const()[name = tensor("op_7712_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208618752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209847616))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209847808)))]; + tensor var_7712_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_7712_dilations_0, groups = var_7712_groups_0, pad = var_7712_pad_0, pad_type = var_7712_pad_type_0, strides = var_7712_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = tensor("op_7712_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = var_7712_cast_fp16, y = inputs_55_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; + tensor hidden_states_97_axes_0 = const()[name = tensor("hidden_states_97_axes_0"), val = tensor([1])]; + tensor hidden_states_97_gamma_0_to_fp16 = const()[name = tensor("hidden_states_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209850432)))]; + tensor hidden_states_97_beta_0_to_fp16 = const()[name = tensor("hidden_states_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209853056)))]; + tensor var_7722_to_fp16 = const()[name = tensor("op_7722_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_97_cast_fp16 = layer_norm(axes = hidden_states_97_axes_0, beta = hidden_states_97_beta_0_to_fp16, epsilon = var_7722_to_fp16, gamma = hidden_states_97_gamma_0_to_fp16, x = inputs_57_cast_fp16)[name = tensor("hidden_states_97_cast_fp16")]; + tensor q_39_pad_type_0 = const()[name = tensor("q_39_pad_type_0"), val = tensor("valid")]; + tensor q_39_strides_0 = const()[name = tensor("q_39_strides_0"), val = tensor([1, 1])]; + tensor q_39_pad_0 = const()[name = tensor("q_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_39_dilations_0 = const()[name = tensor("q_39_dilations_0"), val = tensor([1, 1])]; + tensor q_39_groups_0 = const()[name = tensor("q_39_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209855680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211084544))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_39_cast_fp16 = conv(dilations = q_39_dilations_0, groups = q_39_groups_0, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = q_39_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_97_cast_fp16)[name = tensor("q_39_cast_fp16")]; + tensor k_77_pad_type_0 = const()[name = tensor("k_77_pad_type_0"), val = tensor("valid")]; + tensor k_77_strides_0 = const()[name = tensor("k_77_strides_0"), val = tensor([1, 1])]; + tensor k_77_pad_0 = const()[name = tensor("k_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_77_dilations_0 = const()[name = tensor("k_77_dilations_0"), val = tensor([1, 1])]; + tensor k_77_groups_0 = const()[name = tensor("k_77_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211084736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213050880))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_77_cast_fp16 = conv(dilations = k_77_dilations_0, groups = k_77_groups_0, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = k_77_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_77_cast_fp16")]; + tensor v_39_pad_type_0 = const()[name = tensor("v_39_pad_type_0"), val = tensor("valid")]; + tensor v_39_strides_0 = const()[name = tensor("v_39_strides_0"), val = tensor([1, 1])]; + tensor v_39_pad_0 = const()[name = tensor("v_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_39_dilations_0 = const()[name = tensor("v_39_dilations_0"), val = tensor([1, 1])]; + tensor v_39_groups_0 = const()[name = tensor("v_39_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213051072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215017216))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_39_cast_fp16 = conv(dilations = v_39_dilations_0, groups = v_39_groups_0, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = v_39_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_39_cast_fp16")]; + tensor var_7755_begin_0 = const()[name = tensor("op_7755_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7755_end_0 = const()[name = tensor("op_7755_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_7755_end_mask_0 = const()[name = tensor("op_7755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7755_cast_fp16 = slice_by_index(begin = var_7755_begin_0, end = var_7755_end_0, end_mask = var_7755_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7755_cast_fp16")]; + tensor var_7759_begin_0 = const()[name = tensor("op_7759_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7759_end_0 = const()[name = tensor("op_7759_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_7759_end_mask_0 = const()[name = tensor("op_7759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7759_cast_fp16 = slice_by_index(begin = var_7759_begin_0, end = var_7759_end_0, end_mask = var_7759_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7759_cast_fp16")]; + tensor var_7763_begin_0 = const()[name = tensor("op_7763_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7763_end_0 = const()[name = tensor("op_7763_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_7763_end_mask_0 = const()[name = tensor("op_7763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7763_cast_fp16 = slice_by_index(begin = var_7763_begin_0, end = var_7763_end_0, end_mask = var_7763_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7763_cast_fp16")]; + tensor var_7767_begin_0 = const()[name = tensor("op_7767_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7767_end_0 = const()[name = tensor("op_7767_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_7767_end_mask_0 = const()[name = tensor("op_7767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7767_cast_fp16 = slice_by_index(begin = var_7767_begin_0, end = var_7767_end_0, end_mask = var_7767_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7767_cast_fp16")]; + tensor var_7771_begin_0 = const()[name = tensor("op_7771_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7771_end_0 = const()[name = tensor("op_7771_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_7771_end_mask_0 = const()[name = tensor("op_7771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7771_cast_fp16 = slice_by_index(begin = var_7771_begin_0, end = var_7771_end_0, end_mask = var_7771_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7771_cast_fp16")]; + tensor var_7775_begin_0 = const()[name = tensor("op_7775_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7775_end_0 = const()[name = tensor("op_7775_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_7775_end_mask_0 = const()[name = tensor("op_7775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7775_cast_fp16 = slice_by_index(begin = var_7775_begin_0, end = var_7775_end_0, end_mask = var_7775_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7775_cast_fp16")]; + tensor var_7779_begin_0 = const()[name = tensor("op_7779_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7779_end_0 = const()[name = tensor("op_7779_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_7779_end_mask_0 = const()[name = tensor("op_7779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7779_cast_fp16 = slice_by_index(begin = var_7779_begin_0, end = var_7779_end_0, end_mask = var_7779_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7779_cast_fp16")]; + tensor var_7783_begin_0 = const()[name = tensor("op_7783_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7783_end_0 = const()[name = tensor("op_7783_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_7783_end_mask_0 = const()[name = tensor("op_7783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7783_cast_fp16 = slice_by_index(begin = var_7783_begin_0, end = var_7783_end_0, end_mask = var_7783_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7783_cast_fp16")]; + tensor var_7787_begin_0 = const()[name = tensor("op_7787_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7787_end_0 = const()[name = tensor("op_7787_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_7787_end_mask_0 = const()[name = tensor("op_7787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7787_cast_fp16 = slice_by_index(begin = var_7787_begin_0, end = var_7787_end_0, end_mask = var_7787_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7787_cast_fp16")]; + tensor var_7791_begin_0 = const()[name = tensor("op_7791_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7791_end_0 = const()[name = tensor("op_7791_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_7791_end_mask_0 = const()[name = tensor("op_7791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7791_cast_fp16 = slice_by_index(begin = var_7791_begin_0, end = var_7791_end_0, end_mask = var_7791_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7791_cast_fp16")]; + tensor var_7795_begin_0 = const()[name = tensor("op_7795_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7795_end_0 = const()[name = tensor("op_7795_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_7795_end_mask_0 = const()[name = tensor("op_7795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7795_cast_fp16 = slice_by_index(begin = var_7795_begin_0, end = var_7795_end_0, end_mask = var_7795_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7795_cast_fp16")]; + tensor var_7799_begin_0 = const()[name = tensor("op_7799_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7799_end_0 = const()[name = tensor("op_7799_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_7799_end_mask_0 = const()[name = tensor("op_7799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7799_cast_fp16 = slice_by_index(begin = var_7799_begin_0, end = var_7799_end_0, end_mask = var_7799_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7799_cast_fp16")]; + tensor var_7803_begin_0 = const()[name = tensor("op_7803_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7803_end_0 = const()[name = tensor("op_7803_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_7803_end_mask_0 = const()[name = tensor("op_7803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7803_cast_fp16 = slice_by_index(begin = var_7803_begin_0, end = var_7803_end_0, end_mask = var_7803_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7803_cast_fp16")]; + tensor var_7807_begin_0 = const()[name = tensor("op_7807_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7807_end_0 = const()[name = tensor("op_7807_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_7807_end_mask_0 = const()[name = tensor("op_7807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7807_cast_fp16 = slice_by_index(begin = var_7807_begin_0, end = var_7807_end_0, end_mask = var_7807_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7807_cast_fp16")]; + tensor var_7811_begin_0 = const()[name = tensor("op_7811_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7811_end_0 = const()[name = tensor("op_7811_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_7811_end_mask_0 = const()[name = tensor("op_7811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7811_cast_fp16 = slice_by_index(begin = var_7811_begin_0, end = var_7811_end_0, end_mask = var_7811_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7811_cast_fp16")]; + tensor var_7815_begin_0 = const()[name = tensor("op_7815_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7815_end_0 = const()[name = tensor("op_7815_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7815_end_mask_0 = const()[name = tensor("op_7815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7815_cast_fp16 = slice_by_index(begin = var_7815_begin_0, end = var_7815_end_0, end_mask = var_7815_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7815_cast_fp16")]; + tensor var_7819_begin_0 = const()[name = tensor("op_7819_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7819_end_0 = const()[name = tensor("op_7819_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_7819_end_mask_0 = const()[name = tensor("op_7819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7819_cast_fp16 = slice_by_index(begin = var_7819_begin_0, end = var_7819_end_0, end_mask = var_7819_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7819_cast_fp16")]; + tensor var_7823_begin_0 = const()[name = tensor("op_7823_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7823_end_0 = const()[name = tensor("op_7823_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_7823_end_mask_0 = const()[name = tensor("op_7823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7823_cast_fp16 = slice_by_index(begin = var_7823_begin_0, end = var_7823_end_0, end_mask = var_7823_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7823_cast_fp16")]; + tensor var_7827_begin_0 = const()[name = tensor("op_7827_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7827_end_0 = const()[name = tensor("op_7827_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_7827_end_mask_0 = const()[name = tensor("op_7827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7827_cast_fp16 = slice_by_index(begin = var_7827_begin_0, end = var_7827_end_0, end_mask = var_7827_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7827_cast_fp16")]; + tensor var_7831_begin_0 = const()[name = tensor("op_7831_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7831_end_0 = const()[name = tensor("op_7831_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_7831_end_mask_0 = const()[name = tensor("op_7831_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7831_cast_fp16 = slice_by_index(begin = var_7831_begin_0, end = var_7831_end_0, end_mask = var_7831_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7831_cast_fp16")]; + tensor k_79_perm_0 = const()[name = tensor("k_79_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7838_begin_0 = const()[name = tensor("op_7838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7838_end_0 = const()[name = tensor("op_7838_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_7838_end_mask_0 = const()[name = tensor("op_7838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_79_cast_fp16 = transpose(perm = k_79_perm_0, x = k_77_cast_fp16)[name = tensor("transpose_48")]; + tensor var_7838_cast_fp16 = slice_by_index(begin = var_7838_begin_0, end = var_7838_end_0, end_mask = var_7838_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7838_cast_fp16")]; + tensor var_7842_begin_0 = const()[name = tensor("op_7842_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7842_end_0 = const()[name = tensor("op_7842_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_7842_end_mask_0 = const()[name = tensor("op_7842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7842_cast_fp16 = slice_by_index(begin = var_7842_begin_0, end = var_7842_end_0, end_mask = var_7842_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7842_cast_fp16")]; + tensor var_7846_begin_0 = const()[name = tensor("op_7846_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7846_end_0 = const()[name = tensor("op_7846_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_7846_end_mask_0 = const()[name = tensor("op_7846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7846_cast_fp16 = slice_by_index(begin = var_7846_begin_0, end = var_7846_end_0, end_mask = var_7846_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7846_cast_fp16")]; + tensor var_7850_begin_0 = const()[name = tensor("op_7850_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7850_end_0 = const()[name = tensor("op_7850_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_7850_end_mask_0 = const()[name = tensor("op_7850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7850_cast_fp16 = slice_by_index(begin = var_7850_begin_0, end = var_7850_end_0, end_mask = var_7850_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7850_cast_fp16")]; + tensor var_7854_begin_0 = const()[name = tensor("op_7854_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7854_end_0 = const()[name = tensor("op_7854_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_7854_end_mask_0 = const()[name = tensor("op_7854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7854_cast_fp16 = slice_by_index(begin = var_7854_begin_0, end = var_7854_end_0, end_mask = var_7854_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7854_cast_fp16")]; + tensor var_7858_begin_0 = const()[name = tensor("op_7858_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7858_end_0 = const()[name = tensor("op_7858_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_7858_end_mask_0 = const()[name = tensor("op_7858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7858_cast_fp16 = slice_by_index(begin = var_7858_begin_0, end = var_7858_end_0, end_mask = var_7858_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7858_cast_fp16")]; + tensor var_7862_begin_0 = const()[name = tensor("op_7862_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7862_end_0 = const()[name = tensor("op_7862_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_7862_end_mask_0 = const()[name = tensor("op_7862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7862_cast_fp16 = slice_by_index(begin = var_7862_begin_0, end = var_7862_end_0, end_mask = var_7862_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7862_cast_fp16")]; + tensor var_7866_begin_0 = const()[name = tensor("op_7866_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7866_end_0 = const()[name = tensor("op_7866_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_7866_end_mask_0 = const()[name = tensor("op_7866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7866_cast_fp16 = slice_by_index(begin = var_7866_begin_0, end = var_7866_end_0, end_mask = var_7866_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7866_cast_fp16")]; + tensor var_7870_begin_0 = const()[name = tensor("op_7870_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7870_end_0 = const()[name = tensor("op_7870_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_7870_end_mask_0 = const()[name = tensor("op_7870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7870_cast_fp16 = slice_by_index(begin = var_7870_begin_0, end = var_7870_end_0, end_mask = var_7870_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7870_cast_fp16")]; + tensor var_7874_begin_0 = const()[name = tensor("op_7874_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7874_end_0 = const()[name = tensor("op_7874_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_7874_end_mask_0 = const()[name = tensor("op_7874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7874_cast_fp16 = slice_by_index(begin = var_7874_begin_0, end = var_7874_end_0, end_mask = var_7874_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7874_cast_fp16")]; + tensor var_7878_begin_0 = const()[name = tensor("op_7878_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7878_end_0 = const()[name = tensor("op_7878_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_7878_end_mask_0 = const()[name = tensor("op_7878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7878_cast_fp16 = slice_by_index(begin = var_7878_begin_0, end = var_7878_end_0, end_mask = var_7878_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7878_cast_fp16")]; + tensor var_7882_begin_0 = const()[name = tensor("op_7882_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7882_end_0 = const()[name = tensor("op_7882_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_7882_end_mask_0 = const()[name = tensor("op_7882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7882_cast_fp16 = slice_by_index(begin = var_7882_begin_0, end = var_7882_end_0, end_mask = var_7882_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7882_cast_fp16")]; + tensor var_7886_begin_0 = const()[name = tensor("op_7886_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7886_end_0 = const()[name = tensor("op_7886_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_7886_end_mask_0 = const()[name = tensor("op_7886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7886_cast_fp16 = slice_by_index(begin = var_7886_begin_0, end = var_7886_end_0, end_mask = var_7886_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7886_cast_fp16")]; + tensor var_7890_begin_0 = const()[name = tensor("op_7890_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7890_end_0 = const()[name = tensor("op_7890_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_7890_end_mask_0 = const()[name = tensor("op_7890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7890_cast_fp16 = slice_by_index(begin = var_7890_begin_0, end = var_7890_end_0, end_mask = var_7890_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7890_cast_fp16")]; + tensor var_7894_begin_0 = const()[name = tensor("op_7894_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7894_end_0 = const()[name = tensor("op_7894_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_7894_end_mask_0 = const()[name = tensor("op_7894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7894_cast_fp16 = slice_by_index(begin = var_7894_begin_0, end = var_7894_end_0, end_mask = var_7894_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7894_cast_fp16")]; + tensor var_7898_begin_0 = const()[name = tensor("op_7898_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7898_end_0 = const()[name = tensor("op_7898_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_7898_end_mask_0 = const()[name = tensor("op_7898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7898_cast_fp16 = slice_by_index(begin = var_7898_begin_0, end = var_7898_end_0, end_mask = var_7898_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7898_cast_fp16")]; + tensor var_7902_begin_0 = const()[name = tensor("op_7902_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7902_end_0 = const()[name = tensor("op_7902_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_7902_end_mask_0 = const()[name = tensor("op_7902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7902_cast_fp16 = slice_by_index(begin = var_7902_begin_0, end = var_7902_end_0, end_mask = var_7902_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7902_cast_fp16")]; + tensor var_7906_begin_0 = const()[name = tensor("op_7906_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7906_end_0 = const()[name = tensor("op_7906_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_7906_end_mask_0 = const()[name = tensor("op_7906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7906_cast_fp16 = slice_by_index(begin = var_7906_begin_0, end = var_7906_end_0, end_mask = var_7906_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7906_cast_fp16")]; + tensor var_7910_begin_0 = const()[name = tensor("op_7910_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7910_end_0 = const()[name = tensor("op_7910_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_7910_end_mask_0 = const()[name = tensor("op_7910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7910_cast_fp16 = slice_by_index(begin = var_7910_begin_0, end = var_7910_end_0, end_mask = var_7910_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7910_cast_fp16")]; + tensor var_7914_begin_0 = const()[name = tensor("op_7914_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7914_end_0 = const()[name = tensor("op_7914_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_7914_end_mask_0 = const()[name = tensor("op_7914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7914_cast_fp16 = slice_by_index(begin = var_7914_begin_0, end = var_7914_end_0, end_mask = var_7914_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7914_cast_fp16")]; + tensor var_7916_begin_0 = const()[name = tensor("op_7916_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7916_end_0 = const()[name = tensor("op_7916_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_7916_end_mask_0 = const()[name = tensor("op_7916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7916_cast_fp16 = slice_by_index(begin = var_7916_begin_0, end = var_7916_end_0, end_mask = var_7916_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7916_cast_fp16")]; + tensor var_7920_begin_0 = const()[name = tensor("op_7920_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7920_end_0 = const()[name = tensor("op_7920_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_7920_end_mask_0 = const()[name = tensor("op_7920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7920_cast_fp16 = slice_by_index(begin = var_7920_begin_0, end = var_7920_end_0, end_mask = var_7920_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7920_cast_fp16")]; + tensor var_7924_begin_0 = const()[name = tensor("op_7924_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7924_end_0 = const()[name = tensor("op_7924_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_7924_end_mask_0 = const()[name = tensor("op_7924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7924_cast_fp16 = slice_by_index(begin = var_7924_begin_0, end = var_7924_end_0, end_mask = var_7924_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7924_cast_fp16")]; + tensor var_7928_begin_0 = const()[name = tensor("op_7928_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7928_end_0 = const()[name = tensor("op_7928_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_7928_end_mask_0 = const()[name = tensor("op_7928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7928_cast_fp16 = slice_by_index(begin = var_7928_begin_0, end = var_7928_end_0, end_mask = var_7928_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7928_cast_fp16")]; + tensor var_7932_begin_0 = const()[name = tensor("op_7932_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7932_end_0 = const()[name = tensor("op_7932_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_7932_end_mask_0 = const()[name = tensor("op_7932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7932_cast_fp16 = slice_by_index(begin = var_7932_begin_0, end = var_7932_end_0, end_mask = var_7932_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7932_cast_fp16")]; + tensor var_7936_begin_0 = const()[name = tensor("op_7936_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7936_end_0 = const()[name = tensor("op_7936_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_7936_end_mask_0 = const()[name = tensor("op_7936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7936_cast_fp16 = slice_by_index(begin = var_7936_begin_0, end = var_7936_end_0, end_mask = var_7936_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7936_cast_fp16")]; + tensor var_7940_begin_0 = const()[name = tensor("op_7940_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7940_end_0 = const()[name = tensor("op_7940_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_7940_end_mask_0 = const()[name = tensor("op_7940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7940_cast_fp16 = slice_by_index(begin = var_7940_begin_0, end = var_7940_end_0, end_mask = var_7940_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7940_cast_fp16")]; + tensor var_7944_begin_0 = const()[name = tensor("op_7944_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7944_end_0 = const()[name = tensor("op_7944_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_7944_end_mask_0 = const()[name = tensor("op_7944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7944_cast_fp16 = slice_by_index(begin = var_7944_begin_0, end = var_7944_end_0, end_mask = var_7944_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7944_cast_fp16")]; + tensor var_7948_begin_0 = const()[name = tensor("op_7948_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7948_end_0 = const()[name = tensor("op_7948_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_7948_end_mask_0 = const()[name = tensor("op_7948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7948_cast_fp16 = slice_by_index(begin = var_7948_begin_0, end = var_7948_end_0, end_mask = var_7948_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7948_cast_fp16")]; + tensor var_7952_begin_0 = const()[name = tensor("op_7952_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7952_end_0 = const()[name = tensor("op_7952_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_7952_end_mask_0 = const()[name = tensor("op_7952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7952_cast_fp16 = slice_by_index(begin = var_7952_begin_0, end = var_7952_end_0, end_mask = var_7952_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7952_cast_fp16")]; + tensor var_7956_begin_0 = const()[name = tensor("op_7956_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7956_end_0 = const()[name = tensor("op_7956_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_7956_end_mask_0 = const()[name = tensor("op_7956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7956_cast_fp16 = slice_by_index(begin = var_7956_begin_0, end = var_7956_end_0, end_mask = var_7956_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7956_cast_fp16")]; + tensor var_7960_begin_0 = const()[name = tensor("op_7960_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7960_end_0 = const()[name = tensor("op_7960_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_7960_end_mask_0 = const()[name = tensor("op_7960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7960_cast_fp16 = slice_by_index(begin = var_7960_begin_0, end = var_7960_end_0, end_mask = var_7960_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7960_cast_fp16")]; + tensor var_7964_begin_0 = const()[name = tensor("op_7964_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7964_end_0 = const()[name = tensor("op_7964_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_7964_end_mask_0 = const()[name = tensor("op_7964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7964_cast_fp16 = slice_by_index(begin = var_7964_begin_0, end = var_7964_end_0, end_mask = var_7964_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7964_cast_fp16")]; + tensor var_7968_begin_0 = const()[name = tensor("op_7968_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7968_end_0 = const()[name = tensor("op_7968_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_7968_end_mask_0 = const()[name = tensor("op_7968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7968_cast_fp16 = slice_by_index(begin = var_7968_begin_0, end = var_7968_end_0, end_mask = var_7968_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7968_cast_fp16")]; + tensor var_7972_begin_0 = const()[name = tensor("op_7972_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7972_end_0 = const()[name = tensor("op_7972_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_7972_end_mask_0 = const()[name = tensor("op_7972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7972_cast_fp16 = slice_by_index(begin = var_7972_begin_0, end = var_7972_end_0, end_mask = var_7972_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7972_cast_fp16")]; + tensor var_7976_begin_0 = const()[name = tensor("op_7976_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7976_end_0 = const()[name = tensor("op_7976_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_7976_end_mask_0 = const()[name = tensor("op_7976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7976_cast_fp16 = slice_by_index(begin = var_7976_begin_0, end = var_7976_end_0, end_mask = var_7976_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7976_cast_fp16")]; + tensor var_7980_begin_0 = const()[name = tensor("op_7980_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7980_end_0 = const()[name = tensor("op_7980_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_7980_end_mask_0 = const()[name = tensor("op_7980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7980_cast_fp16 = slice_by_index(begin = var_7980_begin_0, end = var_7980_end_0, end_mask = var_7980_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7980_cast_fp16")]; + tensor var_7984_begin_0 = const()[name = tensor("op_7984_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7984_end_0 = const()[name = tensor("op_7984_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_7984_end_mask_0 = const()[name = tensor("op_7984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7984_cast_fp16 = slice_by_index(begin = var_7984_begin_0, end = var_7984_end_0, end_mask = var_7984_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7984_cast_fp16")]; + tensor var_7988_begin_0 = const()[name = tensor("op_7988_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7988_end_0 = const()[name = tensor("op_7988_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_7988_end_mask_0 = const()[name = tensor("op_7988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7988_cast_fp16 = slice_by_index(begin = var_7988_begin_0, end = var_7988_end_0, end_mask = var_7988_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7988_cast_fp16")]; + tensor var_7992_begin_0 = const()[name = tensor("op_7992_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7992_end_0 = const()[name = tensor("op_7992_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_7992_end_mask_0 = const()[name = tensor("op_7992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7992_cast_fp16 = slice_by_index(begin = var_7992_begin_0, end = var_7992_end_0, end_mask = var_7992_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7992_cast_fp16")]; + tensor var_7996_equation_0 = const()[name = tensor("op_7996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7996_cast_fp16 = einsum(equation = var_7996_equation_0, values = (var_7838_cast_fp16, var_7755_cast_fp16))[name = tensor("op_7996_cast_fp16")]; + tensor var_7997_to_fp16 = const()[name = tensor("op_7997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_601_cast_fp16 = mul(x = var_7996_cast_fp16, y = var_7997_to_fp16)[name = tensor("aw_601_cast_fp16")]; + tensor var_8000_equation_0 = const()[name = tensor("op_8000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8000_cast_fp16 = einsum(equation = var_8000_equation_0, values = (var_7842_cast_fp16, var_7759_cast_fp16))[name = tensor("op_8000_cast_fp16")]; + tensor var_8001_to_fp16 = const()[name = tensor("op_8001_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_603_cast_fp16 = mul(x = var_8000_cast_fp16, y = var_8001_to_fp16)[name = tensor("aw_603_cast_fp16")]; + tensor var_8004_equation_0 = const()[name = tensor("op_8004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8004_cast_fp16 = einsum(equation = var_8004_equation_0, values = (var_7846_cast_fp16, var_7763_cast_fp16))[name = tensor("op_8004_cast_fp16")]; + tensor var_8005_to_fp16 = const()[name = tensor("op_8005_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_605_cast_fp16 = mul(x = var_8004_cast_fp16, y = var_8005_to_fp16)[name = tensor("aw_605_cast_fp16")]; + tensor var_8008_equation_0 = const()[name = tensor("op_8008_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8008_cast_fp16 = einsum(equation = var_8008_equation_0, values = (var_7850_cast_fp16, var_7767_cast_fp16))[name = tensor("op_8008_cast_fp16")]; + tensor var_8009_to_fp16 = const()[name = tensor("op_8009_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_607_cast_fp16 = mul(x = var_8008_cast_fp16, y = var_8009_to_fp16)[name = tensor("aw_607_cast_fp16")]; + tensor var_8012_equation_0 = const()[name = tensor("op_8012_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8012_cast_fp16 = einsum(equation = var_8012_equation_0, values = (var_7854_cast_fp16, var_7771_cast_fp16))[name = tensor("op_8012_cast_fp16")]; + tensor var_8013_to_fp16 = const()[name = tensor("op_8013_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_609_cast_fp16 = mul(x = var_8012_cast_fp16, y = var_8013_to_fp16)[name = tensor("aw_609_cast_fp16")]; + tensor var_8016_equation_0 = const()[name = tensor("op_8016_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8016_cast_fp16 = einsum(equation = var_8016_equation_0, values = (var_7858_cast_fp16, var_7775_cast_fp16))[name = tensor("op_8016_cast_fp16")]; + tensor var_8017_to_fp16 = const()[name = tensor("op_8017_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_611_cast_fp16 = mul(x = var_8016_cast_fp16, y = var_8017_to_fp16)[name = tensor("aw_611_cast_fp16")]; + tensor var_8020_equation_0 = const()[name = tensor("op_8020_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8020_cast_fp16 = einsum(equation = var_8020_equation_0, values = (var_7862_cast_fp16, var_7779_cast_fp16))[name = tensor("op_8020_cast_fp16")]; + tensor var_8021_to_fp16 = const()[name = tensor("op_8021_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_613_cast_fp16 = mul(x = var_8020_cast_fp16, y = var_8021_to_fp16)[name = tensor("aw_613_cast_fp16")]; + tensor var_8024_equation_0 = const()[name = tensor("op_8024_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8024_cast_fp16 = einsum(equation = var_8024_equation_0, values = (var_7866_cast_fp16, var_7783_cast_fp16))[name = tensor("op_8024_cast_fp16")]; + tensor var_8025_to_fp16 = const()[name = tensor("op_8025_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_615_cast_fp16 = mul(x = var_8024_cast_fp16, y = var_8025_to_fp16)[name = tensor("aw_615_cast_fp16")]; + tensor var_8028_equation_0 = const()[name = tensor("op_8028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8028_cast_fp16 = einsum(equation = var_8028_equation_0, values = (var_7870_cast_fp16, var_7787_cast_fp16))[name = tensor("op_8028_cast_fp16")]; + tensor var_8029_to_fp16 = const()[name = tensor("op_8029_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_617_cast_fp16 = mul(x = var_8028_cast_fp16, y = var_8029_to_fp16)[name = tensor("aw_617_cast_fp16")]; + tensor var_8032_equation_0 = const()[name = tensor("op_8032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8032_cast_fp16 = einsum(equation = var_8032_equation_0, values = (var_7874_cast_fp16, var_7791_cast_fp16))[name = tensor("op_8032_cast_fp16")]; + tensor var_8033_to_fp16 = const()[name = tensor("op_8033_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_619_cast_fp16 = mul(x = var_8032_cast_fp16, y = var_8033_to_fp16)[name = tensor("aw_619_cast_fp16")]; + tensor var_8036_equation_0 = const()[name = tensor("op_8036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8036_cast_fp16 = einsum(equation = var_8036_equation_0, values = (var_7878_cast_fp16, var_7795_cast_fp16))[name = tensor("op_8036_cast_fp16")]; + tensor var_8037_to_fp16 = const()[name = tensor("op_8037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_621_cast_fp16 = mul(x = var_8036_cast_fp16, y = var_8037_to_fp16)[name = tensor("aw_621_cast_fp16")]; + tensor var_8040_equation_0 = const()[name = tensor("op_8040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8040_cast_fp16 = einsum(equation = var_8040_equation_0, values = (var_7882_cast_fp16, var_7799_cast_fp16))[name = tensor("op_8040_cast_fp16")]; + tensor var_8041_to_fp16 = const()[name = tensor("op_8041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_623_cast_fp16 = mul(x = var_8040_cast_fp16, y = var_8041_to_fp16)[name = tensor("aw_623_cast_fp16")]; + tensor var_8044_equation_0 = const()[name = tensor("op_8044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8044_cast_fp16 = einsum(equation = var_8044_equation_0, values = (var_7886_cast_fp16, var_7803_cast_fp16))[name = tensor("op_8044_cast_fp16")]; + tensor var_8045_to_fp16 = const()[name = tensor("op_8045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_625_cast_fp16 = mul(x = var_8044_cast_fp16, y = var_8045_to_fp16)[name = tensor("aw_625_cast_fp16")]; + tensor var_8048_equation_0 = const()[name = tensor("op_8048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8048_cast_fp16 = einsum(equation = var_8048_equation_0, values = (var_7890_cast_fp16, var_7807_cast_fp16))[name = tensor("op_8048_cast_fp16")]; + tensor var_8049_to_fp16 = const()[name = tensor("op_8049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_627_cast_fp16 = mul(x = var_8048_cast_fp16, y = var_8049_to_fp16)[name = tensor("aw_627_cast_fp16")]; + tensor var_8052_equation_0 = const()[name = tensor("op_8052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8052_cast_fp16 = einsum(equation = var_8052_equation_0, values = (var_7894_cast_fp16, var_7811_cast_fp16))[name = tensor("op_8052_cast_fp16")]; + tensor var_8053_to_fp16 = const()[name = tensor("op_8053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_629_cast_fp16 = mul(x = var_8052_cast_fp16, y = var_8053_to_fp16)[name = tensor("aw_629_cast_fp16")]; + tensor var_8056_equation_0 = const()[name = tensor("op_8056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8056_cast_fp16 = einsum(equation = var_8056_equation_0, values = (var_7898_cast_fp16, var_7815_cast_fp16))[name = tensor("op_8056_cast_fp16")]; + tensor var_8057_to_fp16 = const()[name = tensor("op_8057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_631_cast_fp16 = mul(x = var_8056_cast_fp16, y = var_8057_to_fp16)[name = tensor("aw_631_cast_fp16")]; + tensor var_8060_equation_0 = const()[name = tensor("op_8060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8060_cast_fp16 = einsum(equation = var_8060_equation_0, values = (var_7902_cast_fp16, var_7819_cast_fp16))[name = tensor("op_8060_cast_fp16")]; + tensor var_8061_to_fp16 = const()[name = tensor("op_8061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_633_cast_fp16 = mul(x = var_8060_cast_fp16, y = var_8061_to_fp16)[name = tensor("aw_633_cast_fp16")]; + tensor var_8064_equation_0 = const()[name = tensor("op_8064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8064_cast_fp16 = einsum(equation = var_8064_equation_0, values = (var_7906_cast_fp16, var_7823_cast_fp16))[name = tensor("op_8064_cast_fp16")]; + tensor var_8065_to_fp16 = const()[name = tensor("op_8065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_635_cast_fp16 = mul(x = var_8064_cast_fp16, y = var_8065_to_fp16)[name = tensor("aw_635_cast_fp16")]; + tensor var_8068_equation_0 = const()[name = tensor("op_8068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8068_cast_fp16 = einsum(equation = var_8068_equation_0, values = (var_7910_cast_fp16, var_7827_cast_fp16))[name = tensor("op_8068_cast_fp16")]; + tensor var_8069_to_fp16 = const()[name = tensor("op_8069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_637_cast_fp16 = mul(x = var_8068_cast_fp16, y = var_8069_to_fp16)[name = tensor("aw_637_cast_fp16")]; + tensor var_8072_equation_0 = const()[name = tensor("op_8072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8072_cast_fp16 = einsum(equation = var_8072_equation_0, values = (var_7914_cast_fp16, var_7831_cast_fp16))[name = tensor("op_8072_cast_fp16")]; + tensor var_8073_to_fp16 = const()[name = tensor("op_8073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_639_cast_fp16 = mul(x = var_8072_cast_fp16, y = var_8073_to_fp16)[name = tensor("aw_639_cast_fp16")]; + tensor var_8075_cast_fp16 = softmax(axis = var_2624, x = aw_601_cast_fp16)[name = tensor("op_8075_cast_fp16")]; + tensor var_8076_cast_fp16 = softmax(axis = var_2624, x = aw_603_cast_fp16)[name = tensor("op_8076_cast_fp16")]; + tensor var_8077_cast_fp16 = softmax(axis = var_2624, x = aw_605_cast_fp16)[name = tensor("op_8077_cast_fp16")]; + tensor var_8078_cast_fp16 = softmax(axis = var_2624, x = aw_607_cast_fp16)[name = tensor("op_8078_cast_fp16")]; + tensor var_8079_cast_fp16 = softmax(axis = var_2624, x = aw_609_cast_fp16)[name = tensor("op_8079_cast_fp16")]; + tensor var_8080_cast_fp16 = softmax(axis = var_2624, x = aw_611_cast_fp16)[name = tensor("op_8080_cast_fp16")]; + tensor var_8081_cast_fp16 = softmax(axis = var_2624, x = aw_613_cast_fp16)[name = tensor("op_8081_cast_fp16")]; + tensor var_8082_cast_fp16 = softmax(axis = var_2624, x = aw_615_cast_fp16)[name = tensor("op_8082_cast_fp16")]; + tensor var_8083_cast_fp16 = softmax(axis = var_2624, x = aw_617_cast_fp16)[name = tensor("op_8083_cast_fp16")]; + tensor var_8084_cast_fp16 = softmax(axis = var_2624, x = aw_619_cast_fp16)[name = tensor("op_8084_cast_fp16")]; + tensor var_8085_cast_fp16 = softmax(axis = var_2624, x = aw_621_cast_fp16)[name = tensor("op_8085_cast_fp16")]; + tensor var_8086_cast_fp16 = softmax(axis = var_2624, x = aw_623_cast_fp16)[name = tensor("op_8086_cast_fp16")]; + tensor var_8087_cast_fp16 = softmax(axis = var_2624, x = aw_625_cast_fp16)[name = tensor("op_8087_cast_fp16")]; + tensor var_8088_cast_fp16 = softmax(axis = var_2624, x = aw_627_cast_fp16)[name = tensor("op_8088_cast_fp16")]; + tensor var_8089_cast_fp16 = softmax(axis = var_2624, x = aw_629_cast_fp16)[name = tensor("op_8089_cast_fp16")]; + tensor var_8090_cast_fp16 = softmax(axis = var_2624, x = aw_631_cast_fp16)[name = tensor("op_8090_cast_fp16")]; + tensor var_8091_cast_fp16 = softmax(axis = var_2624, x = aw_633_cast_fp16)[name = tensor("op_8091_cast_fp16")]; + tensor var_8092_cast_fp16 = softmax(axis = var_2624, x = aw_635_cast_fp16)[name = tensor("op_8092_cast_fp16")]; + tensor var_8093_cast_fp16 = softmax(axis = var_2624, x = aw_637_cast_fp16)[name = tensor("op_8093_cast_fp16")]; + tensor var_8094_cast_fp16 = softmax(axis = var_2624, x = aw_639_cast_fp16)[name = tensor("op_8094_cast_fp16")]; + tensor var_8096_equation_0 = const()[name = tensor("op_8096_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8096_cast_fp16 = einsum(equation = var_8096_equation_0, values = (var_7916_cast_fp16, var_8075_cast_fp16))[name = tensor("op_8096_cast_fp16")]; + tensor var_8098_equation_0 = const()[name = tensor("op_8098_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8098_cast_fp16 = einsum(equation = var_8098_equation_0, values = (var_7920_cast_fp16, var_8076_cast_fp16))[name = tensor("op_8098_cast_fp16")]; + tensor var_8100_equation_0 = const()[name = tensor("op_8100_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8100_cast_fp16 = einsum(equation = var_8100_equation_0, values = (var_7924_cast_fp16, var_8077_cast_fp16))[name = tensor("op_8100_cast_fp16")]; + tensor var_8102_equation_0 = const()[name = tensor("op_8102_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8102_cast_fp16 = einsum(equation = var_8102_equation_0, values = (var_7928_cast_fp16, var_8078_cast_fp16))[name = tensor("op_8102_cast_fp16")]; + tensor var_8104_equation_0 = const()[name = tensor("op_8104_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8104_cast_fp16 = einsum(equation = var_8104_equation_0, values = (var_7932_cast_fp16, var_8079_cast_fp16))[name = tensor("op_8104_cast_fp16")]; + tensor var_8106_equation_0 = const()[name = tensor("op_8106_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8106_cast_fp16 = einsum(equation = var_8106_equation_0, values = (var_7936_cast_fp16, var_8080_cast_fp16))[name = tensor("op_8106_cast_fp16")]; + tensor var_8108_equation_0 = const()[name = tensor("op_8108_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8108_cast_fp16 = einsum(equation = var_8108_equation_0, values = (var_7940_cast_fp16, var_8081_cast_fp16))[name = tensor("op_8108_cast_fp16")]; + tensor var_8110_equation_0 = const()[name = tensor("op_8110_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8110_cast_fp16 = einsum(equation = var_8110_equation_0, values = (var_7944_cast_fp16, var_8082_cast_fp16))[name = tensor("op_8110_cast_fp16")]; + tensor var_8112_equation_0 = const()[name = tensor("op_8112_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8112_cast_fp16 = einsum(equation = var_8112_equation_0, values = (var_7948_cast_fp16, var_8083_cast_fp16))[name = tensor("op_8112_cast_fp16")]; + tensor var_8114_equation_0 = const()[name = tensor("op_8114_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8114_cast_fp16 = einsum(equation = var_8114_equation_0, values = (var_7952_cast_fp16, var_8084_cast_fp16))[name = tensor("op_8114_cast_fp16")]; + tensor var_8116_equation_0 = const()[name = tensor("op_8116_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8116_cast_fp16 = einsum(equation = var_8116_equation_0, values = (var_7956_cast_fp16, var_8085_cast_fp16))[name = tensor("op_8116_cast_fp16")]; + tensor var_8118_equation_0 = const()[name = tensor("op_8118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8118_cast_fp16 = einsum(equation = var_8118_equation_0, values = (var_7960_cast_fp16, var_8086_cast_fp16))[name = tensor("op_8118_cast_fp16")]; + tensor var_8120_equation_0 = const()[name = tensor("op_8120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8120_cast_fp16 = einsum(equation = var_8120_equation_0, values = (var_7964_cast_fp16, var_8087_cast_fp16))[name = tensor("op_8120_cast_fp16")]; + tensor var_8122_equation_0 = const()[name = tensor("op_8122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8122_cast_fp16 = einsum(equation = var_8122_equation_0, values = (var_7968_cast_fp16, var_8088_cast_fp16))[name = tensor("op_8122_cast_fp16")]; + tensor var_8124_equation_0 = const()[name = tensor("op_8124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8124_cast_fp16 = einsum(equation = var_8124_equation_0, values = (var_7972_cast_fp16, var_8089_cast_fp16))[name = tensor("op_8124_cast_fp16")]; + tensor var_8126_equation_0 = const()[name = tensor("op_8126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8126_cast_fp16 = einsum(equation = var_8126_equation_0, values = (var_7976_cast_fp16, var_8090_cast_fp16))[name = tensor("op_8126_cast_fp16")]; + tensor var_8128_equation_0 = const()[name = tensor("op_8128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8128_cast_fp16 = einsum(equation = var_8128_equation_0, values = (var_7980_cast_fp16, var_8091_cast_fp16))[name = tensor("op_8128_cast_fp16")]; + tensor var_8130_equation_0 = const()[name = tensor("op_8130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8130_cast_fp16 = einsum(equation = var_8130_equation_0, values = (var_7984_cast_fp16, var_8092_cast_fp16))[name = tensor("op_8130_cast_fp16")]; + tensor var_8132_equation_0 = const()[name = tensor("op_8132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8132_cast_fp16 = einsum(equation = var_8132_equation_0, values = (var_7988_cast_fp16, var_8093_cast_fp16))[name = tensor("op_8132_cast_fp16")]; + tensor var_8134_equation_0 = const()[name = tensor("op_8134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8134_cast_fp16 = einsum(equation = var_8134_equation_0, values = (var_7992_cast_fp16, var_8094_cast_fp16))[name = tensor("op_8134_cast_fp16")]; + tensor input_173_interleave_0 = const()[name = tensor("input_173_interleave_0"), val = tensor(false)]; + tensor input_173_cast_fp16 = concat(axis = var_2624, interleave = input_173_interleave_0, values = (var_8096_cast_fp16, var_8098_cast_fp16, var_8100_cast_fp16, var_8102_cast_fp16, var_8104_cast_fp16, var_8106_cast_fp16, var_8108_cast_fp16, var_8110_cast_fp16, var_8112_cast_fp16, var_8114_cast_fp16, var_8116_cast_fp16, var_8118_cast_fp16, var_8120_cast_fp16, var_8122_cast_fp16, var_8124_cast_fp16, var_8126_cast_fp16, var_8128_cast_fp16, var_8130_cast_fp16, var_8132_cast_fp16, var_8134_cast_fp16))[name = tensor("input_173_cast_fp16")]; + tensor var_8144_pad_type_0 = const()[name = tensor("op_8144_pad_type_0"), val = tensor("valid")]; + tensor var_8144_strides_0 = const()[name = tensor("op_8144_strides_0"), val = tensor([1, 1])]; + tensor var_8144_pad_0 = const()[name = tensor("op_8144_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8144_dilations_0 = const()[name = tensor("op_8144_dilations_0"), val = tensor([1, 1])]; + tensor var_8144_groups_0 = const()[name = tensor("op_8144_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215017408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216246272))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216246464)))]; + tensor var_8144_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_8144_dilations_0, groups = var_8144_groups_0, pad = var_8144_pad_0, pad_type = var_8144_pad_type_0, strides = var_8144_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("op_8144_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = var_8144_cast_fp16, y = inputs_57_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; + tensor input_175_axes_0 = const()[name = tensor("input_175_axes_0"), val = tensor([1])]; + tensor input_175_gamma_0_to_fp16 = const()[name = tensor("input_175_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216249088)))]; + tensor input_175_beta_0_to_fp16 = const()[name = tensor("input_175_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216251712)))]; + tensor var_8154_to_fp16 = const()[name = tensor("op_8154_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = input_175_beta_0_to_fp16, epsilon = var_8154_to_fp16, gamma = input_175_gamma_0_to_fp16, x = inputs_59_cast_fp16)[name = tensor("input_175_cast_fp16")]; + tensor var_8174_pad_type_0 = const()[name = tensor("op_8174_pad_type_0"), val = tensor("valid")]; + tensor var_8174_strides_0 = const()[name = tensor("op_8174_strides_0"), val = tensor([1, 1])]; + tensor var_8174_pad_0 = const()[name = tensor("op_8174_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8174_dilations_0 = const()[name = tensor("op_8174_dilations_0"), val = tensor([1, 1])]; + tensor var_8174_groups_0 = const()[name = tensor("op_8174_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216254336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226084800))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226084992)))]; + tensor var_8174_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_8174_dilations_0, groups = var_8174_groups_0, pad = var_8174_pad_0, pad_type = var_8174_pad_type_0, strides = var_8174_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = tensor("op_8174_cast_fp16")]; + tensor var_8175_split_sizes_0 = const()[name = tensor("op_8175_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_8175_axis_0 = const()[name = tensor("op_8175_axis_0"), val = tensor(1)]; + tensor var_8175_cast_fp16_0, tensor var_8175_cast_fp16_1 = split(axis = var_8175_axis_0, split_sizes = var_8175_split_sizes_0, x = var_8174_cast_fp16)[name = tensor("op_8175_cast_fp16")]; + tensor var_8177_mode_0 = const()[name = tensor("op_8177_mode_0"), val = tensor("EXACT")]; + tensor var_8177_cast_fp16 = gelu(mode = var_8177_mode_0, x = var_8175_cast_fp16_1)[name = tensor("op_8177_cast_fp16")]; + tensor input_177_cast_fp16 = mul(x = var_8175_cast_fp16_0, y = var_8177_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor var_8185_pad_type_0 = const()[name = tensor("op_8185_pad_type_0"), val = tensor("valid")]; + tensor var_8185_strides_0 = const()[name = tensor("op_8185_strides_0"), val = tensor([1, 1])]; + tensor var_8185_pad_0 = const()[name = tensor("op_8185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8185_dilations_0 = const()[name = tensor("op_8185_dilations_0"), val = tensor([1, 1])]; + tensor var_8185_groups_0 = const()[name = tensor("op_8185_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226105536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231020800))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231020992)))]; + tensor var_8185_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_8185_dilations_0, groups = var_8185_groups_0, pad = var_8185_pad_0, pad_type = var_8185_pad_type_0, strides = var_8185_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor("op_8185_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = var_8185_cast_fp16, y = inputs_59_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; + tensor hidden_states_101_axes_0 = const()[name = tensor("hidden_states_101_axes_0"), val = tensor([1])]; + tensor hidden_states_101_gamma_0_to_fp16 = const()[name = tensor("hidden_states_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231023616)))]; + tensor hidden_states_101_beta_0_to_fp16 = const()[name = tensor("hidden_states_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231026240)))]; + tensor var_8201_to_fp16 = const()[name = tensor("op_8201_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_101_cast_fp16 = layer_norm(axes = hidden_states_101_axes_0, beta = hidden_states_101_beta_0_to_fp16, epsilon = var_8201_to_fp16, gamma = hidden_states_101_gamma_0_to_fp16, x = inputs_61_cast_fp16)[name = tensor("hidden_states_101_cast_fp16")]; + tensor q_41_pad_type_0 = const()[name = tensor("q_41_pad_type_0"), val = tensor("valid")]; + tensor q_41_strides_0 = const()[name = tensor("q_41_strides_0"), val = tensor([1, 1])]; + tensor q_41_pad_0 = const()[name = tensor("q_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_41_dilations_0 = const()[name = tensor("q_41_dilations_0"), val = tensor([1, 1])]; + tensor q_41_groups_0 = const()[name = tensor("q_41_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231028864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232257728))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_41_cast_fp16 = conv(dilations = q_41_dilations_0, groups = q_41_groups_0, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = q_41_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_101_cast_fp16)[name = tensor("q_41_cast_fp16")]; + tensor k_81_pad_type_0 = const()[name = tensor("k_81_pad_type_0"), val = tensor("valid")]; + tensor k_81_strides_0 = const()[name = tensor("k_81_strides_0"), val = tensor([1, 1])]; + tensor k_81_pad_0 = const()[name = tensor("k_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_81_dilations_0 = const()[name = tensor("k_81_dilations_0"), val = tensor([1, 1])]; + tensor k_81_groups_0 = const()[name = tensor("k_81_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232257920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233486784))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_81_cast_fp16 = conv(dilations = k_81_dilations_0, groups = k_81_groups_0, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = k_81_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_101_cast_fp16)[name = tensor("k_81_cast_fp16")]; + tensor v_41_pad_type_0 = const()[name = tensor("v_41_pad_type_0"), val = tensor("valid")]; + tensor v_41_strides_0 = const()[name = tensor("v_41_strides_0"), val = tensor([1, 1])]; + tensor v_41_pad_0 = const()[name = tensor("v_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_41_dilations_0 = const()[name = tensor("v_41_dilations_0"), val = tensor([1, 1])]; + tensor v_41_groups_0 = const()[name = tensor("v_41_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233486976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234715840))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_41_cast_fp16 = conv(dilations = v_41_dilations_0, groups = v_41_groups_0, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = v_41_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_101_cast_fp16)[name = tensor("v_41_cast_fp16")]; + tensor var_8234_begin_0 = const()[name = tensor("op_8234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8234_end_0 = const()[name = tensor("op_8234_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_8234_end_mask_0 = const()[name = tensor("op_8234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8234_cast_fp16 = slice_by_index(begin = var_8234_begin_0, end = var_8234_end_0, end_mask = var_8234_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8234_cast_fp16")]; + tensor var_8238_begin_0 = const()[name = tensor("op_8238_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8238_end_0 = const()[name = tensor("op_8238_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_8238_end_mask_0 = const()[name = tensor("op_8238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8238_cast_fp16 = slice_by_index(begin = var_8238_begin_0, end = var_8238_end_0, end_mask = var_8238_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8238_cast_fp16")]; + tensor var_8242_begin_0 = const()[name = tensor("op_8242_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8242_end_0 = const()[name = tensor("op_8242_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_8242_end_mask_0 = const()[name = tensor("op_8242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8242_cast_fp16 = slice_by_index(begin = var_8242_begin_0, end = var_8242_end_0, end_mask = var_8242_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8242_cast_fp16")]; + tensor var_8246_begin_0 = const()[name = tensor("op_8246_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8246_end_0 = const()[name = tensor("op_8246_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_8246_end_mask_0 = const()[name = tensor("op_8246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8246_cast_fp16 = slice_by_index(begin = var_8246_begin_0, end = var_8246_end_0, end_mask = var_8246_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8246_cast_fp16")]; + tensor var_8250_begin_0 = const()[name = tensor("op_8250_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8250_end_0 = const()[name = tensor("op_8250_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_8250_end_mask_0 = const()[name = tensor("op_8250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8250_cast_fp16 = slice_by_index(begin = var_8250_begin_0, end = var_8250_end_0, end_mask = var_8250_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8250_cast_fp16")]; + tensor var_8254_begin_0 = const()[name = tensor("op_8254_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8254_end_0 = const()[name = tensor("op_8254_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_8254_end_mask_0 = const()[name = tensor("op_8254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8254_cast_fp16 = slice_by_index(begin = var_8254_begin_0, end = var_8254_end_0, end_mask = var_8254_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8254_cast_fp16")]; + tensor var_8258_begin_0 = const()[name = tensor("op_8258_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8258_end_0 = const()[name = tensor("op_8258_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_8258_end_mask_0 = const()[name = tensor("op_8258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8258_cast_fp16 = slice_by_index(begin = var_8258_begin_0, end = var_8258_end_0, end_mask = var_8258_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8258_cast_fp16")]; + tensor var_8262_begin_0 = const()[name = tensor("op_8262_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8262_end_0 = const()[name = tensor("op_8262_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_8262_end_mask_0 = const()[name = tensor("op_8262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8262_cast_fp16 = slice_by_index(begin = var_8262_begin_0, end = var_8262_end_0, end_mask = var_8262_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8262_cast_fp16")]; + tensor var_8266_begin_0 = const()[name = tensor("op_8266_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8266_end_0 = const()[name = tensor("op_8266_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_8266_end_mask_0 = const()[name = tensor("op_8266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8266_cast_fp16 = slice_by_index(begin = var_8266_begin_0, end = var_8266_end_0, end_mask = var_8266_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8266_cast_fp16")]; + tensor var_8270_begin_0 = const()[name = tensor("op_8270_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8270_end_0 = const()[name = tensor("op_8270_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_8270_end_mask_0 = const()[name = tensor("op_8270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8270_cast_fp16 = slice_by_index(begin = var_8270_begin_0, end = var_8270_end_0, end_mask = var_8270_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8270_cast_fp16")]; + tensor var_8274_begin_0 = const()[name = tensor("op_8274_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8274_end_0 = const()[name = tensor("op_8274_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_8274_end_mask_0 = const()[name = tensor("op_8274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8274_cast_fp16 = slice_by_index(begin = var_8274_begin_0, end = var_8274_end_0, end_mask = var_8274_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8274_cast_fp16")]; + tensor var_8278_begin_0 = const()[name = tensor("op_8278_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8278_end_0 = const()[name = tensor("op_8278_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_8278_end_mask_0 = const()[name = tensor("op_8278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8278_cast_fp16 = slice_by_index(begin = var_8278_begin_0, end = var_8278_end_0, end_mask = var_8278_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8278_cast_fp16")]; + tensor var_8282_begin_0 = const()[name = tensor("op_8282_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8282_end_0 = const()[name = tensor("op_8282_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_8282_end_mask_0 = const()[name = tensor("op_8282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8282_cast_fp16 = slice_by_index(begin = var_8282_begin_0, end = var_8282_end_0, end_mask = var_8282_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8282_cast_fp16")]; + tensor var_8286_begin_0 = const()[name = tensor("op_8286_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8286_end_0 = const()[name = tensor("op_8286_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_8286_end_mask_0 = const()[name = tensor("op_8286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8286_cast_fp16 = slice_by_index(begin = var_8286_begin_0, end = var_8286_end_0, end_mask = var_8286_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8286_cast_fp16")]; + tensor var_8290_begin_0 = const()[name = tensor("op_8290_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8290_end_0 = const()[name = tensor("op_8290_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_8290_end_mask_0 = const()[name = tensor("op_8290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8290_cast_fp16 = slice_by_index(begin = var_8290_begin_0, end = var_8290_end_0, end_mask = var_8290_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8290_cast_fp16")]; + tensor var_8294_begin_0 = const()[name = tensor("op_8294_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8294_end_0 = const()[name = tensor("op_8294_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8294_end_mask_0 = const()[name = tensor("op_8294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8294_cast_fp16 = slice_by_index(begin = var_8294_begin_0, end = var_8294_end_0, end_mask = var_8294_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8294_cast_fp16")]; + tensor var_8298_begin_0 = const()[name = tensor("op_8298_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8298_end_0 = const()[name = tensor("op_8298_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_8298_end_mask_0 = const()[name = tensor("op_8298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8298_cast_fp16 = slice_by_index(begin = var_8298_begin_0, end = var_8298_end_0, end_mask = var_8298_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8298_cast_fp16")]; + tensor var_8302_begin_0 = const()[name = tensor("op_8302_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8302_end_0 = const()[name = tensor("op_8302_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_8302_end_mask_0 = const()[name = tensor("op_8302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8302_cast_fp16 = slice_by_index(begin = var_8302_begin_0, end = var_8302_end_0, end_mask = var_8302_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8302_cast_fp16")]; + tensor var_8306_begin_0 = const()[name = tensor("op_8306_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8306_end_0 = const()[name = tensor("op_8306_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_8306_end_mask_0 = const()[name = tensor("op_8306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8306_cast_fp16 = slice_by_index(begin = var_8306_begin_0, end = var_8306_end_0, end_mask = var_8306_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8306_cast_fp16")]; + tensor var_8310_begin_0 = const()[name = tensor("op_8310_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8310_end_0 = const()[name = tensor("op_8310_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_8310_end_mask_0 = const()[name = tensor("op_8310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8310_cast_fp16 = slice_by_index(begin = var_8310_begin_0, end = var_8310_end_0, end_mask = var_8310_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8310_cast_fp16")]; + tensor k_83_perm_0 = const()[name = tensor("k_83_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8317_begin_0 = const()[name = tensor("op_8317_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8317_end_0 = const()[name = tensor("op_8317_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_8317_end_mask_0 = const()[name = tensor("op_8317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_83_cast_fp16 = transpose(perm = k_83_perm_0, x = k_81_cast_fp16)[name = tensor("transpose_47")]; + tensor var_8317_cast_fp16 = slice_by_index(begin = var_8317_begin_0, end = var_8317_end_0, end_mask = var_8317_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8317_cast_fp16")]; + tensor var_8321_begin_0 = const()[name = tensor("op_8321_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8321_end_0 = const()[name = tensor("op_8321_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_8321_end_mask_0 = const()[name = tensor("op_8321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8321_cast_fp16 = slice_by_index(begin = var_8321_begin_0, end = var_8321_end_0, end_mask = var_8321_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8321_cast_fp16")]; + tensor var_8325_begin_0 = const()[name = tensor("op_8325_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8325_end_0 = const()[name = tensor("op_8325_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_8325_end_mask_0 = const()[name = tensor("op_8325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8325_cast_fp16 = slice_by_index(begin = var_8325_begin_0, end = var_8325_end_0, end_mask = var_8325_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8325_cast_fp16")]; + tensor var_8329_begin_0 = const()[name = tensor("op_8329_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8329_end_0 = const()[name = tensor("op_8329_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_8329_end_mask_0 = const()[name = tensor("op_8329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8329_cast_fp16 = slice_by_index(begin = var_8329_begin_0, end = var_8329_end_0, end_mask = var_8329_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8329_cast_fp16")]; + tensor var_8333_begin_0 = const()[name = tensor("op_8333_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8333_end_0 = const()[name = tensor("op_8333_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_8333_end_mask_0 = const()[name = tensor("op_8333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8333_cast_fp16 = slice_by_index(begin = var_8333_begin_0, end = var_8333_end_0, end_mask = var_8333_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8333_cast_fp16")]; + tensor var_8337_begin_0 = const()[name = tensor("op_8337_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8337_end_0 = const()[name = tensor("op_8337_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_8337_end_mask_0 = const()[name = tensor("op_8337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8337_cast_fp16 = slice_by_index(begin = var_8337_begin_0, end = var_8337_end_0, end_mask = var_8337_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8337_cast_fp16")]; + tensor var_8341_begin_0 = const()[name = tensor("op_8341_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8341_end_0 = const()[name = tensor("op_8341_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_8341_end_mask_0 = const()[name = tensor("op_8341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8341_cast_fp16 = slice_by_index(begin = var_8341_begin_0, end = var_8341_end_0, end_mask = var_8341_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8341_cast_fp16")]; + tensor var_8345_begin_0 = const()[name = tensor("op_8345_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8345_end_0 = const()[name = tensor("op_8345_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_8345_end_mask_0 = const()[name = tensor("op_8345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8345_cast_fp16 = slice_by_index(begin = var_8345_begin_0, end = var_8345_end_0, end_mask = var_8345_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8345_cast_fp16")]; + tensor var_8349_begin_0 = const()[name = tensor("op_8349_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8349_end_0 = const()[name = tensor("op_8349_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_8349_end_mask_0 = const()[name = tensor("op_8349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8349_cast_fp16 = slice_by_index(begin = var_8349_begin_0, end = var_8349_end_0, end_mask = var_8349_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8349_cast_fp16")]; + tensor var_8353_begin_0 = const()[name = tensor("op_8353_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8353_end_0 = const()[name = tensor("op_8353_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_8353_end_mask_0 = const()[name = tensor("op_8353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8353_cast_fp16 = slice_by_index(begin = var_8353_begin_0, end = var_8353_end_0, end_mask = var_8353_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8353_cast_fp16")]; + tensor var_8357_begin_0 = const()[name = tensor("op_8357_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8357_end_0 = const()[name = tensor("op_8357_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_8357_end_mask_0 = const()[name = tensor("op_8357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8357_cast_fp16 = slice_by_index(begin = var_8357_begin_0, end = var_8357_end_0, end_mask = var_8357_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8357_cast_fp16")]; + tensor var_8361_begin_0 = const()[name = tensor("op_8361_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8361_end_0 = const()[name = tensor("op_8361_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_8361_end_mask_0 = const()[name = tensor("op_8361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8361_cast_fp16 = slice_by_index(begin = var_8361_begin_0, end = var_8361_end_0, end_mask = var_8361_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8361_cast_fp16")]; + tensor var_8365_begin_0 = const()[name = tensor("op_8365_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8365_end_0 = const()[name = tensor("op_8365_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_8365_end_mask_0 = const()[name = tensor("op_8365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8365_cast_fp16 = slice_by_index(begin = var_8365_begin_0, end = var_8365_end_0, end_mask = var_8365_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8365_cast_fp16")]; + tensor var_8369_begin_0 = const()[name = tensor("op_8369_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8369_end_0 = const()[name = tensor("op_8369_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_8369_end_mask_0 = const()[name = tensor("op_8369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8369_cast_fp16 = slice_by_index(begin = var_8369_begin_0, end = var_8369_end_0, end_mask = var_8369_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8369_cast_fp16")]; + tensor var_8373_begin_0 = const()[name = tensor("op_8373_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8373_end_0 = const()[name = tensor("op_8373_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_8373_end_mask_0 = const()[name = tensor("op_8373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8373_cast_fp16 = slice_by_index(begin = var_8373_begin_0, end = var_8373_end_0, end_mask = var_8373_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8373_cast_fp16")]; + tensor var_8377_begin_0 = const()[name = tensor("op_8377_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8377_end_0 = const()[name = tensor("op_8377_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8377_end_mask_0 = const()[name = tensor("op_8377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8377_cast_fp16 = slice_by_index(begin = var_8377_begin_0, end = var_8377_end_0, end_mask = var_8377_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8377_cast_fp16")]; + tensor var_8381_begin_0 = const()[name = tensor("op_8381_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8381_end_0 = const()[name = tensor("op_8381_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_8381_end_mask_0 = const()[name = tensor("op_8381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8381_cast_fp16 = slice_by_index(begin = var_8381_begin_0, end = var_8381_end_0, end_mask = var_8381_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8381_cast_fp16")]; + tensor var_8385_begin_0 = const()[name = tensor("op_8385_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8385_end_0 = const()[name = tensor("op_8385_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_8385_end_mask_0 = const()[name = tensor("op_8385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8385_cast_fp16 = slice_by_index(begin = var_8385_begin_0, end = var_8385_end_0, end_mask = var_8385_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8385_cast_fp16")]; + tensor var_8389_begin_0 = const()[name = tensor("op_8389_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8389_end_0 = const()[name = tensor("op_8389_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_8389_end_mask_0 = const()[name = tensor("op_8389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8389_cast_fp16 = slice_by_index(begin = var_8389_begin_0, end = var_8389_end_0, end_mask = var_8389_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8389_cast_fp16")]; + tensor var_8393_begin_0 = const()[name = tensor("op_8393_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8393_end_0 = const()[name = tensor("op_8393_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_8393_end_mask_0 = const()[name = tensor("op_8393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8393_cast_fp16 = slice_by_index(begin = var_8393_begin_0, end = var_8393_end_0, end_mask = var_8393_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8393_cast_fp16")]; + tensor var_8395_begin_0 = const()[name = tensor("op_8395_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8395_end_0 = const()[name = tensor("op_8395_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_8395_end_mask_0 = const()[name = tensor("op_8395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8395_cast_fp16 = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8395_cast_fp16")]; + tensor var_8399_begin_0 = const()[name = tensor("op_8399_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8399_end_0 = const()[name = tensor("op_8399_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_8399_end_mask_0 = const()[name = tensor("op_8399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8399_cast_fp16 = slice_by_index(begin = var_8399_begin_0, end = var_8399_end_0, end_mask = var_8399_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8399_cast_fp16")]; + tensor var_8403_begin_0 = const()[name = tensor("op_8403_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8403_end_0 = const()[name = tensor("op_8403_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_8403_end_mask_0 = const()[name = tensor("op_8403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8403_cast_fp16 = slice_by_index(begin = var_8403_begin_0, end = var_8403_end_0, end_mask = var_8403_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8403_cast_fp16")]; + tensor var_8407_begin_0 = const()[name = tensor("op_8407_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8407_end_0 = const()[name = tensor("op_8407_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_8407_end_mask_0 = const()[name = tensor("op_8407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8407_cast_fp16 = slice_by_index(begin = var_8407_begin_0, end = var_8407_end_0, end_mask = var_8407_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8407_cast_fp16")]; + tensor var_8411_begin_0 = const()[name = tensor("op_8411_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8411_end_0 = const()[name = tensor("op_8411_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_8411_end_mask_0 = const()[name = tensor("op_8411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8411_cast_fp16 = slice_by_index(begin = var_8411_begin_0, end = var_8411_end_0, end_mask = var_8411_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8411_cast_fp16")]; + tensor var_8415_begin_0 = const()[name = tensor("op_8415_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8415_end_0 = const()[name = tensor("op_8415_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_8415_end_mask_0 = const()[name = tensor("op_8415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8415_cast_fp16 = slice_by_index(begin = var_8415_begin_0, end = var_8415_end_0, end_mask = var_8415_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8415_cast_fp16")]; + tensor var_8419_begin_0 = const()[name = tensor("op_8419_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8419_end_0 = const()[name = tensor("op_8419_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_8419_end_mask_0 = const()[name = tensor("op_8419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8419_cast_fp16 = slice_by_index(begin = var_8419_begin_0, end = var_8419_end_0, end_mask = var_8419_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8419_cast_fp16")]; + tensor var_8423_begin_0 = const()[name = tensor("op_8423_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8423_end_0 = const()[name = tensor("op_8423_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_8423_end_mask_0 = const()[name = tensor("op_8423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8423_cast_fp16 = slice_by_index(begin = var_8423_begin_0, end = var_8423_end_0, end_mask = var_8423_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8423_cast_fp16")]; + tensor var_8427_begin_0 = const()[name = tensor("op_8427_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8427_end_0 = const()[name = tensor("op_8427_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_8427_end_mask_0 = const()[name = tensor("op_8427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8427_cast_fp16 = slice_by_index(begin = var_8427_begin_0, end = var_8427_end_0, end_mask = var_8427_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8427_cast_fp16")]; + tensor var_8431_begin_0 = const()[name = tensor("op_8431_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8431_end_0 = const()[name = tensor("op_8431_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_8431_end_mask_0 = const()[name = tensor("op_8431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8431_cast_fp16 = slice_by_index(begin = var_8431_begin_0, end = var_8431_end_0, end_mask = var_8431_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8431_cast_fp16")]; + tensor var_8435_begin_0 = const()[name = tensor("op_8435_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8435_end_0 = const()[name = tensor("op_8435_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_8435_end_mask_0 = const()[name = tensor("op_8435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8435_cast_fp16 = slice_by_index(begin = var_8435_begin_0, end = var_8435_end_0, end_mask = var_8435_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8435_cast_fp16")]; + tensor var_8439_begin_0 = const()[name = tensor("op_8439_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8439_end_0 = const()[name = tensor("op_8439_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_8439_end_mask_0 = const()[name = tensor("op_8439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8439_cast_fp16 = slice_by_index(begin = var_8439_begin_0, end = var_8439_end_0, end_mask = var_8439_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8439_cast_fp16")]; + tensor var_8443_begin_0 = const()[name = tensor("op_8443_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8443_end_0 = const()[name = tensor("op_8443_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_8443_end_mask_0 = const()[name = tensor("op_8443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8443_cast_fp16 = slice_by_index(begin = var_8443_begin_0, end = var_8443_end_0, end_mask = var_8443_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8443_cast_fp16")]; + tensor var_8447_begin_0 = const()[name = tensor("op_8447_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8447_end_0 = const()[name = tensor("op_8447_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_8447_end_mask_0 = const()[name = tensor("op_8447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8447_cast_fp16 = slice_by_index(begin = var_8447_begin_0, end = var_8447_end_0, end_mask = var_8447_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8447_cast_fp16")]; + tensor var_8451_begin_0 = const()[name = tensor("op_8451_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8451_end_0 = const()[name = tensor("op_8451_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_8451_end_mask_0 = const()[name = tensor("op_8451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8451_cast_fp16 = slice_by_index(begin = var_8451_begin_0, end = var_8451_end_0, end_mask = var_8451_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8451_cast_fp16")]; + tensor var_8455_begin_0 = const()[name = tensor("op_8455_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8455_end_0 = const()[name = tensor("op_8455_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8455_end_mask_0 = const()[name = tensor("op_8455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8455_cast_fp16 = slice_by_index(begin = var_8455_begin_0, end = var_8455_end_0, end_mask = var_8455_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8455_cast_fp16")]; + tensor var_8459_begin_0 = const()[name = tensor("op_8459_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8459_end_0 = const()[name = tensor("op_8459_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_8459_end_mask_0 = const()[name = tensor("op_8459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8459_cast_fp16 = slice_by_index(begin = var_8459_begin_0, end = var_8459_end_0, end_mask = var_8459_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8459_cast_fp16")]; + tensor var_8463_begin_0 = const()[name = tensor("op_8463_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8463_end_0 = const()[name = tensor("op_8463_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_8463_end_mask_0 = const()[name = tensor("op_8463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8463_cast_fp16 = slice_by_index(begin = var_8463_begin_0, end = var_8463_end_0, end_mask = var_8463_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8463_cast_fp16")]; + tensor var_8467_begin_0 = const()[name = tensor("op_8467_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8467_end_0 = const()[name = tensor("op_8467_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_8467_end_mask_0 = const()[name = tensor("op_8467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8467_cast_fp16 = slice_by_index(begin = var_8467_begin_0, end = var_8467_end_0, end_mask = var_8467_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8467_cast_fp16")]; + tensor var_8471_begin_0 = const()[name = tensor("op_8471_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8471_end_0 = const()[name = tensor("op_8471_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_8471_end_mask_0 = const()[name = tensor("op_8471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8471_cast_fp16 = slice_by_index(begin = var_8471_begin_0, end = var_8471_end_0, end_mask = var_8471_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8471_cast_fp16")]; + tensor var_8475_equation_0 = const()[name = tensor("op_8475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8475_cast_fp16 = einsum(equation = var_8475_equation_0, values = (var_8317_cast_fp16, var_8234_cast_fp16))[name = tensor("op_8475_cast_fp16")]; + tensor var_8476_to_fp16 = const()[name = tensor("op_8476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_641_cast_fp16 = mul(x = var_8475_cast_fp16, y = var_8476_to_fp16)[name = tensor("aw_641_cast_fp16")]; + tensor var_8479_equation_0 = const()[name = tensor("op_8479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8479_cast_fp16 = einsum(equation = var_8479_equation_0, values = (var_8321_cast_fp16, var_8238_cast_fp16))[name = tensor("op_8479_cast_fp16")]; + tensor var_8480_to_fp16 = const()[name = tensor("op_8480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_643_cast_fp16 = mul(x = var_8479_cast_fp16, y = var_8480_to_fp16)[name = tensor("aw_643_cast_fp16")]; + tensor var_8483_equation_0 = const()[name = tensor("op_8483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8483_cast_fp16 = einsum(equation = var_8483_equation_0, values = (var_8325_cast_fp16, var_8242_cast_fp16))[name = tensor("op_8483_cast_fp16")]; + tensor var_8484_to_fp16 = const()[name = tensor("op_8484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_645_cast_fp16 = mul(x = var_8483_cast_fp16, y = var_8484_to_fp16)[name = tensor("aw_645_cast_fp16")]; + tensor var_8487_equation_0 = const()[name = tensor("op_8487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8487_cast_fp16 = einsum(equation = var_8487_equation_0, values = (var_8329_cast_fp16, var_8246_cast_fp16))[name = tensor("op_8487_cast_fp16")]; + tensor var_8488_to_fp16 = const()[name = tensor("op_8488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_647_cast_fp16 = mul(x = var_8487_cast_fp16, y = var_8488_to_fp16)[name = tensor("aw_647_cast_fp16")]; + tensor var_8491_equation_0 = const()[name = tensor("op_8491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8491_cast_fp16 = einsum(equation = var_8491_equation_0, values = (var_8333_cast_fp16, var_8250_cast_fp16))[name = tensor("op_8491_cast_fp16")]; + tensor var_8492_to_fp16 = const()[name = tensor("op_8492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_649_cast_fp16 = mul(x = var_8491_cast_fp16, y = var_8492_to_fp16)[name = tensor("aw_649_cast_fp16")]; + tensor var_8495_equation_0 = const()[name = tensor("op_8495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8495_cast_fp16 = einsum(equation = var_8495_equation_0, values = (var_8337_cast_fp16, var_8254_cast_fp16))[name = tensor("op_8495_cast_fp16")]; + tensor var_8496_to_fp16 = const()[name = tensor("op_8496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_651_cast_fp16 = mul(x = var_8495_cast_fp16, y = var_8496_to_fp16)[name = tensor("aw_651_cast_fp16")]; + tensor var_8499_equation_0 = const()[name = tensor("op_8499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8499_cast_fp16 = einsum(equation = var_8499_equation_0, values = (var_8341_cast_fp16, var_8258_cast_fp16))[name = tensor("op_8499_cast_fp16")]; + tensor var_8500_to_fp16 = const()[name = tensor("op_8500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_653_cast_fp16 = mul(x = var_8499_cast_fp16, y = var_8500_to_fp16)[name = tensor("aw_653_cast_fp16")]; + tensor var_8503_equation_0 = const()[name = tensor("op_8503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8503_cast_fp16 = einsum(equation = var_8503_equation_0, values = (var_8345_cast_fp16, var_8262_cast_fp16))[name = tensor("op_8503_cast_fp16")]; + tensor var_8504_to_fp16 = const()[name = tensor("op_8504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_655_cast_fp16 = mul(x = var_8503_cast_fp16, y = var_8504_to_fp16)[name = tensor("aw_655_cast_fp16")]; + tensor var_8507_equation_0 = const()[name = tensor("op_8507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8507_cast_fp16 = einsum(equation = var_8507_equation_0, values = (var_8349_cast_fp16, var_8266_cast_fp16))[name = tensor("op_8507_cast_fp16")]; + tensor var_8508_to_fp16 = const()[name = tensor("op_8508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_657_cast_fp16 = mul(x = var_8507_cast_fp16, y = var_8508_to_fp16)[name = tensor("aw_657_cast_fp16")]; + tensor var_8511_equation_0 = const()[name = tensor("op_8511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8511_cast_fp16 = einsum(equation = var_8511_equation_0, values = (var_8353_cast_fp16, var_8270_cast_fp16))[name = tensor("op_8511_cast_fp16")]; + tensor var_8512_to_fp16 = const()[name = tensor("op_8512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_659_cast_fp16 = mul(x = var_8511_cast_fp16, y = var_8512_to_fp16)[name = tensor("aw_659_cast_fp16")]; + tensor var_8515_equation_0 = const()[name = tensor("op_8515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8515_cast_fp16 = einsum(equation = var_8515_equation_0, values = (var_8357_cast_fp16, var_8274_cast_fp16))[name = tensor("op_8515_cast_fp16")]; + tensor var_8516_to_fp16 = const()[name = tensor("op_8516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_661_cast_fp16 = mul(x = var_8515_cast_fp16, y = var_8516_to_fp16)[name = tensor("aw_661_cast_fp16")]; + tensor var_8519_equation_0 = const()[name = tensor("op_8519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8519_cast_fp16 = einsum(equation = var_8519_equation_0, values = (var_8361_cast_fp16, var_8278_cast_fp16))[name = tensor("op_8519_cast_fp16")]; + tensor var_8520_to_fp16 = const()[name = tensor("op_8520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_663_cast_fp16 = mul(x = var_8519_cast_fp16, y = var_8520_to_fp16)[name = tensor("aw_663_cast_fp16")]; + tensor var_8523_equation_0 = const()[name = tensor("op_8523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8523_cast_fp16 = einsum(equation = var_8523_equation_0, values = (var_8365_cast_fp16, var_8282_cast_fp16))[name = tensor("op_8523_cast_fp16")]; + tensor var_8524_to_fp16 = const()[name = tensor("op_8524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_665_cast_fp16 = mul(x = var_8523_cast_fp16, y = var_8524_to_fp16)[name = tensor("aw_665_cast_fp16")]; + tensor var_8527_equation_0 = const()[name = tensor("op_8527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8527_cast_fp16 = einsum(equation = var_8527_equation_0, values = (var_8369_cast_fp16, var_8286_cast_fp16))[name = tensor("op_8527_cast_fp16")]; + tensor var_8528_to_fp16 = const()[name = tensor("op_8528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_667_cast_fp16 = mul(x = var_8527_cast_fp16, y = var_8528_to_fp16)[name = tensor("aw_667_cast_fp16")]; + tensor var_8531_equation_0 = const()[name = tensor("op_8531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8531_cast_fp16 = einsum(equation = var_8531_equation_0, values = (var_8373_cast_fp16, var_8290_cast_fp16))[name = tensor("op_8531_cast_fp16")]; + tensor var_8532_to_fp16 = const()[name = tensor("op_8532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_669_cast_fp16 = mul(x = var_8531_cast_fp16, y = var_8532_to_fp16)[name = tensor("aw_669_cast_fp16")]; + tensor var_8535_equation_0 = const()[name = tensor("op_8535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8535_cast_fp16 = einsum(equation = var_8535_equation_0, values = (var_8377_cast_fp16, var_8294_cast_fp16))[name = tensor("op_8535_cast_fp16")]; + tensor var_8536_to_fp16 = const()[name = tensor("op_8536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_671_cast_fp16 = mul(x = var_8535_cast_fp16, y = var_8536_to_fp16)[name = tensor("aw_671_cast_fp16")]; + tensor var_8539_equation_0 = const()[name = tensor("op_8539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8539_cast_fp16 = einsum(equation = var_8539_equation_0, values = (var_8381_cast_fp16, var_8298_cast_fp16))[name = tensor("op_8539_cast_fp16")]; + tensor var_8540_to_fp16 = const()[name = tensor("op_8540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_673_cast_fp16 = mul(x = var_8539_cast_fp16, y = var_8540_to_fp16)[name = tensor("aw_673_cast_fp16")]; + tensor var_8543_equation_0 = const()[name = tensor("op_8543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8543_cast_fp16 = einsum(equation = var_8543_equation_0, values = (var_8385_cast_fp16, var_8302_cast_fp16))[name = tensor("op_8543_cast_fp16")]; + tensor var_8544_to_fp16 = const()[name = tensor("op_8544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_675_cast_fp16 = mul(x = var_8543_cast_fp16, y = var_8544_to_fp16)[name = tensor("aw_675_cast_fp16")]; + tensor var_8547_equation_0 = const()[name = tensor("op_8547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8547_cast_fp16 = einsum(equation = var_8547_equation_0, values = (var_8389_cast_fp16, var_8306_cast_fp16))[name = tensor("op_8547_cast_fp16")]; + tensor var_8548_to_fp16 = const()[name = tensor("op_8548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_677_cast_fp16 = mul(x = var_8547_cast_fp16, y = var_8548_to_fp16)[name = tensor("aw_677_cast_fp16")]; + tensor var_8551_equation_0 = const()[name = tensor("op_8551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8551_cast_fp16 = einsum(equation = var_8551_equation_0, values = (var_8393_cast_fp16, var_8310_cast_fp16))[name = tensor("op_8551_cast_fp16")]; + tensor var_8552_to_fp16 = const()[name = tensor("op_8552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_679_cast_fp16 = mul(x = var_8551_cast_fp16, y = var_8552_to_fp16)[name = tensor("aw_679_cast_fp16")]; + tensor var_8554_cast_fp16 = softmax(axis = var_2624, x = aw_641_cast_fp16)[name = tensor("op_8554_cast_fp16")]; + tensor var_8555_cast_fp16 = softmax(axis = var_2624, x = aw_643_cast_fp16)[name = tensor("op_8555_cast_fp16")]; + tensor var_8556_cast_fp16 = softmax(axis = var_2624, x = aw_645_cast_fp16)[name = tensor("op_8556_cast_fp16")]; + tensor var_8557_cast_fp16 = softmax(axis = var_2624, x = aw_647_cast_fp16)[name = tensor("op_8557_cast_fp16")]; + tensor var_8558_cast_fp16 = softmax(axis = var_2624, x = aw_649_cast_fp16)[name = tensor("op_8558_cast_fp16")]; + tensor var_8559_cast_fp16 = softmax(axis = var_2624, x = aw_651_cast_fp16)[name = tensor("op_8559_cast_fp16")]; + tensor var_8560_cast_fp16 = softmax(axis = var_2624, x = aw_653_cast_fp16)[name = tensor("op_8560_cast_fp16")]; + tensor var_8561_cast_fp16 = softmax(axis = var_2624, x = aw_655_cast_fp16)[name = tensor("op_8561_cast_fp16")]; + tensor var_8562_cast_fp16 = softmax(axis = var_2624, x = aw_657_cast_fp16)[name = tensor("op_8562_cast_fp16")]; + tensor var_8563_cast_fp16 = softmax(axis = var_2624, x = aw_659_cast_fp16)[name = tensor("op_8563_cast_fp16")]; + tensor var_8564_cast_fp16 = softmax(axis = var_2624, x = aw_661_cast_fp16)[name = tensor("op_8564_cast_fp16")]; + tensor var_8565_cast_fp16 = softmax(axis = var_2624, x = aw_663_cast_fp16)[name = tensor("op_8565_cast_fp16")]; + tensor var_8566_cast_fp16 = softmax(axis = var_2624, x = aw_665_cast_fp16)[name = tensor("op_8566_cast_fp16")]; + tensor var_8567_cast_fp16 = softmax(axis = var_2624, x = aw_667_cast_fp16)[name = tensor("op_8567_cast_fp16")]; + tensor var_8568_cast_fp16 = softmax(axis = var_2624, x = aw_669_cast_fp16)[name = tensor("op_8568_cast_fp16")]; + tensor var_8569_cast_fp16 = softmax(axis = var_2624, x = aw_671_cast_fp16)[name = tensor("op_8569_cast_fp16")]; + tensor var_8570_cast_fp16 = softmax(axis = var_2624, x = aw_673_cast_fp16)[name = tensor("op_8570_cast_fp16")]; + tensor var_8571_cast_fp16 = softmax(axis = var_2624, x = aw_675_cast_fp16)[name = tensor("op_8571_cast_fp16")]; + tensor var_8572_cast_fp16 = softmax(axis = var_2624, x = aw_677_cast_fp16)[name = tensor("op_8572_cast_fp16")]; + tensor var_8573_cast_fp16 = softmax(axis = var_2624, x = aw_679_cast_fp16)[name = tensor("op_8573_cast_fp16")]; + tensor var_8575_equation_0 = const()[name = tensor("op_8575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8575_cast_fp16 = einsum(equation = var_8575_equation_0, values = (var_8395_cast_fp16, var_8554_cast_fp16))[name = tensor("op_8575_cast_fp16")]; + tensor var_8577_equation_0 = const()[name = tensor("op_8577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8577_cast_fp16 = einsum(equation = var_8577_equation_0, values = (var_8399_cast_fp16, var_8555_cast_fp16))[name = tensor("op_8577_cast_fp16")]; + tensor var_8579_equation_0 = const()[name = tensor("op_8579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8579_cast_fp16 = einsum(equation = var_8579_equation_0, values = (var_8403_cast_fp16, var_8556_cast_fp16))[name = tensor("op_8579_cast_fp16")]; + tensor var_8581_equation_0 = const()[name = tensor("op_8581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8581_cast_fp16 = einsum(equation = var_8581_equation_0, values = (var_8407_cast_fp16, var_8557_cast_fp16))[name = tensor("op_8581_cast_fp16")]; + tensor var_8583_equation_0 = const()[name = tensor("op_8583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8583_cast_fp16 = einsum(equation = var_8583_equation_0, values = (var_8411_cast_fp16, var_8558_cast_fp16))[name = tensor("op_8583_cast_fp16")]; + tensor var_8585_equation_0 = const()[name = tensor("op_8585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8585_cast_fp16 = einsum(equation = var_8585_equation_0, values = (var_8415_cast_fp16, var_8559_cast_fp16))[name = tensor("op_8585_cast_fp16")]; + tensor var_8587_equation_0 = const()[name = tensor("op_8587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8587_cast_fp16 = einsum(equation = var_8587_equation_0, values = (var_8419_cast_fp16, var_8560_cast_fp16))[name = tensor("op_8587_cast_fp16")]; + tensor var_8589_equation_0 = const()[name = tensor("op_8589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8589_cast_fp16 = einsum(equation = var_8589_equation_0, values = (var_8423_cast_fp16, var_8561_cast_fp16))[name = tensor("op_8589_cast_fp16")]; + tensor var_8591_equation_0 = const()[name = tensor("op_8591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8591_cast_fp16 = einsum(equation = var_8591_equation_0, values = (var_8427_cast_fp16, var_8562_cast_fp16))[name = tensor("op_8591_cast_fp16")]; + tensor var_8593_equation_0 = const()[name = tensor("op_8593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8593_cast_fp16 = einsum(equation = var_8593_equation_0, values = (var_8431_cast_fp16, var_8563_cast_fp16))[name = tensor("op_8593_cast_fp16")]; + tensor var_8595_equation_0 = const()[name = tensor("op_8595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8595_cast_fp16 = einsum(equation = var_8595_equation_0, values = (var_8435_cast_fp16, var_8564_cast_fp16))[name = tensor("op_8595_cast_fp16")]; + tensor var_8597_equation_0 = const()[name = tensor("op_8597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8597_cast_fp16 = einsum(equation = var_8597_equation_0, values = (var_8439_cast_fp16, var_8565_cast_fp16))[name = tensor("op_8597_cast_fp16")]; + tensor var_8599_equation_0 = const()[name = tensor("op_8599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8599_cast_fp16 = einsum(equation = var_8599_equation_0, values = (var_8443_cast_fp16, var_8566_cast_fp16))[name = tensor("op_8599_cast_fp16")]; + tensor var_8601_equation_0 = const()[name = tensor("op_8601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8601_cast_fp16 = einsum(equation = var_8601_equation_0, values = (var_8447_cast_fp16, var_8567_cast_fp16))[name = tensor("op_8601_cast_fp16")]; + tensor var_8603_equation_0 = const()[name = tensor("op_8603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8603_cast_fp16 = einsum(equation = var_8603_equation_0, values = (var_8451_cast_fp16, var_8568_cast_fp16))[name = tensor("op_8603_cast_fp16")]; + tensor var_8605_equation_0 = const()[name = tensor("op_8605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8605_cast_fp16 = einsum(equation = var_8605_equation_0, values = (var_8455_cast_fp16, var_8569_cast_fp16))[name = tensor("op_8605_cast_fp16")]; + tensor var_8607_equation_0 = const()[name = tensor("op_8607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8607_cast_fp16 = einsum(equation = var_8607_equation_0, values = (var_8459_cast_fp16, var_8570_cast_fp16))[name = tensor("op_8607_cast_fp16")]; + tensor var_8609_equation_0 = const()[name = tensor("op_8609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8609_cast_fp16 = einsum(equation = var_8609_equation_0, values = (var_8463_cast_fp16, var_8571_cast_fp16))[name = tensor("op_8609_cast_fp16")]; + tensor var_8611_equation_0 = const()[name = tensor("op_8611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8611_cast_fp16 = einsum(equation = var_8611_equation_0, values = (var_8467_cast_fp16, var_8572_cast_fp16))[name = tensor("op_8611_cast_fp16")]; + tensor var_8613_equation_0 = const()[name = tensor("op_8613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8613_cast_fp16 = einsum(equation = var_8613_equation_0, values = (var_8471_cast_fp16, var_8573_cast_fp16))[name = tensor("op_8613_cast_fp16")]; + tensor input_179_interleave_0 = const()[name = tensor("input_179_interleave_0"), val = tensor(false)]; + tensor input_179_cast_fp16 = concat(axis = var_2624, interleave = input_179_interleave_0, values = (var_8575_cast_fp16, var_8577_cast_fp16, var_8579_cast_fp16, var_8581_cast_fp16, var_8583_cast_fp16, var_8585_cast_fp16, var_8587_cast_fp16, var_8589_cast_fp16, var_8591_cast_fp16, var_8593_cast_fp16, var_8595_cast_fp16, var_8597_cast_fp16, var_8599_cast_fp16, var_8601_cast_fp16, var_8603_cast_fp16, var_8605_cast_fp16, var_8607_cast_fp16, var_8609_cast_fp16, var_8611_cast_fp16, var_8613_cast_fp16))[name = tensor("input_179_cast_fp16")]; + tensor var_8623_pad_type_0 = const()[name = tensor("op_8623_pad_type_0"), val = tensor("valid")]; + tensor var_8623_strides_0 = const()[name = tensor("op_8623_strides_0"), val = tensor([1, 1])]; + tensor var_8623_pad_0 = const()[name = tensor("op_8623_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8623_dilations_0 = const()[name = tensor("op_8623_dilations_0"), val = tensor([1, 1])]; + tensor var_8623_groups_0 = const()[name = tensor("op_8623_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234716032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235944896))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235945088)))]; + tensor var_8623_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_8623_dilations_0, groups = var_8623_groups_0, pad = var_8623_pad_0, pad_type = var_8623_pad_type_0, strides = var_8623_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = tensor("op_8623_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = var_8623_cast_fp16, y = inputs_61_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; + tensor hidden_states_103_axes_0 = const()[name = tensor("hidden_states_103_axes_0"), val = tensor([1])]; + tensor hidden_states_103_gamma_0_to_fp16 = const()[name = tensor("hidden_states_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235947712)))]; + tensor hidden_states_103_beta_0_to_fp16 = const()[name = tensor("hidden_states_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235950336)))]; + tensor var_8633_to_fp16 = const()[name = tensor("op_8633_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_103_cast_fp16 = layer_norm(axes = hidden_states_103_axes_0, beta = hidden_states_103_beta_0_to_fp16, epsilon = var_8633_to_fp16, gamma = hidden_states_103_gamma_0_to_fp16, x = inputs_63_cast_fp16)[name = tensor("hidden_states_103_cast_fp16")]; + tensor q_43_pad_type_0 = const()[name = tensor("q_43_pad_type_0"), val = tensor("valid")]; + tensor q_43_strides_0 = const()[name = tensor("q_43_strides_0"), val = tensor([1, 1])]; + tensor q_43_pad_0 = const()[name = tensor("q_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_43_dilations_0 = const()[name = tensor("q_43_dilations_0"), val = tensor([1, 1])]; + tensor q_43_groups_0 = const()[name = tensor("q_43_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235952960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237181824))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_43_cast_fp16 = conv(dilations = q_43_dilations_0, groups = q_43_groups_0, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = q_43_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_103_cast_fp16)[name = tensor("q_43_cast_fp16")]; + tensor k_85_pad_type_0 = const()[name = tensor("k_85_pad_type_0"), val = tensor("valid")]; + tensor k_85_strides_0 = const()[name = tensor("k_85_strides_0"), val = tensor([1, 1])]; + tensor k_85_pad_0 = const()[name = tensor("k_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_85_dilations_0 = const()[name = tensor("k_85_dilations_0"), val = tensor([1, 1])]; + tensor k_85_groups_0 = const()[name = tensor("k_85_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237182016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239148160))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_85_cast_fp16 = conv(dilations = k_85_dilations_0, groups = k_85_groups_0, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = k_85_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_85_cast_fp16")]; + tensor v_43_pad_type_0 = const()[name = tensor("v_43_pad_type_0"), val = tensor("valid")]; + tensor v_43_strides_0 = const()[name = tensor("v_43_strides_0"), val = tensor([1, 1])]; + tensor v_43_pad_0 = const()[name = tensor("v_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_43_dilations_0 = const()[name = tensor("v_43_dilations_0"), val = tensor([1, 1])]; + tensor v_43_groups_0 = const()[name = tensor("v_43_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239148352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241114496))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_43_cast_fp16 = conv(dilations = v_43_dilations_0, groups = v_43_groups_0, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = v_43_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_43_cast_fp16")]; + tensor var_8666_begin_0 = const()[name = tensor("op_8666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8666_end_0 = const()[name = tensor("op_8666_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_8666_end_mask_0 = const()[name = tensor("op_8666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8666_cast_fp16 = slice_by_index(begin = var_8666_begin_0, end = var_8666_end_0, end_mask = var_8666_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8666_cast_fp16")]; + tensor var_8670_begin_0 = const()[name = tensor("op_8670_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8670_end_0 = const()[name = tensor("op_8670_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_8670_end_mask_0 = const()[name = tensor("op_8670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8670_cast_fp16 = slice_by_index(begin = var_8670_begin_0, end = var_8670_end_0, end_mask = var_8670_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8670_cast_fp16")]; + tensor var_8674_begin_0 = const()[name = tensor("op_8674_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8674_end_0 = const()[name = tensor("op_8674_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_8674_end_mask_0 = const()[name = tensor("op_8674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8674_cast_fp16 = slice_by_index(begin = var_8674_begin_0, end = var_8674_end_0, end_mask = var_8674_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8674_cast_fp16")]; + tensor var_8678_begin_0 = const()[name = tensor("op_8678_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8678_end_0 = const()[name = tensor("op_8678_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_8678_end_mask_0 = const()[name = tensor("op_8678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8678_cast_fp16 = slice_by_index(begin = var_8678_begin_0, end = var_8678_end_0, end_mask = var_8678_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8678_cast_fp16")]; + tensor var_8682_begin_0 = const()[name = tensor("op_8682_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8682_end_0 = const()[name = tensor("op_8682_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_8682_end_mask_0 = const()[name = tensor("op_8682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8682_cast_fp16 = slice_by_index(begin = var_8682_begin_0, end = var_8682_end_0, end_mask = var_8682_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8682_cast_fp16")]; + tensor var_8686_begin_0 = const()[name = tensor("op_8686_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8686_end_0 = const()[name = tensor("op_8686_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_8686_end_mask_0 = const()[name = tensor("op_8686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8686_cast_fp16 = slice_by_index(begin = var_8686_begin_0, end = var_8686_end_0, end_mask = var_8686_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8686_cast_fp16")]; + tensor var_8690_begin_0 = const()[name = tensor("op_8690_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8690_end_0 = const()[name = tensor("op_8690_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_8690_end_mask_0 = const()[name = tensor("op_8690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8690_cast_fp16 = slice_by_index(begin = var_8690_begin_0, end = var_8690_end_0, end_mask = var_8690_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8690_cast_fp16")]; + tensor var_8694_begin_0 = const()[name = tensor("op_8694_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8694_end_0 = const()[name = tensor("op_8694_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_8694_end_mask_0 = const()[name = tensor("op_8694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8694_cast_fp16 = slice_by_index(begin = var_8694_begin_0, end = var_8694_end_0, end_mask = var_8694_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8694_cast_fp16")]; + tensor var_8698_begin_0 = const()[name = tensor("op_8698_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8698_end_0 = const()[name = tensor("op_8698_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_8698_end_mask_0 = const()[name = tensor("op_8698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8698_cast_fp16 = slice_by_index(begin = var_8698_begin_0, end = var_8698_end_0, end_mask = var_8698_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8698_cast_fp16")]; + tensor var_8702_begin_0 = const()[name = tensor("op_8702_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8702_end_0 = const()[name = tensor("op_8702_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_8702_end_mask_0 = const()[name = tensor("op_8702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8702_cast_fp16 = slice_by_index(begin = var_8702_begin_0, end = var_8702_end_0, end_mask = var_8702_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8702_cast_fp16")]; + tensor var_8706_begin_0 = const()[name = tensor("op_8706_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8706_end_0 = const()[name = tensor("op_8706_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_8706_end_mask_0 = const()[name = tensor("op_8706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8706_cast_fp16 = slice_by_index(begin = var_8706_begin_0, end = var_8706_end_0, end_mask = var_8706_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8706_cast_fp16")]; + tensor var_8710_begin_0 = const()[name = tensor("op_8710_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8710_end_0 = const()[name = tensor("op_8710_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_8710_end_mask_0 = const()[name = tensor("op_8710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8710_cast_fp16 = slice_by_index(begin = var_8710_begin_0, end = var_8710_end_0, end_mask = var_8710_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8710_cast_fp16")]; + tensor var_8714_begin_0 = const()[name = tensor("op_8714_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8714_end_0 = const()[name = tensor("op_8714_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_8714_end_mask_0 = const()[name = tensor("op_8714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8714_cast_fp16 = slice_by_index(begin = var_8714_begin_0, end = var_8714_end_0, end_mask = var_8714_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8714_cast_fp16")]; + tensor var_8718_begin_0 = const()[name = tensor("op_8718_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8718_end_0 = const()[name = tensor("op_8718_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_8718_end_mask_0 = const()[name = tensor("op_8718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8718_cast_fp16 = slice_by_index(begin = var_8718_begin_0, end = var_8718_end_0, end_mask = var_8718_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8718_cast_fp16")]; + tensor var_8722_begin_0 = const()[name = tensor("op_8722_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8722_end_0 = const()[name = tensor("op_8722_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_8722_end_mask_0 = const()[name = tensor("op_8722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8722_cast_fp16 = slice_by_index(begin = var_8722_begin_0, end = var_8722_end_0, end_mask = var_8722_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8722_cast_fp16")]; + tensor var_8726_begin_0 = const()[name = tensor("op_8726_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8726_end_0 = const()[name = tensor("op_8726_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8726_end_mask_0 = const()[name = tensor("op_8726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8726_cast_fp16 = slice_by_index(begin = var_8726_begin_0, end = var_8726_end_0, end_mask = var_8726_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8726_cast_fp16")]; + tensor var_8730_begin_0 = const()[name = tensor("op_8730_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8730_end_0 = const()[name = tensor("op_8730_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_8730_end_mask_0 = const()[name = tensor("op_8730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8730_cast_fp16 = slice_by_index(begin = var_8730_begin_0, end = var_8730_end_0, end_mask = var_8730_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8730_cast_fp16")]; + tensor var_8734_begin_0 = const()[name = tensor("op_8734_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8734_end_0 = const()[name = tensor("op_8734_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_8734_end_mask_0 = const()[name = tensor("op_8734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8734_cast_fp16 = slice_by_index(begin = var_8734_begin_0, end = var_8734_end_0, end_mask = var_8734_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8734_cast_fp16")]; + tensor var_8738_begin_0 = const()[name = tensor("op_8738_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8738_end_0 = const()[name = tensor("op_8738_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_8738_end_mask_0 = const()[name = tensor("op_8738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8738_cast_fp16 = slice_by_index(begin = var_8738_begin_0, end = var_8738_end_0, end_mask = var_8738_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8738_cast_fp16")]; + tensor var_8742_begin_0 = const()[name = tensor("op_8742_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8742_end_0 = const()[name = tensor("op_8742_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_8742_end_mask_0 = const()[name = tensor("op_8742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8742_cast_fp16 = slice_by_index(begin = var_8742_begin_0, end = var_8742_end_0, end_mask = var_8742_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8742_cast_fp16")]; + tensor k_87_perm_0 = const()[name = tensor("k_87_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8749_begin_0 = const()[name = tensor("op_8749_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8749_end_0 = const()[name = tensor("op_8749_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_8749_end_mask_0 = const()[name = tensor("op_8749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_87_cast_fp16 = transpose(perm = k_87_perm_0, x = k_85_cast_fp16)[name = tensor("transpose_46")]; + tensor var_8749_cast_fp16 = slice_by_index(begin = var_8749_begin_0, end = var_8749_end_0, end_mask = var_8749_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8749_cast_fp16")]; + tensor var_8753_begin_0 = const()[name = tensor("op_8753_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8753_end_0 = const()[name = tensor("op_8753_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_8753_end_mask_0 = const()[name = tensor("op_8753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8753_cast_fp16 = slice_by_index(begin = var_8753_begin_0, end = var_8753_end_0, end_mask = var_8753_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8753_cast_fp16")]; + tensor var_8757_begin_0 = const()[name = tensor("op_8757_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8757_end_0 = const()[name = tensor("op_8757_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_8757_end_mask_0 = const()[name = tensor("op_8757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8757_cast_fp16 = slice_by_index(begin = var_8757_begin_0, end = var_8757_end_0, end_mask = var_8757_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8757_cast_fp16")]; + tensor var_8761_begin_0 = const()[name = tensor("op_8761_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8761_end_0 = const()[name = tensor("op_8761_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_8761_end_mask_0 = const()[name = tensor("op_8761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8761_cast_fp16 = slice_by_index(begin = var_8761_begin_0, end = var_8761_end_0, end_mask = var_8761_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8761_cast_fp16")]; + tensor var_8765_begin_0 = const()[name = tensor("op_8765_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8765_end_0 = const()[name = tensor("op_8765_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_8765_end_mask_0 = const()[name = tensor("op_8765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8765_cast_fp16 = slice_by_index(begin = var_8765_begin_0, end = var_8765_end_0, end_mask = var_8765_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8765_cast_fp16")]; + tensor var_8769_begin_0 = const()[name = tensor("op_8769_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8769_end_0 = const()[name = tensor("op_8769_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_8769_end_mask_0 = const()[name = tensor("op_8769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8769_cast_fp16 = slice_by_index(begin = var_8769_begin_0, end = var_8769_end_0, end_mask = var_8769_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8769_cast_fp16")]; + tensor var_8773_begin_0 = const()[name = tensor("op_8773_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8773_end_0 = const()[name = tensor("op_8773_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_8773_end_mask_0 = const()[name = tensor("op_8773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8773_cast_fp16 = slice_by_index(begin = var_8773_begin_0, end = var_8773_end_0, end_mask = var_8773_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8773_cast_fp16")]; + tensor var_8777_begin_0 = const()[name = tensor("op_8777_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8777_end_0 = const()[name = tensor("op_8777_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_8777_end_mask_0 = const()[name = tensor("op_8777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8777_cast_fp16 = slice_by_index(begin = var_8777_begin_0, end = var_8777_end_0, end_mask = var_8777_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8777_cast_fp16")]; + tensor var_8781_begin_0 = const()[name = tensor("op_8781_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8781_end_0 = const()[name = tensor("op_8781_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_8781_end_mask_0 = const()[name = tensor("op_8781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8781_cast_fp16 = slice_by_index(begin = var_8781_begin_0, end = var_8781_end_0, end_mask = var_8781_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8781_cast_fp16")]; + tensor var_8785_begin_0 = const()[name = tensor("op_8785_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8785_end_0 = const()[name = tensor("op_8785_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_8785_end_mask_0 = const()[name = tensor("op_8785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8785_cast_fp16 = slice_by_index(begin = var_8785_begin_0, end = var_8785_end_0, end_mask = var_8785_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8785_cast_fp16")]; + tensor var_8789_begin_0 = const()[name = tensor("op_8789_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8789_end_0 = const()[name = tensor("op_8789_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_8789_end_mask_0 = const()[name = tensor("op_8789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8789_cast_fp16 = slice_by_index(begin = var_8789_begin_0, end = var_8789_end_0, end_mask = var_8789_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8789_cast_fp16")]; + tensor var_8793_begin_0 = const()[name = tensor("op_8793_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8793_end_0 = const()[name = tensor("op_8793_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_8793_end_mask_0 = const()[name = tensor("op_8793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8793_cast_fp16 = slice_by_index(begin = var_8793_begin_0, end = var_8793_end_0, end_mask = var_8793_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8793_cast_fp16")]; + tensor var_8797_begin_0 = const()[name = tensor("op_8797_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8797_end_0 = const()[name = tensor("op_8797_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_8797_end_mask_0 = const()[name = tensor("op_8797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8797_cast_fp16 = slice_by_index(begin = var_8797_begin_0, end = var_8797_end_0, end_mask = var_8797_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8797_cast_fp16")]; + tensor var_8801_begin_0 = const()[name = tensor("op_8801_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8801_end_0 = const()[name = tensor("op_8801_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_8801_end_mask_0 = const()[name = tensor("op_8801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8801_cast_fp16 = slice_by_index(begin = var_8801_begin_0, end = var_8801_end_0, end_mask = var_8801_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8801_cast_fp16")]; + tensor var_8805_begin_0 = const()[name = tensor("op_8805_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8805_end_0 = const()[name = tensor("op_8805_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_8805_end_mask_0 = const()[name = tensor("op_8805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8805_cast_fp16 = slice_by_index(begin = var_8805_begin_0, end = var_8805_end_0, end_mask = var_8805_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8805_cast_fp16")]; + tensor var_8809_begin_0 = const()[name = tensor("op_8809_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8809_end_0 = const()[name = tensor("op_8809_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_8809_end_mask_0 = const()[name = tensor("op_8809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8809_cast_fp16 = slice_by_index(begin = var_8809_begin_0, end = var_8809_end_0, end_mask = var_8809_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8809_cast_fp16")]; + tensor var_8813_begin_0 = const()[name = tensor("op_8813_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8813_end_0 = const()[name = tensor("op_8813_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_8813_end_mask_0 = const()[name = tensor("op_8813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8813_cast_fp16 = slice_by_index(begin = var_8813_begin_0, end = var_8813_end_0, end_mask = var_8813_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8813_cast_fp16")]; + tensor var_8817_begin_0 = const()[name = tensor("op_8817_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8817_end_0 = const()[name = tensor("op_8817_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_8817_end_mask_0 = const()[name = tensor("op_8817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8817_cast_fp16 = slice_by_index(begin = var_8817_begin_0, end = var_8817_end_0, end_mask = var_8817_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8817_cast_fp16")]; + tensor var_8821_begin_0 = const()[name = tensor("op_8821_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8821_end_0 = const()[name = tensor("op_8821_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_8821_end_mask_0 = const()[name = tensor("op_8821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8821_cast_fp16 = slice_by_index(begin = var_8821_begin_0, end = var_8821_end_0, end_mask = var_8821_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8821_cast_fp16")]; + tensor var_8825_begin_0 = const()[name = tensor("op_8825_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8825_end_0 = const()[name = tensor("op_8825_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_8825_end_mask_0 = const()[name = tensor("op_8825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8825_cast_fp16 = slice_by_index(begin = var_8825_begin_0, end = var_8825_end_0, end_mask = var_8825_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8825_cast_fp16")]; + tensor var_8827_begin_0 = const()[name = tensor("op_8827_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8827_end_0 = const()[name = tensor("op_8827_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_8827_end_mask_0 = const()[name = tensor("op_8827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8827_cast_fp16 = slice_by_index(begin = var_8827_begin_0, end = var_8827_end_0, end_mask = var_8827_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8827_cast_fp16")]; + tensor var_8831_begin_0 = const()[name = tensor("op_8831_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8831_end_0 = const()[name = tensor("op_8831_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_8831_end_mask_0 = const()[name = tensor("op_8831_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8831_cast_fp16 = slice_by_index(begin = var_8831_begin_0, end = var_8831_end_0, end_mask = var_8831_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8831_cast_fp16")]; + tensor var_8835_begin_0 = const()[name = tensor("op_8835_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8835_end_0 = const()[name = tensor("op_8835_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_8835_end_mask_0 = const()[name = tensor("op_8835_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8835_cast_fp16 = slice_by_index(begin = var_8835_begin_0, end = var_8835_end_0, end_mask = var_8835_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8835_cast_fp16")]; + tensor var_8839_begin_0 = const()[name = tensor("op_8839_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8839_end_0 = const()[name = tensor("op_8839_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_8839_end_mask_0 = const()[name = tensor("op_8839_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8839_cast_fp16 = slice_by_index(begin = var_8839_begin_0, end = var_8839_end_0, end_mask = var_8839_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8839_cast_fp16")]; + tensor var_8843_begin_0 = const()[name = tensor("op_8843_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8843_end_0 = const()[name = tensor("op_8843_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_8843_end_mask_0 = const()[name = tensor("op_8843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8843_cast_fp16 = slice_by_index(begin = var_8843_begin_0, end = var_8843_end_0, end_mask = var_8843_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8843_cast_fp16")]; + tensor var_8847_begin_0 = const()[name = tensor("op_8847_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8847_end_0 = const()[name = tensor("op_8847_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_8847_end_mask_0 = const()[name = tensor("op_8847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8847_cast_fp16 = slice_by_index(begin = var_8847_begin_0, end = var_8847_end_0, end_mask = var_8847_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8847_cast_fp16")]; + tensor var_8851_begin_0 = const()[name = tensor("op_8851_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8851_end_0 = const()[name = tensor("op_8851_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_8851_end_mask_0 = const()[name = tensor("op_8851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8851_cast_fp16 = slice_by_index(begin = var_8851_begin_0, end = var_8851_end_0, end_mask = var_8851_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8851_cast_fp16")]; + tensor var_8855_begin_0 = const()[name = tensor("op_8855_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8855_end_0 = const()[name = tensor("op_8855_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_8855_end_mask_0 = const()[name = tensor("op_8855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8855_cast_fp16 = slice_by_index(begin = var_8855_begin_0, end = var_8855_end_0, end_mask = var_8855_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8855_cast_fp16")]; + tensor var_8859_begin_0 = const()[name = tensor("op_8859_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8859_end_0 = const()[name = tensor("op_8859_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_8859_end_mask_0 = const()[name = tensor("op_8859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8859_cast_fp16 = slice_by_index(begin = var_8859_begin_0, end = var_8859_end_0, end_mask = var_8859_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8859_cast_fp16")]; + tensor var_8863_begin_0 = const()[name = tensor("op_8863_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8863_end_0 = const()[name = tensor("op_8863_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_8863_end_mask_0 = const()[name = tensor("op_8863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8863_cast_fp16 = slice_by_index(begin = var_8863_begin_0, end = var_8863_end_0, end_mask = var_8863_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8863_cast_fp16")]; + tensor var_8867_begin_0 = const()[name = tensor("op_8867_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8867_end_0 = const()[name = tensor("op_8867_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_8867_end_mask_0 = const()[name = tensor("op_8867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8867_cast_fp16 = slice_by_index(begin = var_8867_begin_0, end = var_8867_end_0, end_mask = var_8867_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8867_cast_fp16")]; + tensor var_8871_begin_0 = const()[name = tensor("op_8871_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8871_end_0 = const()[name = tensor("op_8871_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_8871_end_mask_0 = const()[name = tensor("op_8871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8871_cast_fp16 = slice_by_index(begin = var_8871_begin_0, end = var_8871_end_0, end_mask = var_8871_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8871_cast_fp16")]; + tensor var_8875_begin_0 = const()[name = tensor("op_8875_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8875_end_0 = const()[name = tensor("op_8875_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_8875_end_mask_0 = const()[name = tensor("op_8875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8875_cast_fp16 = slice_by_index(begin = var_8875_begin_0, end = var_8875_end_0, end_mask = var_8875_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8875_cast_fp16")]; + tensor var_8879_begin_0 = const()[name = tensor("op_8879_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8879_end_0 = const()[name = tensor("op_8879_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_8879_end_mask_0 = const()[name = tensor("op_8879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8879_cast_fp16 = slice_by_index(begin = var_8879_begin_0, end = var_8879_end_0, end_mask = var_8879_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8879_cast_fp16")]; + tensor var_8883_begin_0 = const()[name = tensor("op_8883_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8883_end_0 = const()[name = tensor("op_8883_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_8883_end_mask_0 = const()[name = tensor("op_8883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8883_cast_fp16 = slice_by_index(begin = var_8883_begin_0, end = var_8883_end_0, end_mask = var_8883_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8883_cast_fp16")]; + tensor var_8887_begin_0 = const()[name = tensor("op_8887_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8887_end_0 = const()[name = tensor("op_8887_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_8887_end_mask_0 = const()[name = tensor("op_8887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8887_cast_fp16 = slice_by_index(begin = var_8887_begin_0, end = var_8887_end_0, end_mask = var_8887_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8887_cast_fp16")]; + tensor var_8891_begin_0 = const()[name = tensor("op_8891_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8891_end_0 = const()[name = tensor("op_8891_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_8891_end_mask_0 = const()[name = tensor("op_8891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8891_cast_fp16 = slice_by_index(begin = var_8891_begin_0, end = var_8891_end_0, end_mask = var_8891_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8891_cast_fp16")]; + tensor var_8895_begin_0 = const()[name = tensor("op_8895_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8895_end_0 = const()[name = tensor("op_8895_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_8895_end_mask_0 = const()[name = tensor("op_8895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8895_cast_fp16 = slice_by_index(begin = var_8895_begin_0, end = var_8895_end_0, end_mask = var_8895_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8895_cast_fp16")]; + tensor var_8899_begin_0 = const()[name = tensor("op_8899_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8899_end_0 = const()[name = tensor("op_8899_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_8899_end_mask_0 = const()[name = tensor("op_8899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8899_cast_fp16 = slice_by_index(begin = var_8899_begin_0, end = var_8899_end_0, end_mask = var_8899_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8899_cast_fp16")]; + tensor var_8903_begin_0 = const()[name = tensor("op_8903_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8903_end_0 = const()[name = tensor("op_8903_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_8903_end_mask_0 = const()[name = tensor("op_8903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8903_cast_fp16 = slice_by_index(begin = var_8903_begin_0, end = var_8903_end_0, end_mask = var_8903_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8903_cast_fp16")]; + tensor var_8907_equation_0 = const()[name = tensor("op_8907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8907_cast_fp16 = einsum(equation = var_8907_equation_0, values = (var_8749_cast_fp16, var_8666_cast_fp16))[name = tensor("op_8907_cast_fp16")]; + tensor var_8908_to_fp16 = const()[name = tensor("op_8908_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_681_cast_fp16 = mul(x = var_8907_cast_fp16, y = var_8908_to_fp16)[name = tensor("aw_681_cast_fp16")]; + tensor var_8911_equation_0 = const()[name = tensor("op_8911_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8911_cast_fp16 = einsum(equation = var_8911_equation_0, values = (var_8753_cast_fp16, var_8670_cast_fp16))[name = tensor("op_8911_cast_fp16")]; + tensor var_8912_to_fp16 = const()[name = tensor("op_8912_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_683_cast_fp16 = mul(x = var_8911_cast_fp16, y = var_8912_to_fp16)[name = tensor("aw_683_cast_fp16")]; + tensor var_8915_equation_0 = const()[name = tensor("op_8915_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8915_cast_fp16 = einsum(equation = var_8915_equation_0, values = (var_8757_cast_fp16, var_8674_cast_fp16))[name = tensor("op_8915_cast_fp16")]; + tensor var_8916_to_fp16 = const()[name = tensor("op_8916_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_685_cast_fp16 = mul(x = var_8915_cast_fp16, y = var_8916_to_fp16)[name = tensor("aw_685_cast_fp16")]; + tensor var_8919_equation_0 = const()[name = tensor("op_8919_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8919_cast_fp16 = einsum(equation = var_8919_equation_0, values = (var_8761_cast_fp16, var_8678_cast_fp16))[name = tensor("op_8919_cast_fp16")]; + tensor var_8920_to_fp16 = const()[name = tensor("op_8920_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_687_cast_fp16 = mul(x = var_8919_cast_fp16, y = var_8920_to_fp16)[name = tensor("aw_687_cast_fp16")]; + tensor var_8923_equation_0 = const()[name = tensor("op_8923_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8923_cast_fp16 = einsum(equation = var_8923_equation_0, values = (var_8765_cast_fp16, var_8682_cast_fp16))[name = tensor("op_8923_cast_fp16")]; + tensor var_8924_to_fp16 = const()[name = tensor("op_8924_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_689_cast_fp16 = mul(x = var_8923_cast_fp16, y = var_8924_to_fp16)[name = tensor("aw_689_cast_fp16")]; + tensor var_8927_equation_0 = const()[name = tensor("op_8927_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8927_cast_fp16 = einsum(equation = var_8927_equation_0, values = (var_8769_cast_fp16, var_8686_cast_fp16))[name = tensor("op_8927_cast_fp16")]; + tensor var_8928_to_fp16 = const()[name = tensor("op_8928_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_691_cast_fp16 = mul(x = var_8927_cast_fp16, y = var_8928_to_fp16)[name = tensor("aw_691_cast_fp16")]; + tensor var_8931_equation_0 = const()[name = tensor("op_8931_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8931_cast_fp16 = einsum(equation = var_8931_equation_0, values = (var_8773_cast_fp16, var_8690_cast_fp16))[name = tensor("op_8931_cast_fp16")]; + tensor var_8932_to_fp16 = const()[name = tensor("op_8932_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_693_cast_fp16 = mul(x = var_8931_cast_fp16, y = var_8932_to_fp16)[name = tensor("aw_693_cast_fp16")]; + tensor var_8935_equation_0 = const()[name = tensor("op_8935_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8935_cast_fp16 = einsum(equation = var_8935_equation_0, values = (var_8777_cast_fp16, var_8694_cast_fp16))[name = tensor("op_8935_cast_fp16")]; + tensor var_8936_to_fp16 = const()[name = tensor("op_8936_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_695_cast_fp16 = mul(x = var_8935_cast_fp16, y = var_8936_to_fp16)[name = tensor("aw_695_cast_fp16")]; + tensor var_8939_equation_0 = const()[name = tensor("op_8939_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8939_cast_fp16 = einsum(equation = var_8939_equation_0, values = (var_8781_cast_fp16, var_8698_cast_fp16))[name = tensor("op_8939_cast_fp16")]; + tensor var_8940_to_fp16 = const()[name = tensor("op_8940_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_697_cast_fp16 = mul(x = var_8939_cast_fp16, y = var_8940_to_fp16)[name = tensor("aw_697_cast_fp16")]; + tensor var_8943_equation_0 = const()[name = tensor("op_8943_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8943_cast_fp16 = einsum(equation = var_8943_equation_0, values = (var_8785_cast_fp16, var_8702_cast_fp16))[name = tensor("op_8943_cast_fp16")]; + tensor var_8944_to_fp16 = const()[name = tensor("op_8944_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_699_cast_fp16 = mul(x = var_8943_cast_fp16, y = var_8944_to_fp16)[name = tensor("aw_699_cast_fp16")]; + tensor var_8947_equation_0 = const()[name = tensor("op_8947_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8947_cast_fp16 = einsum(equation = var_8947_equation_0, values = (var_8789_cast_fp16, var_8706_cast_fp16))[name = tensor("op_8947_cast_fp16")]; + tensor var_8948_to_fp16 = const()[name = tensor("op_8948_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_701_cast_fp16 = mul(x = var_8947_cast_fp16, y = var_8948_to_fp16)[name = tensor("aw_701_cast_fp16")]; + tensor var_8951_equation_0 = const()[name = tensor("op_8951_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8951_cast_fp16 = einsum(equation = var_8951_equation_0, values = (var_8793_cast_fp16, var_8710_cast_fp16))[name = tensor("op_8951_cast_fp16")]; + tensor var_8952_to_fp16 = const()[name = tensor("op_8952_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_703_cast_fp16 = mul(x = var_8951_cast_fp16, y = var_8952_to_fp16)[name = tensor("aw_703_cast_fp16")]; + tensor var_8955_equation_0 = const()[name = tensor("op_8955_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8955_cast_fp16 = einsum(equation = var_8955_equation_0, values = (var_8797_cast_fp16, var_8714_cast_fp16))[name = tensor("op_8955_cast_fp16")]; + tensor var_8956_to_fp16 = const()[name = tensor("op_8956_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_705_cast_fp16 = mul(x = var_8955_cast_fp16, y = var_8956_to_fp16)[name = tensor("aw_705_cast_fp16")]; + tensor var_8959_equation_0 = const()[name = tensor("op_8959_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8959_cast_fp16 = einsum(equation = var_8959_equation_0, values = (var_8801_cast_fp16, var_8718_cast_fp16))[name = tensor("op_8959_cast_fp16")]; + tensor var_8960_to_fp16 = const()[name = tensor("op_8960_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_707_cast_fp16 = mul(x = var_8959_cast_fp16, y = var_8960_to_fp16)[name = tensor("aw_707_cast_fp16")]; + tensor var_8963_equation_0 = const()[name = tensor("op_8963_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8963_cast_fp16 = einsum(equation = var_8963_equation_0, values = (var_8805_cast_fp16, var_8722_cast_fp16))[name = tensor("op_8963_cast_fp16")]; + tensor var_8964_to_fp16 = const()[name = tensor("op_8964_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_709_cast_fp16 = mul(x = var_8963_cast_fp16, y = var_8964_to_fp16)[name = tensor("aw_709_cast_fp16")]; + tensor var_8967_equation_0 = const()[name = tensor("op_8967_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8967_cast_fp16 = einsum(equation = var_8967_equation_0, values = (var_8809_cast_fp16, var_8726_cast_fp16))[name = tensor("op_8967_cast_fp16")]; + tensor var_8968_to_fp16 = const()[name = tensor("op_8968_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_711_cast_fp16 = mul(x = var_8967_cast_fp16, y = var_8968_to_fp16)[name = tensor("aw_711_cast_fp16")]; + tensor var_8971_equation_0 = const()[name = tensor("op_8971_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8971_cast_fp16 = einsum(equation = var_8971_equation_0, values = (var_8813_cast_fp16, var_8730_cast_fp16))[name = tensor("op_8971_cast_fp16")]; + tensor var_8972_to_fp16 = const()[name = tensor("op_8972_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_713_cast_fp16 = mul(x = var_8971_cast_fp16, y = var_8972_to_fp16)[name = tensor("aw_713_cast_fp16")]; + tensor var_8975_equation_0 = const()[name = tensor("op_8975_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8975_cast_fp16 = einsum(equation = var_8975_equation_0, values = (var_8817_cast_fp16, var_8734_cast_fp16))[name = tensor("op_8975_cast_fp16")]; + tensor var_8976_to_fp16 = const()[name = tensor("op_8976_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_715_cast_fp16 = mul(x = var_8975_cast_fp16, y = var_8976_to_fp16)[name = tensor("aw_715_cast_fp16")]; + tensor var_8979_equation_0 = const()[name = tensor("op_8979_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8979_cast_fp16 = einsum(equation = var_8979_equation_0, values = (var_8821_cast_fp16, var_8738_cast_fp16))[name = tensor("op_8979_cast_fp16")]; + tensor var_8980_to_fp16 = const()[name = tensor("op_8980_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_717_cast_fp16 = mul(x = var_8979_cast_fp16, y = var_8980_to_fp16)[name = tensor("aw_717_cast_fp16")]; + tensor var_8983_equation_0 = const()[name = tensor("op_8983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8983_cast_fp16 = einsum(equation = var_8983_equation_0, values = (var_8825_cast_fp16, var_8742_cast_fp16))[name = tensor("op_8983_cast_fp16")]; + tensor var_8984_to_fp16 = const()[name = tensor("op_8984_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_719_cast_fp16 = mul(x = var_8983_cast_fp16, y = var_8984_to_fp16)[name = tensor("aw_719_cast_fp16")]; + tensor var_8986_cast_fp16 = softmax(axis = var_2624, x = aw_681_cast_fp16)[name = tensor("op_8986_cast_fp16")]; + tensor var_8987_cast_fp16 = softmax(axis = var_2624, x = aw_683_cast_fp16)[name = tensor("op_8987_cast_fp16")]; + tensor var_8988_cast_fp16 = softmax(axis = var_2624, x = aw_685_cast_fp16)[name = tensor("op_8988_cast_fp16")]; + tensor var_8989_cast_fp16 = softmax(axis = var_2624, x = aw_687_cast_fp16)[name = tensor("op_8989_cast_fp16")]; + tensor var_8990_cast_fp16 = softmax(axis = var_2624, x = aw_689_cast_fp16)[name = tensor("op_8990_cast_fp16")]; + tensor var_8991_cast_fp16 = softmax(axis = var_2624, x = aw_691_cast_fp16)[name = tensor("op_8991_cast_fp16")]; + tensor var_8992_cast_fp16 = softmax(axis = var_2624, x = aw_693_cast_fp16)[name = tensor("op_8992_cast_fp16")]; + tensor var_8993_cast_fp16 = softmax(axis = var_2624, x = aw_695_cast_fp16)[name = tensor("op_8993_cast_fp16")]; + tensor var_8994_cast_fp16 = softmax(axis = var_2624, x = aw_697_cast_fp16)[name = tensor("op_8994_cast_fp16")]; + tensor var_8995_cast_fp16 = softmax(axis = var_2624, x = aw_699_cast_fp16)[name = tensor("op_8995_cast_fp16")]; + tensor var_8996_cast_fp16 = softmax(axis = var_2624, x = aw_701_cast_fp16)[name = tensor("op_8996_cast_fp16")]; + tensor var_8997_cast_fp16 = softmax(axis = var_2624, x = aw_703_cast_fp16)[name = tensor("op_8997_cast_fp16")]; + tensor var_8998_cast_fp16 = softmax(axis = var_2624, x = aw_705_cast_fp16)[name = tensor("op_8998_cast_fp16")]; + tensor var_8999_cast_fp16 = softmax(axis = var_2624, x = aw_707_cast_fp16)[name = tensor("op_8999_cast_fp16")]; + tensor var_9000_cast_fp16 = softmax(axis = var_2624, x = aw_709_cast_fp16)[name = tensor("op_9000_cast_fp16")]; + tensor var_9001_cast_fp16 = softmax(axis = var_2624, x = aw_711_cast_fp16)[name = tensor("op_9001_cast_fp16")]; + tensor var_9002_cast_fp16 = softmax(axis = var_2624, x = aw_713_cast_fp16)[name = tensor("op_9002_cast_fp16")]; + tensor var_9003_cast_fp16 = softmax(axis = var_2624, x = aw_715_cast_fp16)[name = tensor("op_9003_cast_fp16")]; + tensor var_9004_cast_fp16 = softmax(axis = var_2624, x = aw_717_cast_fp16)[name = tensor("op_9004_cast_fp16")]; + tensor var_9005_cast_fp16 = softmax(axis = var_2624, x = aw_719_cast_fp16)[name = tensor("op_9005_cast_fp16")]; + tensor var_9007_equation_0 = const()[name = tensor("op_9007_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9007_cast_fp16 = einsum(equation = var_9007_equation_0, values = (var_8827_cast_fp16, var_8986_cast_fp16))[name = tensor("op_9007_cast_fp16")]; + tensor var_9009_equation_0 = const()[name = tensor("op_9009_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9009_cast_fp16 = einsum(equation = var_9009_equation_0, values = (var_8831_cast_fp16, var_8987_cast_fp16))[name = tensor("op_9009_cast_fp16")]; + tensor var_9011_equation_0 = const()[name = tensor("op_9011_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9011_cast_fp16 = einsum(equation = var_9011_equation_0, values = (var_8835_cast_fp16, var_8988_cast_fp16))[name = tensor("op_9011_cast_fp16")]; + tensor var_9013_equation_0 = const()[name = tensor("op_9013_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9013_cast_fp16 = einsum(equation = var_9013_equation_0, values = (var_8839_cast_fp16, var_8989_cast_fp16))[name = tensor("op_9013_cast_fp16")]; + tensor var_9015_equation_0 = const()[name = tensor("op_9015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9015_cast_fp16 = einsum(equation = var_9015_equation_0, values = (var_8843_cast_fp16, var_8990_cast_fp16))[name = tensor("op_9015_cast_fp16")]; + tensor var_9017_equation_0 = const()[name = tensor("op_9017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9017_cast_fp16 = einsum(equation = var_9017_equation_0, values = (var_8847_cast_fp16, var_8991_cast_fp16))[name = tensor("op_9017_cast_fp16")]; + tensor var_9019_equation_0 = const()[name = tensor("op_9019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9019_cast_fp16 = einsum(equation = var_9019_equation_0, values = (var_8851_cast_fp16, var_8992_cast_fp16))[name = tensor("op_9019_cast_fp16")]; + tensor var_9021_equation_0 = const()[name = tensor("op_9021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9021_cast_fp16 = einsum(equation = var_9021_equation_0, values = (var_8855_cast_fp16, var_8993_cast_fp16))[name = tensor("op_9021_cast_fp16")]; + tensor var_9023_equation_0 = const()[name = tensor("op_9023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9023_cast_fp16 = einsum(equation = var_9023_equation_0, values = (var_8859_cast_fp16, var_8994_cast_fp16))[name = tensor("op_9023_cast_fp16")]; + tensor var_9025_equation_0 = const()[name = tensor("op_9025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9025_cast_fp16 = einsum(equation = var_9025_equation_0, values = (var_8863_cast_fp16, var_8995_cast_fp16))[name = tensor("op_9025_cast_fp16")]; + tensor var_9027_equation_0 = const()[name = tensor("op_9027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9027_cast_fp16 = einsum(equation = var_9027_equation_0, values = (var_8867_cast_fp16, var_8996_cast_fp16))[name = tensor("op_9027_cast_fp16")]; + tensor var_9029_equation_0 = const()[name = tensor("op_9029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9029_cast_fp16 = einsum(equation = var_9029_equation_0, values = (var_8871_cast_fp16, var_8997_cast_fp16))[name = tensor("op_9029_cast_fp16")]; + tensor var_9031_equation_0 = const()[name = tensor("op_9031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9031_cast_fp16 = einsum(equation = var_9031_equation_0, values = (var_8875_cast_fp16, var_8998_cast_fp16))[name = tensor("op_9031_cast_fp16")]; + tensor var_9033_equation_0 = const()[name = tensor("op_9033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9033_cast_fp16 = einsum(equation = var_9033_equation_0, values = (var_8879_cast_fp16, var_8999_cast_fp16))[name = tensor("op_9033_cast_fp16")]; + tensor var_9035_equation_0 = const()[name = tensor("op_9035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9035_cast_fp16 = einsum(equation = var_9035_equation_0, values = (var_8883_cast_fp16, var_9000_cast_fp16))[name = tensor("op_9035_cast_fp16")]; + tensor var_9037_equation_0 = const()[name = tensor("op_9037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9037_cast_fp16 = einsum(equation = var_9037_equation_0, values = (var_8887_cast_fp16, var_9001_cast_fp16))[name = tensor("op_9037_cast_fp16")]; + tensor var_9039_equation_0 = const()[name = tensor("op_9039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9039_cast_fp16 = einsum(equation = var_9039_equation_0, values = (var_8891_cast_fp16, var_9002_cast_fp16))[name = tensor("op_9039_cast_fp16")]; + tensor var_9041_equation_0 = const()[name = tensor("op_9041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9041_cast_fp16 = einsum(equation = var_9041_equation_0, values = (var_8895_cast_fp16, var_9003_cast_fp16))[name = tensor("op_9041_cast_fp16")]; + tensor var_9043_equation_0 = const()[name = tensor("op_9043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9043_cast_fp16 = einsum(equation = var_9043_equation_0, values = (var_8899_cast_fp16, var_9004_cast_fp16))[name = tensor("op_9043_cast_fp16")]; + tensor var_9045_equation_0 = const()[name = tensor("op_9045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9045_cast_fp16 = einsum(equation = var_9045_equation_0, values = (var_8903_cast_fp16, var_9005_cast_fp16))[name = tensor("op_9045_cast_fp16")]; + tensor input_181_interleave_0 = const()[name = tensor("input_181_interleave_0"), val = tensor(false)]; + tensor input_181_cast_fp16 = concat(axis = var_2624, interleave = input_181_interleave_0, values = (var_9007_cast_fp16, var_9009_cast_fp16, var_9011_cast_fp16, var_9013_cast_fp16, var_9015_cast_fp16, var_9017_cast_fp16, var_9019_cast_fp16, var_9021_cast_fp16, var_9023_cast_fp16, var_9025_cast_fp16, var_9027_cast_fp16, var_9029_cast_fp16, var_9031_cast_fp16, var_9033_cast_fp16, var_9035_cast_fp16, var_9037_cast_fp16, var_9039_cast_fp16, var_9041_cast_fp16, var_9043_cast_fp16, var_9045_cast_fp16))[name = tensor("input_181_cast_fp16")]; + tensor var_9055_pad_type_0 = const()[name = tensor("op_9055_pad_type_0"), val = tensor("valid")]; + tensor var_9055_strides_0 = const()[name = tensor("op_9055_strides_0"), val = tensor([1, 1])]; + tensor var_9055_pad_0 = const()[name = tensor("op_9055_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9055_dilations_0 = const()[name = tensor("op_9055_dilations_0"), val = tensor([1, 1])]; + tensor var_9055_groups_0 = const()[name = tensor("op_9055_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241114688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242343552))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242343744)))]; + tensor var_9055_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_9055_dilations_0, groups = var_9055_groups_0, pad = var_9055_pad_0, pad_type = var_9055_pad_type_0, strides = var_9055_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = tensor("op_9055_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = var_9055_cast_fp16, y = inputs_63_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; + tensor input_183_axes_0 = const()[name = tensor("input_183_axes_0"), val = tensor([1])]; + tensor input_183_gamma_0_to_fp16 = const()[name = tensor("input_183_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242346368)))]; + tensor input_183_beta_0_to_fp16 = const()[name = tensor("input_183_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242348992)))]; + tensor var_9065_to_fp16 = const()[name = tensor("op_9065_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_183_cast_fp16 = layer_norm(axes = input_183_axes_0, beta = input_183_beta_0_to_fp16, epsilon = var_9065_to_fp16, gamma = input_183_gamma_0_to_fp16, x = inputs_65_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor var_9085_pad_type_0 = const()[name = tensor("op_9085_pad_type_0"), val = tensor("valid")]; + tensor var_9085_strides_0 = const()[name = tensor("op_9085_strides_0"), val = tensor([1, 1])]; + tensor var_9085_pad_0 = const()[name = tensor("op_9085_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9085_dilations_0 = const()[name = tensor("op_9085_dilations_0"), val = tensor([1, 1])]; + tensor var_9085_groups_0 = const()[name = tensor("op_9085_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242351616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252182080))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252182272)))]; + tensor var_9085_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_9085_dilations_0, groups = var_9085_groups_0, pad = var_9085_pad_0, pad_type = var_9085_pad_type_0, strides = var_9085_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized, x = input_183_cast_fp16)[name = tensor("op_9085_cast_fp16")]; + tensor var_9086_split_sizes_0 = const()[name = tensor("op_9086_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_9086_axis_0 = const()[name = tensor("op_9086_axis_0"), val = tensor(1)]; + tensor var_9086_cast_fp16_0, tensor var_9086_cast_fp16_1 = split(axis = var_9086_axis_0, split_sizes = var_9086_split_sizes_0, x = var_9085_cast_fp16)[name = tensor("op_9086_cast_fp16")]; + tensor var_9088_mode_0 = const()[name = tensor("op_9088_mode_0"), val = tensor("EXACT")]; + tensor var_9088_cast_fp16 = gelu(mode = var_9088_mode_0, x = var_9086_cast_fp16_1)[name = tensor("op_9088_cast_fp16")]; + tensor input_185_cast_fp16 = mul(x = var_9086_cast_fp16_0, y = var_9088_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor var_9096_pad_type_0 = const()[name = tensor("op_9096_pad_type_0"), val = tensor("valid")]; + tensor var_9096_strides_0 = const()[name = tensor("op_9096_strides_0"), val = tensor([1, 1])]; + tensor var_9096_pad_0 = const()[name = tensor("op_9096_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9096_dilations_0 = const()[name = tensor("op_9096_dilations_0"), val = tensor([1, 1])]; + tensor var_9096_groups_0 = const()[name = tensor("op_9096_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252202816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257118080))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257118272)))]; + tensor var_9096_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_9096_dilations_0, groups = var_9096_groups_0, pad = var_9096_pad_0, pad_type = var_9096_pad_type_0, strides = var_9096_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor("op_9096_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = var_9096_cast_fp16, y = inputs_65_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; + tensor hidden_states_107_axes_0 = const()[name = tensor("hidden_states_107_axes_0"), val = tensor([1])]; + tensor hidden_states_107_gamma_0_to_fp16 = const()[name = tensor("hidden_states_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257120896)))]; + tensor hidden_states_107_beta_0_to_fp16 = const()[name = tensor("hidden_states_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257123520)))]; + tensor var_9112_to_fp16 = const()[name = tensor("op_9112_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_107_cast_fp16 = layer_norm(axes = hidden_states_107_axes_0, beta = hidden_states_107_beta_0_to_fp16, epsilon = var_9112_to_fp16, gamma = hidden_states_107_gamma_0_to_fp16, x = inputs_67_cast_fp16)[name = tensor("hidden_states_107_cast_fp16")]; + tensor q_45_pad_type_0 = const()[name = tensor("q_45_pad_type_0"), val = tensor("valid")]; + tensor q_45_strides_0 = const()[name = tensor("q_45_strides_0"), val = tensor([1, 1])]; + tensor q_45_pad_0 = const()[name = tensor("q_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_45_dilations_0 = const()[name = tensor("q_45_dilations_0"), val = tensor([1, 1])]; + tensor q_45_groups_0 = const()[name = tensor("q_45_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257126144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258355008))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_45_cast_fp16 = conv(dilations = q_45_dilations_0, groups = q_45_groups_0, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = q_45_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_107_cast_fp16)[name = tensor("q_45_cast_fp16")]; + tensor k_89_pad_type_0 = const()[name = tensor("k_89_pad_type_0"), val = tensor("valid")]; + tensor k_89_strides_0 = const()[name = tensor("k_89_strides_0"), val = tensor([1, 1])]; + tensor k_89_pad_0 = const()[name = tensor("k_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_89_dilations_0 = const()[name = tensor("k_89_dilations_0"), val = tensor([1, 1])]; + tensor k_89_groups_0 = const()[name = tensor("k_89_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258355200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259584064))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_89_cast_fp16 = conv(dilations = k_89_dilations_0, groups = k_89_groups_0, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = k_89_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_107_cast_fp16)[name = tensor("k_89_cast_fp16")]; + tensor v_45_pad_type_0 = const()[name = tensor("v_45_pad_type_0"), val = tensor("valid")]; + tensor v_45_strides_0 = const()[name = tensor("v_45_strides_0"), val = tensor([1, 1])]; + tensor v_45_pad_0 = const()[name = tensor("v_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_45_dilations_0 = const()[name = tensor("v_45_dilations_0"), val = tensor([1, 1])]; + tensor v_45_groups_0 = const()[name = tensor("v_45_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259584256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260813120))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_45_cast_fp16 = conv(dilations = v_45_dilations_0, groups = v_45_groups_0, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = v_45_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_107_cast_fp16)[name = tensor("v_45_cast_fp16")]; + tensor var_9145_begin_0 = const()[name = tensor("op_9145_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9145_end_0 = const()[name = tensor("op_9145_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_9145_end_mask_0 = const()[name = tensor("op_9145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9145_cast_fp16 = slice_by_index(begin = var_9145_begin_0, end = var_9145_end_0, end_mask = var_9145_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9145_cast_fp16")]; + tensor var_9149_begin_0 = const()[name = tensor("op_9149_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9149_end_0 = const()[name = tensor("op_9149_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_9149_end_mask_0 = const()[name = tensor("op_9149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9149_cast_fp16 = slice_by_index(begin = var_9149_begin_0, end = var_9149_end_0, end_mask = var_9149_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9149_cast_fp16")]; + tensor var_9153_begin_0 = const()[name = tensor("op_9153_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9153_end_0 = const()[name = tensor("op_9153_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_9153_end_mask_0 = const()[name = tensor("op_9153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9153_cast_fp16 = slice_by_index(begin = var_9153_begin_0, end = var_9153_end_0, end_mask = var_9153_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9153_cast_fp16")]; + tensor var_9157_begin_0 = const()[name = tensor("op_9157_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9157_end_0 = const()[name = tensor("op_9157_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_9157_end_mask_0 = const()[name = tensor("op_9157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9157_cast_fp16 = slice_by_index(begin = var_9157_begin_0, end = var_9157_end_0, end_mask = var_9157_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9157_cast_fp16")]; + tensor var_9161_begin_0 = const()[name = tensor("op_9161_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9161_end_0 = const()[name = tensor("op_9161_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_9161_end_mask_0 = const()[name = tensor("op_9161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9161_cast_fp16 = slice_by_index(begin = var_9161_begin_0, end = var_9161_end_0, end_mask = var_9161_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9161_cast_fp16")]; + tensor var_9165_begin_0 = const()[name = tensor("op_9165_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9165_end_0 = const()[name = tensor("op_9165_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_9165_end_mask_0 = const()[name = tensor("op_9165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9165_cast_fp16 = slice_by_index(begin = var_9165_begin_0, end = var_9165_end_0, end_mask = var_9165_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9165_cast_fp16")]; + tensor var_9169_begin_0 = const()[name = tensor("op_9169_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9169_end_0 = const()[name = tensor("op_9169_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_9169_end_mask_0 = const()[name = tensor("op_9169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9169_cast_fp16 = slice_by_index(begin = var_9169_begin_0, end = var_9169_end_0, end_mask = var_9169_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9169_cast_fp16")]; + tensor var_9173_begin_0 = const()[name = tensor("op_9173_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9173_end_0 = const()[name = tensor("op_9173_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_9173_end_mask_0 = const()[name = tensor("op_9173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9173_cast_fp16 = slice_by_index(begin = var_9173_begin_0, end = var_9173_end_0, end_mask = var_9173_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9173_cast_fp16")]; + tensor var_9177_begin_0 = const()[name = tensor("op_9177_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9177_end_0 = const()[name = tensor("op_9177_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_9177_end_mask_0 = const()[name = tensor("op_9177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9177_cast_fp16 = slice_by_index(begin = var_9177_begin_0, end = var_9177_end_0, end_mask = var_9177_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9177_cast_fp16")]; + tensor var_9181_begin_0 = const()[name = tensor("op_9181_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9181_end_0 = const()[name = tensor("op_9181_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_9181_end_mask_0 = const()[name = tensor("op_9181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9181_cast_fp16 = slice_by_index(begin = var_9181_begin_0, end = var_9181_end_0, end_mask = var_9181_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9181_cast_fp16")]; + tensor var_9185_begin_0 = const()[name = tensor("op_9185_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9185_end_0 = const()[name = tensor("op_9185_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_9185_end_mask_0 = const()[name = tensor("op_9185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9185_cast_fp16 = slice_by_index(begin = var_9185_begin_0, end = var_9185_end_0, end_mask = var_9185_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9185_cast_fp16")]; + tensor var_9189_begin_0 = const()[name = tensor("op_9189_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9189_end_0 = const()[name = tensor("op_9189_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_9189_end_mask_0 = const()[name = tensor("op_9189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9189_cast_fp16 = slice_by_index(begin = var_9189_begin_0, end = var_9189_end_0, end_mask = var_9189_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9189_cast_fp16")]; + tensor var_9193_begin_0 = const()[name = tensor("op_9193_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9193_end_0 = const()[name = tensor("op_9193_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_9193_end_mask_0 = const()[name = tensor("op_9193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9193_cast_fp16 = slice_by_index(begin = var_9193_begin_0, end = var_9193_end_0, end_mask = var_9193_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9193_cast_fp16")]; + tensor var_9197_begin_0 = const()[name = tensor("op_9197_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9197_end_0 = const()[name = tensor("op_9197_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_9197_end_mask_0 = const()[name = tensor("op_9197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9197_cast_fp16 = slice_by_index(begin = var_9197_begin_0, end = var_9197_end_0, end_mask = var_9197_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9197_cast_fp16")]; + tensor var_9201_begin_0 = const()[name = tensor("op_9201_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9201_end_0 = const()[name = tensor("op_9201_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_9201_end_mask_0 = const()[name = tensor("op_9201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9201_cast_fp16 = slice_by_index(begin = var_9201_begin_0, end = var_9201_end_0, end_mask = var_9201_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9201_cast_fp16")]; + tensor var_9205_begin_0 = const()[name = tensor("op_9205_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9205_end_0 = const()[name = tensor("op_9205_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9205_end_mask_0 = const()[name = tensor("op_9205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9205_cast_fp16 = slice_by_index(begin = var_9205_begin_0, end = var_9205_end_0, end_mask = var_9205_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9205_cast_fp16")]; + tensor var_9209_begin_0 = const()[name = tensor("op_9209_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9209_end_0 = const()[name = tensor("op_9209_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_9209_end_mask_0 = const()[name = tensor("op_9209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9209_cast_fp16 = slice_by_index(begin = var_9209_begin_0, end = var_9209_end_0, end_mask = var_9209_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9209_cast_fp16")]; + tensor var_9213_begin_0 = const()[name = tensor("op_9213_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9213_end_0 = const()[name = tensor("op_9213_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_9213_end_mask_0 = const()[name = tensor("op_9213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9213_cast_fp16 = slice_by_index(begin = var_9213_begin_0, end = var_9213_end_0, end_mask = var_9213_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9213_cast_fp16")]; + tensor var_9217_begin_0 = const()[name = tensor("op_9217_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9217_end_0 = const()[name = tensor("op_9217_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_9217_end_mask_0 = const()[name = tensor("op_9217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9217_cast_fp16 = slice_by_index(begin = var_9217_begin_0, end = var_9217_end_0, end_mask = var_9217_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9217_cast_fp16")]; + tensor var_9221_begin_0 = const()[name = tensor("op_9221_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9221_end_0 = const()[name = tensor("op_9221_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_9221_end_mask_0 = const()[name = tensor("op_9221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9221_cast_fp16 = slice_by_index(begin = var_9221_begin_0, end = var_9221_end_0, end_mask = var_9221_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9221_cast_fp16")]; + tensor k_91_perm_0 = const()[name = tensor("k_91_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9228_begin_0 = const()[name = tensor("op_9228_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9228_end_0 = const()[name = tensor("op_9228_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_9228_end_mask_0 = const()[name = tensor("op_9228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_91_cast_fp16 = transpose(perm = k_91_perm_0, x = k_89_cast_fp16)[name = tensor("transpose_45")]; + tensor var_9228_cast_fp16 = slice_by_index(begin = var_9228_begin_0, end = var_9228_end_0, end_mask = var_9228_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9228_cast_fp16")]; + tensor var_9232_begin_0 = const()[name = tensor("op_9232_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9232_end_0 = const()[name = tensor("op_9232_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_9232_end_mask_0 = const()[name = tensor("op_9232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9232_cast_fp16 = slice_by_index(begin = var_9232_begin_0, end = var_9232_end_0, end_mask = var_9232_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9232_cast_fp16")]; + tensor var_9236_begin_0 = const()[name = tensor("op_9236_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9236_end_0 = const()[name = tensor("op_9236_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_9236_end_mask_0 = const()[name = tensor("op_9236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9236_cast_fp16 = slice_by_index(begin = var_9236_begin_0, end = var_9236_end_0, end_mask = var_9236_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9236_cast_fp16")]; + tensor var_9240_begin_0 = const()[name = tensor("op_9240_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9240_end_0 = const()[name = tensor("op_9240_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_9240_end_mask_0 = const()[name = tensor("op_9240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9240_cast_fp16 = slice_by_index(begin = var_9240_begin_0, end = var_9240_end_0, end_mask = var_9240_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9240_cast_fp16")]; + tensor var_9244_begin_0 = const()[name = tensor("op_9244_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9244_end_0 = const()[name = tensor("op_9244_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_9244_end_mask_0 = const()[name = tensor("op_9244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9244_cast_fp16 = slice_by_index(begin = var_9244_begin_0, end = var_9244_end_0, end_mask = var_9244_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9244_cast_fp16")]; + tensor var_9248_begin_0 = const()[name = tensor("op_9248_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9248_end_0 = const()[name = tensor("op_9248_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_9248_end_mask_0 = const()[name = tensor("op_9248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9248_cast_fp16 = slice_by_index(begin = var_9248_begin_0, end = var_9248_end_0, end_mask = var_9248_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9248_cast_fp16")]; + tensor var_9252_begin_0 = const()[name = tensor("op_9252_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9252_end_0 = const()[name = tensor("op_9252_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_9252_end_mask_0 = const()[name = tensor("op_9252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9252_cast_fp16 = slice_by_index(begin = var_9252_begin_0, end = var_9252_end_0, end_mask = var_9252_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9252_cast_fp16")]; + tensor var_9256_begin_0 = const()[name = tensor("op_9256_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9256_end_0 = const()[name = tensor("op_9256_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_9256_end_mask_0 = const()[name = tensor("op_9256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9256_cast_fp16 = slice_by_index(begin = var_9256_begin_0, end = var_9256_end_0, end_mask = var_9256_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9256_cast_fp16")]; + tensor var_9260_begin_0 = const()[name = tensor("op_9260_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9260_end_0 = const()[name = tensor("op_9260_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_9260_end_mask_0 = const()[name = tensor("op_9260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9260_cast_fp16 = slice_by_index(begin = var_9260_begin_0, end = var_9260_end_0, end_mask = var_9260_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9260_cast_fp16")]; + tensor var_9264_begin_0 = const()[name = tensor("op_9264_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9264_end_0 = const()[name = tensor("op_9264_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_9264_end_mask_0 = const()[name = tensor("op_9264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9264_cast_fp16 = slice_by_index(begin = var_9264_begin_0, end = var_9264_end_0, end_mask = var_9264_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9264_cast_fp16")]; + tensor var_9268_begin_0 = const()[name = tensor("op_9268_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9268_end_0 = const()[name = tensor("op_9268_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_9268_end_mask_0 = const()[name = tensor("op_9268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9268_cast_fp16 = slice_by_index(begin = var_9268_begin_0, end = var_9268_end_0, end_mask = var_9268_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9268_cast_fp16")]; + tensor var_9272_begin_0 = const()[name = tensor("op_9272_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9272_end_0 = const()[name = tensor("op_9272_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_9272_end_mask_0 = const()[name = tensor("op_9272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9272_cast_fp16 = slice_by_index(begin = var_9272_begin_0, end = var_9272_end_0, end_mask = var_9272_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9272_cast_fp16")]; + tensor var_9276_begin_0 = const()[name = tensor("op_9276_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_9276_end_0 = const()[name = tensor("op_9276_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_9276_end_mask_0 = const()[name = tensor("op_9276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9276_cast_fp16 = slice_by_index(begin = var_9276_begin_0, end = var_9276_end_0, end_mask = var_9276_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9276_cast_fp16")]; + tensor var_9280_begin_0 = const()[name = tensor("op_9280_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_9280_end_0 = const()[name = tensor("op_9280_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_9280_end_mask_0 = const()[name = tensor("op_9280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9280_cast_fp16 = slice_by_index(begin = var_9280_begin_0, end = var_9280_end_0, end_mask = var_9280_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9280_cast_fp16")]; + tensor var_9284_begin_0 = const()[name = tensor("op_9284_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_9284_end_0 = const()[name = tensor("op_9284_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_9284_end_mask_0 = const()[name = tensor("op_9284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9284_cast_fp16 = slice_by_index(begin = var_9284_begin_0, end = var_9284_end_0, end_mask = var_9284_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9284_cast_fp16")]; + tensor var_9288_begin_0 = const()[name = tensor("op_9288_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_9288_end_0 = const()[name = tensor("op_9288_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9288_end_mask_0 = const()[name = tensor("op_9288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9288_cast_fp16 = slice_by_index(begin = var_9288_begin_0, end = var_9288_end_0, end_mask = var_9288_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9288_cast_fp16")]; + tensor var_9292_begin_0 = const()[name = tensor("op_9292_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_9292_end_0 = const()[name = tensor("op_9292_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_9292_end_mask_0 = const()[name = tensor("op_9292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9292_cast_fp16 = slice_by_index(begin = var_9292_begin_0, end = var_9292_end_0, end_mask = var_9292_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9292_cast_fp16")]; + tensor var_9296_begin_0 = const()[name = tensor("op_9296_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_9296_end_0 = const()[name = tensor("op_9296_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_9296_end_mask_0 = const()[name = tensor("op_9296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9296_cast_fp16 = slice_by_index(begin = var_9296_begin_0, end = var_9296_end_0, end_mask = var_9296_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9296_cast_fp16")]; + tensor var_9300_begin_0 = const()[name = tensor("op_9300_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_9300_end_0 = const()[name = tensor("op_9300_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_9300_end_mask_0 = const()[name = tensor("op_9300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9300_cast_fp16 = slice_by_index(begin = var_9300_begin_0, end = var_9300_end_0, end_mask = var_9300_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9300_cast_fp16")]; + tensor var_9304_begin_0 = const()[name = tensor("op_9304_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_9304_end_0 = const()[name = tensor("op_9304_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_9304_end_mask_0 = const()[name = tensor("op_9304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9304_cast_fp16 = slice_by_index(begin = var_9304_begin_0, end = var_9304_end_0, end_mask = var_9304_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9304_cast_fp16")]; + tensor var_9306_begin_0 = const()[name = tensor("op_9306_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9306_end_0 = const()[name = tensor("op_9306_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_9306_end_mask_0 = const()[name = tensor("op_9306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9306_cast_fp16 = slice_by_index(begin = var_9306_begin_0, end = var_9306_end_0, end_mask = var_9306_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9306_cast_fp16")]; + tensor var_9310_begin_0 = const()[name = tensor("op_9310_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9310_end_0 = const()[name = tensor("op_9310_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_9310_end_mask_0 = const()[name = tensor("op_9310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9310_cast_fp16 = slice_by_index(begin = var_9310_begin_0, end = var_9310_end_0, end_mask = var_9310_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9310_cast_fp16")]; + tensor var_9314_begin_0 = const()[name = tensor("op_9314_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9314_end_0 = const()[name = tensor("op_9314_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_9314_end_mask_0 = const()[name = tensor("op_9314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9314_cast_fp16 = slice_by_index(begin = var_9314_begin_0, end = var_9314_end_0, end_mask = var_9314_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9314_cast_fp16")]; + tensor var_9318_begin_0 = const()[name = tensor("op_9318_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9318_end_0 = const()[name = tensor("op_9318_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_9318_end_mask_0 = const()[name = tensor("op_9318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9318_cast_fp16 = slice_by_index(begin = var_9318_begin_0, end = var_9318_end_0, end_mask = var_9318_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9318_cast_fp16")]; + tensor var_9322_begin_0 = const()[name = tensor("op_9322_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9322_end_0 = const()[name = tensor("op_9322_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_9322_end_mask_0 = const()[name = tensor("op_9322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9322_cast_fp16 = slice_by_index(begin = var_9322_begin_0, end = var_9322_end_0, end_mask = var_9322_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9322_cast_fp16")]; + tensor var_9326_begin_0 = const()[name = tensor("op_9326_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9326_end_0 = const()[name = tensor("op_9326_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_9326_end_mask_0 = const()[name = tensor("op_9326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9326_cast_fp16 = slice_by_index(begin = var_9326_begin_0, end = var_9326_end_0, end_mask = var_9326_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9326_cast_fp16")]; + tensor var_9330_begin_0 = const()[name = tensor("op_9330_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9330_end_0 = const()[name = tensor("op_9330_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_9330_end_mask_0 = const()[name = tensor("op_9330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9330_cast_fp16 = slice_by_index(begin = var_9330_begin_0, end = var_9330_end_0, end_mask = var_9330_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9330_cast_fp16")]; + tensor var_9334_begin_0 = const()[name = tensor("op_9334_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9334_end_0 = const()[name = tensor("op_9334_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_9334_end_mask_0 = const()[name = tensor("op_9334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9334_cast_fp16 = slice_by_index(begin = var_9334_begin_0, end = var_9334_end_0, end_mask = var_9334_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9334_cast_fp16")]; + tensor var_9338_begin_0 = const()[name = tensor("op_9338_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9338_end_0 = const()[name = tensor("op_9338_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_9338_end_mask_0 = const()[name = tensor("op_9338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9338_cast_fp16 = slice_by_index(begin = var_9338_begin_0, end = var_9338_end_0, end_mask = var_9338_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9338_cast_fp16")]; + tensor var_9342_begin_0 = const()[name = tensor("op_9342_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9342_end_0 = const()[name = tensor("op_9342_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_9342_end_mask_0 = const()[name = tensor("op_9342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9342_cast_fp16 = slice_by_index(begin = var_9342_begin_0, end = var_9342_end_0, end_mask = var_9342_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9342_cast_fp16")]; + tensor var_9346_begin_0 = const()[name = tensor("op_9346_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9346_end_0 = const()[name = tensor("op_9346_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_9346_end_mask_0 = const()[name = tensor("op_9346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9346_cast_fp16 = slice_by_index(begin = var_9346_begin_0, end = var_9346_end_0, end_mask = var_9346_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9346_cast_fp16")]; + tensor var_9350_begin_0 = const()[name = tensor("op_9350_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9350_end_0 = const()[name = tensor("op_9350_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_9350_end_mask_0 = const()[name = tensor("op_9350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9350_cast_fp16 = slice_by_index(begin = var_9350_begin_0, end = var_9350_end_0, end_mask = var_9350_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9350_cast_fp16")]; + tensor var_9354_begin_0 = const()[name = tensor("op_9354_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9354_end_0 = const()[name = tensor("op_9354_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_9354_end_mask_0 = const()[name = tensor("op_9354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9354_cast_fp16 = slice_by_index(begin = var_9354_begin_0, end = var_9354_end_0, end_mask = var_9354_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9354_cast_fp16")]; + tensor var_9358_begin_0 = const()[name = tensor("op_9358_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9358_end_0 = const()[name = tensor("op_9358_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_9358_end_mask_0 = const()[name = tensor("op_9358_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9358_cast_fp16 = slice_by_index(begin = var_9358_begin_0, end = var_9358_end_0, end_mask = var_9358_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9358_cast_fp16")]; + tensor var_9362_begin_0 = const()[name = tensor("op_9362_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9362_end_0 = const()[name = tensor("op_9362_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_9362_end_mask_0 = const()[name = tensor("op_9362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9362_cast_fp16 = slice_by_index(begin = var_9362_begin_0, end = var_9362_end_0, end_mask = var_9362_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9362_cast_fp16")]; + tensor var_9366_begin_0 = const()[name = tensor("op_9366_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9366_end_0 = const()[name = tensor("op_9366_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9366_end_mask_0 = const()[name = tensor("op_9366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9366_cast_fp16 = slice_by_index(begin = var_9366_begin_0, end = var_9366_end_0, end_mask = var_9366_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9366_cast_fp16")]; + tensor var_9370_begin_0 = const()[name = tensor("op_9370_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9370_end_0 = const()[name = tensor("op_9370_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_9370_end_mask_0 = const()[name = tensor("op_9370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9370_cast_fp16 = slice_by_index(begin = var_9370_begin_0, end = var_9370_end_0, end_mask = var_9370_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9370_cast_fp16")]; + tensor var_9374_begin_0 = const()[name = tensor("op_9374_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9374_end_0 = const()[name = tensor("op_9374_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_9374_end_mask_0 = const()[name = tensor("op_9374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9374_cast_fp16 = slice_by_index(begin = var_9374_begin_0, end = var_9374_end_0, end_mask = var_9374_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9374_cast_fp16")]; + tensor var_9378_begin_0 = const()[name = tensor("op_9378_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9378_end_0 = const()[name = tensor("op_9378_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_9378_end_mask_0 = const()[name = tensor("op_9378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9378_cast_fp16 = slice_by_index(begin = var_9378_begin_0, end = var_9378_end_0, end_mask = var_9378_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9378_cast_fp16")]; + tensor var_9382_begin_0 = const()[name = tensor("op_9382_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9382_end_0 = const()[name = tensor("op_9382_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_9382_end_mask_0 = const()[name = tensor("op_9382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9382_cast_fp16 = slice_by_index(begin = var_9382_begin_0, end = var_9382_end_0, end_mask = var_9382_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9382_cast_fp16")]; + tensor var_9386_equation_0 = const()[name = tensor("op_9386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9386_cast_fp16 = einsum(equation = var_9386_equation_0, values = (var_9228_cast_fp16, var_9145_cast_fp16))[name = tensor("op_9386_cast_fp16")]; + tensor var_9387_to_fp16 = const()[name = tensor("op_9387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_721_cast_fp16 = mul(x = var_9386_cast_fp16, y = var_9387_to_fp16)[name = tensor("aw_721_cast_fp16")]; + tensor var_9390_equation_0 = const()[name = tensor("op_9390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9390_cast_fp16 = einsum(equation = var_9390_equation_0, values = (var_9232_cast_fp16, var_9149_cast_fp16))[name = tensor("op_9390_cast_fp16")]; + tensor var_9391_to_fp16 = const()[name = tensor("op_9391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_723_cast_fp16 = mul(x = var_9390_cast_fp16, y = var_9391_to_fp16)[name = tensor("aw_723_cast_fp16")]; + tensor var_9394_equation_0 = const()[name = tensor("op_9394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9394_cast_fp16 = einsum(equation = var_9394_equation_0, values = (var_9236_cast_fp16, var_9153_cast_fp16))[name = tensor("op_9394_cast_fp16")]; + tensor var_9395_to_fp16 = const()[name = tensor("op_9395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_725_cast_fp16 = mul(x = var_9394_cast_fp16, y = var_9395_to_fp16)[name = tensor("aw_725_cast_fp16")]; + tensor var_9398_equation_0 = const()[name = tensor("op_9398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9398_cast_fp16 = einsum(equation = var_9398_equation_0, values = (var_9240_cast_fp16, var_9157_cast_fp16))[name = tensor("op_9398_cast_fp16")]; + tensor var_9399_to_fp16 = const()[name = tensor("op_9399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_727_cast_fp16 = mul(x = var_9398_cast_fp16, y = var_9399_to_fp16)[name = tensor("aw_727_cast_fp16")]; + tensor var_9402_equation_0 = const()[name = tensor("op_9402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9402_cast_fp16 = einsum(equation = var_9402_equation_0, values = (var_9244_cast_fp16, var_9161_cast_fp16))[name = tensor("op_9402_cast_fp16")]; + tensor var_9403_to_fp16 = const()[name = tensor("op_9403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_729_cast_fp16 = mul(x = var_9402_cast_fp16, y = var_9403_to_fp16)[name = tensor("aw_729_cast_fp16")]; + tensor var_9406_equation_0 = const()[name = tensor("op_9406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9406_cast_fp16 = einsum(equation = var_9406_equation_0, values = (var_9248_cast_fp16, var_9165_cast_fp16))[name = tensor("op_9406_cast_fp16")]; + tensor var_9407_to_fp16 = const()[name = tensor("op_9407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_731_cast_fp16 = mul(x = var_9406_cast_fp16, y = var_9407_to_fp16)[name = tensor("aw_731_cast_fp16")]; + tensor var_9410_equation_0 = const()[name = tensor("op_9410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9410_cast_fp16 = einsum(equation = var_9410_equation_0, values = (var_9252_cast_fp16, var_9169_cast_fp16))[name = tensor("op_9410_cast_fp16")]; + tensor var_9411_to_fp16 = const()[name = tensor("op_9411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_733_cast_fp16 = mul(x = var_9410_cast_fp16, y = var_9411_to_fp16)[name = tensor("aw_733_cast_fp16")]; + tensor var_9414_equation_0 = const()[name = tensor("op_9414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9414_cast_fp16 = einsum(equation = var_9414_equation_0, values = (var_9256_cast_fp16, var_9173_cast_fp16))[name = tensor("op_9414_cast_fp16")]; + tensor var_9415_to_fp16 = const()[name = tensor("op_9415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_735_cast_fp16 = mul(x = var_9414_cast_fp16, y = var_9415_to_fp16)[name = tensor("aw_735_cast_fp16")]; + tensor var_9418_equation_0 = const()[name = tensor("op_9418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9418_cast_fp16 = einsum(equation = var_9418_equation_0, values = (var_9260_cast_fp16, var_9177_cast_fp16))[name = tensor("op_9418_cast_fp16")]; + tensor var_9419_to_fp16 = const()[name = tensor("op_9419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_737_cast_fp16 = mul(x = var_9418_cast_fp16, y = var_9419_to_fp16)[name = tensor("aw_737_cast_fp16")]; + tensor var_9422_equation_0 = const()[name = tensor("op_9422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9422_cast_fp16 = einsum(equation = var_9422_equation_0, values = (var_9264_cast_fp16, var_9181_cast_fp16))[name = tensor("op_9422_cast_fp16")]; + tensor var_9423_to_fp16 = const()[name = tensor("op_9423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_739_cast_fp16 = mul(x = var_9422_cast_fp16, y = var_9423_to_fp16)[name = tensor("aw_739_cast_fp16")]; + tensor var_9426_equation_0 = const()[name = tensor("op_9426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9426_cast_fp16 = einsum(equation = var_9426_equation_0, values = (var_9268_cast_fp16, var_9185_cast_fp16))[name = tensor("op_9426_cast_fp16")]; + tensor var_9427_to_fp16 = const()[name = tensor("op_9427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_741_cast_fp16 = mul(x = var_9426_cast_fp16, y = var_9427_to_fp16)[name = tensor("aw_741_cast_fp16")]; + tensor var_9430_equation_0 = const()[name = tensor("op_9430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9430_cast_fp16 = einsum(equation = var_9430_equation_0, values = (var_9272_cast_fp16, var_9189_cast_fp16))[name = tensor("op_9430_cast_fp16")]; + tensor var_9431_to_fp16 = const()[name = tensor("op_9431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_743_cast_fp16 = mul(x = var_9430_cast_fp16, y = var_9431_to_fp16)[name = tensor("aw_743_cast_fp16")]; + tensor var_9434_equation_0 = const()[name = tensor("op_9434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9434_cast_fp16 = einsum(equation = var_9434_equation_0, values = (var_9276_cast_fp16, var_9193_cast_fp16))[name = tensor("op_9434_cast_fp16")]; + tensor var_9435_to_fp16 = const()[name = tensor("op_9435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_745_cast_fp16 = mul(x = var_9434_cast_fp16, y = var_9435_to_fp16)[name = tensor("aw_745_cast_fp16")]; + tensor var_9438_equation_0 = const()[name = tensor("op_9438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9438_cast_fp16 = einsum(equation = var_9438_equation_0, values = (var_9280_cast_fp16, var_9197_cast_fp16))[name = tensor("op_9438_cast_fp16")]; + tensor var_9439_to_fp16 = const()[name = tensor("op_9439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_747_cast_fp16 = mul(x = var_9438_cast_fp16, y = var_9439_to_fp16)[name = tensor("aw_747_cast_fp16")]; + tensor var_9442_equation_0 = const()[name = tensor("op_9442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9442_cast_fp16 = einsum(equation = var_9442_equation_0, values = (var_9284_cast_fp16, var_9201_cast_fp16))[name = tensor("op_9442_cast_fp16")]; + tensor var_9443_to_fp16 = const()[name = tensor("op_9443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_749_cast_fp16 = mul(x = var_9442_cast_fp16, y = var_9443_to_fp16)[name = tensor("aw_749_cast_fp16")]; + tensor var_9446_equation_0 = const()[name = tensor("op_9446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9446_cast_fp16 = einsum(equation = var_9446_equation_0, values = (var_9288_cast_fp16, var_9205_cast_fp16))[name = tensor("op_9446_cast_fp16")]; + tensor var_9447_to_fp16 = const()[name = tensor("op_9447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_751_cast_fp16 = mul(x = var_9446_cast_fp16, y = var_9447_to_fp16)[name = tensor("aw_751_cast_fp16")]; + tensor var_9450_equation_0 = const()[name = tensor("op_9450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9450_cast_fp16 = einsum(equation = var_9450_equation_0, values = (var_9292_cast_fp16, var_9209_cast_fp16))[name = tensor("op_9450_cast_fp16")]; + tensor var_9451_to_fp16 = const()[name = tensor("op_9451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_753_cast_fp16 = mul(x = var_9450_cast_fp16, y = var_9451_to_fp16)[name = tensor("aw_753_cast_fp16")]; + tensor var_9454_equation_0 = const()[name = tensor("op_9454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9454_cast_fp16 = einsum(equation = var_9454_equation_0, values = (var_9296_cast_fp16, var_9213_cast_fp16))[name = tensor("op_9454_cast_fp16")]; + tensor var_9455_to_fp16 = const()[name = tensor("op_9455_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_755_cast_fp16 = mul(x = var_9454_cast_fp16, y = var_9455_to_fp16)[name = tensor("aw_755_cast_fp16")]; + tensor var_9458_equation_0 = const()[name = tensor("op_9458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9458_cast_fp16 = einsum(equation = var_9458_equation_0, values = (var_9300_cast_fp16, var_9217_cast_fp16))[name = tensor("op_9458_cast_fp16")]; + tensor var_9459_to_fp16 = const()[name = tensor("op_9459_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_757_cast_fp16 = mul(x = var_9458_cast_fp16, y = var_9459_to_fp16)[name = tensor("aw_757_cast_fp16")]; + tensor var_9462_equation_0 = const()[name = tensor("op_9462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9462_cast_fp16 = einsum(equation = var_9462_equation_0, values = (var_9304_cast_fp16, var_9221_cast_fp16))[name = tensor("op_9462_cast_fp16")]; + tensor var_9463_to_fp16 = const()[name = tensor("op_9463_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_759_cast_fp16 = mul(x = var_9462_cast_fp16, y = var_9463_to_fp16)[name = tensor("aw_759_cast_fp16")]; + tensor var_9465_cast_fp16 = softmax(axis = var_2624, x = aw_721_cast_fp16)[name = tensor("op_9465_cast_fp16")]; + tensor var_9466_cast_fp16 = softmax(axis = var_2624, x = aw_723_cast_fp16)[name = tensor("op_9466_cast_fp16")]; + tensor var_9467_cast_fp16 = softmax(axis = var_2624, x = aw_725_cast_fp16)[name = tensor("op_9467_cast_fp16")]; + tensor var_9468_cast_fp16 = softmax(axis = var_2624, x = aw_727_cast_fp16)[name = tensor("op_9468_cast_fp16")]; + tensor var_9469_cast_fp16 = softmax(axis = var_2624, x = aw_729_cast_fp16)[name = tensor("op_9469_cast_fp16")]; + tensor var_9470_cast_fp16 = softmax(axis = var_2624, x = aw_731_cast_fp16)[name = tensor("op_9470_cast_fp16")]; + tensor var_9471_cast_fp16 = softmax(axis = var_2624, x = aw_733_cast_fp16)[name = tensor("op_9471_cast_fp16")]; + tensor var_9472_cast_fp16 = softmax(axis = var_2624, x = aw_735_cast_fp16)[name = tensor("op_9472_cast_fp16")]; + tensor var_9473_cast_fp16 = softmax(axis = var_2624, x = aw_737_cast_fp16)[name = tensor("op_9473_cast_fp16")]; + tensor var_9474_cast_fp16 = softmax(axis = var_2624, x = aw_739_cast_fp16)[name = tensor("op_9474_cast_fp16")]; + tensor var_9475_cast_fp16 = softmax(axis = var_2624, x = aw_741_cast_fp16)[name = tensor("op_9475_cast_fp16")]; + tensor var_9476_cast_fp16 = softmax(axis = var_2624, x = aw_743_cast_fp16)[name = tensor("op_9476_cast_fp16")]; + tensor var_9477_cast_fp16 = softmax(axis = var_2624, x = aw_745_cast_fp16)[name = tensor("op_9477_cast_fp16")]; + tensor var_9478_cast_fp16 = softmax(axis = var_2624, x = aw_747_cast_fp16)[name = tensor("op_9478_cast_fp16")]; + tensor var_9479_cast_fp16 = softmax(axis = var_2624, x = aw_749_cast_fp16)[name = tensor("op_9479_cast_fp16")]; + tensor var_9480_cast_fp16 = softmax(axis = var_2624, x = aw_751_cast_fp16)[name = tensor("op_9480_cast_fp16")]; + tensor var_9481_cast_fp16 = softmax(axis = var_2624, x = aw_753_cast_fp16)[name = tensor("op_9481_cast_fp16")]; + tensor var_9482_cast_fp16 = softmax(axis = var_2624, x = aw_755_cast_fp16)[name = tensor("op_9482_cast_fp16")]; + tensor var_9483_cast_fp16 = softmax(axis = var_2624, x = aw_757_cast_fp16)[name = tensor("op_9483_cast_fp16")]; + tensor var_9484_cast_fp16 = softmax(axis = var_2624, x = aw_759_cast_fp16)[name = tensor("op_9484_cast_fp16")]; + tensor var_9486_equation_0 = const()[name = tensor("op_9486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9486_cast_fp16 = einsum(equation = var_9486_equation_0, values = (var_9306_cast_fp16, var_9465_cast_fp16))[name = tensor("op_9486_cast_fp16")]; + tensor var_9488_equation_0 = const()[name = tensor("op_9488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9488_cast_fp16 = einsum(equation = var_9488_equation_0, values = (var_9310_cast_fp16, var_9466_cast_fp16))[name = tensor("op_9488_cast_fp16")]; + tensor var_9490_equation_0 = const()[name = tensor("op_9490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9490_cast_fp16 = einsum(equation = var_9490_equation_0, values = (var_9314_cast_fp16, var_9467_cast_fp16))[name = tensor("op_9490_cast_fp16")]; + tensor var_9492_equation_0 = const()[name = tensor("op_9492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9492_cast_fp16 = einsum(equation = var_9492_equation_0, values = (var_9318_cast_fp16, var_9468_cast_fp16))[name = tensor("op_9492_cast_fp16")]; + tensor var_9494_equation_0 = const()[name = tensor("op_9494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9494_cast_fp16 = einsum(equation = var_9494_equation_0, values = (var_9322_cast_fp16, var_9469_cast_fp16))[name = tensor("op_9494_cast_fp16")]; + tensor var_9496_equation_0 = const()[name = tensor("op_9496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9496_cast_fp16 = einsum(equation = var_9496_equation_0, values = (var_9326_cast_fp16, var_9470_cast_fp16))[name = tensor("op_9496_cast_fp16")]; + tensor var_9498_equation_0 = const()[name = tensor("op_9498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9498_cast_fp16 = einsum(equation = var_9498_equation_0, values = (var_9330_cast_fp16, var_9471_cast_fp16))[name = tensor("op_9498_cast_fp16")]; + tensor var_9500_equation_0 = const()[name = tensor("op_9500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9500_cast_fp16 = einsum(equation = var_9500_equation_0, values = (var_9334_cast_fp16, var_9472_cast_fp16))[name = tensor("op_9500_cast_fp16")]; + tensor var_9502_equation_0 = const()[name = tensor("op_9502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9502_cast_fp16 = einsum(equation = var_9502_equation_0, values = (var_9338_cast_fp16, var_9473_cast_fp16))[name = tensor("op_9502_cast_fp16")]; + tensor var_9504_equation_0 = const()[name = tensor("op_9504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9504_cast_fp16 = einsum(equation = var_9504_equation_0, values = (var_9342_cast_fp16, var_9474_cast_fp16))[name = tensor("op_9504_cast_fp16")]; + tensor var_9506_equation_0 = const()[name = tensor("op_9506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9506_cast_fp16 = einsum(equation = var_9506_equation_0, values = (var_9346_cast_fp16, var_9475_cast_fp16))[name = tensor("op_9506_cast_fp16")]; + tensor var_9508_equation_0 = const()[name = tensor("op_9508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9508_cast_fp16 = einsum(equation = var_9508_equation_0, values = (var_9350_cast_fp16, var_9476_cast_fp16))[name = tensor("op_9508_cast_fp16")]; + tensor var_9510_equation_0 = const()[name = tensor("op_9510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9510_cast_fp16 = einsum(equation = var_9510_equation_0, values = (var_9354_cast_fp16, var_9477_cast_fp16))[name = tensor("op_9510_cast_fp16")]; + tensor var_9512_equation_0 = const()[name = tensor("op_9512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9512_cast_fp16 = einsum(equation = var_9512_equation_0, values = (var_9358_cast_fp16, var_9478_cast_fp16))[name = tensor("op_9512_cast_fp16")]; + tensor var_9514_equation_0 = const()[name = tensor("op_9514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9514_cast_fp16 = einsum(equation = var_9514_equation_0, values = (var_9362_cast_fp16, var_9479_cast_fp16))[name = tensor("op_9514_cast_fp16")]; + tensor var_9516_equation_0 = const()[name = tensor("op_9516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9516_cast_fp16 = einsum(equation = var_9516_equation_0, values = (var_9366_cast_fp16, var_9480_cast_fp16))[name = tensor("op_9516_cast_fp16")]; + tensor var_9518_equation_0 = const()[name = tensor("op_9518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9518_cast_fp16 = einsum(equation = var_9518_equation_0, values = (var_9370_cast_fp16, var_9481_cast_fp16))[name = tensor("op_9518_cast_fp16")]; + tensor var_9520_equation_0 = const()[name = tensor("op_9520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9520_cast_fp16 = einsum(equation = var_9520_equation_0, values = (var_9374_cast_fp16, var_9482_cast_fp16))[name = tensor("op_9520_cast_fp16")]; + tensor var_9522_equation_0 = const()[name = tensor("op_9522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9522_cast_fp16 = einsum(equation = var_9522_equation_0, values = (var_9378_cast_fp16, var_9483_cast_fp16))[name = tensor("op_9522_cast_fp16")]; + tensor var_9524_equation_0 = const()[name = tensor("op_9524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9524_cast_fp16 = einsum(equation = var_9524_equation_0, values = (var_9382_cast_fp16, var_9484_cast_fp16))[name = tensor("op_9524_cast_fp16")]; + tensor input_187_interleave_0 = const()[name = tensor("input_187_interleave_0"), val = tensor(false)]; + tensor input_187_cast_fp16 = concat(axis = var_2624, interleave = input_187_interleave_0, values = (var_9486_cast_fp16, var_9488_cast_fp16, var_9490_cast_fp16, var_9492_cast_fp16, var_9494_cast_fp16, var_9496_cast_fp16, var_9498_cast_fp16, var_9500_cast_fp16, var_9502_cast_fp16, var_9504_cast_fp16, var_9506_cast_fp16, var_9508_cast_fp16, var_9510_cast_fp16, var_9512_cast_fp16, var_9514_cast_fp16, var_9516_cast_fp16, var_9518_cast_fp16, var_9520_cast_fp16, var_9522_cast_fp16, var_9524_cast_fp16))[name = tensor("input_187_cast_fp16")]; + tensor var_9534_pad_type_0 = const()[name = tensor("op_9534_pad_type_0"), val = tensor("valid")]; + tensor var_9534_strides_0 = const()[name = tensor("op_9534_strides_0"), val = tensor([1, 1])]; + tensor var_9534_pad_0 = const()[name = tensor("op_9534_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9534_dilations_0 = const()[name = tensor("op_9534_dilations_0"), val = tensor([1, 1])]; + tensor var_9534_groups_0 = const()[name = tensor("op_9534_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260813312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262042176))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262042368)))]; + tensor var_9534_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_9534_dilations_0, groups = var_9534_groups_0, pad = var_9534_pad_0, pad_type = var_9534_pad_type_0, strides = var_9534_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized, x = input_187_cast_fp16)[name = tensor("op_9534_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = var_9534_cast_fp16, y = inputs_67_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; + tensor hidden_states_109_axes_0 = const()[name = tensor("hidden_states_109_axes_0"), val = tensor([1])]; + tensor hidden_states_109_gamma_0_to_fp16 = const()[name = tensor("hidden_states_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262044992)))]; + tensor hidden_states_109_beta_0_to_fp16 = const()[name = tensor("hidden_states_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262047616)))]; + tensor var_9544_to_fp16 = const()[name = tensor("op_9544_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_109_cast_fp16 = layer_norm(axes = hidden_states_109_axes_0, beta = hidden_states_109_beta_0_to_fp16, epsilon = var_9544_to_fp16, gamma = hidden_states_109_gamma_0_to_fp16, x = inputs_69_cast_fp16)[name = tensor("hidden_states_109_cast_fp16")]; + tensor q_47_pad_type_0 = const()[name = tensor("q_47_pad_type_0"), val = tensor("valid")]; + tensor q_47_strides_0 = const()[name = tensor("q_47_strides_0"), val = tensor([1, 1])]; + tensor q_47_pad_0 = const()[name = tensor("q_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_47_dilations_0 = const()[name = tensor("q_47_dilations_0"), val = tensor([1, 1])]; + tensor q_47_groups_0 = const()[name = tensor("q_47_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(262050240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263279104))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_47_cast_fp16 = conv(dilations = q_47_dilations_0, groups = q_47_groups_0, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = q_47_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_109_cast_fp16)[name = tensor("q_47_cast_fp16")]; + tensor k_93_pad_type_0 = const()[name = tensor("k_93_pad_type_0"), val = tensor("valid")]; + tensor k_93_strides_0 = const()[name = tensor("k_93_strides_0"), val = tensor([1, 1])]; + tensor k_93_pad_0 = const()[name = tensor("k_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_93_dilations_0 = const()[name = tensor("k_93_dilations_0"), val = tensor([1, 1])]; + tensor k_93_groups_0 = const()[name = tensor("k_93_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263279296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265245440))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_93_cast_fp16 = conv(dilations = k_93_dilations_0, groups = k_93_groups_0, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = k_93_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_93_cast_fp16")]; + tensor v_47_pad_type_0 = const()[name = tensor("v_47_pad_type_0"), val = tensor("valid")]; + tensor v_47_strides_0 = const()[name = tensor("v_47_strides_0"), val = tensor([1, 1])]; + tensor v_47_pad_0 = const()[name = tensor("v_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_47_dilations_0 = const()[name = tensor("v_47_dilations_0"), val = tensor([1, 1])]; + tensor v_47_groups_0 = const()[name = tensor("v_47_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265245632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267211776))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_47_cast_fp16 = conv(dilations = v_47_dilations_0, groups = v_47_groups_0, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = v_47_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_47_cast_fp16")]; + tensor var_9577_begin_0 = const()[name = tensor("op_9577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9577_end_0 = const()[name = tensor("op_9577_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_9577_end_mask_0 = const()[name = tensor("op_9577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9577_cast_fp16 = slice_by_index(begin = var_9577_begin_0, end = var_9577_end_0, end_mask = var_9577_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9577_cast_fp16")]; + tensor var_9581_begin_0 = const()[name = tensor("op_9581_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9581_end_0 = const()[name = tensor("op_9581_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_9581_end_mask_0 = const()[name = tensor("op_9581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9581_cast_fp16 = slice_by_index(begin = var_9581_begin_0, end = var_9581_end_0, end_mask = var_9581_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9581_cast_fp16")]; + tensor var_9585_begin_0 = const()[name = tensor("op_9585_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9585_end_0 = const()[name = tensor("op_9585_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_9585_end_mask_0 = const()[name = tensor("op_9585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9585_cast_fp16 = slice_by_index(begin = var_9585_begin_0, end = var_9585_end_0, end_mask = var_9585_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9585_cast_fp16")]; + tensor var_9589_begin_0 = const()[name = tensor("op_9589_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9589_end_0 = const()[name = tensor("op_9589_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_9589_end_mask_0 = const()[name = tensor("op_9589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9589_cast_fp16 = slice_by_index(begin = var_9589_begin_0, end = var_9589_end_0, end_mask = var_9589_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9589_cast_fp16")]; + tensor var_9593_begin_0 = const()[name = tensor("op_9593_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9593_end_0 = const()[name = tensor("op_9593_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_9593_end_mask_0 = const()[name = tensor("op_9593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9593_cast_fp16 = slice_by_index(begin = var_9593_begin_0, end = var_9593_end_0, end_mask = var_9593_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9593_cast_fp16")]; + tensor var_9597_begin_0 = const()[name = tensor("op_9597_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9597_end_0 = const()[name = tensor("op_9597_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_9597_end_mask_0 = const()[name = tensor("op_9597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9597_cast_fp16 = slice_by_index(begin = var_9597_begin_0, end = var_9597_end_0, end_mask = var_9597_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9597_cast_fp16")]; + tensor var_9601_begin_0 = const()[name = tensor("op_9601_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9601_end_0 = const()[name = tensor("op_9601_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_9601_end_mask_0 = const()[name = tensor("op_9601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9601_cast_fp16 = slice_by_index(begin = var_9601_begin_0, end = var_9601_end_0, end_mask = var_9601_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9601_cast_fp16")]; + tensor var_9605_begin_0 = const()[name = tensor("op_9605_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9605_end_0 = const()[name = tensor("op_9605_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_9605_end_mask_0 = const()[name = tensor("op_9605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9605_cast_fp16 = slice_by_index(begin = var_9605_begin_0, end = var_9605_end_0, end_mask = var_9605_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9605_cast_fp16")]; + tensor var_9609_begin_0 = const()[name = tensor("op_9609_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9609_end_0 = const()[name = tensor("op_9609_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_9609_end_mask_0 = const()[name = tensor("op_9609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9609_cast_fp16 = slice_by_index(begin = var_9609_begin_0, end = var_9609_end_0, end_mask = var_9609_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9609_cast_fp16")]; + tensor var_9613_begin_0 = const()[name = tensor("op_9613_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9613_end_0 = const()[name = tensor("op_9613_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_9613_end_mask_0 = const()[name = tensor("op_9613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9613_cast_fp16 = slice_by_index(begin = var_9613_begin_0, end = var_9613_end_0, end_mask = var_9613_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9613_cast_fp16")]; + tensor var_9617_begin_0 = const()[name = tensor("op_9617_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9617_end_0 = const()[name = tensor("op_9617_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_9617_end_mask_0 = const()[name = tensor("op_9617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9617_cast_fp16 = slice_by_index(begin = var_9617_begin_0, end = var_9617_end_0, end_mask = var_9617_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9617_cast_fp16")]; + tensor var_9621_begin_0 = const()[name = tensor("op_9621_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9621_end_0 = const()[name = tensor("op_9621_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_9621_end_mask_0 = const()[name = tensor("op_9621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9621_cast_fp16 = slice_by_index(begin = var_9621_begin_0, end = var_9621_end_0, end_mask = var_9621_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9621_cast_fp16")]; + tensor var_9625_begin_0 = const()[name = tensor("op_9625_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9625_end_0 = const()[name = tensor("op_9625_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_9625_end_mask_0 = const()[name = tensor("op_9625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9625_cast_fp16 = slice_by_index(begin = var_9625_begin_0, end = var_9625_end_0, end_mask = var_9625_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9625_cast_fp16")]; + tensor var_9629_begin_0 = const()[name = tensor("op_9629_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9629_end_0 = const()[name = tensor("op_9629_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_9629_end_mask_0 = const()[name = tensor("op_9629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9629_cast_fp16 = slice_by_index(begin = var_9629_begin_0, end = var_9629_end_0, end_mask = var_9629_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9629_cast_fp16")]; + tensor var_9633_begin_0 = const()[name = tensor("op_9633_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9633_end_0 = const()[name = tensor("op_9633_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_9633_end_mask_0 = const()[name = tensor("op_9633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9633_cast_fp16 = slice_by_index(begin = var_9633_begin_0, end = var_9633_end_0, end_mask = var_9633_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9633_cast_fp16")]; + tensor var_9637_begin_0 = const()[name = tensor("op_9637_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9637_end_0 = const()[name = tensor("op_9637_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9637_end_mask_0 = const()[name = tensor("op_9637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9637_cast_fp16 = slice_by_index(begin = var_9637_begin_0, end = var_9637_end_0, end_mask = var_9637_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9637_cast_fp16")]; + tensor var_9641_begin_0 = const()[name = tensor("op_9641_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9641_end_0 = const()[name = tensor("op_9641_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_9641_end_mask_0 = const()[name = tensor("op_9641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9641_cast_fp16 = slice_by_index(begin = var_9641_begin_0, end = var_9641_end_0, end_mask = var_9641_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9641_cast_fp16")]; + tensor var_9645_begin_0 = const()[name = tensor("op_9645_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9645_end_0 = const()[name = tensor("op_9645_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_9645_end_mask_0 = const()[name = tensor("op_9645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9645_cast_fp16 = slice_by_index(begin = var_9645_begin_0, end = var_9645_end_0, end_mask = var_9645_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9645_cast_fp16")]; + tensor var_9649_begin_0 = const()[name = tensor("op_9649_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9649_end_0 = const()[name = tensor("op_9649_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_9649_end_mask_0 = const()[name = tensor("op_9649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9649_cast_fp16 = slice_by_index(begin = var_9649_begin_0, end = var_9649_end_0, end_mask = var_9649_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9649_cast_fp16")]; + tensor var_9653_begin_0 = const()[name = tensor("op_9653_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9653_end_0 = const()[name = tensor("op_9653_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_9653_end_mask_0 = const()[name = tensor("op_9653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9653_cast_fp16 = slice_by_index(begin = var_9653_begin_0, end = var_9653_end_0, end_mask = var_9653_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9653_cast_fp16")]; + tensor k_95_perm_0 = const()[name = tensor("k_95_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9660_begin_0 = const()[name = tensor("op_9660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9660_end_0 = const()[name = tensor("op_9660_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_9660_end_mask_0 = const()[name = tensor("op_9660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_95_cast_fp16 = transpose(perm = k_95_perm_0, x = k_93_cast_fp16)[name = tensor("transpose_44")]; + tensor var_9660_cast_fp16 = slice_by_index(begin = var_9660_begin_0, end = var_9660_end_0, end_mask = var_9660_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9660_cast_fp16")]; + tensor var_9664_begin_0 = const()[name = tensor("op_9664_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9664_end_0 = const()[name = tensor("op_9664_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_9664_end_mask_0 = const()[name = tensor("op_9664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9664_cast_fp16 = slice_by_index(begin = var_9664_begin_0, end = var_9664_end_0, end_mask = var_9664_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9664_cast_fp16")]; + tensor var_9668_begin_0 = const()[name = tensor("op_9668_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9668_end_0 = const()[name = tensor("op_9668_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_9668_end_mask_0 = const()[name = tensor("op_9668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9668_cast_fp16 = slice_by_index(begin = var_9668_begin_0, end = var_9668_end_0, end_mask = var_9668_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9668_cast_fp16")]; + tensor var_9672_begin_0 = const()[name = tensor("op_9672_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9672_end_0 = const()[name = tensor("op_9672_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_9672_end_mask_0 = const()[name = tensor("op_9672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9672_cast_fp16 = slice_by_index(begin = var_9672_begin_0, end = var_9672_end_0, end_mask = var_9672_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9672_cast_fp16")]; + tensor var_9676_begin_0 = const()[name = tensor("op_9676_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9676_end_0 = const()[name = tensor("op_9676_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_9676_end_mask_0 = const()[name = tensor("op_9676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9676_cast_fp16 = slice_by_index(begin = var_9676_begin_0, end = var_9676_end_0, end_mask = var_9676_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9676_cast_fp16")]; + tensor var_9680_begin_0 = const()[name = tensor("op_9680_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9680_end_0 = const()[name = tensor("op_9680_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_9680_end_mask_0 = const()[name = tensor("op_9680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9680_cast_fp16 = slice_by_index(begin = var_9680_begin_0, end = var_9680_end_0, end_mask = var_9680_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9680_cast_fp16")]; + tensor var_9684_begin_0 = const()[name = tensor("op_9684_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9684_end_0 = const()[name = tensor("op_9684_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_9684_end_mask_0 = const()[name = tensor("op_9684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9684_cast_fp16 = slice_by_index(begin = var_9684_begin_0, end = var_9684_end_0, end_mask = var_9684_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9684_cast_fp16")]; + tensor var_9688_begin_0 = const()[name = tensor("op_9688_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9688_end_0 = const()[name = tensor("op_9688_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_9688_end_mask_0 = const()[name = tensor("op_9688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9688_cast_fp16 = slice_by_index(begin = var_9688_begin_0, end = var_9688_end_0, end_mask = var_9688_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9688_cast_fp16")]; + tensor var_9692_begin_0 = const()[name = tensor("op_9692_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9692_end_0 = const()[name = tensor("op_9692_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_9692_end_mask_0 = const()[name = tensor("op_9692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9692_cast_fp16 = slice_by_index(begin = var_9692_begin_0, end = var_9692_end_0, end_mask = var_9692_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9692_cast_fp16")]; + tensor var_9696_begin_0 = const()[name = tensor("op_9696_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9696_end_0 = const()[name = tensor("op_9696_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_9696_end_mask_0 = const()[name = tensor("op_9696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9696_cast_fp16 = slice_by_index(begin = var_9696_begin_0, end = var_9696_end_0, end_mask = var_9696_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9696_cast_fp16")]; + tensor var_9700_begin_0 = const()[name = tensor("op_9700_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9700_end_0 = const()[name = tensor("op_9700_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_9700_end_mask_0 = const()[name = tensor("op_9700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9700_cast_fp16 = slice_by_index(begin = var_9700_begin_0, end = var_9700_end_0, end_mask = var_9700_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9700_cast_fp16")]; + tensor var_9704_begin_0 = const()[name = tensor("op_9704_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9704_end_0 = const()[name = tensor("op_9704_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_9704_end_mask_0 = const()[name = tensor("op_9704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9704_cast_fp16 = slice_by_index(begin = var_9704_begin_0, end = var_9704_end_0, end_mask = var_9704_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9704_cast_fp16")]; + tensor var_9708_begin_0 = const()[name = tensor("op_9708_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_9708_end_0 = const()[name = tensor("op_9708_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_9708_end_mask_0 = const()[name = tensor("op_9708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9708_cast_fp16 = slice_by_index(begin = var_9708_begin_0, end = var_9708_end_0, end_mask = var_9708_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9708_cast_fp16")]; + tensor var_9712_begin_0 = const()[name = tensor("op_9712_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_9712_end_0 = const()[name = tensor("op_9712_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_9712_end_mask_0 = const()[name = tensor("op_9712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9712_cast_fp16 = slice_by_index(begin = var_9712_begin_0, end = var_9712_end_0, end_mask = var_9712_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9712_cast_fp16")]; + tensor var_9716_begin_0 = const()[name = tensor("op_9716_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_9716_end_0 = const()[name = tensor("op_9716_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_9716_end_mask_0 = const()[name = tensor("op_9716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9716_cast_fp16 = slice_by_index(begin = var_9716_begin_0, end = var_9716_end_0, end_mask = var_9716_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9716_cast_fp16")]; + tensor var_9720_begin_0 = const()[name = tensor("op_9720_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_9720_end_0 = const()[name = tensor("op_9720_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_9720_end_mask_0 = const()[name = tensor("op_9720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9720_cast_fp16 = slice_by_index(begin = var_9720_begin_0, end = var_9720_end_0, end_mask = var_9720_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9720_cast_fp16")]; + tensor var_9724_begin_0 = const()[name = tensor("op_9724_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_9724_end_0 = const()[name = tensor("op_9724_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_9724_end_mask_0 = const()[name = tensor("op_9724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9724_cast_fp16 = slice_by_index(begin = var_9724_begin_0, end = var_9724_end_0, end_mask = var_9724_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9724_cast_fp16")]; + tensor var_9728_begin_0 = const()[name = tensor("op_9728_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_9728_end_0 = const()[name = tensor("op_9728_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_9728_end_mask_0 = const()[name = tensor("op_9728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9728_cast_fp16 = slice_by_index(begin = var_9728_begin_0, end = var_9728_end_0, end_mask = var_9728_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9728_cast_fp16")]; + tensor var_9732_begin_0 = const()[name = tensor("op_9732_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_9732_end_0 = const()[name = tensor("op_9732_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_9732_end_mask_0 = const()[name = tensor("op_9732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9732_cast_fp16 = slice_by_index(begin = var_9732_begin_0, end = var_9732_end_0, end_mask = var_9732_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9732_cast_fp16")]; + tensor var_9736_begin_0 = const()[name = tensor("op_9736_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_9736_end_0 = const()[name = tensor("op_9736_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_9736_end_mask_0 = const()[name = tensor("op_9736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9736_cast_fp16 = slice_by_index(begin = var_9736_begin_0, end = var_9736_end_0, end_mask = var_9736_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9736_cast_fp16")]; + tensor var_9738_begin_0 = const()[name = tensor("op_9738_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9738_end_0 = const()[name = tensor("op_9738_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_9738_end_mask_0 = const()[name = tensor("op_9738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9738_cast_fp16 = slice_by_index(begin = var_9738_begin_0, end = var_9738_end_0, end_mask = var_9738_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9738_cast_fp16")]; + tensor var_9742_begin_0 = const()[name = tensor("op_9742_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9742_end_0 = const()[name = tensor("op_9742_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_9742_end_mask_0 = const()[name = tensor("op_9742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9742_cast_fp16 = slice_by_index(begin = var_9742_begin_0, end = var_9742_end_0, end_mask = var_9742_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9742_cast_fp16")]; + tensor var_9746_begin_0 = const()[name = tensor("op_9746_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9746_end_0 = const()[name = tensor("op_9746_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_9746_end_mask_0 = const()[name = tensor("op_9746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9746_cast_fp16 = slice_by_index(begin = var_9746_begin_0, end = var_9746_end_0, end_mask = var_9746_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9746_cast_fp16")]; + tensor var_9750_begin_0 = const()[name = tensor("op_9750_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9750_end_0 = const()[name = tensor("op_9750_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_9750_end_mask_0 = const()[name = tensor("op_9750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9750_cast_fp16 = slice_by_index(begin = var_9750_begin_0, end = var_9750_end_0, end_mask = var_9750_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9750_cast_fp16")]; + tensor var_9754_begin_0 = const()[name = tensor("op_9754_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9754_end_0 = const()[name = tensor("op_9754_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_9754_end_mask_0 = const()[name = tensor("op_9754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9754_cast_fp16 = slice_by_index(begin = var_9754_begin_0, end = var_9754_end_0, end_mask = var_9754_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9754_cast_fp16")]; + tensor var_9758_begin_0 = const()[name = tensor("op_9758_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9758_end_0 = const()[name = tensor("op_9758_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_9758_end_mask_0 = const()[name = tensor("op_9758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9758_cast_fp16 = slice_by_index(begin = var_9758_begin_0, end = var_9758_end_0, end_mask = var_9758_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9758_cast_fp16")]; + tensor var_9762_begin_0 = const()[name = tensor("op_9762_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9762_end_0 = const()[name = tensor("op_9762_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_9762_end_mask_0 = const()[name = tensor("op_9762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9762_cast_fp16 = slice_by_index(begin = var_9762_begin_0, end = var_9762_end_0, end_mask = var_9762_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9762_cast_fp16")]; + tensor var_9766_begin_0 = const()[name = tensor("op_9766_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9766_end_0 = const()[name = tensor("op_9766_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_9766_end_mask_0 = const()[name = tensor("op_9766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9766_cast_fp16 = slice_by_index(begin = var_9766_begin_0, end = var_9766_end_0, end_mask = var_9766_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9766_cast_fp16")]; + tensor var_9770_begin_0 = const()[name = tensor("op_9770_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9770_end_0 = const()[name = tensor("op_9770_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_9770_end_mask_0 = const()[name = tensor("op_9770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9770_cast_fp16 = slice_by_index(begin = var_9770_begin_0, end = var_9770_end_0, end_mask = var_9770_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9770_cast_fp16")]; + tensor var_9774_begin_0 = const()[name = tensor("op_9774_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9774_end_0 = const()[name = tensor("op_9774_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_9774_end_mask_0 = const()[name = tensor("op_9774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9774_cast_fp16 = slice_by_index(begin = var_9774_begin_0, end = var_9774_end_0, end_mask = var_9774_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9774_cast_fp16")]; + tensor var_9778_begin_0 = const()[name = tensor("op_9778_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9778_end_0 = const()[name = tensor("op_9778_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_9778_end_mask_0 = const()[name = tensor("op_9778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9778_cast_fp16 = slice_by_index(begin = var_9778_begin_0, end = var_9778_end_0, end_mask = var_9778_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9778_cast_fp16")]; + tensor var_9782_begin_0 = const()[name = tensor("op_9782_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9782_end_0 = const()[name = tensor("op_9782_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_9782_end_mask_0 = const()[name = tensor("op_9782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9782_cast_fp16 = slice_by_index(begin = var_9782_begin_0, end = var_9782_end_0, end_mask = var_9782_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9782_cast_fp16")]; + tensor var_9786_begin_0 = const()[name = tensor("op_9786_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9786_end_0 = const()[name = tensor("op_9786_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_9786_end_mask_0 = const()[name = tensor("op_9786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9786_cast_fp16 = slice_by_index(begin = var_9786_begin_0, end = var_9786_end_0, end_mask = var_9786_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9786_cast_fp16")]; + tensor var_9790_begin_0 = const()[name = tensor("op_9790_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9790_end_0 = const()[name = tensor("op_9790_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_9790_end_mask_0 = const()[name = tensor("op_9790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9790_cast_fp16 = slice_by_index(begin = var_9790_begin_0, end = var_9790_end_0, end_mask = var_9790_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9790_cast_fp16")]; + tensor var_9794_begin_0 = const()[name = tensor("op_9794_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9794_end_0 = const()[name = tensor("op_9794_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_9794_end_mask_0 = const()[name = tensor("op_9794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9794_cast_fp16 = slice_by_index(begin = var_9794_begin_0, end = var_9794_end_0, end_mask = var_9794_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9794_cast_fp16")]; + tensor var_9798_begin_0 = const()[name = tensor("op_9798_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9798_end_0 = const()[name = tensor("op_9798_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_9798_end_mask_0 = const()[name = tensor("op_9798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9798_cast_fp16 = slice_by_index(begin = var_9798_begin_0, end = var_9798_end_0, end_mask = var_9798_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9798_cast_fp16")]; + tensor var_9802_begin_0 = const()[name = tensor("op_9802_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9802_end_0 = const()[name = tensor("op_9802_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_9802_end_mask_0 = const()[name = tensor("op_9802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9802_cast_fp16 = slice_by_index(begin = var_9802_begin_0, end = var_9802_end_0, end_mask = var_9802_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9802_cast_fp16")]; + tensor var_9806_begin_0 = const()[name = tensor("op_9806_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9806_end_0 = const()[name = tensor("op_9806_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_9806_end_mask_0 = const()[name = tensor("op_9806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9806_cast_fp16 = slice_by_index(begin = var_9806_begin_0, end = var_9806_end_0, end_mask = var_9806_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9806_cast_fp16")]; + tensor var_9810_begin_0 = const()[name = tensor("op_9810_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9810_end_0 = const()[name = tensor("op_9810_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_9810_end_mask_0 = const()[name = tensor("op_9810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9810_cast_fp16 = slice_by_index(begin = var_9810_begin_0, end = var_9810_end_0, end_mask = var_9810_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9810_cast_fp16")]; + tensor var_9814_begin_0 = const()[name = tensor("op_9814_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9814_end_0 = const()[name = tensor("op_9814_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_9814_end_mask_0 = const()[name = tensor("op_9814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9814_cast_fp16 = slice_by_index(begin = var_9814_begin_0, end = var_9814_end_0, end_mask = var_9814_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9814_cast_fp16")]; + tensor var_9818_equation_0 = const()[name = tensor("op_9818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9818_cast_fp16 = einsum(equation = var_9818_equation_0, values = (var_9660_cast_fp16, var_9577_cast_fp16))[name = tensor("op_9818_cast_fp16")]; + tensor var_9819_to_fp16 = const()[name = tensor("op_9819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_761_cast_fp16 = mul(x = var_9818_cast_fp16, y = var_9819_to_fp16)[name = tensor("aw_761_cast_fp16")]; + tensor var_9822_equation_0 = const()[name = tensor("op_9822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9822_cast_fp16 = einsum(equation = var_9822_equation_0, values = (var_9664_cast_fp16, var_9581_cast_fp16))[name = tensor("op_9822_cast_fp16")]; + tensor var_9823_to_fp16 = const()[name = tensor("op_9823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_763_cast_fp16 = mul(x = var_9822_cast_fp16, y = var_9823_to_fp16)[name = tensor("aw_763_cast_fp16")]; + tensor var_9826_equation_0 = const()[name = tensor("op_9826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9826_cast_fp16 = einsum(equation = var_9826_equation_0, values = (var_9668_cast_fp16, var_9585_cast_fp16))[name = tensor("op_9826_cast_fp16")]; + tensor var_9827_to_fp16 = const()[name = tensor("op_9827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_765_cast_fp16 = mul(x = var_9826_cast_fp16, y = var_9827_to_fp16)[name = tensor("aw_765_cast_fp16")]; + tensor var_9830_equation_0 = const()[name = tensor("op_9830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9830_cast_fp16 = einsum(equation = var_9830_equation_0, values = (var_9672_cast_fp16, var_9589_cast_fp16))[name = tensor("op_9830_cast_fp16")]; + tensor var_9831_to_fp16 = const()[name = tensor("op_9831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_767_cast_fp16 = mul(x = var_9830_cast_fp16, y = var_9831_to_fp16)[name = tensor("aw_767_cast_fp16")]; + tensor var_9834_equation_0 = const()[name = tensor("op_9834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9834_cast_fp16 = einsum(equation = var_9834_equation_0, values = (var_9676_cast_fp16, var_9593_cast_fp16))[name = tensor("op_9834_cast_fp16")]; + tensor var_9835_to_fp16 = const()[name = tensor("op_9835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_769_cast_fp16 = mul(x = var_9834_cast_fp16, y = var_9835_to_fp16)[name = tensor("aw_769_cast_fp16")]; + tensor var_9838_equation_0 = const()[name = tensor("op_9838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9838_cast_fp16 = einsum(equation = var_9838_equation_0, values = (var_9680_cast_fp16, var_9597_cast_fp16))[name = tensor("op_9838_cast_fp16")]; + tensor var_9839_to_fp16 = const()[name = tensor("op_9839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_771_cast_fp16 = mul(x = var_9838_cast_fp16, y = var_9839_to_fp16)[name = tensor("aw_771_cast_fp16")]; + tensor var_9842_equation_0 = const()[name = tensor("op_9842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9842_cast_fp16 = einsum(equation = var_9842_equation_0, values = (var_9684_cast_fp16, var_9601_cast_fp16))[name = tensor("op_9842_cast_fp16")]; + tensor var_9843_to_fp16 = const()[name = tensor("op_9843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_773_cast_fp16 = mul(x = var_9842_cast_fp16, y = var_9843_to_fp16)[name = tensor("aw_773_cast_fp16")]; + tensor var_9846_equation_0 = const()[name = tensor("op_9846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9846_cast_fp16 = einsum(equation = var_9846_equation_0, values = (var_9688_cast_fp16, var_9605_cast_fp16))[name = tensor("op_9846_cast_fp16")]; + tensor var_9847_to_fp16 = const()[name = tensor("op_9847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_775_cast_fp16 = mul(x = var_9846_cast_fp16, y = var_9847_to_fp16)[name = tensor("aw_775_cast_fp16")]; + tensor var_9850_equation_0 = const()[name = tensor("op_9850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9850_cast_fp16 = einsum(equation = var_9850_equation_0, values = (var_9692_cast_fp16, var_9609_cast_fp16))[name = tensor("op_9850_cast_fp16")]; + tensor var_9851_to_fp16 = const()[name = tensor("op_9851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_777_cast_fp16 = mul(x = var_9850_cast_fp16, y = var_9851_to_fp16)[name = tensor("aw_777_cast_fp16")]; + tensor var_9854_equation_0 = const()[name = tensor("op_9854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9854_cast_fp16 = einsum(equation = var_9854_equation_0, values = (var_9696_cast_fp16, var_9613_cast_fp16))[name = tensor("op_9854_cast_fp16")]; + tensor var_9855_to_fp16 = const()[name = tensor("op_9855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_779_cast_fp16 = mul(x = var_9854_cast_fp16, y = var_9855_to_fp16)[name = tensor("aw_779_cast_fp16")]; + tensor var_9858_equation_0 = const()[name = tensor("op_9858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9858_cast_fp16 = einsum(equation = var_9858_equation_0, values = (var_9700_cast_fp16, var_9617_cast_fp16))[name = tensor("op_9858_cast_fp16")]; + tensor var_9859_to_fp16 = const()[name = tensor("op_9859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_781_cast_fp16 = mul(x = var_9858_cast_fp16, y = var_9859_to_fp16)[name = tensor("aw_781_cast_fp16")]; + tensor var_9862_equation_0 = const()[name = tensor("op_9862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9862_cast_fp16 = einsum(equation = var_9862_equation_0, values = (var_9704_cast_fp16, var_9621_cast_fp16))[name = tensor("op_9862_cast_fp16")]; + tensor var_9863_to_fp16 = const()[name = tensor("op_9863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_783_cast_fp16 = mul(x = var_9862_cast_fp16, y = var_9863_to_fp16)[name = tensor("aw_783_cast_fp16")]; + tensor var_9866_equation_0 = const()[name = tensor("op_9866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9866_cast_fp16 = einsum(equation = var_9866_equation_0, values = (var_9708_cast_fp16, var_9625_cast_fp16))[name = tensor("op_9866_cast_fp16")]; + tensor var_9867_to_fp16 = const()[name = tensor("op_9867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_785_cast_fp16 = mul(x = var_9866_cast_fp16, y = var_9867_to_fp16)[name = tensor("aw_785_cast_fp16")]; + tensor var_9870_equation_0 = const()[name = tensor("op_9870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9870_cast_fp16 = einsum(equation = var_9870_equation_0, values = (var_9712_cast_fp16, var_9629_cast_fp16))[name = tensor("op_9870_cast_fp16")]; + tensor var_9871_to_fp16 = const()[name = tensor("op_9871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_787_cast_fp16 = mul(x = var_9870_cast_fp16, y = var_9871_to_fp16)[name = tensor("aw_787_cast_fp16")]; + tensor var_9874_equation_0 = const()[name = tensor("op_9874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9874_cast_fp16 = einsum(equation = var_9874_equation_0, values = (var_9716_cast_fp16, var_9633_cast_fp16))[name = tensor("op_9874_cast_fp16")]; + tensor var_9875_to_fp16 = const()[name = tensor("op_9875_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_789_cast_fp16 = mul(x = var_9874_cast_fp16, y = var_9875_to_fp16)[name = tensor("aw_789_cast_fp16")]; + tensor var_9878_equation_0 = const()[name = tensor("op_9878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9878_cast_fp16 = einsum(equation = var_9878_equation_0, values = (var_9720_cast_fp16, var_9637_cast_fp16))[name = tensor("op_9878_cast_fp16")]; + tensor var_9879_to_fp16 = const()[name = tensor("op_9879_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_791_cast_fp16 = mul(x = var_9878_cast_fp16, y = var_9879_to_fp16)[name = tensor("aw_791_cast_fp16")]; + tensor var_9882_equation_0 = const()[name = tensor("op_9882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9882_cast_fp16 = einsum(equation = var_9882_equation_0, values = (var_9724_cast_fp16, var_9641_cast_fp16))[name = tensor("op_9882_cast_fp16")]; + tensor var_9883_to_fp16 = const()[name = tensor("op_9883_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_793_cast_fp16 = mul(x = var_9882_cast_fp16, y = var_9883_to_fp16)[name = tensor("aw_793_cast_fp16")]; + tensor var_9886_equation_0 = const()[name = tensor("op_9886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9886_cast_fp16 = einsum(equation = var_9886_equation_0, values = (var_9728_cast_fp16, var_9645_cast_fp16))[name = tensor("op_9886_cast_fp16")]; + tensor var_9887_to_fp16 = const()[name = tensor("op_9887_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_795_cast_fp16 = mul(x = var_9886_cast_fp16, y = var_9887_to_fp16)[name = tensor("aw_795_cast_fp16")]; + tensor var_9890_equation_0 = const()[name = tensor("op_9890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9890_cast_fp16 = einsum(equation = var_9890_equation_0, values = (var_9732_cast_fp16, var_9649_cast_fp16))[name = tensor("op_9890_cast_fp16")]; + tensor var_9891_to_fp16 = const()[name = tensor("op_9891_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_797_cast_fp16 = mul(x = var_9890_cast_fp16, y = var_9891_to_fp16)[name = tensor("aw_797_cast_fp16")]; + tensor var_9894_equation_0 = const()[name = tensor("op_9894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9894_cast_fp16 = einsum(equation = var_9894_equation_0, values = (var_9736_cast_fp16, var_9653_cast_fp16))[name = tensor("op_9894_cast_fp16")]; + tensor var_9895_to_fp16 = const()[name = tensor("op_9895_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_799_cast_fp16 = mul(x = var_9894_cast_fp16, y = var_9895_to_fp16)[name = tensor("aw_799_cast_fp16")]; + tensor var_9897_cast_fp16 = softmax(axis = var_2624, x = aw_761_cast_fp16)[name = tensor("op_9897_cast_fp16")]; + tensor var_9898_cast_fp16 = softmax(axis = var_2624, x = aw_763_cast_fp16)[name = tensor("op_9898_cast_fp16")]; + tensor var_9899_cast_fp16 = softmax(axis = var_2624, x = aw_765_cast_fp16)[name = tensor("op_9899_cast_fp16")]; + tensor var_9900_cast_fp16 = softmax(axis = var_2624, x = aw_767_cast_fp16)[name = tensor("op_9900_cast_fp16")]; + tensor var_9901_cast_fp16 = softmax(axis = var_2624, x = aw_769_cast_fp16)[name = tensor("op_9901_cast_fp16")]; + tensor var_9902_cast_fp16 = softmax(axis = var_2624, x = aw_771_cast_fp16)[name = tensor("op_9902_cast_fp16")]; + tensor var_9903_cast_fp16 = softmax(axis = var_2624, x = aw_773_cast_fp16)[name = tensor("op_9903_cast_fp16")]; + tensor var_9904_cast_fp16 = softmax(axis = var_2624, x = aw_775_cast_fp16)[name = tensor("op_9904_cast_fp16")]; + tensor var_9905_cast_fp16 = softmax(axis = var_2624, x = aw_777_cast_fp16)[name = tensor("op_9905_cast_fp16")]; + tensor var_9906_cast_fp16 = softmax(axis = var_2624, x = aw_779_cast_fp16)[name = tensor("op_9906_cast_fp16")]; + tensor var_9907_cast_fp16 = softmax(axis = var_2624, x = aw_781_cast_fp16)[name = tensor("op_9907_cast_fp16")]; + tensor var_9908_cast_fp16 = softmax(axis = var_2624, x = aw_783_cast_fp16)[name = tensor("op_9908_cast_fp16")]; + tensor var_9909_cast_fp16 = softmax(axis = var_2624, x = aw_785_cast_fp16)[name = tensor("op_9909_cast_fp16")]; + tensor var_9910_cast_fp16 = softmax(axis = var_2624, x = aw_787_cast_fp16)[name = tensor("op_9910_cast_fp16")]; + tensor var_9911_cast_fp16 = softmax(axis = var_2624, x = aw_789_cast_fp16)[name = tensor("op_9911_cast_fp16")]; + tensor var_9912_cast_fp16 = softmax(axis = var_2624, x = aw_791_cast_fp16)[name = tensor("op_9912_cast_fp16")]; + tensor var_9913_cast_fp16 = softmax(axis = var_2624, x = aw_793_cast_fp16)[name = tensor("op_9913_cast_fp16")]; + tensor var_9914_cast_fp16 = softmax(axis = var_2624, x = aw_795_cast_fp16)[name = tensor("op_9914_cast_fp16")]; + tensor var_9915_cast_fp16 = softmax(axis = var_2624, x = aw_797_cast_fp16)[name = tensor("op_9915_cast_fp16")]; + tensor var_9916_cast_fp16 = softmax(axis = var_2624, x = aw_799_cast_fp16)[name = tensor("op_9916_cast_fp16")]; + tensor var_9918_equation_0 = const()[name = tensor("op_9918_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9918_cast_fp16 = einsum(equation = var_9918_equation_0, values = (var_9738_cast_fp16, var_9897_cast_fp16))[name = tensor("op_9918_cast_fp16")]; + tensor var_9920_equation_0 = const()[name = tensor("op_9920_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9920_cast_fp16 = einsum(equation = var_9920_equation_0, values = (var_9742_cast_fp16, var_9898_cast_fp16))[name = tensor("op_9920_cast_fp16")]; + tensor var_9922_equation_0 = const()[name = tensor("op_9922_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9922_cast_fp16 = einsum(equation = var_9922_equation_0, values = (var_9746_cast_fp16, var_9899_cast_fp16))[name = tensor("op_9922_cast_fp16")]; + tensor var_9924_equation_0 = const()[name = tensor("op_9924_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9924_cast_fp16 = einsum(equation = var_9924_equation_0, values = (var_9750_cast_fp16, var_9900_cast_fp16))[name = tensor("op_9924_cast_fp16")]; + tensor var_9926_equation_0 = const()[name = tensor("op_9926_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9926_cast_fp16 = einsum(equation = var_9926_equation_0, values = (var_9754_cast_fp16, var_9901_cast_fp16))[name = tensor("op_9926_cast_fp16")]; + tensor var_9928_equation_0 = const()[name = tensor("op_9928_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9928_cast_fp16 = einsum(equation = var_9928_equation_0, values = (var_9758_cast_fp16, var_9902_cast_fp16))[name = tensor("op_9928_cast_fp16")]; + tensor var_9930_equation_0 = const()[name = tensor("op_9930_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9930_cast_fp16 = einsum(equation = var_9930_equation_0, values = (var_9762_cast_fp16, var_9903_cast_fp16))[name = tensor("op_9930_cast_fp16")]; + tensor var_9932_equation_0 = const()[name = tensor("op_9932_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9932_cast_fp16 = einsum(equation = var_9932_equation_0, values = (var_9766_cast_fp16, var_9904_cast_fp16))[name = tensor("op_9932_cast_fp16")]; + tensor var_9934_equation_0 = const()[name = tensor("op_9934_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9934_cast_fp16 = einsum(equation = var_9934_equation_0, values = (var_9770_cast_fp16, var_9905_cast_fp16))[name = tensor("op_9934_cast_fp16")]; + tensor var_9936_equation_0 = const()[name = tensor("op_9936_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9936_cast_fp16 = einsum(equation = var_9936_equation_0, values = (var_9774_cast_fp16, var_9906_cast_fp16))[name = tensor("op_9936_cast_fp16")]; + tensor var_9938_equation_0 = const()[name = tensor("op_9938_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9938_cast_fp16 = einsum(equation = var_9938_equation_0, values = (var_9778_cast_fp16, var_9907_cast_fp16))[name = tensor("op_9938_cast_fp16")]; + tensor var_9940_equation_0 = const()[name = tensor("op_9940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9940_cast_fp16 = einsum(equation = var_9940_equation_0, values = (var_9782_cast_fp16, var_9908_cast_fp16))[name = tensor("op_9940_cast_fp16")]; + tensor var_9942_equation_0 = const()[name = tensor("op_9942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9942_cast_fp16 = einsum(equation = var_9942_equation_0, values = (var_9786_cast_fp16, var_9909_cast_fp16))[name = tensor("op_9942_cast_fp16")]; + tensor var_9944_equation_0 = const()[name = tensor("op_9944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9944_cast_fp16 = einsum(equation = var_9944_equation_0, values = (var_9790_cast_fp16, var_9910_cast_fp16))[name = tensor("op_9944_cast_fp16")]; + tensor var_9946_equation_0 = const()[name = tensor("op_9946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9946_cast_fp16 = einsum(equation = var_9946_equation_0, values = (var_9794_cast_fp16, var_9911_cast_fp16))[name = tensor("op_9946_cast_fp16")]; + tensor var_9948_equation_0 = const()[name = tensor("op_9948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9948_cast_fp16 = einsum(equation = var_9948_equation_0, values = (var_9798_cast_fp16, var_9912_cast_fp16))[name = tensor("op_9948_cast_fp16")]; + tensor var_9950_equation_0 = const()[name = tensor("op_9950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9950_cast_fp16 = einsum(equation = var_9950_equation_0, values = (var_9802_cast_fp16, var_9913_cast_fp16))[name = tensor("op_9950_cast_fp16")]; + tensor var_9952_equation_0 = const()[name = tensor("op_9952_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9952_cast_fp16 = einsum(equation = var_9952_equation_0, values = (var_9806_cast_fp16, var_9914_cast_fp16))[name = tensor("op_9952_cast_fp16")]; + tensor var_9954_equation_0 = const()[name = tensor("op_9954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9954_cast_fp16 = einsum(equation = var_9954_equation_0, values = (var_9810_cast_fp16, var_9915_cast_fp16))[name = tensor("op_9954_cast_fp16")]; + tensor var_9956_equation_0 = const()[name = tensor("op_9956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9956_cast_fp16 = einsum(equation = var_9956_equation_0, values = (var_9814_cast_fp16, var_9916_cast_fp16))[name = tensor("op_9956_cast_fp16")]; + tensor input_189_interleave_0 = const()[name = tensor("input_189_interleave_0"), val = tensor(false)]; + tensor input_189_cast_fp16 = concat(axis = var_2624, interleave = input_189_interleave_0, values = (var_9918_cast_fp16, var_9920_cast_fp16, var_9922_cast_fp16, var_9924_cast_fp16, var_9926_cast_fp16, var_9928_cast_fp16, var_9930_cast_fp16, var_9932_cast_fp16, var_9934_cast_fp16, var_9936_cast_fp16, var_9938_cast_fp16, var_9940_cast_fp16, var_9942_cast_fp16, var_9944_cast_fp16, var_9946_cast_fp16, var_9948_cast_fp16, var_9950_cast_fp16, var_9952_cast_fp16, var_9954_cast_fp16, var_9956_cast_fp16))[name = tensor("input_189_cast_fp16")]; + tensor var_9966_pad_type_0 = const()[name = tensor("op_9966_pad_type_0"), val = tensor("valid")]; + tensor var_9966_strides_0 = const()[name = tensor("op_9966_strides_0"), val = tensor([1, 1])]; + tensor var_9966_pad_0 = const()[name = tensor("op_9966_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9966_dilations_0 = const()[name = tensor("op_9966_dilations_0"), val = tensor([1, 1])]; + tensor var_9966_groups_0 = const()[name = tensor("op_9966_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267211968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268440832))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268441024)))]; + tensor var_9966_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_9966_dilations_0, groups = var_9966_groups_0, pad = var_9966_pad_0, pad_type = var_9966_pad_type_0, strides = var_9966_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor("op_9966_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = var_9966_cast_fp16, y = inputs_69_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; + tensor input_191_axes_0 = const()[name = tensor("input_191_axes_0"), val = tensor([1])]; + tensor input_191_gamma_0_to_fp16 = const()[name = tensor("input_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268443648)))]; + tensor input_191_beta_0_to_fp16 = const()[name = tensor("input_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268446272)))]; + tensor var_9976_to_fp16 = const()[name = tensor("op_9976_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_191_cast_fp16 = layer_norm(axes = input_191_axes_0, beta = input_191_beta_0_to_fp16, epsilon = var_9976_to_fp16, gamma = input_191_gamma_0_to_fp16, x = inputs_71_cast_fp16)[name = tensor("input_191_cast_fp16")]; + tensor var_9996_pad_type_0 = const()[name = tensor("op_9996_pad_type_0"), val = tensor("valid")]; + tensor var_9996_strides_0 = const()[name = tensor("op_9996_strides_0"), val = tensor([1, 1])]; + tensor var_9996_pad_0 = const()[name = tensor("op_9996_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9996_dilations_0 = const()[name = tensor("op_9996_dilations_0"), val = tensor([1, 1])]; + tensor var_9996_groups_0 = const()[name = tensor("op_9996_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268448896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278279360))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278279552)))]; + tensor var_9996_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_9996_dilations_0, groups = var_9996_groups_0, pad = var_9996_pad_0, pad_type = var_9996_pad_type_0, strides = var_9996_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = tensor("op_9996_cast_fp16")]; + tensor var_9997_split_sizes_0 = const()[name = tensor("op_9997_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_9997_axis_0 = const()[name = tensor("op_9997_axis_0"), val = tensor(1)]; + tensor var_9997_cast_fp16_0, tensor var_9997_cast_fp16_1 = split(axis = var_9997_axis_0, split_sizes = var_9997_split_sizes_0, x = var_9996_cast_fp16)[name = tensor("op_9997_cast_fp16")]; + tensor var_9999_mode_0 = const()[name = tensor("op_9999_mode_0"), val = tensor("EXACT")]; + tensor var_9999_cast_fp16 = gelu(mode = var_9999_mode_0, x = var_9997_cast_fp16_1)[name = tensor("op_9999_cast_fp16")]; + tensor input_193_cast_fp16 = mul(x = var_9997_cast_fp16_0, y = var_9999_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor var_10007_pad_type_0 = const()[name = tensor("op_10007_pad_type_0"), val = tensor("valid")]; + tensor var_10007_strides_0 = const()[name = tensor("op_10007_strides_0"), val = tensor([1, 1])]; + tensor var_10007_pad_0 = const()[name = tensor("op_10007_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10007_dilations_0 = const()[name = tensor("op_10007_dilations_0"), val = tensor([1, 1])]; + tensor var_10007_groups_0 = const()[name = tensor("op_10007_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278300096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283215360))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283215552)))]; + tensor var_10007_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_10007_dilations_0, groups = var_10007_groups_0, pad = var_10007_pad_0, pad_type = var_10007_pad_type_0, strides = var_10007_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("op_10007_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = var_10007_cast_fp16, y = inputs_71_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; + tensor hidden_states_113_axes_0 = const()[name = tensor("hidden_states_113_axes_0"), val = tensor([1])]; + tensor hidden_states_113_gamma_0_to_fp16 = const()[name = tensor("hidden_states_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283218176)))]; + tensor hidden_states_113_beta_0_to_fp16 = const()[name = tensor("hidden_states_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283220800)))]; + tensor var_10023_to_fp16 = const()[name = tensor("op_10023_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_113_cast_fp16 = layer_norm(axes = hidden_states_113_axes_0, beta = hidden_states_113_beta_0_to_fp16, epsilon = var_10023_to_fp16, gamma = hidden_states_113_gamma_0_to_fp16, x = inputs_73_cast_fp16)[name = tensor("hidden_states_113_cast_fp16")]; + tensor q_49_pad_type_0 = const()[name = tensor("q_49_pad_type_0"), val = tensor("valid")]; + tensor q_49_strides_0 = const()[name = tensor("q_49_strides_0"), val = tensor([1, 1])]; + tensor q_49_pad_0 = const()[name = tensor("q_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_49_dilations_0 = const()[name = tensor("q_49_dilations_0"), val = tensor([1, 1])]; + tensor q_49_groups_0 = const()[name = tensor("q_49_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(283223424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284452288))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_49_cast_fp16 = conv(dilations = q_49_dilations_0, groups = q_49_groups_0, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = q_49_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_113_cast_fp16)[name = tensor("q_49_cast_fp16")]; + tensor k_97_pad_type_0 = const()[name = tensor("k_97_pad_type_0"), val = tensor("valid")]; + tensor k_97_strides_0 = const()[name = tensor("k_97_strides_0"), val = tensor([1, 1])]; + tensor k_97_pad_0 = const()[name = tensor("k_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_97_dilations_0 = const()[name = tensor("k_97_dilations_0"), val = tensor([1, 1])]; + tensor k_97_groups_0 = const()[name = tensor("k_97_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284452480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285681344))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_97_cast_fp16 = conv(dilations = k_97_dilations_0, groups = k_97_groups_0, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = k_97_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_113_cast_fp16)[name = tensor("k_97_cast_fp16")]; + tensor v_49_pad_type_0 = const()[name = tensor("v_49_pad_type_0"), val = tensor("valid")]; + tensor v_49_strides_0 = const()[name = tensor("v_49_strides_0"), val = tensor([1, 1])]; + tensor v_49_pad_0 = const()[name = tensor("v_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_49_dilations_0 = const()[name = tensor("v_49_dilations_0"), val = tensor([1, 1])]; + tensor v_49_groups_0 = const()[name = tensor("v_49_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(285681536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286910400))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_49_cast_fp16 = conv(dilations = v_49_dilations_0, groups = v_49_groups_0, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = v_49_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_113_cast_fp16)[name = tensor("v_49_cast_fp16")]; + tensor var_10056_begin_0 = const()[name = tensor("op_10056_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10056_end_0 = const()[name = tensor("op_10056_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10056_end_mask_0 = const()[name = tensor("op_10056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10056_cast_fp16 = slice_by_index(begin = var_10056_begin_0, end = var_10056_end_0, end_mask = var_10056_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10056_cast_fp16")]; + tensor var_10060_begin_0 = const()[name = tensor("op_10060_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10060_end_0 = const()[name = tensor("op_10060_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10060_end_mask_0 = const()[name = tensor("op_10060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10060_cast_fp16 = slice_by_index(begin = var_10060_begin_0, end = var_10060_end_0, end_mask = var_10060_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10060_cast_fp16")]; + tensor var_10064_begin_0 = const()[name = tensor("op_10064_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10064_end_0 = const()[name = tensor("op_10064_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10064_end_mask_0 = const()[name = tensor("op_10064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10064_cast_fp16 = slice_by_index(begin = var_10064_begin_0, end = var_10064_end_0, end_mask = var_10064_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10064_cast_fp16")]; + tensor var_10068_begin_0 = const()[name = tensor("op_10068_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10068_end_0 = const()[name = tensor("op_10068_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10068_end_mask_0 = const()[name = tensor("op_10068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10068_cast_fp16 = slice_by_index(begin = var_10068_begin_0, end = var_10068_end_0, end_mask = var_10068_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10068_cast_fp16")]; + tensor var_10072_begin_0 = const()[name = tensor("op_10072_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10072_end_0 = const()[name = tensor("op_10072_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10072_end_mask_0 = const()[name = tensor("op_10072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10072_cast_fp16 = slice_by_index(begin = var_10072_begin_0, end = var_10072_end_0, end_mask = var_10072_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10072_cast_fp16")]; + tensor var_10076_begin_0 = const()[name = tensor("op_10076_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10076_end_0 = const()[name = tensor("op_10076_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10076_end_mask_0 = const()[name = tensor("op_10076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10076_cast_fp16 = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10076_cast_fp16")]; + tensor var_10080_begin_0 = const()[name = tensor("op_10080_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10080_end_0 = const()[name = tensor("op_10080_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10080_end_mask_0 = const()[name = tensor("op_10080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10080_cast_fp16 = slice_by_index(begin = var_10080_begin_0, end = var_10080_end_0, end_mask = var_10080_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10080_cast_fp16")]; + tensor var_10084_begin_0 = const()[name = tensor("op_10084_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10084_end_0 = const()[name = tensor("op_10084_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10084_end_mask_0 = const()[name = tensor("op_10084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10084_cast_fp16 = slice_by_index(begin = var_10084_begin_0, end = var_10084_end_0, end_mask = var_10084_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10084_cast_fp16")]; + tensor var_10088_begin_0 = const()[name = tensor("op_10088_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10088_end_0 = const()[name = tensor("op_10088_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10088_end_mask_0 = const()[name = tensor("op_10088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10088_cast_fp16 = slice_by_index(begin = var_10088_begin_0, end = var_10088_end_0, end_mask = var_10088_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10088_cast_fp16")]; + tensor var_10092_begin_0 = const()[name = tensor("op_10092_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10092_end_0 = const()[name = tensor("op_10092_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10092_end_mask_0 = const()[name = tensor("op_10092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10092_cast_fp16 = slice_by_index(begin = var_10092_begin_0, end = var_10092_end_0, end_mask = var_10092_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10092_cast_fp16")]; + tensor var_10096_begin_0 = const()[name = tensor("op_10096_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10096_end_0 = const()[name = tensor("op_10096_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_10096_end_mask_0 = const()[name = tensor("op_10096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10096_cast_fp16 = slice_by_index(begin = var_10096_begin_0, end = var_10096_end_0, end_mask = var_10096_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10096_cast_fp16")]; + tensor var_10100_begin_0 = const()[name = tensor("op_10100_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10100_end_0 = const()[name = tensor("op_10100_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_10100_end_mask_0 = const()[name = tensor("op_10100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10100_cast_fp16 = slice_by_index(begin = var_10100_begin_0, end = var_10100_end_0, end_mask = var_10100_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10100_cast_fp16")]; + tensor var_10104_begin_0 = const()[name = tensor("op_10104_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10104_end_0 = const()[name = tensor("op_10104_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_10104_end_mask_0 = const()[name = tensor("op_10104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10104_cast_fp16 = slice_by_index(begin = var_10104_begin_0, end = var_10104_end_0, end_mask = var_10104_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10104_cast_fp16")]; + tensor var_10108_begin_0 = const()[name = tensor("op_10108_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10108_end_0 = const()[name = tensor("op_10108_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_10108_end_mask_0 = const()[name = tensor("op_10108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10108_cast_fp16 = slice_by_index(begin = var_10108_begin_0, end = var_10108_end_0, end_mask = var_10108_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10108_cast_fp16")]; + tensor var_10112_begin_0 = const()[name = tensor("op_10112_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10112_end_0 = const()[name = tensor("op_10112_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_10112_end_mask_0 = const()[name = tensor("op_10112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10112_cast_fp16 = slice_by_index(begin = var_10112_begin_0, end = var_10112_end_0, end_mask = var_10112_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10112_cast_fp16")]; + tensor var_10116_begin_0 = const()[name = tensor("op_10116_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10116_end_0 = const()[name = tensor("op_10116_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10116_end_mask_0 = const()[name = tensor("op_10116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10116_cast_fp16 = slice_by_index(begin = var_10116_begin_0, end = var_10116_end_0, end_mask = var_10116_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10116_cast_fp16")]; + tensor var_10120_begin_0 = const()[name = tensor("op_10120_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10120_end_0 = const()[name = tensor("op_10120_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_10120_end_mask_0 = const()[name = tensor("op_10120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10120_cast_fp16 = slice_by_index(begin = var_10120_begin_0, end = var_10120_end_0, end_mask = var_10120_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10120_cast_fp16")]; + tensor var_10124_begin_0 = const()[name = tensor("op_10124_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10124_end_0 = const()[name = tensor("op_10124_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_10124_end_mask_0 = const()[name = tensor("op_10124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10124_cast_fp16 = slice_by_index(begin = var_10124_begin_0, end = var_10124_end_0, end_mask = var_10124_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10124_cast_fp16")]; + tensor var_10128_begin_0 = const()[name = tensor("op_10128_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10128_end_0 = const()[name = tensor("op_10128_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_10128_end_mask_0 = const()[name = tensor("op_10128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10128_cast_fp16 = slice_by_index(begin = var_10128_begin_0, end = var_10128_end_0, end_mask = var_10128_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10128_cast_fp16")]; + tensor var_10132_begin_0 = const()[name = tensor("op_10132_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10132_end_0 = const()[name = tensor("op_10132_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_10132_end_mask_0 = const()[name = tensor("op_10132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10132_cast_fp16 = slice_by_index(begin = var_10132_begin_0, end = var_10132_end_0, end_mask = var_10132_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10132_cast_fp16")]; + tensor k_99_perm_0 = const()[name = tensor("k_99_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10139_begin_0 = const()[name = tensor("op_10139_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10139_end_0 = const()[name = tensor("op_10139_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_10139_end_mask_0 = const()[name = tensor("op_10139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_99_cast_fp16 = transpose(perm = k_99_perm_0, x = k_97_cast_fp16)[name = tensor("transpose_43")]; + tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10139_cast_fp16")]; + tensor var_10143_begin_0 = const()[name = tensor("op_10143_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10143_end_0 = const()[name = tensor("op_10143_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_10143_end_mask_0 = const()[name = tensor("op_10143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10143_cast_fp16 = slice_by_index(begin = var_10143_begin_0, end = var_10143_end_0, end_mask = var_10143_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10143_cast_fp16")]; + tensor var_10147_begin_0 = const()[name = tensor("op_10147_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10147_end_0 = const()[name = tensor("op_10147_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_10147_end_mask_0 = const()[name = tensor("op_10147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10147_cast_fp16 = slice_by_index(begin = var_10147_begin_0, end = var_10147_end_0, end_mask = var_10147_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10147_cast_fp16")]; + tensor var_10151_begin_0 = const()[name = tensor("op_10151_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10151_end_0 = const()[name = tensor("op_10151_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_10151_end_mask_0 = const()[name = tensor("op_10151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10151_cast_fp16 = slice_by_index(begin = var_10151_begin_0, end = var_10151_end_0, end_mask = var_10151_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10151_cast_fp16")]; + tensor var_10155_begin_0 = const()[name = tensor("op_10155_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10155_end_0 = const()[name = tensor("op_10155_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_10155_end_mask_0 = const()[name = tensor("op_10155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10155_cast_fp16 = slice_by_index(begin = var_10155_begin_0, end = var_10155_end_0, end_mask = var_10155_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10155_cast_fp16")]; + tensor var_10159_begin_0 = const()[name = tensor("op_10159_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10159_end_0 = const()[name = tensor("op_10159_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_10159_end_mask_0 = const()[name = tensor("op_10159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10159_cast_fp16 = slice_by_index(begin = var_10159_begin_0, end = var_10159_end_0, end_mask = var_10159_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10159_cast_fp16")]; + tensor var_10163_begin_0 = const()[name = tensor("op_10163_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10163_end_0 = const()[name = tensor("op_10163_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_10163_end_mask_0 = const()[name = tensor("op_10163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10163_cast_fp16 = slice_by_index(begin = var_10163_begin_0, end = var_10163_end_0, end_mask = var_10163_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10163_cast_fp16")]; + tensor var_10167_begin_0 = const()[name = tensor("op_10167_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10167_end_0 = const()[name = tensor("op_10167_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_10167_end_mask_0 = const()[name = tensor("op_10167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10167_cast_fp16 = slice_by_index(begin = var_10167_begin_0, end = var_10167_end_0, end_mask = var_10167_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10167_cast_fp16")]; + tensor var_10171_begin_0 = const()[name = tensor("op_10171_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10171_end_0 = const()[name = tensor("op_10171_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_10171_end_mask_0 = const()[name = tensor("op_10171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10171_cast_fp16 = slice_by_index(begin = var_10171_begin_0, end = var_10171_end_0, end_mask = var_10171_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10171_cast_fp16")]; + tensor var_10175_begin_0 = const()[name = tensor("op_10175_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10175_end_0 = const()[name = tensor("op_10175_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_10175_end_mask_0 = const()[name = tensor("op_10175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10175_cast_fp16 = slice_by_index(begin = var_10175_begin_0, end = var_10175_end_0, end_mask = var_10175_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10175_cast_fp16")]; + tensor var_10179_begin_0 = const()[name = tensor("op_10179_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10179_end_0 = const()[name = tensor("op_10179_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_10179_end_mask_0 = const()[name = tensor("op_10179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10179_cast_fp16 = slice_by_index(begin = var_10179_begin_0, end = var_10179_end_0, end_mask = var_10179_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10179_cast_fp16")]; + tensor var_10183_begin_0 = const()[name = tensor("op_10183_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10183_end_0 = const()[name = tensor("op_10183_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_10183_end_mask_0 = const()[name = tensor("op_10183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10183_cast_fp16 = slice_by_index(begin = var_10183_begin_0, end = var_10183_end_0, end_mask = var_10183_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10183_cast_fp16")]; + tensor var_10187_begin_0 = const()[name = tensor("op_10187_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_10187_end_0 = const()[name = tensor("op_10187_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_10187_end_mask_0 = const()[name = tensor("op_10187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10187_cast_fp16 = slice_by_index(begin = var_10187_begin_0, end = var_10187_end_0, end_mask = var_10187_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10187_cast_fp16")]; + tensor var_10191_begin_0 = const()[name = tensor("op_10191_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_10191_end_0 = const()[name = tensor("op_10191_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_10191_end_mask_0 = const()[name = tensor("op_10191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10191_cast_fp16 = slice_by_index(begin = var_10191_begin_0, end = var_10191_end_0, end_mask = var_10191_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10191_cast_fp16")]; + tensor var_10195_begin_0 = const()[name = tensor("op_10195_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_10195_end_0 = const()[name = tensor("op_10195_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_10195_end_mask_0 = const()[name = tensor("op_10195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10195_cast_fp16 = slice_by_index(begin = var_10195_begin_0, end = var_10195_end_0, end_mask = var_10195_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10195_cast_fp16")]; + tensor var_10199_begin_0 = const()[name = tensor("op_10199_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_10199_end_0 = const()[name = tensor("op_10199_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10199_end_mask_0 = const()[name = tensor("op_10199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10199_cast_fp16 = slice_by_index(begin = var_10199_begin_0, end = var_10199_end_0, end_mask = var_10199_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10199_cast_fp16")]; + tensor var_10203_begin_0 = const()[name = tensor("op_10203_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_10203_end_0 = const()[name = tensor("op_10203_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_10203_end_mask_0 = const()[name = tensor("op_10203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10203_cast_fp16 = slice_by_index(begin = var_10203_begin_0, end = var_10203_end_0, end_mask = var_10203_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10203_cast_fp16")]; + tensor var_10207_begin_0 = const()[name = tensor("op_10207_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_10207_end_0 = const()[name = tensor("op_10207_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_10207_end_mask_0 = const()[name = tensor("op_10207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10207_cast_fp16 = slice_by_index(begin = var_10207_begin_0, end = var_10207_end_0, end_mask = var_10207_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10207_cast_fp16")]; + tensor var_10211_begin_0 = const()[name = tensor("op_10211_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_10211_end_0 = const()[name = tensor("op_10211_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_10211_end_mask_0 = const()[name = tensor("op_10211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10211_cast_fp16 = slice_by_index(begin = var_10211_begin_0, end = var_10211_end_0, end_mask = var_10211_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10211_cast_fp16")]; + tensor var_10215_begin_0 = const()[name = tensor("op_10215_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_10215_end_0 = const()[name = tensor("op_10215_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_10215_end_mask_0 = const()[name = tensor("op_10215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10215_cast_fp16 = slice_by_index(begin = var_10215_begin_0, end = var_10215_end_0, end_mask = var_10215_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10215_cast_fp16")]; + tensor var_10217_begin_0 = const()[name = tensor("op_10217_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10217_end_0 = const()[name = tensor("op_10217_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10217_end_mask_0 = const()[name = tensor("op_10217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10217_cast_fp16 = slice_by_index(begin = var_10217_begin_0, end = var_10217_end_0, end_mask = var_10217_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10217_cast_fp16")]; + tensor var_10221_begin_0 = const()[name = tensor("op_10221_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10221_end_0 = const()[name = tensor("op_10221_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10221_end_mask_0 = const()[name = tensor("op_10221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10221_cast_fp16 = slice_by_index(begin = var_10221_begin_0, end = var_10221_end_0, end_mask = var_10221_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10221_cast_fp16")]; + tensor var_10225_begin_0 = const()[name = tensor("op_10225_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10225_end_0 = const()[name = tensor("op_10225_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10225_end_mask_0 = const()[name = tensor("op_10225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10225_cast_fp16 = slice_by_index(begin = var_10225_begin_0, end = var_10225_end_0, end_mask = var_10225_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10225_cast_fp16")]; + tensor var_10229_begin_0 = const()[name = tensor("op_10229_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10229_end_0 = const()[name = tensor("op_10229_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10229_end_mask_0 = const()[name = tensor("op_10229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10229_cast_fp16 = slice_by_index(begin = var_10229_begin_0, end = var_10229_end_0, end_mask = var_10229_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10229_cast_fp16")]; + tensor var_10233_begin_0 = const()[name = tensor("op_10233_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10233_end_0 = const()[name = tensor("op_10233_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10233_end_mask_0 = const()[name = tensor("op_10233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10233_cast_fp16 = slice_by_index(begin = var_10233_begin_0, end = var_10233_end_0, end_mask = var_10233_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10233_cast_fp16")]; + tensor var_10237_begin_0 = const()[name = tensor("op_10237_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10237_end_0 = const()[name = tensor("op_10237_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10237_end_mask_0 = const()[name = tensor("op_10237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10237_cast_fp16 = slice_by_index(begin = var_10237_begin_0, end = var_10237_end_0, end_mask = var_10237_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10237_cast_fp16")]; + tensor var_10241_begin_0 = const()[name = tensor("op_10241_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10241_end_0 = const()[name = tensor("op_10241_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10241_end_mask_0 = const()[name = tensor("op_10241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10241_cast_fp16 = slice_by_index(begin = var_10241_begin_0, end = var_10241_end_0, end_mask = var_10241_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10241_cast_fp16")]; + tensor var_10245_begin_0 = const()[name = tensor("op_10245_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10245_end_0 = const()[name = tensor("op_10245_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10245_end_mask_0 = const()[name = tensor("op_10245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10245_cast_fp16 = slice_by_index(begin = var_10245_begin_0, end = var_10245_end_0, end_mask = var_10245_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10245_cast_fp16")]; + tensor var_10249_begin_0 = const()[name = tensor("op_10249_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10249_end_0 = const()[name = tensor("op_10249_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10249_end_mask_0 = const()[name = tensor("op_10249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10249_cast_fp16 = slice_by_index(begin = var_10249_begin_0, end = var_10249_end_0, end_mask = var_10249_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10249_cast_fp16")]; + tensor var_10253_begin_0 = const()[name = tensor("op_10253_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10253_end_0 = const()[name = tensor("op_10253_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10253_end_mask_0 = const()[name = tensor("op_10253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10253_cast_fp16 = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10253_cast_fp16")]; + tensor var_10257_begin_0 = const()[name = tensor("op_10257_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10257_end_0 = const()[name = tensor("op_10257_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_10257_end_mask_0 = const()[name = tensor("op_10257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10257_cast_fp16 = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10257_cast_fp16")]; + tensor var_10261_begin_0 = const()[name = tensor("op_10261_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10261_end_0 = const()[name = tensor("op_10261_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_10261_end_mask_0 = const()[name = tensor("op_10261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10261_cast_fp16 = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10261_cast_fp16")]; + tensor var_10265_begin_0 = const()[name = tensor("op_10265_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10265_end_0 = const()[name = tensor("op_10265_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_10265_end_mask_0 = const()[name = tensor("op_10265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10265_cast_fp16 = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10265_cast_fp16")]; + tensor var_10269_begin_0 = const()[name = tensor("op_10269_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10269_end_0 = const()[name = tensor("op_10269_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_10269_end_mask_0 = const()[name = tensor("op_10269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10269_cast_fp16 = slice_by_index(begin = var_10269_begin_0, end = var_10269_end_0, end_mask = var_10269_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10269_cast_fp16")]; + tensor var_10273_begin_0 = const()[name = tensor("op_10273_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10273_end_0 = const()[name = tensor("op_10273_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_10273_end_mask_0 = const()[name = tensor("op_10273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10273_cast_fp16 = slice_by_index(begin = var_10273_begin_0, end = var_10273_end_0, end_mask = var_10273_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10273_cast_fp16")]; + tensor var_10277_begin_0 = const()[name = tensor("op_10277_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10277_end_0 = const()[name = tensor("op_10277_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10277_end_mask_0 = const()[name = tensor("op_10277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10277_cast_fp16 = slice_by_index(begin = var_10277_begin_0, end = var_10277_end_0, end_mask = var_10277_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10277_cast_fp16")]; + tensor var_10281_begin_0 = const()[name = tensor("op_10281_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10281_end_0 = const()[name = tensor("op_10281_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_10281_end_mask_0 = const()[name = tensor("op_10281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10281_cast_fp16 = slice_by_index(begin = var_10281_begin_0, end = var_10281_end_0, end_mask = var_10281_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10281_cast_fp16")]; + tensor var_10285_begin_0 = const()[name = tensor("op_10285_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10285_end_0 = const()[name = tensor("op_10285_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_10285_end_mask_0 = const()[name = tensor("op_10285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10285_cast_fp16 = slice_by_index(begin = var_10285_begin_0, end = var_10285_end_0, end_mask = var_10285_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10285_cast_fp16")]; + tensor var_10289_begin_0 = const()[name = tensor("op_10289_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10289_end_0 = const()[name = tensor("op_10289_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_10289_end_mask_0 = const()[name = tensor("op_10289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10289_cast_fp16 = slice_by_index(begin = var_10289_begin_0, end = var_10289_end_0, end_mask = var_10289_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10289_cast_fp16")]; + tensor var_10293_begin_0 = const()[name = tensor("op_10293_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10293_end_0 = const()[name = tensor("op_10293_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_10293_end_mask_0 = const()[name = tensor("op_10293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10293_cast_fp16 = slice_by_index(begin = var_10293_begin_0, end = var_10293_end_0, end_mask = var_10293_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10293_cast_fp16")]; + tensor var_10297_equation_0 = const()[name = tensor("op_10297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10297_cast_fp16 = einsum(equation = var_10297_equation_0, values = (var_10139_cast_fp16, var_10056_cast_fp16))[name = tensor("op_10297_cast_fp16")]; + tensor var_10298_to_fp16 = const()[name = tensor("op_10298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_801_cast_fp16 = mul(x = var_10297_cast_fp16, y = var_10298_to_fp16)[name = tensor("aw_801_cast_fp16")]; + tensor var_10301_equation_0 = const()[name = tensor("op_10301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10301_cast_fp16 = einsum(equation = var_10301_equation_0, values = (var_10143_cast_fp16, var_10060_cast_fp16))[name = tensor("op_10301_cast_fp16")]; + tensor var_10302_to_fp16 = const()[name = tensor("op_10302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_803_cast_fp16 = mul(x = var_10301_cast_fp16, y = var_10302_to_fp16)[name = tensor("aw_803_cast_fp16")]; + tensor var_10305_equation_0 = const()[name = tensor("op_10305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10305_cast_fp16 = einsum(equation = var_10305_equation_0, values = (var_10147_cast_fp16, var_10064_cast_fp16))[name = tensor("op_10305_cast_fp16")]; + tensor var_10306_to_fp16 = const()[name = tensor("op_10306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_805_cast_fp16 = mul(x = var_10305_cast_fp16, y = var_10306_to_fp16)[name = tensor("aw_805_cast_fp16")]; + tensor var_10309_equation_0 = const()[name = tensor("op_10309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10309_cast_fp16 = einsum(equation = var_10309_equation_0, values = (var_10151_cast_fp16, var_10068_cast_fp16))[name = tensor("op_10309_cast_fp16")]; + tensor var_10310_to_fp16 = const()[name = tensor("op_10310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_807_cast_fp16 = mul(x = var_10309_cast_fp16, y = var_10310_to_fp16)[name = tensor("aw_807_cast_fp16")]; + tensor var_10313_equation_0 = const()[name = tensor("op_10313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10313_cast_fp16 = einsum(equation = var_10313_equation_0, values = (var_10155_cast_fp16, var_10072_cast_fp16))[name = tensor("op_10313_cast_fp16")]; + tensor var_10314_to_fp16 = const()[name = tensor("op_10314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_809_cast_fp16 = mul(x = var_10313_cast_fp16, y = var_10314_to_fp16)[name = tensor("aw_809_cast_fp16")]; + tensor var_10317_equation_0 = const()[name = tensor("op_10317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10317_cast_fp16 = einsum(equation = var_10317_equation_0, values = (var_10159_cast_fp16, var_10076_cast_fp16))[name = tensor("op_10317_cast_fp16")]; + tensor var_10318_to_fp16 = const()[name = tensor("op_10318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_811_cast_fp16 = mul(x = var_10317_cast_fp16, y = var_10318_to_fp16)[name = tensor("aw_811_cast_fp16")]; + tensor var_10321_equation_0 = const()[name = tensor("op_10321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10321_cast_fp16 = einsum(equation = var_10321_equation_0, values = (var_10163_cast_fp16, var_10080_cast_fp16))[name = tensor("op_10321_cast_fp16")]; + tensor var_10322_to_fp16 = const()[name = tensor("op_10322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_813_cast_fp16 = mul(x = var_10321_cast_fp16, y = var_10322_to_fp16)[name = tensor("aw_813_cast_fp16")]; + tensor var_10325_equation_0 = const()[name = tensor("op_10325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10325_cast_fp16 = einsum(equation = var_10325_equation_0, values = (var_10167_cast_fp16, var_10084_cast_fp16))[name = tensor("op_10325_cast_fp16")]; + tensor var_10326_to_fp16 = const()[name = tensor("op_10326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_815_cast_fp16 = mul(x = var_10325_cast_fp16, y = var_10326_to_fp16)[name = tensor("aw_815_cast_fp16")]; + tensor var_10329_equation_0 = const()[name = tensor("op_10329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10329_cast_fp16 = einsum(equation = var_10329_equation_0, values = (var_10171_cast_fp16, var_10088_cast_fp16))[name = tensor("op_10329_cast_fp16")]; + tensor var_10330_to_fp16 = const()[name = tensor("op_10330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_817_cast_fp16 = mul(x = var_10329_cast_fp16, y = var_10330_to_fp16)[name = tensor("aw_817_cast_fp16")]; + tensor var_10333_equation_0 = const()[name = tensor("op_10333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10333_cast_fp16 = einsum(equation = var_10333_equation_0, values = (var_10175_cast_fp16, var_10092_cast_fp16))[name = tensor("op_10333_cast_fp16")]; + tensor var_10334_to_fp16 = const()[name = tensor("op_10334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_819_cast_fp16 = mul(x = var_10333_cast_fp16, y = var_10334_to_fp16)[name = tensor("aw_819_cast_fp16")]; + tensor var_10337_equation_0 = const()[name = tensor("op_10337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10337_cast_fp16 = einsum(equation = var_10337_equation_0, values = (var_10179_cast_fp16, var_10096_cast_fp16))[name = tensor("op_10337_cast_fp16")]; + tensor var_10338_to_fp16 = const()[name = tensor("op_10338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_821_cast_fp16 = mul(x = var_10337_cast_fp16, y = var_10338_to_fp16)[name = tensor("aw_821_cast_fp16")]; + tensor var_10341_equation_0 = const()[name = tensor("op_10341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10341_cast_fp16 = einsum(equation = var_10341_equation_0, values = (var_10183_cast_fp16, var_10100_cast_fp16))[name = tensor("op_10341_cast_fp16")]; + tensor var_10342_to_fp16 = const()[name = tensor("op_10342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_823_cast_fp16 = mul(x = var_10341_cast_fp16, y = var_10342_to_fp16)[name = tensor("aw_823_cast_fp16")]; + tensor var_10345_equation_0 = const()[name = tensor("op_10345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10345_cast_fp16 = einsum(equation = var_10345_equation_0, values = (var_10187_cast_fp16, var_10104_cast_fp16))[name = tensor("op_10345_cast_fp16")]; + tensor var_10346_to_fp16 = const()[name = tensor("op_10346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_825_cast_fp16 = mul(x = var_10345_cast_fp16, y = var_10346_to_fp16)[name = tensor("aw_825_cast_fp16")]; + tensor var_10349_equation_0 = const()[name = tensor("op_10349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10349_cast_fp16 = einsum(equation = var_10349_equation_0, values = (var_10191_cast_fp16, var_10108_cast_fp16))[name = tensor("op_10349_cast_fp16")]; + tensor var_10350_to_fp16 = const()[name = tensor("op_10350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_827_cast_fp16 = mul(x = var_10349_cast_fp16, y = var_10350_to_fp16)[name = tensor("aw_827_cast_fp16")]; + tensor var_10353_equation_0 = const()[name = tensor("op_10353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10353_cast_fp16 = einsum(equation = var_10353_equation_0, values = (var_10195_cast_fp16, var_10112_cast_fp16))[name = tensor("op_10353_cast_fp16")]; + tensor var_10354_to_fp16 = const()[name = tensor("op_10354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_829_cast_fp16 = mul(x = var_10353_cast_fp16, y = var_10354_to_fp16)[name = tensor("aw_829_cast_fp16")]; + tensor var_10357_equation_0 = const()[name = tensor("op_10357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10357_cast_fp16 = einsum(equation = var_10357_equation_0, values = (var_10199_cast_fp16, var_10116_cast_fp16))[name = tensor("op_10357_cast_fp16")]; + tensor var_10358_to_fp16 = const()[name = tensor("op_10358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_831_cast_fp16 = mul(x = var_10357_cast_fp16, y = var_10358_to_fp16)[name = tensor("aw_831_cast_fp16")]; + tensor var_10361_equation_0 = const()[name = tensor("op_10361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10361_cast_fp16 = einsum(equation = var_10361_equation_0, values = (var_10203_cast_fp16, var_10120_cast_fp16))[name = tensor("op_10361_cast_fp16")]; + tensor var_10362_to_fp16 = const()[name = tensor("op_10362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_833_cast_fp16 = mul(x = var_10361_cast_fp16, y = var_10362_to_fp16)[name = tensor("aw_833_cast_fp16")]; + tensor var_10365_equation_0 = const()[name = tensor("op_10365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10365_cast_fp16 = einsum(equation = var_10365_equation_0, values = (var_10207_cast_fp16, var_10124_cast_fp16))[name = tensor("op_10365_cast_fp16")]; + tensor var_10366_to_fp16 = const()[name = tensor("op_10366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_835_cast_fp16 = mul(x = var_10365_cast_fp16, y = var_10366_to_fp16)[name = tensor("aw_835_cast_fp16")]; + tensor var_10369_equation_0 = const()[name = tensor("op_10369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10369_cast_fp16 = einsum(equation = var_10369_equation_0, values = (var_10211_cast_fp16, var_10128_cast_fp16))[name = tensor("op_10369_cast_fp16")]; + tensor var_10370_to_fp16 = const()[name = tensor("op_10370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_837_cast_fp16 = mul(x = var_10369_cast_fp16, y = var_10370_to_fp16)[name = tensor("aw_837_cast_fp16")]; + tensor var_10373_equation_0 = const()[name = tensor("op_10373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10373_cast_fp16 = einsum(equation = var_10373_equation_0, values = (var_10215_cast_fp16, var_10132_cast_fp16))[name = tensor("op_10373_cast_fp16")]; + tensor var_10374_to_fp16 = const()[name = tensor("op_10374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_839_cast_fp16 = mul(x = var_10373_cast_fp16, y = var_10374_to_fp16)[name = tensor("aw_839_cast_fp16")]; + tensor var_10376_cast_fp16 = softmax(axis = var_2624, x = aw_801_cast_fp16)[name = tensor("op_10376_cast_fp16")]; + tensor var_10377_cast_fp16 = softmax(axis = var_2624, x = aw_803_cast_fp16)[name = tensor("op_10377_cast_fp16")]; + tensor var_10378_cast_fp16 = softmax(axis = var_2624, x = aw_805_cast_fp16)[name = tensor("op_10378_cast_fp16")]; + tensor var_10379_cast_fp16 = softmax(axis = var_2624, x = aw_807_cast_fp16)[name = tensor("op_10379_cast_fp16")]; + tensor var_10380_cast_fp16 = softmax(axis = var_2624, x = aw_809_cast_fp16)[name = tensor("op_10380_cast_fp16")]; + tensor var_10381_cast_fp16 = softmax(axis = var_2624, x = aw_811_cast_fp16)[name = tensor("op_10381_cast_fp16")]; + tensor var_10382_cast_fp16 = softmax(axis = var_2624, x = aw_813_cast_fp16)[name = tensor("op_10382_cast_fp16")]; + tensor var_10383_cast_fp16 = softmax(axis = var_2624, x = aw_815_cast_fp16)[name = tensor("op_10383_cast_fp16")]; + tensor var_10384_cast_fp16 = softmax(axis = var_2624, x = aw_817_cast_fp16)[name = tensor("op_10384_cast_fp16")]; + tensor var_10385_cast_fp16 = softmax(axis = var_2624, x = aw_819_cast_fp16)[name = tensor("op_10385_cast_fp16")]; + tensor var_10386_cast_fp16 = softmax(axis = var_2624, x = aw_821_cast_fp16)[name = tensor("op_10386_cast_fp16")]; + tensor var_10387_cast_fp16 = softmax(axis = var_2624, x = aw_823_cast_fp16)[name = tensor("op_10387_cast_fp16")]; + tensor var_10388_cast_fp16 = softmax(axis = var_2624, x = aw_825_cast_fp16)[name = tensor("op_10388_cast_fp16")]; + tensor var_10389_cast_fp16 = softmax(axis = var_2624, x = aw_827_cast_fp16)[name = tensor("op_10389_cast_fp16")]; + tensor var_10390_cast_fp16 = softmax(axis = var_2624, x = aw_829_cast_fp16)[name = tensor("op_10390_cast_fp16")]; + tensor var_10391_cast_fp16 = softmax(axis = var_2624, x = aw_831_cast_fp16)[name = tensor("op_10391_cast_fp16")]; + tensor var_10392_cast_fp16 = softmax(axis = var_2624, x = aw_833_cast_fp16)[name = tensor("op_10392_cast_fp16")]; + tensor var_10393_cast_fp16 = softmax(axis = var_2624, x = aw_835_cast_fp16)[name = tensor("op_10393_cast_fp16")]; + tensor var_10394_cast_fp16 = softmax(axis = var_2624, x = aw_837_cast_fp16)[name = tensor("op_10394_cast_fp16")]; + tensor var_10395_cast_fp16 = softmax(axis = var_2624, x = aw_839_cast_fp16)[name = tensor("op_10395_cast_fp16")]; + tensor var_10397_equation_0 = const()[name = tensor("op_10397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10397_cast_fp16 = einsum(equation = var_10397_equation_0, values = (var_10217_cast_fp16, var_10376_cast_fp16))[name = tensor("op_10397_cast_fp16")]; + tensor var_10399_equation_0 = const()[name = tensor("op_10399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10399_cast_fp16 = einsum(equation = var_10399_equation_0, values = (var_10221_cast_fp16, var_10377_cast_fp16))[name = tensor("op_10399_cast_fp16")]; + tensor var_10401_equation_0 = const()[name = tensor("op_10401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10401_cast_fp16 = einsum(equation = var_10401_equation_0, values = (var_10225_cast_fp16, var_10378_cast_fp16))[name = tensor("op_10401_cast_fp16")]; + tensor var_10403_equation_0 = const()[name = tensor("op_10403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10403_cast_fp16 = einsum(equation = var_10403_equation_0, values = (var_10229_cast_fp16, var_10379_cast_fp16))[name = tensor("op_10403_cast_fp16")]; + tensor var_10405_equation_0 = const()[name = tensor("op_10405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10405_cast_fp16 = einsum(equation = var_10405_equation_0, values = (var_10233_cast_fp16, var_10380_cast_fp16))[name = tensor("op_10405_cast_fp16")]; + tensor var_10407_equation_0 = const()[name = tensor("op_10407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10407_cast_fp16 = einsum(equation = var_10407_equation_0, values = (var_10237_cast_fp16, var_10381_cast_fp16))[name = tensor("op_10407_cast_fp16")]; + tensor var_10409_equation_0 = const()[name = tensor("op_10409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10409_cast_fp16 = einsum(equation = var_10409_equation_0, values = (var_10241_cast_fp16, var_10382_cast_fp16))[name = tensor("op_10409_cast_fp16")]; + tensor var_10411_equation_0 = const()[name = tensor("op_10411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10411_cast_fp16 = einsum(equation = var_10411_equation_0, values = (var_10245_cast_fp16, var_10383_cast_fp16))[name = tensor("op_10411_cast_fp16")]; + tensor var_10413_equation_0 = const()[name = tensor("op_10413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10413_cast_fp16 = einsum(equation = var_10413_equation_0, values = (var_10249_cast_fp16, var_10384_cast_fp16))[name = tensor("op_10413_cast_fp16")]; + tensor var_10415_equation_0 = const()[name = tensor("op_10415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10415_cast_fp16 = einsum(equation = var_10415_equation_0, values = (var_10253_cast_fp16, var_10385_cast_fp16))[name = tensor("op_10415_cast_fp16")]; + tensor var_10417_equation_0 = const()[name = tensor("op_10417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10417_cast_fp16 = einsum(equation = var_10417_equation_0, values = (var_10257_cast_fp16, var_10386_cast_fp16))[name = tensor("op_10417_cast_fp16")]; + tensor var_10419_equation_0 = const()[name = tensor("op_10419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10419_cast_fp16 = einsum(equation = var_10419_equation_0, values = (var_10261_cast_fp16, var_10387_cast_fp16))[name = tensor("op_10419_cast_fp16")]; + tensor var_10421_equation_0 = const()[name = tensor("op_10421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10421_cast_fp16 = einsum(equation = var_10421_equation_0, values = (var_10265_cast_fp16, var_10388_cast_fp16))[name = tensor("op_10421_cast_fp16")]; + tensor var_10423_equation_0 = const()[name = tensor("op_10423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10423_cast_fp16 = einsum(equation = var_10423_equation_0, values = (var_10269_cast_fp16, var_10389_cast_fp16))[name = tensor("op_10423_cast_fp16")]; + tensor var_10425_equation_0 = const()[name = tensor("op_10425_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10425_cast_fp16 = einsum(equation = var_10425_equation_0, values = (var_10273_cast_fp16, var_10390_cast_fp16))[name = tensor("op_10425_cast_fp16")]; + tensor var_10427_equation_0 = const()[name = tensor("op_10427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10427_cast_fp16 = einsum(equation = var_10427_equation_0, values = (var_10277_cast_fp16, var_10391_cast_fp16))[name = tensor("op_10427_cast_fp16")]; + tensor var_10429_equation_0 = const()[name = tensor("op_10429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10429_cast_fp16 = einsum(equation = var_10429_equation_0, values = (var_10281_cast_fp16, var_10392_cast_fp16))[name = tensor("op_10429_cast_fp16")]; + tensor var_10431_equation_0 = const()[name = tensor("op_10431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10431_cast_fp16 = einsum(equation = var_10431_equation_0, values = (var_10285_cast_fp16, var_10393_cast_fp16))[name = tensor("op_10431_cast_fp16")]; + tensor var_10433_equation_0 = const()[name = tensor("op_10433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10433_cast_fp16 = einsum(equation = var_10433_equation_0, values = (var_10289_cast_fp16, var_10394_cast_fp16))[name = tensor("op_10433_cast_fp16")]; + tensor var_10435_equation_0 = const()[name = tensor("op_10435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10435_cast_fp16 = einsum(equation = var_10435_equation_0, values = (var_10293_cast_fp16, var_10395_cast_fp16))[name = tensor("op_10435_cast_fp16")]; + tensor input_195_interleave_0 = const()[name = tensor("input_195_interleave_0"), val = tensor(false)]; + tensor input_195_cast_fp16 = concat(axis = var_2624, interleave = input_195_interleave_0, values = (var_10397_cast_fp16, var_10399_cast_fp16, var_10401_cast_fp16, var_10403_cast_fp16, var_10405_cast_fp16, var_10407_cast_fp16, var_10409_cast_fp16, var_10411_cast_fp16, var_10413_cast_fp16, var_10415_cast_fp16, var_10417_cast_fp16, var_10419_cast_fp16, var_10421_cast_fp16, var_10423_cast_fp16, var_10425_cast_fp16, var_10427_cast_fp16, var_10429_cast_fp16, var_10431_cast_fp16, var_10433_cast_fp16, var_10435_cast_fp16))[name = tensor("input_195_cast_fp16")]; + tensor var_10445_pad_type_0 = const()[name = tensor("op_10445_pad_type_0"), val = tensor("valid")]; + tensor var_10445_strides_0 = const()[name = tensor("op_10445_strides_0"), val = tensor([1, 1])]; + tensor var_10445_pad_0 = const()[name = tensor("op_10445_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10445_dilations_0 = const()[name = tensor("op_10445_dilations_0"), val = tensor([1, 1])]; + tensor var_10445_groups_0 = const()[name = tensor("op_10445_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286910592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288139456))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288139648)))]; + tensor var_10445_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_10445_dilations_0, groups = var_10445_groups_0, pad = var_10445_pad_0, pad_type = var_10445_pad_type_0, strides = var_10445_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = tensor("op_10445_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = var_10445_cast_fp16, y = inputs_73_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; + tensor hidden_states_115_axes_0 = const()[name = tensor("hidden_states_115_axes_0"), val = tensor([1])]; + tensor hidden_states_115_gamma_0_to_fp16 = const()[name = tensor("hidden_states_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288142272)))]; + tensor hidden_states_115_beta_0_to_fp16 = const()[name = tensor("hidden_states_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288144896)))]; + tensor var_10455_to_fp16 = const()[name = tensor("op_10455_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_115_cast_fp16 = layer_norm(axes = hidden_states_115_axes_0, beta = hidden_states_115_beta_0_to_fp16, epsilon = var_10455_to_fp16, gamma = hidden_states_115_gamma_0_to_fp16, x = inputs_75_cast_fp16)[name = tensor("hidden_states_115_cast_fp16")]; + tensor q_51_pad_type_0 = const()[name = tensor("q_51_pad_type_0"), val = tensor("valid")]; + tensor q_51_strides_0 = const()[name = tensor("q_51_strides_0"), val = tensor([1, 1])]; + tensor q_51_pad_0 = const()[name = tensor("q_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_51_dilations_0 = const()[name = tensor("q_51_dilations_0"), val = tensor([1, 1])]; + tensor q_51_groups_0 = const()[name = tensor("q_51_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288147520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289376384))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_51_cast_fp16 = conv(dilations = q_51_dilations_0, groups = q_51_groups_0, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = q_51_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_115_cast_fp16)[name = tensor("q_51_cast_fp16")]; + tensor k_101_pad_type_0 = const()[name = tensor("k_101_pad_type_0"), val = tensor("valid")]; + tensor k_101_strides_0 = const()[name = tensor("k_101_strides_0"), val = tensor([1, 1])]; + tensor k_101_pad_0 = const()[name = tensor("k_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_101_dilations_0 = const()[name = tensor("k_101_dilations_0"), val = tensor([1, 1])]; + tensor k_101_groups_0 = const()[name = tensor("k_101_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289376576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291342720))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_101_cast_fp16 = conv(dilations = k_101_dilations_0, groups = k_101_groups_0, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = k_101_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_101_cast_fp16")]; + tensor v_51_pad_type_0 = const()[name = tensor("v_51_pad_type_0"), val = tensor("valid")]; + tensor v_51_strides_0 = const()[name = tensor("v_51_strides_0"), val = tensor([1, 1])]; + tensor v_51_pad_0 = const()[name = tensor("v_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_51_dilations_0 = const()[name = tensor("v_51_dilations_0"), val = tensor([1, 1])]; + tensor v_51_groups_0 = const()[name = tensor("v_51_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291342912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293309056))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_51_cast_fp16 = conv(dilations = v_51_dilations_0, groups = v_51_groups_0, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = v_51_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_51_cast_fp16")]; + tensor var_10488_begin_0 = const()[name = tensor("op_10488_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10488_end_0 = const()[name = tensor("op_10488_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10488_end_mask_0 = const()[name = tensor("op_10488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10488_cast_fp16 = slice_by_index(begin = var_10488_begin_0, end = var_10488_end_0, end_mask = var_10488_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10488_cast_fp16")]; + tensor var_10492_begin_0 = const()[name = tensor("op_10492_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10492_end_0 = const()[name = tensor("op_10492_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10492_end_mask_0 = const()[name = tensor("op_10492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10492_cast_fp16 = slice_by_index(begin = var_10492_begin_0, end = var_10492_end_0, end_mask = var_10492_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10492_cast_fp16")]; + tensor var_10496_begin_0 = const()[name = tensor("op_10496_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10496_end_0 = const()[name = tensor("op_10496_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10496_end_mask_0 = const()[name = tensor("op_10496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10496_cast_fp16 = slice_by_index(begin = var_10496_begin_0, end = var_10496_end_0, end_mask = var_10496_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10496_cast_fp16")]; + tensor var_10500_begin_0 = const()[name = tensor("op_10500_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10500_end_0 = const()[name = tensor("op_10500_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10500_end_mask_0 = const()[name = tensor("op_10500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10500_cast_fp16 = slice_by_index(begin = var_10500_begin_0, end = var_10500_end_0, end_mask = var_10500_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10500_cast_fp16")]; + tensor var_10504_begin_0 = const()[name = tensor("op_10504_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10504_end_0 = const()[name = tensor("op_10504_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10504_end_mask_0 = const()[name = tensor("op_10504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10504_cast_fp16 = slice_by_index(begin = var_10504_begin_0, end = var_10504_end_0, end_mask = var_10504_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10504_cast_fp16")]; + tensor var_10508_begin_0 = const()[name = tensor("op_10508_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10508_end_0 = const()[name = tensor("op_10508_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10508_end_mask_0 = const()[name = tensor("op_10508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10508_cast_fp16 = slice_by_index(begin = var_10508_begin_0, end = var_10508_end_0, end_mask = var_10508_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10508_cast_fp16")]; + tensor var_10512_begin_0 = const()[name = tensor("op_10512_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10512_end_0 = const()[name = tensor("op_10512_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10512_end_mask_0 = const()[name = tensor("op_10512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10512_cast_fp16 = slice_by_index(begin = var_10512_begin_0, end = var_10512_end_0, end_mask = var_10512_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10512_cast_fp16")]; + tensor var_10516_begin_0 = const()[name = tensor("op_10516_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10516_end_0 = const()[name = tensor("op_10516_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10516_end_mask_0 = const()[name = tensor("op_10516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10516_cast_fp16 = slice_by_index(begin = var_10516_begin_0, end = var_10516_end_0, end_mask = var_10516_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10516_cast_fp16")]; + tensor var_10520_begin_0 = const()[name = tensor("op_10520_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10520_end_0 = const()[name = tensor("op_10520_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10520_end_mask_0 = const()[name = tensor("op_10520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10520_cast_fp16 = slice_by_index(begin = var_10520_begin_0, end = var_10520_end_0, end_mask = var_10520_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10520_cast_fp16")]; + tensor var_10524_begin_0 = const()[name = tensor("op_10524_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10524_end_0 = const()[name = tensor("op_10524_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10524_end_mask_0 = const()[name = tensor("op_10524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10524_cast_fp16 = slice_by_index(begin = var_10524_begin_0, end = var_10524_end_0, end_mask = var_10524_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10524_cast_fp16")]; + tensor var_10528_begin_0 = const()[name = tensor("op_10528_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10528_end_0 = const()[name = tensor("op_10528_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_10528_end_mask_0 = const()[name = tensor("op_10528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10528_cast_fp16 = slice_by_index(begin = var_10528_begin_0, end = var_10528_end_0, end_mask = var_10528_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10528_cast_fp16")]; + tensor var_10532_begin_0 = const()[name = tensor("op_10532_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10532_end_0 = const()[name = tensor("op_10532_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_10532_end_mask_0 = const()[name = tensor("op_10532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10532_cast_fp16 = slice_by_index(begin = var_10532_begin_0, end = var_10532_end_0, end_mask = var_10532_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10532_cast_fp16")]; + tensor var_10536_begin_0 = const()[name = tensor("op_10536_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10536_end_0 = const()[name = tensor("op_10536_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_10536_end_mask_0 = const()[name = tensor("op_10536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10536_cast_fp16 = slice_by_index(begin = var_10536_begin_0, end = var_10536_end_0, end_mask = var_10536_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10536_cast_fp16")]; + tensor var_10540_begin_0 = const()[name = tensor("op_10540_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10540_end_0 = const()[name = tensor("op_10540_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_10540_end_mask_0 = const()[name = tensor("op_10540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10540_cast_fp16 = slice_by_index(begin = var_10540_begin_0, end = var_10540_end_0, end_mask = var_10540_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10540_cast_fp16")]; + tensor var_10544_begin_0 = const()[name = tensor("op_10544_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10544_end_0 = const()[name = tensor("op_10544_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_10544_end_mask_0 = const()[name = tensor("op_10544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10544_cast_fp16 = slice_by_index(begin = var_10544_begin_0, end = var_10544_end_0, end_mask = var_10544_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10544_cast_fp16")]; + tensor var_10548_begin_0 = const()[name = tensor("op_10548_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10548_end_0 = const()[name = tensor("op_10548_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10548_end_mask_0 = const()[name = tensor("op_10548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10548_cast_fp16 = slice_by_index(begin = var_10548_begin_0, end = var_10548_end_0, end_mask = var_10548_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10548_cast_fp16")]; + tensor var_10552_begin_0 = const()[name = tensor("op_10552_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10552_end_0 = const()[name = tensor("op_10552_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_10552_end_mask_0 = const()[name = tensor("op_10552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10552_cast_fp16 = slice_by_index(begin = var_10552_begin_0, end = var_10552_end_0, end_mask = var_10552_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10552_cast_fp16")]; + tensor var_10556_begin_0 = const()[name = tensor("op_10556_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10556_end_0 = const()[name = tensor("op_10556_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_10556_end_mask_0 = const()[name = tensor("op_10556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10556_cast_fp16 = slice_by_index(begin = var_10556_begin_0, end = var_10556_end_0, end_mask = var_10556_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10556_cast_fp16")]; + tensor var_10560_begin_0 = const()[name = tensor("op_10560_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10560_end_0 = const()[name = tensor("op_10560_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_10560_end_mask_0 = const()[name = tensor("op_10560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10560_cast_fp16 = slice_by_index(begin = var_10560_begin_0, end = var_10560_end_0, end_mask = var_10560_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10560_cast_fp16")]; + tensor var_10564_begin_0 = const()[name = tensor("op_10564_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10564_end_0 = const()[name = tensor("op_10564_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_10564_end_mask_0 = const()[name = tensor("op_10564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10564_cast_fp16 = slice_by_index(begin = var_10564_begin_0, end = var_10564_end_0, end_mask = var_10564_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10564_cast_fp16")]; + tensor k_103_perm_0 = const()[name = tensor("k_103_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10571_begin_0 = const()[name = tensor("op_10571_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10571_end_0 = const()[name = tensor("op_10571_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_10571_end_mask_0 = const()[name = tensor("op_10571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_103_cast_fp16 = transpose(perm = k_103_perm_0, x = k_101_cast_fp16)[name = tensor("transpose_42")]; + tensor var_10571_cast_fp16 = slice_by_index(begin = var_10571_begin_0, end = var_10571_end_0, end_mask = var_10571_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10571_cast_fp16")]; + tensor var_10575_begin_0 = const()[name = tensor("op_10575_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10575_end_0 = const()[name = tensor("op_10575_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_10575_end_mask_0 = const()[name = tensor("op_10575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10575_cast_fp16 = slice_by_index(begin = var_10575_begin_0, end = var_10575_end_0, end_mask = var_10575_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10575_cast_fp16")]; + tensor var_10579_begin_0 = const()[name = tensor("op_10579_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10579_end_0 = const()[name = tensor("op_10579_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_10579_end_mask_0 = const()[name = tensor("op_10579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10579_cast_fp16 = slice_by_index(begin = var_10579_begin_0, end = var_10579_end_0, end_mask = var_10579_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10579_cast_fp16")]; + tensor var_10583_begin_0 = const()[name = tensor("op_10583_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10583_end_0 = const()[name = tensor("op_10583_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_10583_end_mask_0 = const()[name = tensor("op_10583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10583_cast_fp16 = slice_by_index(begin = var_10583_begin_0, end = var_10583_end_0, end_mask = var_10583_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10583_cast_fp16")]; + tensor var_10587_begin_0 = const()[name = tensor("op_10587_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10587_end_0 = const()[name = tensor("op_10587_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_10587_end_mask_0 = const()[name = tensor("op_10587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10587_cast_fp16 = slice_by_index(begin = var_10587_begin_0, end = var_10587_end_0, end_mask = var_10587_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10587_cast_fp16")]; + tensor var_10591_begin_0 = const()[name = tensor("op_10591_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10591_end_0 = const()[name = tensor("op_10591_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_10591_end_mask_0 = const()[name = tensor("op_10591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10591_cast_fp16 = slice_by_index(begin = var_10591_begin_0, end = var_10591_end_0, end_mask = var_10591_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10591_cast_fp16")]; + tensor var_10595_begin_0 = const()[name = tensor("op_10595_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10595_end_0 = const()[name = tensor("op_10595_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_10595_end_mask_0 = const()[name = tensor("op_10595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10595_cast_fp16 = slice_by_index(begin = var_10595_begin_0, end = var_10595_end_0, end_mask = var_10595_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10595_cast_fp16")]; + tensor var_10599_begin_0 = const()[name = tensor("op_10599_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10599_end_0 = const()[name = tensor("op_10599_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_10599_end_mask_0 = const()[name = tensor("op_10599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10599_cast_fp16 = slice_by_index(begin = var_10599_begin_0, end = var_10599_end_0, end_mask = var_10599_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10599_cast_fp16")]; + tensor var_10603_begin_0 = const()[name = tensor("op_10603_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10603_end_0 = const()[name = tensor("op_10603_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_10603_end_mask_0 = const()[name = tensor("op_10603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10603_cast_fp16 = slice_by_index(begin = var_10603_begin_0, end = var_10603_end_0, end_mask = var_10603_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10603_cast_fp16")]; + tensor var_10607_begin_0 = const()[name = tensor("op_10607_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10607_end_0 = const()[name = tensor("op_10607_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_10607_end_mask_0 = const()[name = tensor("op_10607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10607_cast_fp16 = slice_by_index(begin = var_10607_begin_0, end = var_10607_end_0, end_mask = var_10607_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10607_cast_fp16")]; + tensor var_10611_begin_0 = const()[name = tensor("op_10611_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10611_end_0 = const()[name = tensor("op_10611_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_10611_end_mask_0 = const()[name = tensor("op_10611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10611_cast_fp16 = slice_by_index(begin = var_10611_begin_0, end = var_10611_end_0, end_mask = var_10611_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10611_cast_fp16")]; + tensor var_10615_begin_0 = const()[name = tensor("op_10615_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10615_end_0 = const()[name = tensor("op_10615_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_10615_end_mask_0 = const()[name = tensor("op_10615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10615_cast_fp16 = slice_by_index(begin = var_10615_begin_0, end = var_10615_end_0, end_mask = var_10615_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10615_cast_fp16")]; + tensor var_10619_begin_0 = const()[name = tensor("op_10619_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_10619_end_0 = const()[name = tensor("op_10619_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_10619_end_mask_0 = const()[name = tensor("op_10619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10619_cast_fp16 = slice_by_index(begin = var_10619_begin_0, end = var_10619_end_0, end_mask = var_10619_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10619_cast_fp16")]; + tensor var_10623_begin_0 = const()[name = tensor("op_10623_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_10623_end_0 = const()[name = tensor("op_10623_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_10623_end_mask_0 = const()[name = tensor("op_10623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10623_cast_fp16 = slice_by_index(begin = var_10623_begin_0, end = var_10623_end_0, end_mask = var_10623_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10623_cast_fp16")]; + tensor var_10627_begin_0 = const()[name = tensor("op_10627_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_10627_end_0 = const()[name = tensor("op_10627_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_10627_end_mask_0 = const()[name = tensor("op_10627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10627_cast_fp16 = slice_by_index(begin = var_10627_begin_0, end = var_10627_end_0, end_mask = var_10627_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10627_cast_fp16")]; + tensor var_10631_begin_0 = const()[name = tensor("op_10631_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_10631_end_0 = const()[name = tensor("op_10631_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_10631_end_mask_0 = const()[name = tensor("op_10631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10631_cast_fp16 = slice_by_index(begin = var_10631_begin_0, end = var_10631_end_0, end_mask = var_10631_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10631_cast_fp16")]; + tensor var_10635_begin_0 = const()[name = tensor("op_10635_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_10635_end_0 = const()[name = tensor("op_10635_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_10635_end_mask_0 = const()[name = tensor("op_10635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10635_cast_fp16 = slice_by_index(begin = var_10635_begin_0, end = var_10635_end_0, end_mask = var_10635_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10635_cast_fp16")]; + tensor var_10639_begin_0 = const()[name = tensor("op_10639_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_10639_end_0 = const()[name = tensor("op_10639_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_10639_end_mask_0 = const()[name = tensor("op_10639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10639_cast_fp16 = slice_by_index(begin = var_10639_begin_0, end = var_10639_end_0, end_mask = var_10639_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10639_cast_fp16")]; + tensor var_10643_begin_0 = const()[name = tensor("op_10643_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_10643_end_0 = const()[name = tensor("op_10643_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_10643_end_mask_0 = const()[name = tensor("op_10643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10643_cast_fp16 = slice_by_index(begin = var_10643_begin_0, end = var_10643_end_0, end_mask = var_10643_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10643_cast_fp16")]; + tensor var_10647_begin_0 = const()[name = tensor("op_10647_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_10647_end_0 = const()[name = tensor("op_10647_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_10647_end_mask_0 = const()[name = tensor("op_10647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10647_cast_fp16 = slice_by_index(begin = var_10647_begin_0, end = var_10647_end_0, end_mask = var_10647_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10647_cast_fp16")]; + tensor var_10649_begin_0 = const()[name = tensor("op_10649_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10649_end_0 = const()[name = tensor("op_10649_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_10649_end_mask_0 = const()[name = tensor("op_10649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10649_cast_fp16 = slice_by_index(begin = var_10649_begin_0, end = var_10649_end_0, end_mask = var_10649_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10649_cast_fp16")]; + tensor var_10653_begin_0 = const()[name = tensor("op_10653_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10653_end_0 = const()[name = tensor("op_10653_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_10653_end_mask_0 = const()[name = tensor("op_10653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10653_cast_fp16 = slice_by_index(begin = var_10653_begin_0, end = var_10653_end_0, end_mask = var_10653_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10653_cast_fp16")]; + tensor var_10657_begin_0 = const()[name = tensor("op_10657_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10657_end_0 = const()[name = tensor("op_10657_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_10657_end_mask_0 = const()[name = tensor("op_10657_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10657_cast_fp16 = slice_by_index(begin = var_10657_begin_0, end = var_10657_end_0, end_mask = var_10657_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10657_cast_fp16")]; + tensor var_10661_begin_0 = const()[name = tensor("op_10661_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10661_end_0 = const()[name = tensor("op_10661_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_10661_end_mask_0 = const()[name = tensor("op_10661_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10661_cast_fp16 = slice_by_index(begin = var_10661_begin_0, end = var_10661_end_0, end_mask = var_10661_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10661_cast_fp16")]; + tensor var_10665_begin_0 = const()[name = tensor("op_10665_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10665_end_0 = const()[name = tensor("op_10665_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_10665_end_mask_0 = const()[name = tensor("op_10665_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10665_cast_fp16 = slice_by_index(begin = var_10665_begin_0, end = var_10665_end_0, end_mask = var_10665_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10665_cast_fp16")]; + tensor var_10669_begin_0 = const()[name = tensor("op_10669_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10669_end_0 = const()[name = tensor("op_10669_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_10669_end_mask_0 = const()[name = tensor("op_10669_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10669_cast_fp16 = slice_by_index(begin = var_10669_begin_0, end = var_10669_end_0, end_mask = var_10669_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10669_cast_fp16")]; + tensor var_10673_begin_0 = const()[name = tensor("op_10673_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10673_end_0 = const()[name = tensor("op_10673_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_10673_end_mask_0 = const()[name = tensor("op_10673_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10673_cast_fp16 = slice_by_index(begin = var_10673_begin_0, end = var_10673_end_0, end_mask = var_10673_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10673_cast_fp16")]; + tensor var_10677_begin_0 = const()[name = tensor("op_10677_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10677_end_0 = const()[name = tensor("op_10677_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_10677_end_mask_0 = const()[name = tensor("op_10677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10677_cast_fp16 = slice_by_index(begin = var_10677_begin_0, end = var_10677_end_0, end_mask = var_10677_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10677_cast_fp16")]; + tensor var_10681_begin_0 = const()[name = tensor("op_10681_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10681_end_0 = const()[name = tensor("op_10681_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_10681_end_mask_0 = const()[name = tensor("op_10681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10681_cast_fp16 = slice_by_index(begin = var_10681_begin_0, end = var_10681_end_0, end_mask = var_10681_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10681_cast_fp16")]; + tensor var_10685_begin_0 = const()[name = tensor("op_10685_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10685_end_0 = const()[name = tensor("op_10685_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_10685_end_mask_0 = const()[name = tensor("op_10685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10685_cast_fp16 = slice_by_index(begin = var_10685_begin_0, end = var_10685_end_0, end_mask = var_10685_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10685_cast_fp16")]; + tensor var_10689_begin_0 = const()[name = tensor("op_10689_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10689_end_0 = const()[name = tensor("op_10689_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_10689_end_mask_0 = const()[name = tensor("op_10689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10689_cast_fp16 = slice_by_index(begin = var_10689_begin_0, end = var_10689_end_0, end_mask = var_10689_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10689_cast_fp16")]; + tensor var_10693_begin_0 = const()[name = tensor("op_10693_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10693_end_0 = const()[name = tensor("op_10693_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_10693_end_mask_0 = const()[name = tensor("op_10693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10693_cast_fp16 = slice_by_index(begin = var_10693_begin_0, end = var_10693_end_0, end_mask = var_10693_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10693_cast_fp16")]; + tensor var_10697_begin_0 = const()[name = tensor("op_10697_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10697_end_0 = const()[name = tensor("op_10697_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_10697_end_mask_0 = const()[name = tensor("op_10697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10697_cast_fp16 = slice_by_index(begin = var_10697_begin_0, end = var_10697_end_0, end_mask = var_10697_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10697_cast_fp16")]; + tensor var_10701_begin_0 = const()[name = tensor("op_10701_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10701_end_0 = const()[name = tensor("op_10701_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_10701_end_mask_0 = const()[name = tensor("op_10701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10701_cast_fp16 = slice_by_index(begin = var_10701_begin_0, end = var_10701_end_0, end_mask = var_10701_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10701_cast_fp16")]; + tensor var_10705_begin_0 = const()[name = tensor("op_10705_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10705_end_0 = const()[name = tensor("op_10705_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_10705_end_mask_0 = const()[name = tensor("op_10705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10705_cast_fp16 = slice_by_index(begin = var_10705_begin_0, end = var_10705_end_0, end_mask = var_10705_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10705_cast_fp16")]; + tensor var_10709_begin_0 = const()[name = tensor("op_10709_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10709_end_0 = const()[name = tensor("op_10709_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_10709_end_mask_0 = const()[name = tensor("op_10709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10709_cast_fp16 = slice_by_index(begin = var_10709_begin_0, end = var_10709_end_0, end_mask = var_10709_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10709_cast_fp16")]; + tensor var_10713_begin_0 = const()[name = tensor("op_10713_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10713_end_0 = const()[name = tensor("op_10713_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_10713_end_mask_0 = const()[name = tensor("op_10713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10713_cast_fp16 = slice_by_index(begin = var_10713_begin_0, end = var_10713_end_0, end_mask = var_10713_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10713_cast_fp16")]; + tensor var_10717_begin_0 = const()[name = tensor("op_10717_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10717_end_0 = const()[name = tensor("op_10717_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_10717_end_mask_0 = const()[name = tensor("op_10717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10717_cast_fp16 = slice_by_index(begin = var_10717_begin_0, end = var_10717_end_0, end_mask = var_10717_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10717_cast_fp16")]; + tensor var_10721_begin_0 = const()[name = tensor("op_10721_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10721_end_0 = const()[name = tensor("op_10721_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_10721_end_mask_0 = const()[name = tensor("op_10721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10721_cast_fp16 = slice_by_index(begin = var_10721_begin_0, end = var_10721_end_0, end_mask = var_10721_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10721_cast_fp16")]; + tensor var_10725_begin_0 = const()[name = tensor("op_10725_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10725_end_0 = const()[name = tensor("op_10725_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_10725_end_mask_0 = const()[name = tensor("op_10725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10725_cast_fp16 = slice_by_index(begin = var_10725_begin_0, end = var_10725_end_0, end_mask = var_10725_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10725_cast_fp16")]; + tensor var_10729_equation_0 = const()[name = tensor("op_10729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10729_cast_fp16 = einsum(equation = var_10729_equation_0, values = (var_10571_cast_fp16, var_10488_cast_fp16))[name = tensor("op_10729_cast_fp16")]; + tensor var_10730_to_fp16 = const()[name = tensor("op_10730_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_841_cast_fp16 = mul(x = var_10729_cast_fp16, y = var_10730_to_fp16)[name = tensor("aw_841_cast_fp16")]; + tensor var_10733_equation_0 = const()[name = tensor("op_10733_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10733_cast_fp16 = einsum(equation = var_10733_equation_0, values = (var_10575_cast_fp16, var_10492_cast_fp16))[name = tensor("op_10733_cast_fp16")]; + tensor var_10734_to_fp16 = const()[name = tensor("op_10734_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_843_cast_fp16 = mul(x = var_10733_cast_fp16, y = var_10734_to_fp16)[name = tensor("aw_843_cast_fp16")]; + tensor var_10737_equation_0 = const()[name = tensor("op_10737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10737_cast_fp16 = einsum(equation = var_10737_equation_0, values = (var_10579_cast_fp16, var_10496_cast_fp16))[name = tensor("op_10737_cast_fp16")]; + tensor var_10738_to_fp16 = const()[name = tensor("op_10738_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_845_cast_fp16 = mul(x = var_10737_cast_fp16, y = var_10738_to_fp16)[name = tensor("aw_845_cast_fp16")]; + tensor var_10741_equation_0 = const()[name = tensor("op_10741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10741_cast_fp16 = einsum(equation = var_10741_equation_0, values = (var_10583_cast_fp16, var_10500_cast_fp16))[name = tensor("op_10741_cast_fp16")]; + tensor var_10742_to_fp16 = const()[name = tensor("op_10742_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_847_cast_fp16 = mul(x = var_10741_cast_fp16, y = var_10742_to_fp16)[name = tensor("aw_847_cast_fp16")]; + tensor var_10745_equation_0 = const()[name = tensor("op_10745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10745_cast_fp16 = einsum(equation = var_10745_equation_0, values = (var_10587_cast_fp16, var_10504_cast_fp16))[name = tensor("op_10745_cast_fp16")]; + tensor var_10746_to_fp16 = const()[name = tensor("op_10746_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_849_cast_fp16 = mul(x = var_10745_cast_fp16, y = var_10746_to_fp16)[name = tensor("aw_849_cast_fp16")]; + tensor var_10749_equation_0 = const()[name = tensor("op_10749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10749_cast_fp16 = einsum(equation = var_10749_equation_0, values = (var_10591_cast_fp16, var_10508_cast_fp16))[name = tensor("op_10749_cast_fp16")]; + tensor var_10750_to_fp16 = const()[name = tensor("op_10750_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_851_cast_fp16 = mul(x = var_10749_cast_fp16, y = var_10750_to_fp16)[name = tensor("aw_851_cast_fp16")]; + tensor var_10753_equation_0 = const()[name = tensor("op_10753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10753_cast_fp16 = einsum(equation = var_10753_equation_0, values = (var_10595_cast_fp16, var_10512_cast_fp16))[name = tensor("op_10753_cast_fp16")]; + tensor var_10754_to_fp16 = const()[name = tensor("op_10754_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_853_cast_fp16 = mul(x = var_10753_cast_fp16, y = var_10754_to_fp16)[name = tensor("aw_853_cast_fp16")]; + tensor var_10757_equation_0 = const()[name = tensor("op_10757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10757_cast_fp16 = einsum(equation = var_10757_equation_0, values = (var_10599_cast_fp16, var_10516_cast_fp16))[name = tensor("op_10757_cast_fp16")]; + tensor var_10758_to_fp16 = const()[name = tensor("op_10758_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_855_cast_fp16 = mul(x = var_10757_cast_fp16, y = var_10758_to_fp16)[name = tensor("aw_855_cast_fp16")]; + tensor var_10761_equation_0 = const()[name = tensor("op_10761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10761_cast_fp16 = einsum(equation = var_10761_equation_0, values = (var_10603_cast_fp16, var_10520_cast_fp16))[name = tensor("op_10761_cast_fp16")]; + tensor var_10762_to_fp16 = const()[name = tensor("op_10762_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_857_cast_fp16 = mul(x = var_10761_cast_fp16, y = var_10762_to_fp16)[name = tensor("aw_857_cast_fp16")]; + tensor var_10765_equation_0 = const()[name = tensor("op_10765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10765_cast_fp16 = einsum(equation = var_10765_equation_0, values = (var_10607_cast_fp16, var_10524_cast_fp16))[name = tensor("op_10765_cast_fp16")]; + tensor var_10766_to_fp16 = const()[name = tensor("op_10766_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_859_cast_fp16 = mul(x = var_10765_cast_fp16, y = var_10766_to_fp16)[name = tensor("aw_859_cast_fp16")]; + tensor var_10769_equation_0 = const()[name = tensor("op_10769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10769_cast_fp16 = einsum(equation = var_10769_equation_0, values = (var_10611_cast_fp16, var_10528_cast_fp16))[name = tensor("op_10769_cast_fp16")]; + tensor var_10770_to_fp16 = const()[name = tensor("op_10770_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_861_cast_fp16 = mul(x = var_10769_cast_fp16, y = var_10770_to_fp16)[name = tensor("aw_861_cast_fp16")]; + tensor var_10773_equation_0 = const()[name = tensor("op_10773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10773_cast_fp16 = einsum(equation = var_10773_equation_0, values = (var_10615_cast_fp16, var_10532_cast_fp16))[name = tensor("op_10773_cast_fp16")]; + tensor var_10774_to_fp16 = const()[name = tensor("op_10774_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_863_cast_fp16 = mul(x = var_10773_cast_fp16, y = var_10774_to_fp16)[name = tensor("aw_863_cast_fp16")]; + tensor var_10777_equation_0 = const()[name = tensor("op_10777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10777_cast_fp16 = einsum(equation = var_10777_equation_0, values = (var_10619_cast_fp16, var_10536_cast_fp16))[name = tensor("op_10777_cast_fp16")]; + tensor var_10778_to_fp16 = const()[name = tensor("op_10778_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_865_cast_fp16 = mul(x = var_10777_cast_fp16, y = var_10778_to_fp16)[name = tensor("aw_865_cast_fp16")]; + tensor var_10781_equation_0 = const()[name = tensor("op_10781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10781_cast_fp16 = einsum(equation = var_10781_equation_0, values = (var_10623_cast_fp16, var_10540_cast_fp16))[name = tensor("op_10781_cast_fp16")]; + tensor var_10782_to_fp16 = const()[name = tensor("op_10782_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_867_cast_fp16 = mul(x = var_10781_cast_fp16, y = var_10782_to_fp16)[name = tensor("aw_867_cast_fp16")]; + tensor var_10785_equation_0 = const()[name = tensor("op_10785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10785_cast_fp16 = einsum(equation = var_10785_equation_0, values = (var_10627_cast_fp16, var_10544_cast_fp16))[name = tensor("op_10785_cast_fp16")]; + tensor var_10786_to_fp16 = const()[name = tensor("op_10786_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_869_cast_fp16 = mul(x = var_10785_cast_fp16, y = var_10786_to_fp16)[name = tensor("aw_869_cast_fp16")]; + tensor var_10789_equation_0 = const()[name = tensor("op_10789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10789_cast_fp16 = einsum(equation = var_10789_equation_0, values = (var_10631_cast_fp16, var_10548_cast_fp16))[name = tensor("op_10789_cast_fp16")]; + tensor var_10790_to_fp16 = const()[name = tensor("op_10790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_871_cast_fp16 = mul(x = var_10789_cast_fp16, y = var_10790_to_fp16)[name = tensor("aw_871_cast_fp16")]; + tensor var_10793_equation_0 = const()[name = tensor("op_10793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10793_cast_fp16 = einsum(equation = var_10793_equation_0, values = (var_10635_cast_fp16, var_10552_cast_fp16))[name = tensor("op_10793_cast_fp16")]; + tensor var_10794_to_fp16 = const()[name = tensor("op_10794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_873_cast_fp16 = mul(x = var_10793_cast_fp16, y = var_10794_to_fp16)[name = tensor("aw_873_cast_fp16")]; + tensor var_10797_equation_0 = const()[name = tensor("op_10797_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10797_cast_fp16 = einsum(equation = var_10797_equation_0, values = (var_10639_cast_fp16, var_10556_cast_fp16))[name = tensor("op_10797_cast_fp16")]; + tensor var_10798_to_fp16 = const()[name = tensor("op_10798_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_875_cast_fp16 = mul(x = var_10797_cast_fp16, y = var_10798_to_fp16)[name = tensor("aw_875_cast_fp16")]; + tensor var_10801_equation_0 = const()[name = tensor("op_10801_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10801_cast_fp16 = einsum(equation = var_10801_equation_0, values = (var_10643_cast_fp16, var_10560_cast_fp16))[name = tensor("op_10801_cast_fp16")]; + tensor var_10802_to_fp16 = const()[name = tensor("op_10802_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_877_cast_fp16 = mul(x = var_10801_cast_fp16, y = var_10802_to_fp16)[name = tensor("aw_877_cast_fp16")]; + tensor var_10805_equation_0 = const()[name = tensor("op_10805_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10805_cast_fp16 = einsum(equation = var_10805_equation_0, values = (var_10647_cast_fp16, var_10564_cast_fp16))[name = tensor("op_10805_cast_fp16")]; + tensor var_10806_to_fp16 = const()[name = tensor("op_10806_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_879_cast_fp16 = mul(x = var_10805_cast_fp16, y = var_10806_to_fp16)[name = tensor("aw_879_cast_fp16")]; + tensor var_10808_cast_fp16 = softmax(axis = var_2624, x = aw_841_cast_fp16)[name = tensor("op_10808_cast_fp16")]; + tensor var_10809_cast_fp16 = softmax(axis = var_2624, x = aw_843_cast_fp16)[name = tensor("op_10809_cast_fp16")]; + tensor var_10810_cast_fp16 = softmax(axis = var_2624, x = aw_845_cast_fp16)[name = tensor("op_10810_cast_fp16")]; + tensor var_10811_cast_fp16 = softmax(axis = var_2624, x = aw_847_cast_fp16)[name = tensor("op_10811_cast_fp16")]; + tensor var_10812_cast_fp16 = softmax(axis = var_2624, x = aw_849_cast_fp16)[name = tensor("op_10812_cast_fp16")]; + tensor var_10813_cast_fp16 = softmax(axis = var_2624, x = aw_851_cast_fp16)[name = tensor("op_10813_cast_fp16")]; + tensor var_10814_cast_fp16 = softmax(axis = var_2624, x = aw_853_cast_fp16)[name = tensor("op_10814_cast_fp16")]; + tensor var_10815_cast_fp16 = softmax(axis = var_2624, x = aw_855_cast_fp16)[name = tensor("op_10815_cast_fp16")]; + tensor var_10816_cast_fp16 = softmax(axis = var_2624, x = aw_857_cast_fp16)[name = tensor("op_10816_cast_fp16")]; + tensor var_10817_cast_fp16 = softmax(axis = var_2624, x = aw_859_cast_fp16)[name = tensor("op_10817_cast_fp16")]; + tensor var_10818_cast_fp16 = softmax(axis = var_2624, x = aw_861_cast_fp16)[name = tensor("op_10818_cast_fp16")]; + tensor var_10819_cast_fp16 = softmax(axis = var_2624, x = aw_863_cast_fp16)[name = tensor("op_10819_cast_fp16")]; + tensor var_10820_cast_fp16 = softmax(axis = var_2624, x = aw_865_cast_fp16)[name = tensor("op_10820_cast_fp16")]; + tensor var_10821_cast_fp16 = softmax(axis = var_2624, x = aw_867_cast_fp16)[name = tensor("op_10821_cast_fp16")]; + tensor var_10822_cast_fp16 = softmax(axis = var_2624, x = aw_869_cast_fp16)[name = tensor("op_10822_cast_fp16")]; + tensor var_10823_cast_fp16 = softmax(axis = var_2624, x = aw_871_cast_fp16)[name = tensor("op_10823_cast_fp16")]; + tensor var_10824_cast_fp16 = softmax(axis = var_2624, x = aw_873_cast_fp16)[name = tensor("op_10824_cast_fp16")]; + tensor var_10825_cast_fp16 = softmax(axis = var_2624, x = aw_875_cast_fp16)[name = tensor("op_10825_cast_fp16")]; + tensor var_10826_cast_fp16 = softmax(axis = var_2624, x = aw_877_cast_fp16)[name = tensor("op_10826_cast_fp16")]; + tensor var_10827_cast_fp16 = softmax(axis = var_2624, x = aw_879_cast_fp16)[name = tensor("op_10827_cast_fp16")]; + tensor var_10829_equation_0 = const()[name = tensor("op_10829_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10829_cast_fp16 = einsum(equation = var_10829_equation_0, values = (var_10649_cast_fp16, var_10808_cast_fp16))[name = tensor("op_10829_cast_fp16")]; + tensor var_10831_equation_0 = const()[name = tensor("op_10831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10831_cast_fp16 = einsum(equation = var_10831_equation_0, values = (var_10653_cast_fp16, var_10809_cast_fp16))[name = tensor("op_10831_cast_fp16")]; + tensor var_10833_equation_0 = const()[name = tensor("op_10833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10833_cast_fp16 = einsum(equation = var_10833_equation_0, values = (var_10657_cast_fp16, var_10810_cast_fp16))[name = tensor("op_10833_cast_fp16")]; + tensor var_10835_equation_0 = const()[name = tensor("op_10835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10835_cast_fp16 = einsum(equation = var_10835_equation_0, values = (var_10661_cast_fp16, var_10811_cast_fp16))[name = tensor("op_10835_cast_fp16")]; + tensor var_10837_equation_0 = const()[name = tensor("op_10837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10837_cast_fp16 = einsum(equation = var_10837_equation_0, values = (var_10665_cast_fp16, var_10812_cast_fp16))[name = tensor("op_10837_cast_fp16")]; + tensor var_10839_equation_0 = const()[name = tensor("op_10839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10839_cast_fp16 = einsum(equation = var_10839_equation_0, values = (var_10669_cast_fp16, var_10813_cast_fp16))[name = tensor("op_10839_cast_fp16")]; + tensor var_10841_equation_0 = const()[name = tensor("op_10841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10841_cast_fp16 = einsum(equation = var_10841_equation_0, values = (var_10673_cast_fp16, var_10814_cast_fp16))[name = tensor("op_10841_cast_fp16")]; + tensor var_10843_equation_0 = const()[name = tensor("op_10843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10843_cast_fp16 = einsum(equation = var_10843_equation_0, values = (var_10677_cast_fp16, var_10815_cast_fp16))[name = tensor("op_10843_cast_fp16")]; + tensor var_10845_equation_0 = const()[name = tensor("op_10845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10845_cast_fp16 = einsum(equation = var_10845_equation_0, values = (var_10681_cast_fp16, var_10816_cast_fp16))[name = tensor("op_10845_cast_fp16")]; + tensor var_10847_equation_0 = const()[name = tensor("op_10847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10847_cast_fp16 = einsum(equation = var_10847_equation_0, values = (var_10685_cast_fp16, var_10817_cast_fp16))[name = tensor("op_10847_cast_fp16")]; + tensor var_10849_equation_0 = const()[name = tensor("op_10849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10849_cast_fp16 = einsum(equation = var_10849_equation_0, values = (var_10689_cast_fp16, var_10818_cast_fp16))[name = tensor("op_10849_cast_fp16")]; + tensor var_10851_equation_0 = const()[name = tensor("op_10851_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10851_cast_fp16 = einsum(equation = var_10851_equation_0, values = (var_10693_cast_fp16, var_10819_cast_fp16))[name = tensor("op_10851_cast_fp16")]; + tensor var_10853_equation_0 = const()[name = tensor("op_10853_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10853_cast_fp16 = einsum(equation = var_10853_equation_0, values = (var_10697_cast_fp16, var_10820_cast_fp16))[name = tensor("op_10853_cast_fp16")]; + tensor var_10855_equation_0 = const()[name = tensor("op_10855_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10855_cast_fp16 = einsum(equation = var_10855_equation_0, values = (var_10701_cast_fp16, var_10821_cast_fp16))[name = tensor("op_10855_cast_fp16")]; + tensor var_10857_equation_0 = const()[name = tensor("op_10857_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10857_cast_fp16 = einsum(equation = var_10857_equation_0, values = (var_10705_cast_fp16, var_10822_cast_fp16))[name = tensor("op_10857_cast_fp16")]; + tensor var_10859_equation_0 = const()[name = tensor("op_10859_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10859_cast_fp16 = einsum(equation = var_10859_equation_0, values = (var_10709_cast_fp16, var_10823_cast_fp16))[name = tensor("op_10859_cast_fp16")]; + tensor var_10861_equation_0 = const()[name = tensor("op_10861_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10861_cast_fp16 = einsum(equation = var_10861_equation_0, values = (var_10713_cast_fp16, var_10824_cast_fp16))[name = tensor("op_10861_cast_fp16")]; + tensor var_10863_equation_0 = const()[name = tensor("op_10863_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10863_cast_fp16 = einsum(equation = var_10863_equation_0, values = (var_10717_cast_fp16, var_10825_cast_fp16))[name = tensor("op_10863_cast_fp16")]; + tensor var_10865_equation_0 = const()[name = tensor("op_10865_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10865_cast_fp16 = einsum(equation = var_10865_equation_0, values = (var_10721_cast_fp16, var_10826_cast_fp16))[name = tensor("op_10865_cast_fp16")]; + tensor var_10867_equation_0 = const()[name = tensor("op_10867_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10867_cast_fp16 = einsum(equation = var_10867_equation_0, values = (var_10725_cast_fp16, var_10827_cast_fp16))[name = tensor("op_10867_cast_fp16")]; + tensor input_197_interleave_0 = const()[name = tensor("input_197_interleave_0"), val = tensor(false)]; + tensor input_197_cast_fp16 = concat(axis = var_2624, interleave = input_197_interleave_0, values = (var_10829_cast_fp16, var_10831_cast_fp16, var_10833_cast_fp16, var_10835_cast_fp16, var_10837_cast_fp16, var_10839_cast_fp16, var_10841_cast_fp16, var_10843_cast_fp16, var_10845_cast_fp16, var_10847_cast_fp16, var_10849_cast_fp16, var_10851_cast_fp16, var_10853_cast_fp16, var_10855_cast_fp16, var_10857_cast_fp16, var_10859_cast_fp16, var_10861_cast_fp16, var_10863_cast_fp16, var_10865_cast_fp16, var_10867_cast_fp16))[name = tensor("input_197_cast_fp16")]; + tensor var_10877_pad_type_0 = const()[name = tensor("op_10877_pad_type_0"), val = tensor("valid")]; + tensor var_10877_strides_0 = const()[name = tensor("op_10877_strides_0"), val = tensor([1, 1])]; + tensor var_10877_pad_0 = const()[name = tensor("op_10877_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10877_dilations_0 = const()[name = tensor("op_10877_dilations_0"), val = tensor([1, 1])]; + tensor var_10877_groups_0 = const()[name = tensor("op_10877_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293309248))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294538112))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294538304)))]; + tensor var_10877_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_10877_dilations_0, groups = var_10877_groups_0, pad = var_10877_pad_0, pad_type = var_10877_pad_type_0, strides = var_10877_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = tensor("op_10877_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = var_10877_cast_fp16, y = inputs_75_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; + tensor input_199_axes_0 = const()[name = tensor("input_199_axes_0"), val = tensor([1])]; + tensor input_199_gamma_0_to_fp16 = const()[name = tensor("input_199_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294540928)))]; + tensor input_199_beta_0_to_fp16 = const()[name = tensor("input_199_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294543552)))]; + tensor var_10887_to_fp16 = const()[name = tensor("op_10887_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_199_cast_fp16 = layer_norm(axes = input_199_axes_0, beta = input_199_beta_0_to_fp16, epsilon = var_10887_to_fp16, gamma = input_199_gamma_0_to_fp16, x = inputs_77_cast_fp16)[name = tensor("input_199_cast_fp16")]; + tensor var_10907_pad_type_0 = const()[name = tensor("op_10907_pad_type_0"), val = tensor("valid")]; + tensor var_10907_strides_0 = const()[name = tensor("op_10907_strides_0"), val = tensor([1, 1])]; + tensor var_10907_pad_0 = const()[name = tensor("op_10907_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10907_dilations_0 = const()[name = tensor("op_10907_dilations_0"), val = tensor([1, 1])]; + tensor var_10907_groups_0 = const()[name = tensor("op_10907_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294546176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304376640))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304376832)))]; + tensor var_10907_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_10907_dilations_0, groups = var_10907_groups_0, pad = var_10907_pad_0, pad_type = var_10907_pad_type_0, strides = var_10907_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor("op_10907_cast_fp16")]; + tensor var_10908_split_sizes_0 = const()[name = tensor("op_10908_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_10908_axis_0 = const()[name = tensor("op_10908_axis_0"), val = tensor(1)]; + tensor var_10908_cast_fp16_0, tensor var_10908_cast_fp16_1 = split(axis = var_10908_axis_0, split_sizes = var_10908_split_sizes_0, x = var_10907_cast_fp16)[name = tensor("op_10908_cast_fp16")]; + tensor var_10910_mode_0 = const()[name = tensor("op_10910_mode_0"), val = tensor("EXACT")]; + tensor var_10910_cast_fp16 = gelu(mode = var_10910_mode_0, x = var_10908_cast_fp16_1)[name = tensor("op_10910_cast_fp16")]; + tensor input_201_cast_fp16 = mul(x = var_10908_cast_fp16_0, y = var_10910_cast_fp16)[name = tensor("input_201_cast_fp16")]; + tensor var_10918_pad_type_0 = const()[name = tensor("op_10918_pad_type_0"), val = tensor("valid")]; + tensor var_10918_strides_0 = const()[name = tensor("op_10918_strides_0"), val = tensor([1, 1])]; + tensor var_10918_pad_0 = const()[name = tensor("op_10918_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10918_dilations_0 = const()[name = tensor("op_10918_dilations_0"), val = tensor([1, 1])]; + tensor var_10918_groups_0 = const()[name = tensor("op_10918_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304397376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(309312640))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(309312832)))]; + tensor var_10918_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_10918_dilations_0, groups = var_10918_groups_0, pad = var_10918_pad_0, pad_type = var_10918_pad_type_0, strides = var_10918_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized, x = input_201_cast_fp16)[name = tensor("op_10918_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = var_10918_cast_fp16, y = inputs_77_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; + tensor hidden_states_119_axes_0 = const()[name = tensor("hidden_states_119_axes_0"), val = tensor([1])]; + tensor hidden_states_119_gamma_0_to_fp16 = const()[name = tensor("hidden_states_119_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(309315456)))]; + tensor hidden_states_119_beta_0_to_fp16 = const()[name = tensor("hidden_states_119_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(309318080)))]; + tensor var_10934_to_fp16 = const()[name = tensor("op_10934_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_119_cast_fp16 = layer_norm(axes = hidden_states_119_axes_0, beta = hidden_states_119_beta_0_to_fp16, epsilon = var_10934_to_fp16, gamma = hidden_states_119_gamma_0_to_fp16, x = inputs_79_cast_fp16)[name = tensor("hidden_states_119_cast_fp16")]; + tensor q_53_pad_type_0 = const()[name = tensor("q_53_pad_type_0"), val = tensor("valid")]; + tensor q_53_strides_0 = const()[name = tensor("q_53_strides_0"), val = tensor([1, 1])]; + tensor q_53_pad_0 = const()[name = tensor("q_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_53_dilations_0 = const()[name = tensor("q_53_dilations_0"), val = tensor([1, 1])]; + tensor q_53_groups_0 = const()[name = tensor("q_53_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(309320704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310549568))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_53_cast_fp16 = conv(dilations = q_53_dilations_0, groups = q_53_groups_0, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = q_53_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_119_cast_fp16)[name = tensor("q_53_cast_fp16")]; + tensor k_105_pad_type_0 = const()[name = tensor("k_105_pad_type_0"), val = tensor("valid")]; + tensor k_105_strides_0 = const()[name = tensor("k_105_strides_0"), val = tensor([1, 1])]; + tensor k_105_pad_0 = const()[name = tensor("k_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_105_dilations_0 = const()[name = tensor("k_105_dilations_0"), val = tensor([1, 1])]; + tensor k_105_groups_0 = const()[name = tensor("k_105_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310549760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311778624))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_105_cast_fp16 = conv(dilations = k_105_dilations_0, groups = k_105_groups_0, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = k_105_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_119_cast_fp16)[name = tensor("k_105_cast_fp16")]; + tensor v_53_pad_type_0 = const()[name = tensor("v_53_pad_type_0"), val = tensor("valid")]; + tensor v_53_strides_0 = const()[name = tensor("v_53_strides_0"), val = tensor([1, 1])]; + tensor v_53_pad_0 = const()[name = tensor("v_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_53_dilations_0 = const()[name = tensor("v_53_dilations_0"), val = tensor([1, 1])]; + tensor v_53_groups_0 = const()[name = tensor("v_53_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311778816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313007680))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_53_cast_fp16 = conv(dilations = v_53_dilations_0, groups = v_53_groups_0, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = v_53_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_119_cast_fp16)[name = tensor("v_53_cast_fp16")]; + tensor var_10967_begin_0 = const()[name = tensor("op_10967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10967_end_0 = const()[name = tensor("op_10967_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10967_end_mask_0 = const()[name = tensor("op_10967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10967_cast_fp16 = slice_by_index(begin = var_10967_begin_0, end = var_10967_end_0, end_mask = var_10967_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10967_cast_fp16")]; + tensor var_10971_begin_0 = const()[name = tensor("op_10971_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10971_end_0 = const()[name = tensor("op_10971_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10971_end_mask_0 = const()[name = tensor("op_10971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10971_cast_fp16 = slice_by_index(begin = var_10971_begin_0, end = var_10971_end_0, end_mask = var_10971_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10971_cast_fp16")]; + tensor var_10975_begin_0 = const()[name = tensor("op_10975_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10975_end_0 = const()[name = tensor("op_10975_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10975_end_mask_0 = const()[name = tensor("op_10975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10975_cast_fp16 = slice_by_index(begin = var_10975_begin_0, end = var_10975_end_0, end_mask = var_10975_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10975_cast_fp16")]; + tensor var_10979_begin_0 = const()[name = tensor("op_10979_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10979_end_0 = const()[name = tensor("op_10979_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10979_end_mask_0 = const()[name = tensor("op_10979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10979_cast_fp16 = slice_by_index(begin = var_10979_begin_0, end = var_10979_end_0, end_mask = var_10979_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10979_cast_fp16")]; + tensor var_10983_begin_0 = const()[name = tensor("op_10983_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10983_end_0 = const()[name = tensor("op_10983_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10983_end_mask_0 = const()[name = tensor("op_10983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10983_cast_fp16 = slice_by_index(begin = var_10983_begin_0, end = var_10983_end_0, end_mask = var_10983_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10983_cast_fp16")]; + tensor var_10987_begin_0 = const()[name = tensor("op_10987_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10987_end_0 = const()[name = tensor("op_10987_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10987_end_mask_0 = const()[name = tensor("op_10987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10987_cast_fp16 = slice_by_index(begin = var_10987_begin_0, end = var_10987_end_0, end_mask = var_10987_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10987_cast_fp16")]; + tensor var_10991_begin_0 = const()[name = tensor("op_10991_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10991_end_0 = const()[name = tensor("op_10991_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10991_end_mask_0 = const()[name = tensor("op_10991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10991_cast_fp16 = slice_by_index(begin = var_10991_begin_0, end = var_10991_end_0, end_mask = var_10991_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10991_cast_fp16")]; + tensor var_10995_begin_0 = const()[name = tensor("op_10995_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10995_end_0 = const()[name = tensor("op_10995_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10995_end_mask_0 = const()[name = tensor("op_10995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10995_cast_fp16 = slice_by_index(begin = var_10995_begin_0, end = var_10995_end_0, end_mask = var_10995_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10995_cast_fp16")]; + tensor var_10999_begin_0 = const()[name = tensor("op_10999_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10999_end_0 = const()[name = tensor("op_10999_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10999_end_mask_0 = const()[name = tensor("op_10999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10999_cast_fp16 = slice_by_index(begin = var_10999_begin_0, end = var_10999_end_0, end_mask = var_10999_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10999_cast_fp16")]; + tensor var_11003_begin_0 = const()[name = tensor("op_11003_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11003_end_0 = const()[name = tensor("op_11003_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11003_end_mask_0 = const()[name = tensor("op_11003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11003_cast_fp16 = slice_by_index(begin = var_11003_begin_0, end = var_11003_end_0, end_mask = var_11003_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11003_cast_fp16")]; + tensor var_11007_begin_0 = const()[name = tensor("op_11007_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11007_end_0 = const()[name = tensor("op_11007_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_11007_end_mask_0 = const()[name = tensor("op_11007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11007_cast_fp16 = slice_by_index(begin = var_11007_begin_0, end = var_11007_end_0, end_mask = var_11007_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11007_cast_fp16")]; + tensor var_11011_begin_0 = const()[name = tensor("op_11011_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11011_end_0 = const()[name = tensor("op_11011_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_11011_end_mask_0 = const()[name = tensor("op_11011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11011_cast_fp16 = slice_by_index(begin = var_11011_begin_0, end = var_11011_end_0, end_mask = var_11011_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11011_cast_fp16")]; + tensor var_11015_begin_0 = const()[name = tensor("op_11015_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11015_end_0 = const()[name = tensor("op_11015_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_11015_end_mask_0 = const()[name = tensor("op_11015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11015_cast_fp16 = slice_by_index(begin = var_11015_begin_0, end = var_11015_end_0, end_mask = var_11015_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11015_cast_fp16")]; + tensor var_11019_begin_0 = const()[name = tensor("op_11019_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11019_end_0 = const()[name = tensor("op_11019_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_11019_end_mask_0 = const()[name = tensor("op_11019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11019_cast_fp16 = slice_by_index(begin = var_11019_begin_0, end = var_11019_end_0, end_mask = var_11019_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11019_cast_fp16")]; + tensor var_11023_begin_0 = const()[name = tensor("op_11023_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11023_end_0 = const()[name = tensor("op_11023_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_11023_end_mask_0 = const()[name = tensor("op_11023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11023_cast_fp16 = slice_by_index(begin = var_11023_begin_0, end = var_11023_end_0, end_mask = var_11023_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11023_cast_fp16")]; + tensor var_11027_begin_0 = const()[name = tensor("op_11027_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11027_end_0 = const()[name = tensor("op_11027_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11027_end_mask_0 = const()[name = tensor("op_11027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11027_cast_fp16 = slice_by_index(begin = var_11027_begin_0, end = var_11027_end_0, end_mask = var_11027_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11027_cast_fp16")]; + tensor var_11031_begin_0 = const()[name = tensor("op_11031_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11031_end_0 = const()[name = tensor("op_11031_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_11031_end_mask_0 = const()[name = tensor("op_11031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11031_cast_fp16 = slice_by_index(begin = var_11031_begin_0, end = var_11031_end_0, end_mask = var_11031_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11031_cast_fp16")]; + tensor var_11035_begin_0 = const()[name = tensor("op_11035_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11035_end_0 = const()[name = tensor("op_11035_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_11035_end_mask_0 = const()[name = tensor("op_11035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11035_cast_fp16 = slice_by_index(begin = var_11035_begin_0, end = var_11035_end_0, end_mask = var_11035_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11035_cast_fp16")]; + tensor var_11039_begin_0 = const()[name = tensor("op_11039_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11039_end_0 = const()[name = tensor("op_11039_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_11039_end_mask_0 = const()[name = tensor("op_11039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11039_cast_fp16 = slice_by_index(begin = var_11039_begin_0, end = var_11039_end_0, end_mask = var_11039_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11039_cast_fp16")]; + tensor var_11043_begin_0 = const()[name = tensor("op_11043_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11043_end_0 = const()[name = tensor("op_11043_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_11043_end_mask_0 = const()[name = tensor("op_11043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11043_cast_fp16 = slice_by_index(begin = var_11043_begin_0, end = var_11043_end_0, end_mask = var_11043_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11043_cast_fp16")]; + tensor k_107_perm_0 = const()[name = tensor("k_107_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11050_begin_0 = const()[name = tensor("op_11050_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11050_end_0 = const()[name = tensor("op_11050_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_11050_end_mask_0 = const()[name = tensor("op_11050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_107_cast_fp16 = transpose(perm = k_107_perm_0, x = k_105_cast_fp16)[name = tensor("transpose_41")]; + tensor var_11050_cast_fp16 = slice_by_index(begin = var_11050_begin_0, end = var_11050_end_0, end_mask = var_11050_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11050_cast_fp16")]; + tensor var_11054_begin_0 = const()[name = tensor("op_11054_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11054_end_0 = const()[name = tensor("op_11054_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_11054_end_mask_0 = const()[name = tensor("op_11054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11054_cast_fp16 = slice_by_index(begin = var_11054_begin_0, end = var_11054_end_0, end_mask = var_11054_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11054_cast_fp16")]; + tensor var_11058_begin_0 = const()[name = tensor("op_11058_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11058_end_0 = const()[name = tensor("op_11058_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_11058_end_mask_0 = const()[name = tensor("op_11058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11058_cast_fp16 = slice_by_index(begin = var_11058_begin_0, end = var_11058_end_0, end_mask = var_11058_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11058_cast_fp16")]; + tensor var_11062_begin_0 = const()[name = tensor("op_11062_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11062_end_0 = const()[name = tensor("op_11062_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_11062_end_mask_0 = const()[name = tensor("op_11062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11062_cast_fp16 = slice_by_index(begin = var_11062_begin_0, end = var_11062_end_0, end_mask = var_11062_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11062_cast_fp16")]; + tensor var_11066_begin_0 = const()[name = tensor("op_11066_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11066_end_0 = const()[name = tensor("op_11066_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_11066_end_mask_0 = const()[name = tensor("op_11066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11066_cast_fp16 = slice_by_index(begin = var_11066_begin_0, end = var_11066_end_0, end_mask = var_11066_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11066_cast_fp16")]; + tensor var_11070_begin_0 = const()[name = tensor("op_11070_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11070_end_0 = const()[name = tensor("op_11070_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_11070_end_mask_0 = const()[name = tensor("op_11070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11070_cast_fp16 = slice_by_index(begin = var_11070_begin_0, end = var_11070_end_0, end_mask = var_11070_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11070_cast_fp16")]; + tensor var_11074_begin_0 = const()[name = tensor("op_11074_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11074_end_0 = const()[name = tensor("op_11074_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_11074_end_mask_0 = const()[name = tensor("op_11074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11074_cast_fp16 = slice_by_index(begin = var_11074_begin_0, end = var_11074_end_0, end_mask = var_11074_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11074_cast_fp16")]; + tensor var_11078_begin_0 = const()[name = tensor("op_11078_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11078_end_0 = const()[name = tensor("op_11078_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_11078_end_mask_0 = const()[name = tensor("op_11078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11078_cast_fp16 = slice_by_index(begin = var_11078_begin_0, end = var_11078_end_0, end_mask = var_11078_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11078_cast_fp16")]; + tensor var_11082_begin_0 = const()[name = tensor("op_11082_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11082_end_0 = const()[name = tensor("op_11082_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_11082_end_mask_0 = const()[name = tensor("op_11082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11082_cast_fp16 = slice_by_index(begin = var_11082_begin_0, end = var_11082_end_0, end_mask = var_11082_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11082_cast_fp16")]; + tensor var_11086_begin_0 = const()[name = tensor("op_11086_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11086_end_0 = const()[name = tensor("op_11086_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_11086_end_mask_0 = const()[name = tensor("op_11086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11086_cast_fp16 = slice_by_index(begin = var_11086_begin_0, end = var_11086_end_0, end_mask = var_11086_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11086_cast_fp16")]; + tensor var_11090_begin_0 = const()[name = tensor("op_11090_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11090_end_0 = const()[name = tensor("op_11090_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_11090_end_mask_0 = const()[name = tensor("op_11090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11090_cast_fp16 = slice_by_index(begin = var_11090_begin_0, end = var_11090_end_0, end_mask = var_11090_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11090_cast_fp16")]; + tensor var_11094_begin_0 = const()[name = tensor("op_11094_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11094_end_0 = const()[name = tensor("op_11094_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_11094_end_mask_0 = const()[name = tensor("op_11094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11094_cast_fp16 = slice_by_index(begin = var_11094_begin_0, end = var_11094_end_0, end_mask = var_11094_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11094_cast_fp16")]; + tensor var_11098_begin_0 = const()[name = tensor("op_11098_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_11098_end_0 = const()[name = tensor("op_11098_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_11098_end_mask_0 = const()[name = tensor("op_11098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11098_cast_fp16 = slice_by_index(begin = var_11098_begin_0, end = var_11098_end_0, end_mask = var_11098_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11098_cast_fp16")]; + tensor var_11102_begin_0 = const()[name = tensor("op_11102_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_11102_end_0 = const()[name = tensor("op_11102_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_11102_end_mask_0 = const()[name = tensor("op_11102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11102_cast_fp16 = slice_by_index(begin = var_11102_begin_0, end = var_11102_end_0, end_mask = var_11102_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11102_cast_fp16")]; + tensor var_11106_begin_0 = const()[name = tensor("op_11106_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_11106_end_0 = const()[name = tensor("op_11106_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_11106_end_mask_0 = const()[name = tensor("op_11106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11106_cast_fp16 = slice_by_index(begin = var_11106_begin_0, end = var_11106_end_0, end_mask = var_11106_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11106_cast_fp16")]; + tensor var_11110_begin_0 = const()[name = tensor("op_11110_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_11110_end_0 = const()[name = tensor("op_11110_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11110_end_mask_0 = const()[name = tensor("op_11110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11110_cast_fp16 = slice_by_index(begin = var_11110_begin_0, end = var_11110_end_0, end_mask = var_11110_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11110_cast_fp16")]; + tensor var_11114_begin_0 = const()[name = tensor("op_11114_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_11114_end_0 = const()[name = tensor("op_11114_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_11114_end_mask_0 = const()[name = tensor("op_11114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11114_cast_fp16 = slice_by_index(begin = var_11114_begin_0, end = var_11114_end_0, end_mask = var_11114_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11114_cast_fp16")]; + tensor var_11118_begin_0 = const()[name = tensor("op_11118_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_11118_end_0 = const()[name = tensor("op_11118_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_11118_end_mask_0 = const()[name = tensor("op_11118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11118_cast_fp16 = slice_by_index(begin = var_11118_begin_0, end = var_11118_end_0, end_mask = var_11118_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11118_cast_fp16")]; + tensor var_11122_begin_0 = const()[name = tensor("op_11122_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_11122_end_0 = const()[name = tensor("op_11122_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_11122_end_mask_0 = const()[name = tensor("op_11122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11122_cast_fp16 = slice_by_index(begin = var_11122_begin_0, end = var_11122_end_0, end_mask = var_11122_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11122_cast_fp16")]; + tensor var_11126_begin_0 = const()[name = tensor("op_11126_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_11126_end_0 = const()[name = tensor("op_11126_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_11126_end_mask_0 = const()[name = tensor("op_11126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11126_cast_fp16 = slice_by_index(begin = var_11126_begin_0, end = var_11126_end_0, end_mask = var_11126_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11126_cast_fp16")]; + tensor var_11128_begin_0 = const()[name = tensor("op_11128_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11128_end_0 = const()[name = tensor("op_11128_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11128_end_mask_0 = const()[name = tensor("op_11128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11128_cast_fp16 = slice_by_index(begin = var_11128_begin_0, end = var_11128_end_0, end_mask = var_11128_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11128_cast_fp16")]; + tensor var_11132_begin_0 = const()[name = tensor("op_11132_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11132_end_0 = const()[name = tensor("op_11132_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11132_end_mask_0 = const()[name = tensor("op_11132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11132_cast_fp16 = slice_by_index(begin = var_11132_begin_0, end = var_11132_end_0, end_mask = var_11132_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11132_cast_fp16")]; + tensor var_11136_begin_0 = const()[name = tensor("op_11136_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11136_end_0 = const()[name = tensor("op_11136_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11136_end_mask_0 = const()[name = tensor("op_11136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11136_cast_fp16 = slice_by_index(begin = var_11136_begin_0, end = var_11136_end_0, end_mask = var_11136_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11136_cast_fp16")]; + tensor var_11140_begin_0 = const()[name = tensor("op_11140_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11140_end_0 = const()[name = tensor("op_11140_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11140_end_mask_0 = const()[name = tensor("op_11140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11140_cast_fp16 = slice_by_index(begin = var_11140_begin_0, end = var_11140_end_0, end_mask = var_11140_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11140_cast_fp16")]; + tensor var_11144_begin_0 = const()[name = tensor("op_11144_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11144_end_0 = const()[name = tensor("op_11144_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11144_end_mask_0 = const()[name = tensor("op_11144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11144_cast_fp16 = slice_by_index(begin = var_11144_begin_0, end = var_11144_end_0, end_mask = var_11144_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11144_cast_fp16")]; + tensor var_11148_begin_0 = const()[name = tensor("op_11148_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11148_end_0 = const()[name = tensor("op_11148_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11148_end_mask_0 = const()[name = tensor("op_11148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11148_cast_fp16 = slice_by_index(begin = var_11148_begin_0, end = var_11148_end_0, end_mask = var_11148_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11148_cast_fp16")]; + tensor var_11152_begin_0 = const()[name = tensor("op_11152_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11152_end_0 = const()[name = tensor("op_11152_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11152_end_mask_0 = const()[name = tensor("op_11152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11152_cast_fp16 = slice_by_index(begin = var_11152_begin_0, end = var_11152_end_0, end_mask = var_11152_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11152_cast_fp16")]; + tensor var_11156_begin_0 = const()[name = tensor("op_11156_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11156_end_0 = const()[name = tensor("op_11156_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11156_end_mask_0 = const()[name = tensor("op_11156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11156_cast_fp16 = slice_by_index(begin = var_11156_begin_0, end = var_11156_end_0, end_mask = var_11156_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11156_cast_fp16")]; + tensor var_11160_begin_0 = const()[name = tensor("op_11160_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11160_end_0 = const()[name = tensor("op_11160_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_11160_end_mask_0 = const()[name = tensor("op_11160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11160_cast_fp16 = slice_by_index(begin = var_11160_begin_0, end = var_11160_end_0, end_mask = var_11160_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11160_cast_fp16")]; + tensor var_11164_begin_0 = const()[name = tensor("op_11164_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11164_end_0 = const()[name = tensor("op_11164_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11164_end_mask_0 = const()[name = tensor("op_11164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11164_cast_fp16 = slice_by_index(begin = var_11164_begin_0, end = var_11164_end_0, end_mask = var_11164_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11164_cast_fp16")]; + tensor var_11168_begin_0 = const()[name = tensor("op_11168_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11168_end_0 = const()[name = tensor("op_11168_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_11168_end_mask_0 = const()[name = tensor("op_11168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11168_cast_fp16 = slice_by_index(begin = var_11168_begin_0, end = var_11168_end_0, end_mask = var_11168_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11168_cast_fp16")]; + tensor var_11172_begin_0 = const()[name = tensor("op_11172_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11172_end_0 = const()[name = tensor("op_11172_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_11172_end_mask_0 = const()[name = tensor("op_11172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11172_cast_fp16 = slice_by_index(begin = var_11172_begin_0, end = var_11172_end_0, end_mask = var_11172_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11172_cast_fp16")]; + tensor var_11176_begin_0 = const()[name = tensor("op_11176_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11176_end_0 = const()[name = tensor("op_11176_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_11176_end_mask_0 = const()[name = tensor("op_11176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11176_cast_fp16 = slice_by_index(begin = var_11176_begin_0, end = var_11176_end_0, end_mask = var_11176_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11176_cast_fp16")]; + tensor var_11180_begin_0 = const()[name = tensor("op_11180_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11180_end_0 = const()[name = tensor("op_11180_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_11180_end_mask_0 = const()[name = tensor("op_11180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11180_cast_fp16 = slice_by_index(begin = var_11180_begin_0, end = var_11180_end_0, end_mask = var_11180_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11180_cast_fp16")]; + tensor var_11184_begin_0 = const()[name = tensor("op_11184_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11184_end_0 = const()[name = tensor("op_11184_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_11184_end_mask_0 = const()[name = tensor("op_11184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11184_cast_fp16 = slice_by_index(begin = var_11184_begin_0, end = var_11184_end_0, end_mask = var_11184_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11184_cast_fp16")]; + tensor var_11188_begin_0 = const()[name = tensor("op_11188_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11188_end_0 = const()[name = tensor("op_11188_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11188_end_mask_0 = const()[name = tensor("op_11188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11188_cast_fp16 = slice_by_index(begin = var_11188_begin_0, end = var_11188_end_0, end_mask = var_11188_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11188_cast_fp16")]; + tensor var_11192_begin_0 = const()[name = tensor("op_11192_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11192_end_0 = const()[name = tensor("op_11192_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_11192_end_mask_0 = const()[name = tensor("op_11192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11192_cast_fp16 = slice_by_index(begin = var_11192_begin_0, end = var_11192_end_0, end_mask = var_11192_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11192_cast_fp16")]; + tensor var_11196_begin_0 = const()[name = tensor("op_11196_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11196_end_0 = const()[name = tensor("op_11196_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_11196_end_mask_0 = const()[name = tensor("op_11196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11196_cast_fp16 = slice_by_index(begin = var_11196_begin_0, end = var_11196_end_0, end_mask = var_11196_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11196_cast_fp16")]; + tensor var_11200_begin_0 = const()[name = tensor("op_11200_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11200_end_0 = const()[name = tensor("op_11200_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_11200_end_mask_0 = const()[name = tensor("op_11200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11200_cast_fp16 = slice_by_index(begin = var_11200_begin_0, end = var_11200_end_0, end_mask = var_11200_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11200_cast_fp16")]; + tensor var_11204_begin_0 = const()[name = tensor("op_11204_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11204_end_0 = const()[name = tensor("op_11204_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_11204_end_mask_0 = const()[name = tensor("op_11204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11204_cast_fp16 = slice_by_index(begin = var_11204_begin_0, end = var_11204_end_0, end_mask = var_11204_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11204_cast_fp16")]; + tensor var_11208_equation_0 = const()[name = tensor("op_11208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11208_cast_fp16 = einsum(equation = var_11208_equation_0, values = (var_11050_cast_fp16, var_10967_cast_fp16))[name = tensor("op_11208_cast_fp16")]; + tensor var_11209_to_fp16 = const()[name = tensor("op_11209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_881_cast_fp16 = mul(x = var_11208_cast_fp16, y = var_11209_to_fp16)[name = tensor("aw_881_cast_fp16")]; + tensor var_11212_equation_0 = const()[name = tensor("op_11212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11212_cast_fp16 = einsum(equation = var_11212_equation_0, values = (var_11054_cast_fp16, var_10971_cast_fp16))[name = tensor("op_11212_cast_fp16")]; + tensor var_11213_to_fp16 = const()[name = tensor("op_11213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_883_cast_fp16 = mul(x = var_11212_cast_fp16, y = var_11213_to_fp16)[name = tensor("aw_883_cast_fp16")]; + tensor var_11216_equation_0 = const()[name = tensor("op_11216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11216_cast_fp16 = einsum(equation = var_11216_equation_0, values = (var_11058_cast_fp16, var_10975_cast_fp16))[name = tensor("op_11216_cast_fp16")]; + tensor var_11217_to_fp16 = const()[name = tensor("op_11217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_885_cast_fp16 = mul(x = var_11216_cast_fp16, y = var_11217_to_fp16)[name = tensor("aw_885_cast_fp16")]; + tensor var_11220_equation_0 = const()[name = tensor("op_11220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11220_cast_fp16 = einsum(equation = var_11220_equation_0, values = (var_11062_cast_fp16, var_10979_cast_fp16))[name = tensor("op_11220_cast_fp16")]; + tensor var_11221_to_fp16 = const()[name = tensor("op_11221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_887_cast_fp16 = mul(x = var_11220_cast_fp16, y = var_11221_to_fp16)[name = tensor("aw_887_cast_fp16")]; + tensor var_11224_equation_0 = const()[name = tensor("op_11224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11224_cast_fp16 = einsum(equation = var_11224_equation_0, values = (var_11066_cast_fp16, var_10983_cast_fp16))[name = tensor("op_11224_cast_fp16")]; + tensor var_11225_to_fp16 = const()[name = tensor("op_11225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_889_cast_fp16 = mul(x = var_11224_cast_fp16, y = var_11225_to_fp16)[name = tensor("aw_889_cast_fp16")]; + tensor var_11228_equation_0 = const()[name = tensor("op_11228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11228_cast_fp16 = einsum(equation = var_11228_equation_0, values = (var_11070_cast_fp16, var_10987_cast_fp16))[name = tensor("op_11228_cast_fp16")]; + tensor var_11229_to_fp16 = const()[name = tensor("op_11229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_891_cast_fp16 = mul(x = var_11228_cast_fp16, y = var_11229_to_fp16)[name = tensor("aw_891_cast_fp16")]; + tensor var_11232_equation_0 = const()[name = tensor("op_11232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11232_cast_fp16 = einsum(equation = var_11232_equation_0, values = (var_11074_cast_fp16, var_10991_cast_fp16))[name = tensor("op_11232_cast_fp16")]; + tensor var_11233_to_fp16 = const()[name = tensor("op_11233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_893_cast_fp16 = mul(x = var_11232_cast_fp16, y = var_11233_to_fp16)[name = tensor("aw_893_cast_fp16")]; + tensor var_11236_equation_0 = const()[name = tensor("op_11236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11236_cast_fp16 = einsum(equation = var_11236_equation_0, values = (var_11078_cast_fp16, var_10995_cast_fp16))[name = tensor("op_11236_cast_fp16")]; + tensor var_11237_to_fp16 = const()[name = tensor("op_11237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_895_cast_fp16 = mul(x = var_11236_cast_fp16, y = var_11237_to_fp16)[name = tensor("aw_895_cast_fp16")]; + tensor var_11240_equation_0 = const()[name = tensor("op_11240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11240_cast_fp16 = einsum(equation = var_11240_equation_0, values = (var_11082_cast_fp16, var_10999_cast_fp16))[name = tensor("op_11240_cast_fp16")]; + tensor var_11241_to_fp16 = const()[name = tensor("op_11241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_897_cast_fp16 = mul(x = var_11240_cast_fp16, y = var_11241_to_fp16)[name = tensor("aw_897_cast_fp16")]; + tensor var_11244_equation_0 = const()[name = tensor("op_11244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11244_cast_fp16 = einsum(equation = var_11244_equation_0, values = (var_11086_cast_fp16, var_11003_cast_fp16))[name = tensor("op_11244_cast_fp16")]; + tensor var_11245_to_fp16 = const()[name = tensor("op_11245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_899_cast_fp16 = mul(x = var_11244_cast_fp16, y = var_11245_to_fp16)[name = tensor("aw_899_cast_fp16")]; + tensor var_11248_equation_0 = const()[name = tensor("op_11248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11248_cast_fp16 = einsum(equation = var_11248_equation_0, values = (var_11090_cast_fp16, var_11007_cast_fp16))[name = tensor("op_11248_cast_fp16")]; + tensor var_11249_to_fp16 = const()[name = tensor("op_11249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_901_cast_fp16 = mul(x = var_11248_cast_fp16, y = var_11249_to_fp16)[name = tensor("aw_901_cast_fp16")]; + tensor var_11252_equation_0 = const()[name = tensor("op_11252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11252_cast_fp16 = einsum(equation = var_11252_equation_0, values = (var_11094_cast_fp16, var_11011_cast_fp16))[name = tensor("op_11252_cast_fp16")]; + tensor var_11253_to_fp16 = const()[name = tensor("op_11253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_903_cast_fp16 = mul(x = var_11252_cast_fp16, y = var_11253_to_fp16)[name = tensor("aw_903_cast_fp16")]; + tensor var_11256_equation_0 = const()[name = tensor("op_11256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11256_cast_fp16 = einsum(equation = var_11256_equation_0, values = (var_11098_cast_fp16, var_11015_cast_fp16))[name = tensor("op_11256_cast_fp16")]; + tensor var_11257_to_fp16 = const()[name = tensor("op_11257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_905_cast_fp16 = mul(x = var_11256_cast_fp16, y = var_11257_to_fp16)[name = tensor("aw_905_cast_fp16")]; + tensor var_11260_equation_0 = const()[name = tensor("op_11260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11260_cast_fp16 = einsum(equation = var_11260_equation_0, values = (var_11102_cast_fp16, var_11019_cast_fp16))[name = tensor("op_11260_cast_fp16")]; + tensor var_11261_to_fp16 = const()[name = tensor("op_11261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_907_cast_fp16 = mul(x = var_11260_cast_fp16, y = var_11261_to_fp16)[name = tensor("aw_907_cast_fp16")]; + tensor var_11264_equation_0 = const()[name = tensor("op_11264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11264_cast_fp16 = einsum(equation = var_11264_equation_0, values = (var_11106_cast_fp16, var_11023_cast_fp16))[name = tensor("op_11264_cast_fp16")]; + tensor var_11265_to_fp16 = const()[name = tensor("op_11265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_909_cast_fp16 = mul(x = var_11264_cast_fp16, y = var_11265_to_fp16)[name = tensor("aw_909_cast_fp16")]; + tensor var_11268_equation_0 = const()[name = tensor("op_11268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11268_cast_fp16 = einsum(equation = var_11268_equation_0, values = (var_11110_cast_fp16, var_11027_cast_fp16))[name = tensor("op_11268_cast_fp16")]; + tensor var_11269_to_fp16 = const()[name = tensor("op_11269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_911_cast_fp16 = mul(x = var_11268_cast_fp16, y = var_11269_to_fp16)[name = tensor("aw_911_cast_fp16")]; + tensor var_11272_equation_0 = const()[name = tensor("op_11272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11272_cast_fp16 = einsum(equation = var_11272_equation_0, values = (var_11114_cast_fp16, var_11031_cast_fp16))[name = tensor("op_11272_cast_fp16")]; + tensor var_11273_to_fp16 = const()[name = tensor("op_11273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_913_cast_fp16 = mul(x = var_11272_cast_fp16, y = var_11273_to_fp16)[name = tensor("aw_913_cast_fp16")]; + tensor var_11276_equation_0 = const()[name = tensor("op_11276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11276_cast_fp16 = einsum(equation = var_11276_equation_0, values = (var_11118_cast_fp16, var_11035_cast_fp16))[name = tensor("op_11276_cast_fp16")]; + tensor var_11277_to_fp16 = const()[name = tensor("op_11277_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_915_cast_fp16 = mul(x = var_11276_cast_fp16, y = var_11277_to_fp16)[name = tensor("aw_915_cast_fp16")]; + tensor var_11280_equation_0 = const()[name = tensor("op_11280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11280_cast_fp16 = einsum(equation = var_11280_equation_0, values = (var_11122_cast_fp16, var_11039_cast_fp16))[name = tensor("op_11280_cast_fp16")]; + tensor var_11281_to_fp16 = const()[name = tensor("op_11281_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_917_cast_fp16 = mul(x = var_11280_cast_fp16, y = var_11281_to_fp16)[name = tensor("aw_917_cast_fp16")]; + tensor var_11284_equation_0 = const()[name = tensor("op_11284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11284_cast_fp16 = einsum(equation = var_11284_equation_0, values = (var_11126_cast_fp16, var_11043_cast_fp16))[name = tensor("op_11284_cast_fp16")]; + tensor var_11285_to_fp16 = const()[name = tensor("op_11285_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_919_cast_fp16 = mul(x = var_11284_cast_fp16, y = var_11285_to_fp16)[name = tensor("aw_919_cast_fp16")]; + tensor var_11287_cast_fp16 = softmax(axis = var_2624, x = aw_881_cast_fp16)[name = tensor("op_11287_cast_fp16")]; + tensor var_11288_cast_fp16 = softmax(axis = var_2624, x = aw_883_cast_fp16)[name = tensor("op_11288_cast_fp16")]; + tensor var_11289_cast_fp16 = softmax(axis = var_2624, x = aw_885_cast_fp16)[name = tensor("op_11289_cast_fp16")]; + tensor var_11290_cast_fp16 = softmax(axis = var_2624, x = aw_887_cast_fp16)[name = tensor("op_11290_cast_fp16")]; + tensor var_11291_cast_fp16 = softmax(axis = var_2624, x = aw_889_cast_fp16)[name = tensor("op_11291_cast_fp16")]; + tensor var_11292_cast_fp16 = softmax(axis = var_2624, x = aw_891_cast_fp16)[name = tensor("op_11292_cast_fp16")]; + tensor var_11293_cast_fp16 = softmax(axis = var_2624, x = aw_893_cast_fp16)[name = tensor("op_11293_cast_fp16")]; + tensor var_11294_cast_fp16 = softmax(axis = var_2624, x = aw_895_cast_fp16)[name = tensor("op_11294_cast_fp16")]; + tensor var_11295_cast_fp16 = softmax(axis = var_2624, x = aw_897_cast_fp16)[name = tensor("op_11295_cast_fp16")]; + tensor var_11296_cast_fp16 = softmax(axis = var_2624, x = aw_899_cast_fp16)[name = tensor("op_11296_cast_fp16")]; + tensor var_11297_cast_fp16 = softmax(axis = var_2624, x = aw_901_cast_fp16)[name = tensor("op_11297_cast_fp16")]; + tensor var_11298_cast_fp16 = softmax(axis = var_2624, x = aw_903_cast_fp16)[name = tensor("op_11298_cast_fp16")]; + tensor var_11299_cast_fp16 = softmax(axis = var_2624, x = aw_905_cast_fp16)[name = tensor("op_11299_cast_fp16")]; + tensor var_11300_cast_fp16 = softmax(axis = var_2624, x = aw_907_cast_fp16)[name = tensor("op_11300_cast_fp16")]; + tensor var_11301_cast_fp16 = softmax(axis = var_2624, x = aw_909_cast_fp16)[name = tensor("op_11301_cast_fp16")]; + tensor var_11302_cast_fp16 = softmax(axis = var_2624, x = aw_911_cast_fp16)[name = tensor("op_11302_cast_fp16")]; + tensor var_11303_cast_fp16 = softmax(axis = var_2624, x = aw_913_cast_fp16)[name = tensor("op_11303_cast_fp16")]; + tensor var_11304_cast_fp16 = softmax(axis = var_2624, x = aw_915_cast_fp16)[name = tensor("op_11304_cast_fp16")]; + tensor var_11305_cast_fp16 = softmax(axis = var_2624, x = aw_917_cast_fp16)[name = tensor("op_11305_cast_fp16")]; + tensor var_11306_cast_fp16 = softmax(axis = var_2624, x = aw_919_cast_fp16)[name = tensor("op_11306_cast_fp16")]; + tensor var_11308_equation_0 = const()[name = tensor("op_11308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11308_cast_fp16 = einsum(equation = var_11308_equation_0, values = (var_11128_cast_fp16, var_11287_cast_fp16))[name = tensor("op_11308_cast_fp16")]; + tensor var_11310_equation_0 = const()[name = tensor("op_11310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11310_cast_fp16 = einsum(equation = var_11310_equation_0, values = (var_11132_cast_fp16, var_11288_cast_fp16))[name = tensor("op_11310_cast_fp16")]; + tensor var_11312_equation_0 = const()[name = tensor("op_11312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11312_cast_fp16 = einsum(equation = var_11312_equation_0, values = (var_11136_cast_fp16, var_11289_cast_fp16))[name = tensor("op_11312_cast_fp16")]; + tensor var_11314_equation_0 = const()[name = tensor("op_11314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11314_cast_fp16 = einsum(equation = var_11314_equation_0, values = (var_11140_cast_fp16, var_11290_cast_fp16))[name = tensor("op_11314_cast_fp16")]; + tensor var_11316_equation_0 = const()[name = tensor("op_11316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11316_cast_fp16 = einsum(equation = var_11316_equation_0, values = (var_11144_cast_fp16, var_11291_cast_fp16))[name = tensor("op_11316_cast_fp16")]; + tensor var_11318_equation_0 = const()[name = tensor("op_11318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11318_cast_fp16 = einsum(equation = var_11318_equation_0, values = (var_11148_cast_fp16, var_11292_cast_fp16))[name = tensor("op_11318_cast_fp16")]; + tensor var_11320_equation_0 = const()[name = tensor("op_11320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11320_cast_fp16 = einsum(equation = var_11320_equation_0, values = (var_11152_cast_fp16, var_11293_cast_fp16))[name = tensor("op_11320_cast_fp16")]; + tensor var_11322_equation_0 = const()[name = tensor("op_11322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11322_cast_fp16 = einsum(equation = var_11322_equation_0, values = (var_11156_cast_fp16, var_11294_cast_fp16))[name = tensor("op_11322_cast_fp16")]; + tensor var_11324_equation_0 = const()[name = tensor("op_11324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11324_cast_fp16 = einsum(equation = var_11324_equation_0, values = (var_11160_cast_fp16, var_11295_cast_fp16))[name = tensor("op_11324_cast_fp16")]; + tensor var_11326_equation_0 = const()[name = tensor("op_11326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11326_cast_fp16 = einsum(equation = var_11326_equation_0, values = (var_11164_cast_fp16, var_11296_cast_fp16))[name = tensor("op_11326_cast_fp16")]; + tensor var_11328_equation_0 = const()[name = tensor("op_11328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11328_cast_fp16 = einsum(equation = var_11328_equation_0, values = (var_11168_cast_fp16, var_11297_cast_fp16))[name = tensor("op_11328_cast_fp16")]; + tensor var_11330_equation_0 = const()[name = tensor("op_11330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11330_cast_fp16 = einsum(equation = var_11330_equation_0, values = (var_11172_cast_fp16, var_11298_cast_fp16))[name = tensor("op_11330_cast_fp16")]; + tensor var_11332_equation_0 = const()[name = tensor("op_11332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11332_cast_fp16 = einsum(equation = var_11332_equation_0, values = (var_11176_cast_fp16, var_11299_cast_fp16))[name = tensor("op_11332_cast_fp16")]; + tensor var_11334_equation_0 = const()[name = tensor("op_11334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11334_cast_fp16 = einsum(equation = var_11334_equation_0, values = (var_11180_cast_fp16, var_11300_cast_fp16))[name = tensor("op_11334_cast_fp16")]; + tensor var_11336_equation_0 = const()[name = tensor("op_11336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11336_cast_fp16 = einsum(equation = var_11336_equation_0, values = (var_11184_cast_fp16, var_11301_cast_fp16))[name = tensor("op_11336_cast_fp16")]; + tensor var_11338_equation_0 = const()[name = tensor("op_11338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11338_cast_fp16 = einsum(equation = var_11338_equation_0, values = (var_11188_cast_fp16, var_11302_cast_fp16))[name = tensor("op_11338_cast_fp16")]; + tensor var_11340_equation_0 = const()[name = tensor("op_11340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11340_cast_fp16 = einsum(equation = var_11340_equation_0, values = (var_11192_cast_fp16, var_11303_cast_fp16))[name = tensor("op_11340_cast_fp16")]; + tensor var_11342_equation_0 = const()[name = tensor("op_11342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11342_cast_fp16 = einsum(equation = var_11342_equation_0, values = (var_11196_cast_fp16, var_11304_cast_fp16))[name = tensor("op_11342_cast_fp16")]; + tensor var_11344_equation_0 = const()[name = tensor("op_11344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11344_cast_fp16 = einsum(equation = var_11344_equation_0, values = (var_11200_cast_fp16, var_11305_cast_fp16))[name = tensor("op_11344_cast_fp16")]; + tensor var_11346_equation_0 = const()[name = tensor("op_11346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11346_cast_fp16 = einsum(equation = var_11346_equation_0, values = (var_11204_cast_fp16, var_11306_cast_fp16))[name = tensor("op_11346_cast_fp16")]; + tensor input_203_interleave_0 = const()[name = tensor("input_203_interleave_0"), val = tensor(false)]; + tensor input_203_cast_fp16 = concat(axis = var_2624, interleave = input_203_interleave_0, values = (var_11308_cast_fp16, var_11310_cast_fp16, var_11312_cast_fp16, var_11314_cast_fp16, var_11316_cast_fp16, var_11318_cast_fp16, var_11320_cast_fp16, var_11322_cast_fp16, var_11324_cast_fp16, var_11326_cast_fp16, var_11328_cast_fp16, var_11330_cast_fp16, var_11332_cast_fp16, var_11334_cast_fp16, var_11336_cast_fp16, var_11338_cast_fp16, var_11340_cast_fp16, var_11342_cast_fp16, var_11344_cast_fp16, var_11346_cast_fp16))[name = tensor("input_203_cast_fp16")]; + tensor var_11356_pad_type_0 = const()[name = tensor("op_11356_pad_type_0"), val = tensor("valid")]; + tensor var_11356_strides_0 = const()[name = tensor("op_11356_strides_0"), val = tensor([1, 1])]; + tensor var_11356_pad_0 = const()[name = tensor("op_11356_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11356_dilations_0 = const()[name = tensor("op_11356_dilations_0"), val = tensor([1, 1])]; + tensor var_11356_groups_0 = const()[name = tensor("op_11356_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313007872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314236736))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314236928)))]; + tensor var_11356_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_11356_dilations_0, groups = var_11356_groups_0, pad = var_11356_pad_0, pad_type = var_11356_pad_type_0, strides = var_11356_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = tensor("op_11356_cast_fp16")]; + tensor inputs_81_cast_fp16 = add(x = var_11356_cast_fp16, y = inputs_79_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; + tensor hidden_states_121_axes_0 = const()[name = tensor("hidden_states_121_axes_0"), val = tensor([1])]; + tensor hidden_states_121_gamma_0_to_fp16 = const()[name = tensor("hidden_states_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314239552)))]; + tensor hidden_states_121_beta_0_to_fp16 = const()[name = tensor("hidden_states_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314242176)))]; + tensor var_11366_to_fp16 = const()[name = tensor("op_11366_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = hidden_states_121_beta_0_to_fp16, epsilon = var_11366_to_fp16, gamma = hidden_states_121_gamma_0_to_fp16, x = inputs_81_cast_fp16)[name = tensor("hidden_states_121_cast_fp16")]; + tensor q_55_pad_type_0 = const()[name = tensor("q_55_pad_type_0"), val = tensor("valid")]; + tensor q_55_strides_0 = const()[name = tensor("q_55_strides_0"), val = tensor([1, 1])]; + tensor q_55_pad_0 = const()[name = tensor("q_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_55_dilations_0 = const()[name = tensor("q_55_dilations_0"), val = tensor([1, 1])]; + tensor q_55_groups_0 = const()[name = tensor("q_55_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314244800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315473664))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_55_cast_fp16 = conv(dilations = q_55_dilations_0, groups = q_55_groups_0, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = q_55_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_121_cast_fp16)[name = tensor("q_55_cast_fp16")]; + tensor k_109_pad_type_0 = const()[name = tensor("k_109_pad_type_0"), val = tensor("valid")]; + tensor k_109_strides_0 = const()[name = tensor("k_109_strides_0"), val = tensor([1, 1])]; + tensor k_109_pad_0 = const()[name = tensor("k_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_109_dilations_0 = const()[name = tensor("k_109_dilations_0"), val = tensor([1, 1])]; + tensor k_109_groups_0 = const()[name = tensor("k_109_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315473856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(317440000))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_109_cast_fp16 = conv(dilations = k_109_dilations_0, groups = k_109_groups_0, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = k_109_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_109_cast_fp16")]; + tensor v_55_pad_type_0 = const()[name = tensor("v_55_pad_type_0"), val = tensor("valid")]; + tensor v_55_strides_0 = const()[name = tensor("v_55_strides_0"), val = tensor([1, 1])]; + tensor v_55_pad_0 = const()[name = tensor("v_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_55_dilations_0 = const()[name = tensor("v_55_dilations_0"), val = tensor([1, 1])]; + tensor v_55_groups_0 = const()[name = tensor("v_55_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(317440192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319406336))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_55_cast_fp16 = conv(dilations = v_55_dilations_0, groups = v_55_groups_0, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = v_55_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_55_cast_fp16")]; + tensor var_11399_begin_0 = const()[name = tensor("op_11399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11399_end_0 = const()[name = tensor("op_11399_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11399_end_mask_0 = const()[name = tensor("op_11399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11399_cast_fp16 = slice_by_index(begin = var_11399_begin_0, end = var_11399_end_0, end_mask = var_11399_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11399_cast_fp16")]; + tensor var_11403_begin_0 = const()[name = tensor("op_11403_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11403_end_0 = const()[name = tensor("op_11403_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11403_end_mask_0 = const()[name = tensor("op_11403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11403_cast_fp16 = slice_by_index(begin = var_11403_begin_0, end = var_11403_end_0, end_mask = var_11403_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11403_cast_fp16")]; + tensor var_11407_begin_0 = const()[name = tensor("op_11407_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11407_end_0 = const()[name = tensor("op_11407_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11407_end_mask_0 = const()[name = tensor("op_11407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11407_cast_fp16 = slice_by_index(begin = var_11407_begin_0, end = var_11407_end_0, end_mask = var_11407_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11407_cast_fp16")]; + tensor var_11411_begin_0 = const()[name = tensor("op_11411_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11411_end_0 = const()[name = tensor("op_11411_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11411_end_mask_0 = const()[name = tensor("op_11411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11411_cast_fp16 = slice_by_index(begin = var_11411_begin_0, end = var_11411_end_0, end_mask = var_11411_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11411_cast_fp16")]; + tensor var_11415_begin_0 = const()[name = tensor("op_11415_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11415_end_0 = const()[name = tensor("op_11415_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11415_end_mask_0 = const()[name = tensor("op_11415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11415_cast_fp16 = slice_by_index(begin = var_11415_begin_0, end = var_11415_end_0, end_mask = var_11415_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11415_cast_fp16")]; + tensor var_11419_begin_0 = const()[name = tensor("op_11419_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11419_end_0 = const()[name = tensor("op_11419_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11419_end_mask_0 = const()[name = tensor("op_11419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11419_cast_fp16 = slice_by_index(begin = var_11419_begin_0, end = var_11419_end_0, end_mask = var_11419_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11419_cast_fp16")]; + tensor var_11423_begin_0 = const()[name = tensor("op_11423_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11423_end_0 = const()[name = tensor("op_11423_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11423_end_mask_0 = const()[name = tensor("op_11423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11423_cast_fp16 = slice_by_index(begin = var_11423_begin_0, end = var_11423_end_0, end_mask = var_11423_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11423_cast_fp16")]; + tensor var_11427_begin_0 = const()[name = tensor("op_11427_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11427_end_0 = const()[name = tensor("op_11427_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11427_end_mask_0 = const()[name = tensor("op_11427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11427_cast_fp16 = slice_by_index(begin = var_11427_begin_0, end = var_11427_end_0, end_mask = var_11427_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11427_cast_fp16")]; + tensor var_11431_begin_0 = const()[name = tensor("op_11431_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11431_end_0 = const()[name = tensor("op_11431_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_11431_end_mask_0 = const()[name = tensor("op_11431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11431_cast_fp16 = slice_by_index(begin = var_11431_begin_0, end = var_11431_end_0, end_mask = var_11431_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11431_cast_fp16")]; + tensor var_11435_begin_0 = const()[name = tensor("op_11435_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11435_end_0 = const()[name = tensor("op_11435_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11435_end_mask_0 = const()[name = tensor("op_11435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11435_cast_fp16 = slice_by_index(begin = var_11435_begin_0, end = var_11435_end_0, end_mask = var_11435_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11435_cast_fp16")]; + tensor var_11439_begin_0 = const()[name = tensor("op_11439_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11439_end_0 = const()[name = tensor("op_11439_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_11439_end_mask_0 = const()[name = tensor("op_11439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11439_cast_fp16 = slice_by_index(begin = var_11439_begin_0, end = var_11439_end_0, end_mask = var_11439_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11439_cast_fp16")]; + tensor var_11443_begin_0 = const()[name = tensor("op_11443_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11443_end_0 = const()[name = tensor("op_11443_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_11443_end_mask_0 = const()[name = tensor("op_11443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11443_cast_fp16 = slice_by_index(begin = var_11443_begin_0, end = var_11443_end_0, end_mask = var_11443_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11443_cast_fp16")]; + tensor var_11447_begin_0 = const()[name = tensor("op_11447_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11447_end_0 = const()[name = tensor("op_11447_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_11447_end_mask_0 = const()[name = tensor("op_11447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11447_cast_fp16 = slice_by_index(begin = var_11447_begin_0, end = var_11447_end_0, end_mask = var_11447_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11447_cast_fp16")]; + tensor var_11451_begin_0 = const()[name = tensor("op_11451_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11451_end_0 = const()[name = tensor("op_11451_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_11451_end_mask_0 = const()[name = tensor("op_11451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11451_cast_fp16 = slice_by_index(begin = var_11451_begin_0, end = var_11451_end_0, end_mask = var_11451_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11451_cast_fp16")]; + tensor var_11455_begin_0 = const()[name = tensor("op_11455_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11455_end_0 = const()[name = tensor("op_11455_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_11455_end_mask_0 = const()[name = tensor("op_11455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11455_cast_fp16 = slice_by_index(begin = var_11455_begin_0, end = var_11455_end_0, end_mask = var_11455_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11455_cast_fp16")]; + tensor var_11459_begin_0 = const()[name = tensor("op_11459_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11459_end_0 = const()[name = tensor("op_11459_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11459_end_mask_0 = const()[name = tensor("op_11459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11459_cast_fp16 = slice_by_index(begin = var_11459_begin_0, end = var_11459_end_0, end_mask = var_11459_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11459_cast_fp16")]; + tensor var_11463_begin_0 = const()[name = tensor("op_11463_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11463_end_0 = const()[name = tensor("op_11463_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_11463_end_mask_0 = const()[name = tensor("op_11463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11463_cast_fp16 = slice_by_index(begin = var_11463_begin_0, end = var_11463_end_0, end_mask = var_11463_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11463_cast_fp16")]; + tensor var_11467_begin_0 = const()[name = tensor("op_11467_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11467_end_0 = const()[name = tensor("op_11467_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_11467_end_mask_0 = const()[name = tensor("op_11467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11467_cast_fp16 = slice_by_index(begin = var_11467_begin_0, end = var_11467_end_0, end_mask = var_11467_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11467_cast_fp16")]; + tensor var_11471_begin_0 = const()[name = tensor("op_11471_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11471_end_0 = const()[name = tensor("op_11471_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_11471_end_mask_0 = const()[name = tensor("op_11471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11471_cast_fp16 = slice_by_index(begin = var_11471_begin_0, end = var_11471_end_0, end_mask = var_11471_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11471_cast_fp16")]; + tensor var_11475_begin_0 = const()[name = tensor("op_11475_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11475_end_0 = const()[name = tensor("op_11475_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_11475_end_mask_0 = const()[name = tensor("op_11475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11475_cast_fp16 = slice_by_index(begin = var_11475_begin_0, end = var_11475_end_0, end_mask = var_11475_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11475_cast_fp16")]; + tensor k_111_perm_0 = const()[name = tensor("k_111_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11482_begin_0 = const()[name = tensor("op_11482_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11482_end_0 = const()[name = tensor("op_11482_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_11482_end_mask_0 = const()[name = tensor("op_11482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_111_cast_fp16 = transpose(perm = k_111_perm_0, x = k_109_cast_fp16)[name = tensor("transpose_40")]; + tensor var_11482_cast_fp16 = slice_by_index(begin = var_11482_begin_0, end = var_11482_end_0, end_mask = var_11482_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11482_cast_fp16")]; + tensor var_11486_begin_0 = const()[name = tensor("op_11486_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11486_end_0 = const()[name = tensor("op_11486_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_11486_end_mask_0 = const()[name = tensor("op_11486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11486_cast_fp16 = slice_by_index(begin = var_11486_begin_0, end = var_11486_end_0, end_mask = var_11486_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11486_cast_fp16")]; + tensor var_11490_begin_0 = const()[name = tensor("op_11490_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11490_end_0 = const()[name = tensor("op_11490_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_11490_end_mask_0 = const()[name = tensor("op_11490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11490_cast_fp16 = slice_by_index(begin = var_11490_begin_0, end = var_11490_end_0, end_mask = var_11490_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11490_cast_fp16")]; + tensor var_11494_begin_0 = const()[name = tensor("op_11494_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11494_end_0 = const()[name = tensor("op_11494_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_11494_end_mask_0 = const()[name = tensor("op_11494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11494_cast_fp16 = slice_by_index(begin = var_11494_begin_0, end = var_11494_end_0, end_mask = var_11494_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11494_cast_fp16")]; + tensor var_11498_begin_0 = const()[name = tensor("op_11498_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11498_end_0 = const()[name = tensor("op_11498_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_11498_end_mask_0 = const()[name = tensor("op_11498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11498_cast_fp16 = slice_by_index(begin = var_11498_begin_0, end = var_11498_end_0, end_mask = var_11498_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11498_cast_fp16")]; + tensor var_11502_begin_0 = const()[name = tensor("op_11502_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11502_end_0 = const()[name = tensor("op_11502_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_11502_end_mask_0 = const()[name = tensor("op_11502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11502_cast_fp16 = slice_by_index(begin = var_11502_begin_0, end = var_11502_end_0, end_mask = var_11502_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11502_cast_fp16")]; + tensor var_11506_begin_0 = const()[name = tensor("op_11506_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11506_end_0 = const()[name = tensor("op_11506_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_11506_end_mask_0 = const()[name = tensor("op_11506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11506_cast_fp16 = slice_by_index(begin = var_11506_begin_0, end = var_11506_end_0, end_mask = var_11506_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11506_cast_fp16")]; + tensor var_11510_begin_0 = const()[name = tensor("op_11510_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11510_end_0 = const()[name = tensor("op_11510_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_11510_end_mask_0 = const()[name = tensor("op_11510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11510_cast_fp16 = slice_by_index(begin = var_11510_begin_0, end = var_11510_end_0, end_mask = var_11510_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11510_cast_fp16")]; + tensor var_11514_begin_0 = const()[name = tensor("op_11514_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11514_end_0 = const()[name = tensor("op_11514_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_11514_end_mask_0 = const()[name = tensor("op_11514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11514_cast_fp16 = slice_by_index(begin = var_11514_begin_0, end = var_11514_end_0, end_mask = var_11514_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11514_cast_fp16")]; + tensor var_11518_begin_0 = const()[name = tensor("op_11518_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11518_end_0 = const()[name = tensor("op_11518_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_11518_end_mask_0 = const()[name = tensor("op_11518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11518_cast_fp16 = slice_by_index(begin = var_11518_begin_0, end = var_11518_end_0, end_mask = var_11518_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11518_cast_fp16")]; + tensor var_11522_begin_0 = const()[name = tensor("op_11522_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11522_end_0 = const()[name = tensor("op_11522_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_11522_end_mask_0 = const()[name = tensor("op_11522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11522_cast_fp16 = slice_by_index(begin = var_11522_begin_0, end = var_11522_end_0, end_mask = var_11522_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11522_cast_fp16")]; + tensor var_11526_begin_0 = const()[name = tensor("op_11526_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11526_end_0 = const()[name = tensor("op_11526_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_11526_end_mask_0 = const()[name = tensor("op_11526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11526_cast_fp16 = slice_by_index(begin = var_11526_begin_0, end = var_11526_end_0, end_mask = var_11526_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11526_cast_fp16")]; + tensor var_11530_begin_0 = const()[name = tensor("op_11530_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_11530_end_0 = const()[name = tensor("op_11530_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_11530_end_mask_0 = const()[name = tensor("op_11530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11530_cast_fp16 = slice_by_index(begin = var_11530_begin_0, end = var_11530_end_0, end_mask = var_11530_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11530_cast_fp16")]; + tensor var_11534_begin_0 = const()[name = tensor("op_11534_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_11534_end_0 = const()[name = tensor("op_11534_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_11534_end_mask_0 = const()[name = tensor("op_11534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11534_cast_fp16 = slice_by_index(begin = var_11534_begin_0, end = var_11534_end_0, end_mask = var_11534_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11534_cast_fp16")]; + tensor var_11538_begin_0 = const()[name = tensor("op_11538_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_11538_end_0 = const()[name = tensor("op_11538_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_11538_end_mask_0 = const()[name = tensor("op_11538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11538_cast_fp16 = slice_by_index(begin = var_11538_begin_0, end = var_11538_end_0, end_mask = var_11538_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11538_cast_fp16")]; + tensor var_11542_begin_0 = const()[name = tensor("op_11542_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_11542_end_0 = const()[name = tensor("op_11542_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_11542_end_mask_0 = const()[name = tensor("op_11542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11542_cast_fp16 = slice_by_index(begin = var_11542_begin_0, end = var_11542_end_0, end_mask = var_11542_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11542_cast_fp16")]; + tensor var_11546_begin_0 = const()[name = tensor("op_11546_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_11546_end_0 = const()[name = tensor("op_11546_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_11546_end_mask_0 = const()[name = tensor("op_11546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11546_cast_fp16 = slice_by_index(begin = var_11546_begin_0, end = var_11546_end_0, end_mask = var_11546_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11546_cast_fp16")]; + tensor var_11550_begin_0 = const()[name = tensor("op_11550_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_11550_end_0 = const()[name = tensor("op_11550_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_11550_end_mask_0 = const()[name = tensor("op_11550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11550_cast_fp16 = slice_by_index(begin = var_11550_begin_0, end = var_11550_end_0, end_mask = var_11550_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11550_cast_fp16")]; + tensor var_11554_begin_0 = const()[name = tensor("op_11554_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_11554_end_0 = const()[name = tensor("op_11554_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_11554_end_mask_0 = const()[name = tensor("op_11554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11554_cast_fp16 = slice_by_index(begin = var_11554_begin_0, end = var_11554_end_0, end_mask = var_11554_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11554_cast_fp16")]; + tensor var_11558_begin_0 = const()[name = tensor("op_11558_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_11558_end_0 = const()[name = tensor("op_11558_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_11558_end_mask_0 = const()[name = tensor("op_11558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11558_cast_fp16 = slice_by_index(begin = var_11558_begin_0, end = var_11558_end_0, end_mask = var_11558_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11558_cast_fp16")]; + tensor var_11560_begin_0 = const()[name = tensor("op_11560_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11560_end_0 = const()[name = tensor("op_11560_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_11560_end_mask_0 = const()[name = tensor("op_11560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11560_cast_fp16 = slice_by_index(begin = var_11560_begin_0, end = var_11560_end_0, end_mask = var_11560_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11560_cast_fp16")]; + tensor var_11564_begin_0 = const()[name = tensor("op_11564_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11564_end_0 = const()[name = tensor("op_11564_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_11564_end_mask_0 = const()[name = tensor("op_11564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11564_cast_fp16 = slice_by_index(begin = var_11564_begin_0, end = var_11564_end_0, end_mask = var_11564_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11564_cast_fp16")]; + tensor var_11568_begin_0 = const()[name = tensor("op_11568_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11568_end_0 = const()[name = tensor("op_11568_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_11568_end_mask_0 = const()[name = tensor("op_11568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11568_cast_fp16 = slice_by_index(begin = var_11568_begin_0, end = var_11568_end_0, end_mask = var_11568_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11568_cast_fp16")]; + tensor var_11572_begin_0 = const()[name = tensor("op_11572_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11572_end_0 = const()[name = tensor("op_11572_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_11572_end_mask_0 = const()[name = tensor("op_11572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11572_cast_fp16 = slice_by_index(begin = var_11572_begin_0, end = var_11572_end_0, end_mask = var_11572_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11572_cast_fp16")]; + tensor var_11576_begin_0 = const()[name = tensor("op_11576_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11576_end_0 = const()[name = tensor("op_11576_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_11576_end_mask_0 = const()[name = tensor("op_11576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11576_cast_fp16 = slice_by_index(begin = var_11576_begin_0, end = var_11576_end_0, end_mask = var_11576_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11576_cast_fp16")]; + tensor var_11580_begin_0 = const()[name = tensor("op_11580_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11580_end_0 = const()[name = tensor("op_11580_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_11580_end_mask_0 = const()[name = tensor("op_11580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11580_cast_fp16 = slice_by_index(begin = var_11580_begin_0, end = var_11580_end_0, end_mask = var_11580_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11580_cast_fp16")]; + tensor var_11584_begin_0 = const()[name = tensor("op_11584_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11584_end_0 = const()[name = tensor("op_11584_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_11584_end_mask_0 = const()[name = tensor("op_11584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11584_cast_fp16 = slice_by_index(begin = var_11584_begin_0, end = var_11584_end_0, end_mask = var_11584_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11584_cast_fp16")]; + tensor var_11588_begin_0 = const()[name = tensor("op_11588_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11588_end_0 = const()[name = tensor("op_11588_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_11588_end_mask_0 = const()[name = tensor("op_11588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11588_cast_fp16 = slice_by_index(begin = var_11588_begin_0, end = var_11588_end_0, end_mask = var_11588_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11588_cast_fp16")]; + tensor var_11592_begin_0 = const()[name = tensor("op_11592_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11592_end_0 = const()[name = tensor("op_11592_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_11592_end_mask_0 = const()[name = tensor("op_11592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11592_cast_fp16 = slice_by_index(begin = var_11592_begin_0, end = var_11592_end_0, end_mask = var_11592_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11592_cast_fp16")]; + tensor var_11596_begin_0 = const()[name = tensor("op_11596_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11596_end_0 = const()[name = tensor("op_11596_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_11596_end_mask_0 = const()[name = tensor("op_11596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11596_cast_fp16 = slice_by_index(begin = var_11596_begin_0, end = var_11596_end_0, end_mask = var_11596_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11596_cast_fp16")]; + tensor var_11600_begin_0 = const()[name = tensor("op_11600_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11600_end_0 = const()[name = tensor("op_11600_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_11600_end_mask_0 = const()[name = tensor("op_11600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11600_cast_fp16 = slice_by_index(begin = var_11600_begin_0, end = var_11600_end_0, end_mask = var_11600_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11600_cast_fp16")]; + tensor var_11604_begin_0 = const()[name = tensor("op_11604_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11604_end_0 = const()[name = tensor("op_11604_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_11604_end_mask_0 = const()[name = tensor("op_11604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11604_cast_fp16 = slice_by_index(begin = var_11604_begin_0, end = var_11604_end_0, end_mask = var_11604_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11604_cast_fp16")]; + tensor var_11608_begin_0 = const()[name = tensor("op_11608_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11608_end_0 = const()[name = tensor("op_11608_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_11608_end_mask_0 = const()[name = tensor("op_11608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11608_cast_fp16 = slice_by_index(begin = var_11608_begin_0, end = var_11608_end_0, end_mask = var_11608_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11608_cast_fp16")]; + tensor var_11612_begin_0 = const()[name = tensor("op_11612_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11612_end_0 = const()[name = tensor("op_11612_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_11612_end_mask_0 = const()[name = tensor("op_11612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11612_cast_fp16 = slice_by_index(begin = var_11612_begin_0, end = var_11612_end_0, end_mask = var_11612_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11612_cast_fp16")]; + tensor var_11616_begin_0 = const()[name = tensor("op_11616_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11616_end_0 = const()[name = tensor("op_11616_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_11616_end_mask_0 = const()[name = tensor("op_11616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11616_cast_fp16 = slice_by_index(begin = var_11616_begin_0, end = var_11616_end_0, end_mask = var_11616_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11616_cast_fp16")]; + tensor var_11620_begin_0 = const()[name = tensor("op_11620_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11620_end_0 = const()[name = tensor("op_11620_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_11620_end_mask_0 = const()[name = tensor("op_11620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11620_cast_fp16 = slice_by_index(begin = var_11620_begin_0, end = var_11620_end_0, end_mask = var_11620_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11620_cast_fp16")]; + tensor var_11624_begin_0 = const()[name = tensor("op_11624_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11624_end_0 = const()[name = tensor("op_11624_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_11624_end_mask_0 = const()[name = tensor("op_11624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11624_cast_fp16 = slice_by_index(begin = var_11624_begin_0, end = var_11624_end_0, end_mask = var_11624_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11624_cast_fp16")]; + tensor var_11628_begin_0 = const()[name = tensor("op_11628_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11628_end_0 = const()[name = tensor("op_11628_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_11628_end_mask_0 = const()[name = tensor("op_11628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11628_cast_fp16 = slice_by_index(begin = var_11628_begin_0, end = var_11628_end_0, end_mask = var_11628_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11628_cast_fp16")]; + tensor var_11632_begin_0 = const()[name = tensor("op_11632_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11632_end_0 = const()[name = tensor("op_11632_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_11632_end_mask_0 = const()[name = tensor("op_11632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11632_cast_fp16 = slice_by_index(begin = var_11632_begin_0, end = var_11632_end_0, end_mask = var_11632_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11632_cast_fp16")]; + tensor var_11636_begin_0 = const()[name = tensor("op_11636_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11636_end_0 = const()[name = tensor("op_11636_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_11636_end_mask_0 = const()[name = tensor("op_11636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11636_cast_fp16 = slice_by_index(begin = var_11636_begin_0, end = var_11636_end_0, end_mask = var_11636_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11636_cast_fp16")]; + tensor var_11640_equation_0 = const()[name = tensor("op_11640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11640_cast_fp16 = einsum(equation = var_11640_equation_0, values = (var_11482_cast_fp16, var_11399_cast_fp16))[name = tensor("op_11640_cast_fp16")]; + tensor var_11641_to_fp16 = const()[name = tensor("op_11641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_921_cast_fp16 = mul(x = var_11640_cast_fp16, y = var_11641_to_fp16)[name = tensor("aw_921_cast_fp16")]; + tensor var_11644_equation_0 = const()[name = tensor("op_11644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11644_cast_fp16 = einsum(equation = var_11644_equation_0, values = (var_11486_cast_fp16, var_11403_cast_fp16))[name = tensor("op_11644_cast_fp16")]; + tensor var_11645_to_fp16 = const()[name = tensor("op_11645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_923_cast_fp16 = mul(x = var_11644_cast_fp16, y = var_11645_to_fp16)[name = tensor("aw_923_cast_fp16")]; + tensor var_11648_equation_0 = const()[name = tensor("op_11648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11648_cast_fp16 = einsum(equation = var_11648_equation_0, values = (var_11490_cast_fp16, var_11407_cast_fp16))[name = tensor("op_11648_cast_fp16")]; + tensor var_11649_to_fp16 = const()[name = tensor("op_11649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_925_cast_fp16 = mul(x = var_11648_cast_fp16, y = var_11649_to_fp16)[name = tensor("aw_925_cast_fp16")]; + tensor var_11652_equation_0 = const()[name = tensor("op_11652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11652_cast_fp16 = einsum(equation = var_11652_equation_0, values = (var_11494_cast_fp16, var_11411_cast_fp16))[name = tensor("op_11652_cast_fp16")]; + tensor var_11653_to_fp16 = const()[name = tensor("op_11653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_927_cast_fp16 = mul(x = var_11652_cast_fp16, y = var_11653_to_fp16)[name = tensor("aw_927_cast_fp16")]; + tensor var_11656_equation_0 = const()[name = tensor("op_11656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11656_cast_fp16 = einsum(equation = var_11656_equation_0, values = (var_11498_cast_fp16, var_11415_cast_fp16))[name = tensor("op_11656_cast_fp16")]; + tensor var_11657_to_fp16 = const()[name = tensor("op_11657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_929_cast_fp16 = mul(x = var_11656_cast_fp16, y = var_11657_to_fp16)[name = tensor("aw_929_cast_fp16")]; + tensor var_11660_equation_0 = const()[name = tensor("op_11660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11660_cast_fp16 = einsum(equation = var_11660_equation_0, values = (var_11502_cast_fp16, var_11419_cast_fp16))[name = tensor("op_11660_cast_fp16")]; + tensor var_11661_to_fp16 = const()[name = tensor("op_11661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_931_cast_fp16 = mul(x = var_11660_cast_fp16, y = var_11661_to_fp16)[name = tensor("aw_931_cast_fp16")]; + tensor var_11664_equation_0 = const()[name = tensor("op_11664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11664_cast_fp16 = einsum(equation = var_11664_equation_0, values = (var_11506_cast_fp16, var_11423_cast_fp16))[name = tensor("op_11664_cast_fp16")]; + tensor var_11665_to_fp16 = const()[name = tensor("op_11665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_933_cast_fp16 = mul(x = var_11664_cast_fp16, y = var_11665_to_fp16)[name = tensor("aw_933_cast_fp16")]; + tensor var_11668_equation_0 = const()[name = tensor("op_11668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11668_cast_fp16 = einsum(equation = var_11668_equation_0, values = (var_11510_cast_fp16, var_11427_cast_fp16))[name = tensor("op_11668_cast_fp16")]; + tensor var_11669_to_fp16 = const()[name = tensor("op_11669_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_935_cast_fp16 = mul(x = var_11668_cast_fp16, y = var_11669_to_fp16)[name = tensor("aw_935_cast_fp16")]; + tensor var_11672_equation_0 = const()[name = tensor("op_11672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11672_cast_fp16 = einsum(equation = var_11672_equation_0, values = (var_11514_cast_fp16, var_11431_cast_fp16))[name = tensor("op_11672_cast_fp16")]; + tensor var_11673_to_fp16 = const()[name = tensor("op_11673_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_937_cast_fp16 = mul(x = var_11672_cast_fp16, y = var_11673_to_fp16)[name = tensor("aw_937_cast_fp16")]; + tensor var_11676_equation_0 = const()[name = tensor("op_11676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11676_cast_fp16 = einsum(equation = var_11676_equation_0, values = (var_11518_cast_fp16, var_11435_cast_fp16))[name = tensor("op_11676_cast_fp16")]; + tensor var_11677_to_fp16 = const()[name = tensor("op_11677_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_939_cast_fp16 = mul(x = var_11676_cast_fp16, y = var_11677_to_fp16)[name = tensor("aw_939_cast_fp16")]; + tensor var_11680_equation_0 = const()[name = tensor("op_11680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11680_cast_fp16 = einsum(equation = var_11680_equation_0, values = (var_11522_cast_fp16, var_11439_cast_fp16))[name = tensor("op_11680_cast_fp16")]; + tensor var_11681_to_fp16 = const()[name = tensor("op_11681_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_941_cast_fp16 = mul(x = var_11680_cast_fp16, y = var_11681_to_fp16)[name = tensor("aw_941_cast_fp16")]; + tensor var_11684_equation_0 = const()[name = tensor("op_11684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11684_cast_fp16 = einsum(equation = var_11684_equation_0, values = (var_11526_cast_fp16, var_11443_cast_fp16))[name = tensor("op_11684_cast_fp16")]; + tensor var_11685_to_fp16 = const()[name = tensor("op_11685_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_943_cast_fp16 = mul(x = var_11684_cast_fp16, y = var_11685_to_fp16)[name = tensor("aw_943_cast_fp16")]; + tensor var_11688_equation_0 = const()[name = tensor("op_11688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11688_cast_fp16 = einsum(equation = var_11688_equation_0, values = (var_11530_cast_fp16, var_11447_cast_fp16))[name = tensor("op_11688_cast_fp16")]; + tensor var_11689_to_fp16 = const()[name = tensor("op_11689_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_945_cast_fp16 = mul(x = var_11688_cast_fp16, y = var_11689_to_fp16)[name = tensor("aw_945_cast_fp16")]; + tensor var_11692_equation_0 = const()[name = tensor("op_11692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11692_cast_fp16 = einsum(equation = var_11692_equation_0, values = (var_11534_cast_fp16, var_11451_cast_fp16))[name = tensor("op_11692_cast_fp16")]; + tensor var_11693_to_fp16 = const()[name = tensor("op_11693_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_947_cast_fp16 = mul(x = var_11692_cast_fp16, y = var_11693_to_fp16)[name = tensor("aw_947_cast_fp16")]; + tensor var_11696_equation_0 = const()[name = tensor("op_11696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11696_cast_fp16 = einsum(equation = var_11696_equation_0, values = (var_11538_cast_fp16, var_11455_cast_fp16))[name = tensor("op_11696_cast_fp16")]; + tensor var_11697_to_fp16 = const()[name = tensor("op_11697_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_949_cast_fp16 = mul(x = var_11696_cast_fp16, y = var_11697_to_fp16)[name = tensor("aw_949_cast_fp16")]; + tensor var_11700_equation_0 = const()[name = tensor("op_11700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11700_cast_fp16 = einsum(equation = var_11700_equation_0, values = (var_11542_cast_fp16, var_11459_cast_fp16))[name = tensor("op_11700_cast_fp16")]; + tensor var_11701_to_fp16 = const()[name = tensor("op_11701_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_951_cast_fp16 = mul(x = var_11700_cast_fp16, y = var_11701_to_fp16)[name = tensor("aw_951_cast_fp16")]; + tensor var_11704_equation_0 = const()[name = tensor("op_11704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11704_cast_fp16 = einsum(equation = var_11704_equation_0, values = (var_11546_cast_fp16, var_11463_cast_fp16))[name = tensor("op_11704_cast_fp16")]; + tensor var_11705_to_fp16 = const()[name = tensor("op_11705_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_953_cast_fp16 = mul(x = var_11704_cast_fp16, y = var_11705_to_fp16)[name = tensor("aw_953_cast_fp16")]; + tensor var_11708_equation_0 = const()[name = tensor("op_11708_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11708_cast_fp16 = einsum(equation = var_11708_equation_0, values = (var_11550_cast_fp16, var_11467_cast_fp16))[name = tensor("op_11708_cast_fp16")]; + tensor var_11709_to_fp16 = const()[name = tensor("op_11709_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_955_cast_fp16 = mul(x = var_11708_cast_fp16, y = var_11709_to_fp16)[name = tensor("aw_955_cast_fp16")]; + tensor var_11712_equation_0 = const()[name = tensor("op_11712_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11712_cast_fp16 = einsum(equation = var_11712_equation_0, values = (var_11554_cast_fp16, var_11471_cast_fp16))[name = tensor("op_11712_cast_fp16")]; + tensor var_11713_to_fp16 = const()[name = tensor("op_11713_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_957_cast_fp16 = mul(x = var_11712_cast_fp16, y = var_11713_to_fp16)[name = tensor("aw_957_cast_fp16")]; + tensor var_11716_equation_0 = const()[name = tensor("op_11716_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11716_cast_fp16 = einsum(equation = var_11716_equation_0, values = (var_11558_cast_fp16, var_11475_cast_fp16))[name = tensor("op_11716_cast_fp16")]; + tensor var_11717_to_fp16 = const()[name = tensor("op_11717_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_959_cast_fp16 = mul(x = var_11716_cast_fp16, y = var_11717_to_fp16)[name = tensor("aw_959_cast_fp16")]; + tensor var_11719_cast_fp16 = softmax(axis = var_2624, x = aw_921_cast_fp16)[name = tensor("op_11719_cast_fp16")]; + tensor var_11720_cast_fp16 = softmax(axis = var_2624, x = aw_923_cast_fp16)[name = tensor("op_11720_cast_fp16")]; + tensor var_11721_cast_fp16 = softmax(axis = var_2624, x = aw_925_cast_fp16)[name = tensor("op_11721_cast_fp16")]; + tensor var_11722_cast_fp16 = softmax(axis = var_2624, x = aw_927_cast_fp16)[name = tensor("op_11722_cast_fp16")]; + tensor var_11723_cast_fp16 = softmax(axis = var_2624, x = aw_929_cast_fp16)[name = tensor("op_11723_cast_fp16")]; + tensor var_11724_cast_fp16 = softmax(axis = var_2624, x = aw_931_cast_fp16)[name = tensor("op_11724_cast_fp16")]; + tensor var_11725_cast_fp16 = softmax(axis = var_2624, x = aw_933_cast_fp16)[name = tensor("op_11725_cast_fp16")]; + tensor var_11726_cast_fp16 = softmax(axis = var_2624, x = aw_935_cast_fp16)[name = tensor("op_11726_cast_fp16")]; + tensor var_11727_cast_fp16 = softmax(axis = var_2624, x = aw_937_cast_fp16)[name = tensor("op_11727_cast_fp16")]; + tensor var_11728_cast_fp16 = softmax(axis = var_2624, x = aw_939_cast_fp16)[name = tensor("op_11728_cast_fp16")]; + tensor var_11729_cast_fp16 = softmax(axis = var_2624, x = aw_941_cast_fp16)[name = tensor("op_11729_cast_fp16")]; + tensor var_11730_cast_fp16 = softmax(axis = var_2624, x = aw_943_cast_fp16)[name = tensor("op_11730_cast_fp16")]; + tensor var_11731_cast_fp16 = softmax(axis = var_2624, x = aw_945_cast_fp16)[name = tensor("op_11731_cast_fp16")]; + tensor var_11732_cast_fp16 = softmax(axis = var_2624, x = aw_947_cast_fp16)[name = tensor("op_11732_cast_fp16")]; + tensor var_11733_cast_fp16 = softmax(axis = var_2624, x = aw_949_cast_fp16)[name = tensor("op_11733_cast_fp16")]; + tensor var_11734_cast_fp16 = softmax(axis = var_2624, x = aw_951_cast_fp16)[name = tensor("op_11734_cast_fp16")]; + tensor var_11735_cast_fp16 = softmax(axis = var_2624, x = aw_953_cast_fp16)[name = tensor("op_11735_cast_fp16")]; + tensor var_11736_cast_fp16 = softmax(axis = var_2624, x = aw_955_cast_fp16)[name = tensor("op_11736_cast_fp16")]; + tensor var_11737_cast_fp16 = softmax(axis = var_2624, x = aw_957_cast_fp16)[name = tensor("op_11737_cast_fp16")]; + tensor var_11738_cast_fp16 = softmax(axis = var_2624, x = aw_959_cast_fp16)[name = tensor("op_11738_cast_fp16")]; + tensor var_11740_equation_0 = const()[name = tensor("op_11740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11740_cast_fp16 = einsum(equation = var_11740_equation_0, values = (var_11560_cast_fp16, var_11719_cast_fp16))[name = tensor("op_11740_cast_fp16")]; + tensor var_11742_equation_0 = const()[name = tensor("op_11742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11742_cast_fp16 = einsum(equation = var_11742_equation_0, values = (var_11564_cast_fp16, var_11720_cast_fp16))[name = tensor("op_11742_cast_fp16")]; + tensor var_11744_equation_0 = const()[name = tensor("op_11744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11744_cast_fp16 = einsum(equation = var_11744_equation_0, values = (var_11568_cast_fp16, var_11721_cast_fp16))[name = tensor("op_11744_cast_fp16")]; + tensor var_11746_equation_0 = const()[name = tensor("op_11746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11746_cast_fp16 = einsum(equation = var_11746_equation_0, values = (var_11572_cast_fp16, var_11722_cast_fp16))[name = tensor("op_11746_cast_fp16")]; + tensor var_11748_equation_0 = const()[name = tensor("op_11748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11748_cast_fp16 = einsum(equation = var_11748_equation_0, values = (var_11576_cast_fp16, var_11723_cast_fp16))[name = tensor("op_11748_cast_fp16")]; + tensor var_11750_equation_0 = const()[name = tensor("op_11750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11750_cast_fp16 = einsum(equation = var_11750_equation_0, values = (var_11580_cast_fp16, var_11724_cast_fp16))[name = tensor("op_11750_cast_fp16")]; + tensor var_11752_equation_0 = const()[name = tensor("op_11752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11752_cast_fp16 = einsum(equation = var_11752_equation_0, values = (var_11584_cast_fp16, var_11725_cast_fp16))[name = tensor("op_11752_cast_fp16")]; + tensor var_11754_equation_0 = const()[name = tensor("op_11754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11754_cast_fp16 = einsum(equation = var_11754_equation_0, values = (var_11588_cast_fp16, var_11726_cast_fp16))[name = tensor("op_11754_cast_fp16")]; + tensor var_11756_equation_0 = const()[name = tensor("op_11756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11756_cast_fp16 = einsum(equation = var_11756_equation_0, values = (var_11592_cast_fp16, var_11727_cast_fp16))[name = tensor("op_11756_cast_fp16")]; + tensor var_11758_equation_0 = const()[name = tensor("op_11758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11758_cast_fp16 = einsum(equation = var_11758_equation_0, values = (var_11596_cast_fp16, var_11728_cast_fp16))[name = tensor("op_11758_cast_fp16")]; + tensor var_11760_equation_0 = const()[name = tensor("op_11760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11760_cast_fp16 = einsum(equation = var_11760_equation_0, values = (var_11600_cast_fp16, var_11729_cast_fp16))[name = tensor("op_11760_cast_fp16")]; + tensor var_11762_equation_0 = const()[name = tensor("op_11762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11762_cast_fp16 = einsum(equation = var_11762_equation_0, values = (var_11604_cast_fp16, var_11730_cast_fp16))[name = tensor("op_11762_cast_fp16")]; + tensor var_11764_equation_0 = const()[name = tensor("op_11764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11764_cast_fp16 = einsum(equation = var_11764_equation_0, values = (var_11608_cast_fp16, var_11731_cast_fp16))[name = tensor("op_11764_cast_fp16")]; + tensor var_11766_equation_0 = const()[name = tensor("op_11766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11766_cast_fp16 = einsum(equation = var_11766_equation_0, values = (var_11612_cast_fp16, var_11732_cast_fp16))[name = tensor("op_11766_cast_fp16")]; + tensor var_11768_equation_0 = const()[name = tensor("op_11768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11768_cast_fp16 = einsum(equation = var_11768_equation_0, values = (var_11616_cast_fp16, var_11733_cast_fp16))[name = tensor("op_11768_cast_fp16")]; + tensor var_11770_equation_0 = const()[name = tensor("op_11770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11770_cast_fp16 = einsum(equation = var_11770_equation_0, values = (var_11620_cast_fp16, var_11734_cast_fp16))[name = tensor("op_11770_cast_fp16")]; + tensor var_11772_equation_0 = const()[name = tensor("op_11772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11772_cast_fp16 = einsum(equation = var_11772_equation_0, values = (var_11624_cast_fp16, var_11735_cast_fp16))[name = tensor("op_11772_cast_fp16")]; + tensor var_11774_equation_0 = const()[name = tensor("op_11774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11774_cast_fp16 = einsum(equation = var_11774_equation_0, values = (var_11628_cast_fp16, var_11736_cast_fp16))[name = tensor("op_11774_cast_fp16")]; + tensor var_11776_equation_0 = const()[name = tensor("op_11776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11776_cast_fp16 = einsum(equation = var_11776_equation_0, values = (var_11632_cast_fp16, var_11737_cast_fp16))[name = tensor("op_11776_cast_fp16")]; + tensor var_11778_equation_0 = const()[name = tensor("op_11778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11778_cast_fp16 = einsum(equation = var_11778_equation_0, values = (var_11636_cast_fp16, var_11738_cast_fp16))[name = tensor("op_11778_cast_fp16")]; + tensor input_205_interleave_0 = const()[name = tensor("input_205_interleave_0"), val = tensor(false)]; + tensor input_205_cast_fp16 = concat(axis = var_2624, interleave = input_205_interleave_0, values = (var_11740_cast_fp16, var_11742_cast_fp16, var_11744_cast_fp16, var_11746_cast_fp16, var_11748_cast_fp16, var_11750_cast_fp16, var_11752_cast_fp16, var_11754_cast_fp16, var_11756_cast_fp16, var_11758_cast_fp16, var_11760_cast_fp16, var_11762_cast_fp16, var_11764_cast_fp16, var_11766_cast_fp16, var_11768_cast_fp16, var_11770_cast_fp16, var_11772_cast_fp16, var_11774_cast_fp16, var_11776_cast_fp16, var_11778_cast_fp16))[name = tensor("input_205_cast_fp16")]; + tensor var_11788_pad_type_0 = const()[name = tensor("op_11788_pad_type_0"), val = tensor("valid")]; + tensor var_11788_strides_0 = const()[name = tensor("op_11788_strides_0"), val = tensor([1, 1])]; + tensor var_11788_pad_0 = const()[name = tensor("op_11788_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11788_dilations_0 = const()[name = tensor("op_11788_dilations_0"), val = tensor([1, 1])]; + tensor var_11788_groups_0 = const()[name = tensor("op_11788_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319406528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320635392))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320635584)))]; + tensor var_11788_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_11788_dilations_0, groups = var_11788_groups_0, pad = var_11788_pad_0, pad_type = var_11788_pad_type_0, strides = var_11788_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor("op_11788_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = var_11788_cast_fp16, y = inputs_81_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; + tensor input_207_axes_0 = const()[name = tensor("input_207_axes_0"), val = tensor([1])]; + tensor input_207_gamma_0_to_fp16 = const()[name = tensor("input_207_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320638208)))]; + tensor input_207_beta_0_to_fp16 = const()[name = tensor("input_207_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320640832)))]; + tensor var_11798_to_fp16 = const()[name = tensor("op_11798_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_207_cast_fp16 = layer_norm(axes = input_207_axes_0, beta = input_207_beta_0_to_fp16, epsilon = var_11798_to_fp16, gamma = input_207_gamma_0_to_fp16, x = inputs_83_cast_fp16)[name = tensor("input_207_cast_fp16")]; + tensor var_11818_pad_type_0 = const()[name = tensor("op_11818_pad_type_0"), val = tensor("valid")]; + tensor var_11818_strides_0 = const()[name = tensor("op_11818_strides_0"), val = tensor([1, 1])]; + tensor var_11818_pad_0 = const()[name = tensor("op_11818_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11818_dilations_0 = const()[name = tensor("op_11818_dilations_0"), val = tensor([1, 1])]; + tensor var_11818_groups_0 = const()[name = tensor("op_11818_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320643456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330473920))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330474112)))]; + tensor var_11818_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_11818_dilations_0, groups = var_11818_groups_0, pad = var_11818_pad_0, pad_type = var_11818_pad_type_0, strides = var_11818_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = tensor("op_11818_cast_fp16")]; + tensor var_11819_split_sizes_0 = const()[name = tensor("op_11819_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_11819_axis_0 = const()[name = tensor("op_11819_axis_0"), val = tensor(1)]; + tensor var_11819_cast_fp16_0, tensor var_11819_cast_fp16_1 = split(axis = var_11819_axis_0, split_sizes = var_11819_split_sizes_0, x = var_11818_cast_fp16)[name = tensor("op_11819_cast_fp16")]; + tensor var_11821_mode_0 = const()[name = tensor("op_11821_mode_0"), val = tensor("EXACT")]; + tensor var_11821_cast_fp16 = gelu(mode = var_11821_mode_0, x = var_11819_cast_fp16_1)[name = tensor("op_11821_cast_fp16")]; + tensor input_209_cast_fp16 = mul(x = var_11819_cast_fp16_0, y = var_11821_cast_fp16)[name = tensor("input_209_cast_fp16")]; + tensor var_11829_pad_type_0 = const()[name = tensor("op_11829_pad_type_0"), val = tensor("valid")]; + tensor var_11829_strides_0 = const()[name = tensor("op_11829_strides_0"), val = tensor([1, 1])]; + tensor var_11829_pad_0 = const()[name = tensor("op_11829_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11829_dilations_0 = const()[name = tensor("op_11829_dilations_0"), val = tensor([1, 1])]; + tensor var_11829_groups_0 = const()[name = tensor("op_11829_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330494656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335409920))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335410112)))]; + tensor var_11829_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_11829_dilations_0, groups = var_11829_groups_0, pad = var_11829_pad_0, pad_type = var_11829_pad_type_0, strides = var_11829_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = tensor("op_11829_cast_fp16")]; + tensor hidden_states_125_cast_fp16 = add(x = var_11829_cast_fp16, y = inputs_83_cast_fp16)[name = tensor("hidden_states_125_cast_fp16")]; + tensor var_11831 = const()[name = tensor("op_11831"), val = tensor([2, 1280, 32, 32])]; + tensor input_211_cast_fp16 = reshape(shape = var_11831, x = hidden_states_125_cast_fp16)[name = tensor("input_211_cast_fp16")]; + tensor hidden_states_127_pad_type_0 = const()[name = tensor("hidden_states_127_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_127_strides_0 = const()[name = tensor("hidden_states_127_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_127_pad_0 = const()[name = tensor("hidden_states_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_127_dilations_0 = const()[name = tensor("hidden_states_127_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_127_groups_0 = const()[name = tensor("hidden_states_127_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335412736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336641600))), name = tensor("down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336641792)))]; + tensor hidden_states_127_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_127_dilations_0, groups = hidden_states_127_groups_0, pad = hidden_states_127_pad_0, pad_type = hidden_states_127_pad_type_0, strides = hidden_states_127_strides_0, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized, x = input_211_cast_fp16)[name = tensor("hidden_states_127_cast_fp16")]; + tensor input_213_cast_fp16_1 = add(x = hidden_states_127_cast_fp16, y = hidden_states_61_cast_fp16)[name = tensor("input_213_cast_fp16")]; + tensor reshape_52_shape_0 = const()[name = tensor("reshape_52_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_52_cast_fp16 = reshape(shape = reshape_52_shape_0, x = input_213_cast_fp16_1)[name = tensor("reshape_52_cast_fp16")]; + tensor reduce_mean_39_axes_0 = const()[name = tensor("reduce_mean_39_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_39_keep_dims_0 = const()[name = tensor("reduce_mean_39_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_39_cast_fp16 = reduce_mean(axes = reduce_mean_39_axes_0, keep_dims = reduce_mean_39_keep_dims_0, x = reshape_52_cast_fp16)[name = tensor("reduce_mean_39_cast_fp16")]; + tensor sub_26_cast_fp16 = sub(x = reshape_52_cast_fp16, y = reduce_mean_39_cast_fp16)[name = tensor("sub_26_cast_fp16")]; + tensor square_13_cast_fp16 = square(x = sub_26_cast_fp16)[name = tensor("square_13_cast_fp16")]; + tensor reduce_mean_41_axes_0 = const()[name = tensor("reduce_mean_41_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_41_keep_dims_0 = const()[name = tensor("reduce_mean_41_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_41_cast_fp16 = reduce_mean(axes = reduce_mean_41_axes_0, keep_dims = reduce_mean_41_keep_dims_0, x = square_13_cast_fp16)[name = tensor("reduce_mean_41_cast_fp16")]; + tensor add_26_y_0_to_fp16 = const()[name = tensor("add_26_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_26_cast_fp16 = add(x = reduce_mean_41_cast_fp16, y = add_26_y_0_to_fp16)[name = tensor("add_26_cast_fp16")]; + tensor sqrt_13_cast_fp16 = sqrt(x = add_26_cast_fp16)[name = tensor("sqrt_13_cast_fp16")]; + tensor real_div_13_cast_fp16 = real_div(x = sub_26_cast_fp16, y = sqrt_13_cast_fp16)[name = tensor("real_div_13_cast_fp16")]; + tensor reshape_53_shape_0 = const()[name = tensor("reshape_53_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_53_cast_fp16 = reshape(shape = reshape_53_shape_0, x = real_div_13_cast_fp16)[name = tensor("reshape_53_cast_fp16")]; + tensor add_27_gamma_0_to_fp16 = const()[name = tensor("add_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336644416)))]; + tensor add_27_beta_0_to_fp16 = const()[name = tensor("add_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336647040)))]; + tensor add_27_epsilon_0_to_fp16 = const()[name = tensor("add_27_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_27_cast_fp16 = batch_norm(beta = add_27_beta_0_to_fp16, epsilon = add_27_epsilon_0_to_fp16, gamma = add_27_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_53_cast_fp16)[name = tensor("add_27_cast_fp16")]; + tensor input_217_cast_fp16 = silu(x = add_27_cast_fp16)[name = tensor("input_217_cast_fp16")]; + tensor hidden_states_129_pad_type_0 = const()[name = tensor("hidden_states_129_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_129_pad_0 = const()[name = tensor("hidden_states_129_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_129_strides_0 = const()[name = tensor("hidden_states_129_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_129_dilations_0 = const()[name = tensor("hidden_states_129_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_129_groups_0 = const()[name = tensor("hidden_states_129_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336649664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(347708928))), name = tensor("down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(347709120)))]; + tensor hidden_states_129_cast_fp16 = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_129_dilations_0, groups = hidden_states_129_groups_0, pad = hidden_states_129_pad_0, pad_type = hidden_states_129_pad_type_0, strides = hidden_states_129_strides_0, weight = down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = tensor("hidden_states_129_cast_fp16")]; + tensor temb_11_pad_type_0 = const()[name = tensor("temb_11_pad_type_0"), val = tensor("valid")]; + tensor temb_11_strides_0 = const()[name = tensor("temb_11_strides_0"), val = tensor([1, 1])]; + tensor temb_11_pad_0 = const()[name = tensor("temb_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_11_dilations_0 = const()[name = tensor("temb_11_dilations_0"), val = tensor([1, 1])]; + tensor temb_11_groups_0 = const()[name = tensor("temb_11_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(347711744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348940608))), name = tensor("down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348940800)))]; + tensor temb_11_cast_fp16 = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_11_dilations_0, groups = temb_11_groups_0, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = temb_11_strides_0, weight = down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_11_cast_fp16")]; + tensor input_221_cast_fp16 = add(x = hidden_states_129_cast_fp16, y = temb_11_cast_fp16)[name = tensor("input_221_cast_fp16")]; + tensor reshape_56_shape_0 = const()[name = tensor("reshape_56_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_56_cast_fp16 = reshape(shape = reshape_56_shape_0, x = input_221_cast_fp16)[name = tensor("reshape_56_cast_fp16")]; + tensor reduce_mean_42_axes_0 = const()[name = tensor("reduce_mean_42_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_42_keep_dims_0 = const()[name = tensor("reduce_mean_42_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_42_cast_fp16 = reduce_mean(axes = reduce_mean_42_axes_0, keep_dims = reduce_mean_42_keep_dims_0, x = reshape_56_cast_fp16)[name = tensor("reduce_mean_42_cast_fp16")]; + tensor sub_28_cast_fp16 = sub(x = reshape_56_cast_fp16, y = reduce_mean_42_cast_fp16)[name = tensor("sub_28_cast_fp16")]; + tensor square_14_cast_fp16 = square(x = sub_28_cast_fp16)[name = tensor("square_14_cast_fp16")]; + tensor reduce_mean_44_axes_0 = const()[name = tensor("reduce_mean_44_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_44_keep_dims_0 = const()[name = tensor("reduce_mean_44_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_44_cast_fp16 = reduce_mean(axes = reduce_mean_44_axes_0, keep_dims = reduce_mean_44_keep_dims_0, x = square_14_cast_fp16)[name = tensor("reduce_mean_44_cast_fp16")]; + tensor add_28_y_0_to_fp16 = const()[name = tensor("add_28_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_28_cast_fp16 = add(x = reduce_mean_44_cast_fp16, y = add_28_y_0_to_fp16)[name = tensor("add_28_cast_fp16")]; + tensor sqrt_14_cast_fp16 = sqrt(x = add_28_cast_fp16)[name = tensor("sqrt_14_cast_fp16")]; + tensor real_div_14_cast_fp16 = real_div(x = sub_28_cast_fp16, y = sqrt_14_cast_fp16)[name = tensor("real_div_14_cast_fp16")]; + tensor reshape_57_shape_0 = const()[name = tensor("reshape_57_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_57_cast_fp16 = reshape(shape = reshape_57_shape_0, x = real_div_14_cast_fp16)[name = tensor("reshape_57_cast_fp16")]; + tensor add_29_gamma_0_to_fp16 = const()[name = tensor("add_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348943424)))]; + tensor add_29_beta_0_to_fp16 = const()[name = tensor("add_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348946048)))]; + tensor add_29_epsilon_0_to_fp16 = const()[name = tensor("add_29_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_29_cast_fp16 = batch_norm(beta = add_29_beta_0_to_fp16, epsilon = add_29_epsilon_0_to_fp16, gamma = add_29_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_57_cast_fp16)[name = tensor("add_29_cast_fp16")]; + tensor input_225_cast_fp16 = silu(x = add_29_cast_fp16)[name = tensor("input_225_cast_fp16")]; + tensor hidden_states_131_pad_type_0 = const()[name = tensor("hidden_states_131_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_131_pad_0 = const()[name = tensor("hidden_states_131_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_131_strides_0 = const()[name = tensor("hidden_states_131_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_131_dilations_0 = const()[name = tensor("hidden_states_131_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_131_groups_0 = const()[name = tensor("hidden_states_131_groups_0"), val = tensor(1)]; + tensor down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348948672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360007936))), name = tensor("down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360008128)))]; + tensor hidden_states_131_cast_fp16 = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_131_dilations_0, groups = hidden_states_131_groups_0, pad = hidden_states_131_pad_0, pad_type = hidden_states_131_pad_type_0, strides = hidden_states_131_strides_0, weight = down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor("hidden_states_131_cast_fp16")]; + tensor hidden_states_133_cast_fp16 = add(x = input_213_cast_fp16_1, y = hidden_states_131_cast_fp16)[name = tensor("hidden_states_133_cast_fp16")]; + tensor reshape_60_shape_0 = const()[name = tensor("reshape_60_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_60_cast_fp16 = reshape(shape = reshape_60_shape_0, x = hidden_states_133_cast_fp16)[name = tensor("reshape_60_cast_fp16")]; + tensor reduce_mean_45_axes_0 = const()[name = tensor("reduce_mean_45_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_45_keep_dims_0 = const()[name = tensor("reduce_mean_45_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_45_cast_fp16 = reduce_mean(axes = reduce_mean_45_axes_0, keep_dims = reduce_mean_45_keep_dims_0, x = reshape_60_cast_fp16)[name = tensor("reduce_mean_45_cast_fp16")]; + tensor sub_30_cast_fp16 = sub(x = reshape_60_cast_fp16, y = reduce_mean_45_cast_fp16)[name = tensor("sub_30_cast_fp16")]; + tensor square_15_cast_fp16 = square(x = sub_30_cast_fp16)[name = tensor("square_15_cast_fp16")]; + tensor reduce_mean_47_axes_0 = const()[name = tensor("reduce_mean_47_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_47_keep_dims_0 = const()[name = tensor("reduce_mean_47_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_47_cast_fp16 = reduce_mean(axes = reduce_mean_47_axes_0, keep_dims = reduce_mean_47_keep_dims_0, x = square_15_cast_fp16)[name = tensor("reduce_mean_47_cast_fp16")]; + tensor add_30_y_0_to_fp16 = const()[name = tensor("add_30_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_30_cast_fp16 = add(x = reduce_mean_47_cast_fp16, y = add_30_y_0_to_fp16)[name = tensor("add_30_cast_fp16")]; + tensor sqrt_15_cast_fp16 = sqrt(x = add_30_cast_fp16)[name = tensor("sqrt_15_cast_fp16")]; + tensor real_div_15_cast_fp16 = real_div(x = sub_30_cast_fp16, y = sqrt_15_cast_fp16)[name = tensor("real_div_15_cast_fp16")]; + tensor reshape_61_shape_0 = const()[name = tensor("reshape_61_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_61_cast_fp16 = reshape(shape = reshape_61_shape_0, x = real_div_15_cast_fp16)[name = tensor("reshape_61_cast_fp16")]; + tensor add_31_gamma_0_to_fp16 = const()[name = tensor("add_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360010752)))]; + tensor add_31_beta_0_to_fp16 = const()[name = tensor("add_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360013376)))]; + tensor add_31_epsilon_0_to_fp16 = const()[name = tensor("add_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_31_cast_fp16 = batch_norm(beta = add_31_beta_0_to_fp16, epsilon = add_31_epsilon_0_to_fp16, gamma = add_31_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_61_cast_fp16)[name = tensor("add_31_cast_fp16")]; + tensor hidden_states_135_pad_type_0 = const()[name = tensor("hidden_states_135_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_135_strides_0 = const()[name = tensor("hidden_states_135_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_135_pad_0 = const()[name = tensor("hidden_states_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_135_dilations_0 = const()[name = tensor("hidden_states_135_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_135_groups_0 = const()[name = tensor("hidden_states_135_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360016000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361244864))), name = tensor("down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361245056)))]; + tensor hidden_states_135_cast_fp16 = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_135_dilations_0, groups = hidden_states_135_groups_0, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = hidden_states_135_strides_0, weight = down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized, x = add_31_cast_fp16)[name = tensor("hidden_states_135_cast_fp16")]; + tensor var_11919 = const()[name = tensor("op_11919"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_85_cast_fp16 = reshape(shape = var_11919, x = hidden_states_135_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; + tensor hidden_states_137_axes_0 = const()[name = tensor("hidden_states_137_axes_0"), val = tensor([1])]; + tensor hidden_states_137_gamma_0_to_fp16 = const()[name = tensor("hidden_states_137_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361247680)))]; + tensor hidden_states_137_beta_0_to_fp16 = const()[name = tensor("hidden_states_137_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361250304)))]; + tensor var_11935_to_fp16 = const()[name = tensor("op_11935_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_137_cast_fp16 = layer_norm(axes = hidden_states_137_axes_0, beta = hidden_states_137_beta_0_to_fp16, epsilon = var_11935_to_fp16, gamma = hidden_states_137_gamma_0_to_fp16, x = inputs_85_cast_fp16)[name = tensor("hidden_states_137_cast_fp16")]; + tensor q_57_pad_type_0 = const()[name = tensor("q_57_pad_type_0"), val = tensor("valid")]; + tensor q_57_strides_0 = const()[name = tensor("q_57_strides_0"), val = tensor([1, 1])]; + tensor q_57_pad_0 = const()[name = tensor("q_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_57_dilations_0 = const()[name = tensor("q_57_dilations_0"), val = tensor([1, 1])]; + tensor q_57_groups_0 = const()[name = tensor("q_57_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361252928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362481792))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_57_cast_fp16 = conv(dilations = q_57_dilations_0, groups = q_57_groups_0, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = q_57_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_137_cast_fp16)[name = tensor("q_57_cast_fp16")]; + tensor k_113_pad_type_0 = const()[name = tensor("k_113_pad_type_0"), val = tensor("valid")]; + tensor k_113_strides_0 = const()[name = tensor("k_113_strides_0"), val = tensor([1, 1])]; + tensor k_113_pad_0 = const()[name = tensor("k_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_113_dilations_0 = const()[name = tensor("k_113_dilations_0"), val = tensor([1, 1])]; + tensor k_113_groups_0 = const()[name = tensor("k_113_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362481984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(363710848))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_113_cast_fp16 = conv(dilations = k_113_dilations_0, groups = k_113_groups_0, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = k_113_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_137_cast_fp16)[name = tensor("k_113_cast_fp16")]; + tensor v_57_pad_type_0 = const()[name = tensor("v_57_pad_type_0"), val = tensor("valid")]; + tensor v_57_strides_0 = const()[name = tensor("v_57_strides_0"), val = tensor([1, 1])]; + tensor v_57_pad_0 = const()[name = tensor("v_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_57_dilations_0 = const()[name = tensor("v_57_dilations_0"), val = tensor([1, 1])]; + tensor v_57_groups_0 = const()[name = tensor("v_57_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(363711040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364939904))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_57_cast_fp16 = conv(dilations = v_57_dilations_0, groups = v_57_groups_0, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = v_57_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_137_cast_fp16)[name = tensor("v_57_cast_fp16")]; + tensor var_11968_begin_0 = const()[name = tensor("op_11968_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11968_end_0 = const()[name = tensor("op_11968_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11968_end_mask_0 = const()[name = tensor("op_11968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11968_cast_fp16 = slice_by_index(begin = var_11968_begin_0, end = var_11968_end_0, end_mask = var_11968_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11968_cast_fp16")]; + tensor var_11972_begin_0 = const()[name = tensor("op_11972_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11972_end_0 = const()[name = tensor("op_11972_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11972_end_mask_0 = const()[name = tensor("op_11972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11972_cast_fp16 = slice_by_index(begin = var_11972_begin_0, end = var_11972_end_0, end_mask = var_11972_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11972_cast_fp16")]; + tensor var_11976_begin_0 = const()[name = tensor("op_11976_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11976_end_0 = const()[name = tensor("op_11976_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11976_end_mask_0 = const()[name = tensor("op_11976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11976_cast_fp16 = slice_by_index(begin = var_11976_begin_0, end = var_11976_end_0, end_mask = var_11976_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11976_cast_fp16")]; + tensor var_11980_begin_0 = const()[name = tensor("op_11980_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11980_end_0 = const()[name = tensor("op_11980_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11980_end_mask_0 = const()[name = tensor("op_11980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11980_cast_fp16 = slice_by_index(begin = var_11980_begin_0, end = var_11980_end_0, end_mask = var_11980_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11980_cast_fp16")]; + tensor var_11984_begin_0 = const()[name = tensor("op_11984_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11984_end_0 = const()[name = tensor("op_11984_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11984_end_mask_0 = const()[name = tensor("op_11984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11984_cast_fp16 = slice_by_index(begin = var_11984_begin_0, end = var_11984_end_0, end_mask = var_11984_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11984_cast_fp16")]; + tensor var_11988_begin_0 = const()[name = tensor("op_11988_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11988_end_0 = const()[name = tensor("op_11988_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11988_end_mask_0 = const()[name = tensor("op_11988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11988_cast_fp16 = slice_by_index(begin = var_11988_begin_0, end = var_11988_end_0, end_mask = var_11988_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11988_cast_fp16")]; + tensor var_11992_begin_0 = const()[name = tensor("op_11992_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11992_end_0 = const()[name = tensor("op_11992_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11992_end_mask_0 = const()[name = tensor("op_11992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11992_cast_fp16 = slice_by_index(begin = var_11992_begin_0, end = var_11992_end_0, end_mask = var_11992_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11992_cast_fp16")]; + tensor var_11996_begin_0 = const()[name = tensor("op_11996_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11996_end_0 = const()[name = tensor("op_11996_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11996_end_mask_0 = const()[name = tensor("op_11996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11996_cast_fp16 = slice_by_index(begin = var_11996_begin_0, end = var_11996_end_0, end_mask = var_11996_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11996_cast_fp16")]; + tensor var_12000_begin_0 = const()[name = tensor("op_12000_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12000_end_0 = const()[name = tensor("op_12000_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12000_end_mask_0 = const()[name = tensor("op_12000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12000_cast_fp16 = slice_by_index(begin = var_12000_begin_0, end = var_12000_end_0, end_mask = var_12000_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12000_cast_fp16")]; + tensor var_12004_begin_0 = const()[name = tensor("op_12004_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12004_end_0 = const()[name = tensor("op_12004_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12004_end_mask_0 = const()[name = tensor("op_12004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12004_cast_fp16 = slice_by_index(begin = var_12004_begin_0, end = var_12004_end_0, end_mask = var_12004_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12004_cast_fp16")]; + tensor var_12008_begin_0 = const()[name = tensor("op_12008_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12008_end_0 = const()[name = tensor("op_12008_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12008_end_mask_0 = const()[name = tensor("op_12008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12008_cast_fp16 = slice_by_index(begin = var_12008_begin_0, end = var_12008_end_0, end_mask = var_12008_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12008_cast_fp16")]; + tensor var_12012_begin_0 = const()[name = tensor("op_12012_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12012_end_0 = const()[name = tensor("op_12012_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12012_end_mask_0 = const()[name = tensor("op_12012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12012_cast_fp16 = slice_by_index(begin = var_12012_begin_0, end = var_12012_end_0, end_mask = var_12012_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12012_cast_fp16")]; + tensor var_12016_begin_0 = const()[name = tensor("op_12016_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12016_end_0 = const()[name = tensor("op_12016_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12016_end_mask_0 = const()[name = tensor("op_12016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12016_cast_fp16 = slice_by_index(begin = var_12016_begin_0, end = var_12016_end_0, end_mask = var_12016_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12016_cast_fp16")]; + tensor var_12020_begin_0 = const()[name = tensor("op_12020_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12020_end_0 = const()[name = tensor("op_12020_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12020_end_mask_0 = const()[name = tensor("op_12020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12020_cast_fp16 = slice_by_index(begin = var_12020_begin_0, end = var_12020_end_0, end_mask = var_12020_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12020_cast_fp16")]; + tensor var_12024_begin_0 = const()[name = tensor("op_12024_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12024_end_0 = const()[name = tensor("op_12024_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12024_end_mask_0 = const()[name = tensor("op_12024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12024_cast_fp16 = slice_by_index(begin = var_12024_begin_0, end = var_12024_end_0, end_mask = var_12024_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12024_cast_fp16")]; + tensor var_12028_begin_0 = const()[name = tensor("op_12028_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12028_end_0 = const()[name = tensor("op_12028_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12028_end_mask_0 = const()[name = tensor("op_12028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12028_cast_fp16 = slice_by_index(begin = var_12028_begin_0, end = var_12028_end_0, end_mask = var_12028_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12028_cast_fp16")]; + tensor var_12032_begin_0 = const()[name = tensor("op_12032_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12032_end_0 = const()[name = tensor("op_12032_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12032_end_mask_0 = const()[name = tensor("op_12032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12032_cast_fp16 = slice_by_index(begin = var_12032_begin_0, end = var_12032_end_0, end_mask = var_12032_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12032_cast_fp16")]; + tensor var_12036_begin_0 = const()[name = tensor("op_12036_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12036_end_0 = const()[name = tensor("op_12036_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12036_end_mask_0 = const()[name = tensor("op_12036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12036_cast_fp16 = slice_by_index(begin = var_12036_begin_0, end = var_12036_end_0, end_mask = var_12036_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12036_cast_fp16")]; + tensor var_12040_begin_0 = const()[name = tensor("op_12040_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12040_end_0 = const()[name = tensor("op_12040_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12040_end_mask_0 = const()[name = tensor("op_12040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12040_cast_fp16 = slice_by_index(begin = var_12040_begin_0, end = var_12040_end_0, end_mask = var_12040_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12040_cast_fp16")]; + tensor var_12044_begin_0 = const()[name = tensor("op_12044_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12044_end_0 = const()[name = tensor("op_12044_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12044_end_mask_0 = const()[name = tensor("op_12044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12044_cast_fp16 = slice_by_index(begin = var_12044_begin_0, end = var_12044_end_0, end_mask = var_12044_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12044_cast_fp16")]; + tensor k_115_perm_0 = const()[name = tensor("k_115_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12051_begin_0 = const()[name = tensor("op_12051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12051_end_0 = const()[name = tensor("op_12051_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_12051_end_mask_0 = const()[name = tensor("op_12051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_115_cast_fp16 = transpose(perm = k_115_perm_0, x = k_113_cast_fp16)[name = tensor("transpose_39")]; + tensor var_12051_cast_fp16 = slice_by_index(begin = var_12051_begin_0, end = var_12051_end_0, end_mask = var_12051_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12051_cast_fp16")]; + tensor var_12055_begin_0 = const()[name = tensor("op_12055_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12055_end_0 = const()[name = tensor("op_12055_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_12055_end_mask_0 = const()[name = tensor("op_12055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12055_cast_fp16 = slice_by_index(begin = var_12055_begin_0, end = var_12055_end_0, end_mask = var_12055_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12055_cast_fp16")]; + tensor var_12059_begin_0 = const()[name = tensor("op_12059_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12059_end_0 = const()[name = tensor("op_12059_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_12059_end_mask_0 = const()[name = tensor("op_12059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12059_cast_fp16 = slice_by_index(begin = var_12059_begin_0, end = var_12059_end_0, end_mask = var_12059_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12059_cast_fp16")]; + tensor var_12063_begin_0 = const()[name = tensor("op_12063_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12063_end_0 = const()[name = tensor("op_12063_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_12063_end_mask_0 = const()[name = tensor("op_12063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12063_cast_fp16 = slice_by_index(begin = var_12063_begin_0, end = var_12063_end_0, end_mask = var_12063_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12063_cast_fp16")]; + tensor var_12067_begin_0 = const()[name = tensor("op_12067_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12067_end_0 = const()[name = tensor("op_12067_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_12067_end_mask_0 = const()[name = tensor("op_12067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12067_cast_fp16 = slice_by_index(begin = var_12067_begin_0, end = var_12067_end_0, end_mask = var_12067_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12067_cast_fp16")]; + tensor var_12071_begin_0 = const()[name = tensor("op_12071_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12071_end_0 = const()[name = tensor("op_12071_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_12071_end_mask_0 = const()[name = tensor("op_12071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12071_cast_fp16 = slice_by_index(begin = var_12071_begin_0, end = var_12071_end_0, end_mask = var_12071_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12071_cast_fp16")]; + tensor var_12075_begin_0 = const()[name = tensor("op_12075_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12075_end_0 = const()[name = tensor("op_12075_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_12075_end_mask_0 = const()[name = tensor("op_12075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12075_cast_fp16 = slice_by_index(begin = var_12075_begin_0, end = var_12075_end_0, end_mask = var_12075_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12075_cast_fp16")]; + tensor var_12079_begin_0 = const()[name = tensor("op_12079_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12079_end_0 = const()[name = tensor("op_12079_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_12079_end_mask_0 = const()[name = tensor("op_12079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12079_cast_fp16 = slice_by_index(begin = var_12079_begin_0, end = var_12079_end_0, end_mask = var_12079_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12079_cast_fp16")]; + tensor var_12083_begin_0 = const()[name = tensor("op_12083_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12083_end_0 = const()[name = tensor("op_12083_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_12083_end_mask_0 = const()[name = tensor("op_12083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12083_cast_fp16 = slice_by_index(begin = var_12083_begin_0, end = var_12083_end_0, end_mask = var_12083_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12083_cast_fp16")]; + tensor var_12087_begin_0 = const()[name = tensor("op_12087_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12087_end_0 = const()[name = tensor("op_12087_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_12087_end_mask_0 = const()[name = tensor("op_12087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12087_cast_fp16 = slice_by_index(begin = var_12087_begin_0, end = var_12087_end_0, end_mask = var_12087_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12087_cast_fp16")]; + tensor var_12091_begin_0 = const()[name = tensor("op_12091_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_12091_end_0 = const()[name = tensor("op_12091_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_12091_end_mask_0 = const()[name = tensor("op_12091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12091_cast_fp16 = slice_by_index(begin = var_12091_begin_0, end = var_12091_end_0, end_mask = var_12091_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12091_cast_fp16")]; + tensor var_12095_begin_0 = const()[name = tensor("op_12095_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_12095_end_0 = const()[name = tensor("op_12095_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_12095_end_mask_0 = const()[name = tensor("op_12095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12095_cast_fp16 = slice_by_index(begin = var_12095_begin_0, end = var_12095_end_0, end_mask = var_12095_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12095_cast_fp16")]; + tensor var_12099_begin_0 = const()[name = tensor("op_12099_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_12099_end_0 = const()[name = tensor("op_12099_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_12099_end_mask_0 = const()[name = tensor("op_12099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12099_cast_fp16 = slice_by_index(begin = var_12099_begin_0, end = var_12099_end_0, end_mask = var_12099_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12099_cast_fp16")]; + tensor var_12103_begin_0 = const()[name = tensor("op_12103_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_12103_end_0 = const()[name = tensor("op_12103_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_12103_end_mask_0 = const()[name = tensor("op_12103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12103_cast_fp16 = slice_by_index(begin = var_12103_begin_0, end = var_12103_end_0, end_mask = var_12103_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12103_cast_fp16")]; + tensor var_12107_begin_0 = const()[name = tensor("op_12107_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_12107_end_0 = const()[name = tensor("op_12107_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_12107_end_mask_0 = const()[name = tensor("op_12107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12107_cast_fp16 = slice_by_index(begin = var_12107_begin_0, end = var_12107_end_0, end_mask = var_12107_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12107_cast_fp16")]; + tensor var_12111_begin_0 = const()[name = tensor("op_12111_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_12111_end_0 = const()[name = tensor("op_12111_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12111_end_mask_0 = const()[name = tensor("op_12111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12111_cast_fp16 = slice_by_index(begin = var_12111_begin_0, end = var_12111_end_0, end_mask = var_12111_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12111_cast_fp16")]; + tensor var_12115_begin_0 = const()[name = tensor("op_12115_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12115_end_0 = const()[name = tensor("op_12115_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_12115_end_mask_0 = const()[name = tensor("op_12115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12115_cast_fp16 = slice_by_index(begin = var_12115_begin_0, end = var_12115_end_0, end_mask = var_12115_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12115_cast_fp16")]; + tensor var_12119_begin_0 = const()[name = tensor("op_12119_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_12119_end_0 = const()[name = tensor("op_12119_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_12119_end_mask_0 = const()[name = tensor("op_12119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12119_cast_fp16 = slice_by_index(begin = var_12119_begin_0, end = var_12119_end_0, end_mask = var_12119_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12119_cast_fp16")]; + tensor var_12123_begin_0 = const()[name = tensor("op_12123_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_12123_end_0 = const()[name = tensor("op_12123_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_12123_end_mask_0 = const()[name = tensor("op_12123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12123_cast_fp16 = slice_by_index(begin = var_12123_begin_0, end = var_12123_end_0, end_mask = var_12123_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12123_cast_fp16")]; + tensor var_12127_begin_0 = const()[name = tensor("op_12127_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_12127_end_0 = const()[name = tensor("op_12127_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_12127_end_mask_0 = const()[name = tensor("op_12127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12127_cast_fp16 = slice_by_index(begin = var_12127_begin_0, end = var_12127_end_0, end_mask = var_12127_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12127_cast_fp16")]; + tensor var_12129_begin_0 = const()[name = tensor("op_12129_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12129_end_0 = const()[name = tensor("op_12129_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12129_end_mask_0 = const()[name = tensor("op_12129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12129_cast_fp16 = slice_by_index(begin = var_12129_begin_0, end = var_12129_end_0, end_mask = var_12129_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12129_cast_fp16")]; + tensor var_12133_begin_0 = const()[name = tensor("op_12133_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12133_end_0 = const()[name = tensor("op_12133_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12133_end_mask_0 = const()[name = tensor("op_12133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12133_cast_fp16 = slice_by_index(begin = var_12133_begin_0, end = var_12133_end_0, end_mask = var_12133_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12133_cast_fp16")]; + tensor var_12137_begin_0 = const()[name = tensor("op_12137_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12137_end_0 = const()[name = tensor("op_12137_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12137_end_mask_0 = const()[name = tensor("op_12137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12137_cast_fp16 = slice_by_index(begin = var_12137_begin_0, end = var_12137_end_0, end_mask = var_12137_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12137_cast_fp16")]; + tensor var_12141_begin_0 = const()[name = tensor("op_12141_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12141_end_0 = const()[name = tensor("op_12141_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12141_end_mask_0 = const()[name = tensor("op_12141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12141_cast_fp16 = slice_by_index(begin = var_12141_begin_0, end = var_12141_end_0, end_mask = var_12141_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12141_cast_fp16")]; + tensor var_12145_begin_0 = const()[name = tensor("op_12145_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12145_end_0 = const()[name = tensor("op_12145_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12145_end_mask_0 = const()[name = tensor("op_12145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12145_cast_fp16 = slice_by_index(begin = var_12145_begin_0, end = var_12145_end_0, end_mask = var_12145_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12145_cast_fp16")]; + tensor var_12149_begin_0 = const()[name = tensor("op_12149_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12149_end_0 = const()[name = tensor("op_12149_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12149_end_mask_0 = const()[name = tensor("op_12149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12149_cast_fp16 = slice_by_index(begin = var_12149_begin_0, end = var_12149_end_0, end_mask = var_12149_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12149_cast_fp16")]; + tensor var_12153_begin_0 = const()[name = tensor("op_12153_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12153_end_0 = const()[name = tensor("op_12153_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12153_end_mask_0 = const()[name = tensor("op_12153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12153_cast_fp16 = slice_by_index(begin = var_12153_begin_0, end = var_12153_end_0, end_mask = var_12153_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12153_cast_fp16")]; + tensor var_12157_begin_0 = const()[name = tensor("op_12157_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12157_end_0 = const()[name = tensor("op_12157_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12157_end_mask_0 = const()[name = tensor("op_12157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12157_cast_fp16 = slice_by_index(begin = var_12157_begin_0, end = var_12157_end_0, end_mask = var_12157_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12157_cast_fp16")]; + tensor var_12161_begin_0 = const()[name = tensor("op_12161_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12161_end_0 = const()[name = tensor("op_12161_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12161_end_mask_0 = const()[name = tensor("op_12161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12161_cast_fp16 = slice_by_index(begin = var_12161_begin_0, end = var_12161_end_0, end_mask = var_12161_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12161_cast_fp16")]; + tensor var_12165_begin_0 = const()[name = tensor("op_12165_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12165_end_0 = const()[name = tensor("op_12165_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12165_end_mask_0 = const()[name = tensor("op_12165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12165_cast_fp16 = slice_by_index(begin = var_12165_begin_0, end = var_12165_end_0, end_mask = var_12165_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12165_cast_fp16")]; + tensor var_12169_begin_0 = const()[name = tensor("op_12169_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12169_end_0 = const()[name = tensor("op_12169_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12169_end_mask_0 = const()[name = tensor("op_12169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12169_cast_fp16 = slice_by_index(begin = var_12169_begin_0, end = var_12169_end_0, end_mask = var_12169_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12169_cast_fp16")]; + tensor var_12173_begin_0 = const()[name = tensor("op_12173_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12173_end_0 = const()[name = tensor("op_12173_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12173_end_mask_0 = const()[name = tensor("op_12173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12173_cast_fp16 = slice_by_index(begin = var_12173_begin_0, end = var_12173_end_0, end_mask = var_12173_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12173_cast_fp16")]; + tensor var_12177_begin_0 = const()[name = tensor("op_12177_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12177_end_0 = const()[name = tensor("op_12177_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12177_end_mask_0 = const()[name = tensor("op_12177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12177_cast_fp16 = slice_by_index(begin = var_12177_begin_0, end = var_12177_end_0, end_mask = var_12177_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12177_cast_fp16")]; + tensor var_12181_begin_0 = const()[name = tensor("op_12181_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12181_end_0 = const()[name = tensor("op_12181_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12181_end_mask_0 = const()[name = tensor("op_12181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12181_cast_fp16 = slice_by_index(begin = var_12181_begin_0, end = var_12181_end_0, end_mask = var_12181_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12181_cast_fp16")]; + tensor var_12185_begin_0 = const()[name = tensor("op_12185_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12185_end_0 = const()[name = tensor("op_12185_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12185_end_mask_0 = const()[name = tensor("op_12185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12185_cast_fp16 = slice_by_index(begin = var_12185_begin_0, end = var_12185_end_0, end_mask = var_12185_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12185_cast_fp16")]; + tensor var_12189_begin_0 = const()[name = tensor("op_12189_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12189_end_0 = const()[name = tensor("op_12189_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12189_end_mask_0 = const()[name = tensor("op_12189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12189_cast_fp16 = slice_by_index(begin = var_12189_begin_0, end = var_12189_end_0, end_mask = var_12189_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12189_cast_fp16")]; + tensor var_12193_begin_0 = const()[name = tensor("op_12193_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12193_end_0 = const()[name = tensor("op_12193_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12193_end_mask_0 = const()[name = tensor("op_12193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12193_cast_fp16 = slice_by_index(begin = var_12193_begin_0, end = var_12193_end_0, end_mask = var_12193_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12193_cast_fp16")]; + tensor var_12197_begin_0 = const()[name = tensor("op_12197_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12197_end_0 = const()[name = tensor("op_12197_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12197_end_mask_0 = const()[name = tensor("op_12197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12197_cast_fp16 = slice_by_index(begin = var_12197_begin_0, end = var_12197_end_0, end_mask = var_12197_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12197_cast_fp16")]; + tensor var_12201_begin_0 = const()[name = tensor("op_12201_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12201_end_0 = const()[name = tensor("op_12201_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12201_end_mask_0 = const()[name = tensor("op_12201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12201_cast_fp16 = slice_by_index(begin = var_12201_begin_0, end = var_12201_end_0, end_mask = var_12201_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12201_cast_fp16")]; + tensor var_12205_begin_0 = const()[name = tensor("op_12205_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12205_end_0 = const()[name = tensor("op_12205_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12205_end_mask_0 = const()[name = tensor("op_12205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12205_cast_fp16 = slice_by_index(begin = var_12205_begin_0, end = var_12205_end_0, end_mask = var_12205_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12205_cast_fp16")]; + tensor var_12209_equation_0 = const()[name = tensor("op_12209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12209_cast_fp16 = einsum(equation = var_12209_equation_0, values = (var_12051_cast_fp16, var_11968_cast_fp16))[name = tensor("op_12209_cast_fp16")]; + tensor var_12210_to_fp16 = const()[name = tensor("op_12210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_961_cast_fp16 = mul(x = var_12209_cast_fp16, y = var_12210_to_fp16)[name = tensor("aw_961_cast_fp16")]; + tensor var_12213_equation_0 = const()[name = tensor("op_12213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12213_cast_fp16 = einsum(equation = var_12213_equation_0, values = (var_12055_cast_fp16, var_11972_cast_fp16))[name = tensor("op_12213_cast_fp16")]; + tensor var_12214_to_fp16 = const()[name = tensor("op_12214_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_963_cast_fp16 = mul(x = var_12213_cast_fp16, y = var_12214_to_fp16)[name = tensor("aw_963_cast_fp16")]; + tensor var_12217_equation_0 = const()[name = tensor("op_12217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12217_cast_fp16 = einsum(equation = var_12217_equation_0, values = (var_12059_cast_fp16, var_11976_cast_fp16))[name = tensor("op_12217_cast_fp16")]; + tensor var_12218_to_fp16 = const()[name = tensor("op_12218_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_965_cast_fp16 = mul(x = var_12217_cast_fp16, y = var_12218_to_fp16)[name = tensor("aw_965_cast_fp16")]; + tensor var_12221_equation_0 = const()[name = tensor("op_12221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12221_cast_fp16 = einsum(equation = var_12221_equation_0, values = (var_12063_cast_fp16, var_11980_cast_fp16))[name = tensor("op_12221_cast_fp16")]; + tensor var_12222_to_fp16 = const()[name = tensor("op_12222_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_967_cast_fp16 = mul(x = var_12221_cast_fp16, y = var_12222_to_fp16)[name = tensor("aw_967_cast_fp16")]; + tensor var_12225_equation_0 = const()[name = tensor("op_12225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12225_cast_fp16 = einsum(equation = var_12225_equation_0, values = (var_12067_cast_fp16, var_11984_cast_fp16))[name = tensor("op_12225_cast_fp16")]; + tensor var_12226_to_fp16 = const()[name = tensor("op_12226_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_969_cast_fp16 = mul(x = var_12225_cast_fp16, y = var_12226_to_fp16)[name = tensor("aw_969_cast_fp16")]; + tensor var_12229_equation_0 = const()[name = tensor("op_12229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12229_cast_fp16 = einsum(equation = var_12229_equation_0, values = (var_12071_cast_fp16, var_11988_cast_fp16))[name = tensor("op_12229_cast_fp16")]; + tensor var_12230_to_fp16 = const()[name = tensor("op_12230_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_971_cast_fp16 = mul(x = var_12229_cast_fp16, y = var_12230_to_fp16)[name = tensor("aw_971_cast_fp16")]; + tensor var_12233_equation_0 = const()[name = tensor("op_12233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12233_cast_fp16 = einsum(equation = var_12233_equation_0, values = (var_12075_cast_fp16, var_11992_cast_fp16))[name = tensor("op_12233_cast_fp16")]; + tensor var_12234_to_fp16 = const()[name = tensor("op_12234_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_973_cast_fp16 = mul(x = var_12233_cast_fp16, y = var_12234_to_fp16)[name = tensor("aw_973_cast_fp16")]; + tensor var_12237_equation_0 = const()[name = tensor("op_12237_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12237_cast_fp16 = einsum(equation = var_12237_equation_0, values = (var_12079_cast_fp16, var_11996_cast_fp16))[name = tensor("op_12237_cast_fp16")]; + tensor var_12238_to_fp16 = const()[name = tensor("op_12238_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_975_cast_fp16 = mul(x = var_12237_cast_fp16, y = var_12238_to_fp16)[name = tensor("aw_975_cast_fp16")]; + tensor var_12241_equation_0 = const()[name = tensor("op_12241_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12241_cast_fp16 = einsum(equation = var_12241_equation_0, values = (var_12083_cast_fp16, var_12000_cast_fp16))[name = tensor("op_12241_cast_fp16")]; + tensor var_12242_to_fp16 = const()[name = tensor("op_12242_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_977_cast_fp16 = mul(x = var_12241_cast_fp16, y = var_12242_to_fp16)[name = tensor("aw_977_cast_fp16")]; + tensor var_12245_equation_0 = const()[name = tensor("op_12245_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12245_cast_fp16 = einsum(equation = var_12245_equation_0, values = (var_12087_cast_fp16, var_12004_cast_fp16))[name = tensor("op_12245_cast_fp16")]; + tensor var_12246_to_fp16 = const()[name = tensor("op_12246_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_979_cast_fp16 = mul(x = var_12245_cast_fp16, y = var_12246_to_fp16)[name = tensor("aw_979_cast_fp16")]; + tensor var_12249_equation_0 = const()[name = tensor("op_12249_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12249_cast_fp16 = einsum(equation = var_12249_equation_0, values = (var_12091_cast_fp16, var_12008_cast_fp16))[name = tensor("op_12249_cast_fp16")]; + tensor var_12250_to_fp16 = const()[name = tensor("op_12250_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_981_cast_fp16 = mul(x = var_12249_cast_fp16, y = var_12250_to_fp16)[name = tensor("aw_981_cast_fp16")]; + tensor var_12253_equation_0 = const()[name = tensor("op_12253_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12253_cast_fp16 = einsum(equation = var_12253_equation_0, values = (var_12095_cast_fp16, var_12012_cast_fp16))[name = tensor("op_12253_cast_fp16")]; + tensor var_12254_to_fp16 = const()[name = tensor("op_12254_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_983_cast_fp16 = mul(x = var_12253_cast_fp16, y = var_12254_to_fp16)[name = tensor("aw_983_cast_fp16")]; + tensor var_12257_equation_0 = const()[name = tensor("op_12257_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12257_cast_fp16 = einsum(equation = var_12257_equation_0, values = (var_12099_cast_fp16, var_12016_cast_fp16))[name = tensor("op_12257_cast_fp16")]; + tensor var_12258_to_fp16 = const()[name = tensor("op_12258_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_985_cast_fp16 = mul(x = var_12257_cast_fp16, y = var_12258_to_fp16)[name = tensor("aw_985_cast_fp16")]; + tensor var_12261_equation_0 = const()[name = tensor("op_12261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12261_cast_fp16 = einsum(equation = var_12261_equation_0, values = (var_12103_cast_fp16, var_12020_cast_fp16))[name = tensor("op_12261_cast_fp16")]; + tensor var_12262_to_fp16 = const()[name = tensor("op_12262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_987_cast_fp16 = mul(x = var_12261_cast_fp16, y = var_12262_to_fp16)[name = tensor("aw_987_cast_fp16")]; + tensor var_12265_equation_0 = const()[name = tensor("op_12265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12265_cast_fp16 = einsum(equation = var_12265_equation_0, values = (var_12107_cast_fp16, var_12024_cast_fp16))[name = tensor("op_12265_cast_fp16")]; + tensor var_12266_to_fp16 = const()[name = tensor("op_12266_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_989_cast_fp16 = mul(x = var_12265_cast_fp16, y = var_12266_to_fp16)[name = tensor("aw_989_cast_fp16")]; + tensor var_12269_equation_0 = const()[name = tensor("op_12269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12269_cast_fp16 = einsum(equation = var_12269_equation_0, values = (var_12111_cast_fp16, var_12028_cast_fp16))[name = tensor("op_12269_cast_fp16")]; + tensor var_12270_to_fp16 = const()[name = tensor("op_12270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_991_cast_fp16 = mul(x = var_12269_cast_fp16, y = var_12270_to_fp16)[name = tensor("aw_991_cast_fp16")]; + tensor var_12273_equation_0 = const()[name = tensor("op_12273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12273_cast_fp16 = einsum(equation = var_12273_equation_0, values = (var_12115_cast_fp16, var_12032_cast_fp16))[name = tensor("op_12273_cast_fp16")]; + tensor var_12274_to_fp16 = const()[name = tensor("op_12274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_993_cast_fp16 = mul(x = var_12273_cast_fp16, y = var_12274_to_fp16)[name = tensor("aw_993_cast_fp16")]; + tensor var_12277_equation_0 = const()[name = tensor("op_12277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12277_cast_fp16 = einsum(equation = var_12277_equation_0, values = (var_12119_cast_fp16, var_12036_cast_fp16))[name = tensor("op_12277_cast_fp16")]; + tensor var_12278_to_fp16 = const()[name = tensor("op_12278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_995_cast_fp16 = mul(x = var_12277_cast_fp16, y = var_12278_to_fp16)[name = tensor("aw_995_cast_fp16")]; + tensor var_12281_equation_0 = const()[name = tensor("op_12281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12281_cast_fp16 = einsum(equation = var_12281_equation_0, values = (var_12123_cast_fp16, var_12040_cast_fp16))[name = tensor("op_12281_cast_fp16")]; + tensor var_12282_to_fp16 = const()[name = tensor("op_12282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_997_cast_fp16 = mul(x = var_12281_cast_fp16, y = var_12282_to_fp16)[name = tensor("aw_997_cast_fp16")]; + tensor var_12285_equation_0 = const()[name = tensor("op_12285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12285_cast_fp16 = einsum(equation = var_12285_equation_0, values = (var_12127_cast_fp16, var_12044_cast_fp16))[name = tensor("op_12285_cast_fp16")]; + tensor var_12286_to_fp16 = const()[name = tensor("op_12286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_999_cast_fp16 = mul(x = var_12285_cast_fp16, y = var_12286_to_fp16)[name = tensor("aw_999_cast_fp16")]; + tensor var_12288_cast_fp16 = softmax(axis = var_2624, x = aw_961_cast_fp16)[name = tensor("op_12288_cast_fp16")]; + tensor var_12289_cast_fp16 = softmax(axis = var_2624, x = aw_963_cast_fp16)[name = tensor("op_12289_cast_fp16")]; + tensor var_12290_cast_fp16 = softmax(axis = var_2624, x = aw_965_cast_fp16)[name = tensor("op_12290_cast_fp16")]; + tensor var_12291_cast_fp16 = softmax(axis = var_2624, x = aw_967_cast_fp16)[name = tensor("op_12291_cast_fp16")]; + tensor var_12292_cast_fp16 = softmax(axis = var_2624, x = aw_969_cast_fp16)[name = tensor("op_12292_cast_fp16")]; + tensor var_12293_cast_fp16 = softmax(axis = var_2624, x = aw_971_cast_fp16)[name = tensor("op_12293_cast_fp16")]; + tensor var_12294_cast_fp16 = softmax(axis = var_2624, x = aw_973_cast_fp16)[name = tensor("op_12294_cast_fp16")]; + tensor var_12295_cast_fp16 = softmax(axis = var_2624, x = aw_975_cast_fp16)[name = tensor("op_12295_cast_fp16")]; + tensor var_12296_cast_fp16 = softmax(axis = var_2624, x = aw_977_cast_fp16)[name = tensor("op_12296_cast_fp16")]; + tensor var_12297_cast_fp16 = softmax(axis = var_2624, x = aw_979_cast_fp16)[name = tensor("op_12297_cast_fp16")]; + tensor var_12298_cast_fp16 = softmax(axis = var_2624, x = aw_981_cast_fp16)[name = tensor("op_12298_cast_fp16")]; + tensor var_12299_cast_fp16 = softmax(axis = var_2624, x = aw_983_cast_fp16)[name = tensor("op_12299_cast_fp16")]; + tensor var_12300_cast_fp16 = softmax(axis = var_2624, x = aw_985_cast_fp16)[name = tensor("op_12300_cast_fp16")]; + tensor var_12301_cast_fp16 = softmax(axis = var_2624, x = aw_987_cast_fp16)[name = tensor("op_12301_cast_fp16")]; + tensor var_12302_cast_fp16 = softmax(axis = var_2624, x = aw_989_cast_fp16)[name = tensor("op_12302_cast_fp16")]; + tensor var_12303_cast_fp16 = softmax(axis = var_2624, x = aw_991_cast_fp16)[name = tensor("op_12303_cast_fp16")]; + tensor var_12304_cast_fp16 = softmax(axis = var_2624, x = aw_993_cast_fp16)[name = tensor("op_12304_cast_fp16")]; + tensor var_12305_cast_fp16 = softmax(axis = var_2624, x = aw_995_cast_fp16)[name = tensor("op_12305_cast_fp16")]; + tensor var_12306_cast_fp16 = softmax(axis = var_2624, x = aw_997_cast_fp16)[name = tensor("op_12306_cast_fp16")]; + tensor var_12307_cast_fp16 = softmax(axis = var_2624, x = aw_999_cast_fp16)[name = tensor("op_12307_cast_fp16")]; + tensor var_12309_equation_0 = const()[name = tensor("op_12309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12309_cast_fp16 = einsum(equation = var_12309_equation_0, values = (var_12129_cast_fp16, var_12288_cast_fp16))[name = tensor("op_12309_cast_fp16")]; + tensor var_12311_equation_0 = const()[name = tensor("op_12311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12311_cast_fp16 = einsum(equation = var_12311_equation_0, values = (var_12133_cast_fp16, var_12289_cast_fp16))[name = tensor("op_12311_cast_fp16")]; + tensor var_12313_equation_0 = const()[name = tensor("op_12313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12313_cast_fp16 = einsum(equation = var_12313_equation_0, values = (var_12137_cast_fp16, var_12290_cast_fp16))[name = tensor("op_12313_cast_fp16")]; + tensor var_12315_equation_0 = const()[name = tensor("op_12315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12315_cast_fp16 = einsum(equation = var_12315_equation_0, values = (var_12141_cast_fp16, var_12291_cast_fp16))[name = tensor("op_12315_cast_fp16")]; + tensor var_12317_equation_0 = const()[name = tensor("op_12317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12317_cast_fp16 = einsum(equation = var_12317_equation_0, values = (var_12145_cast_fp16, var_12292_cast_fp16))[name = tensor("op_12317_cast_fp16")]; + tensor var_12319_equation_0 = const()[name = tensor("op_12319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12319_cast_fp16 = einsum(equation = var_12319_equation_0, values = (var_12149_cast_fp16, var_12293_cast_fp16))[name = tensor("op_12319_cast_fp16")]; + tensor var_12321_equation_0 = const()[name = tensor("op_12321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12321_cast_fp16 = einsum(equation = var_12321_equation_0, values = (var_12153_cast_fp16, var_12294_cast_fp16))[name = tensor("op_12321_cast_fp16")]; + tensor var_12323_equation_0 = const()[name = tensor("op_12323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12323_cast_fp16 = einsum(equation = var_12323_equation_0, values = (var_12157_cast_fp16, var_12295_cast_fp16))[name = tensor("op_12323_cast_fp16")]; + tensor var_12325_equation_0 = const()[name = tensor("op_12325_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12325_cast_fp16 = einsum(equation = var_12325_equation_0, values = (var_12161_cast_fp16, var_12296_cast_fp16))[name = tensor("op_12325_cast_fp16")]; + tensor var_12327_equation_0 = const()[name = tensor("op_12327_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12327_cast_fp16 = einsum(equation = var_12327_equation_0, values = (var_12165_cast_fp16, var_12297_cast_fp16))[name = tensor("op_12327_cast_fp16")]; + tensor var_12329_equation_0 = const()[name = tensor("op_12329_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12329_cast_fp16 = einsum(equation = var_12329_equation_0, values = (var_12169_cast_fp16, var_12298_cast_fp16))[name = tensor("op_12329_cast_fp16")]; + tensor var_12331_equation_0 = const()[name = tensor("op_12331_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12331_cast_fp16 = einsum(equation = var_12331_equation_0, values = (var_12173_cast_fp16, var_12299_cast_fp16))[name = tensor("op_12331_cast_fp16")]; + tensor var_12333_equation_0 = const()[name = tensor("op_12333_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12333_cast_fp16 = einsum(equation = var_12333_equation_0, values = (var_12177_cast_fp16, var_12300_cast_fp16))[name = tensor("op_12333_cast_fp16")]; + tensor var_12335_equation_0 = const()[name = tensor("op_12335_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12335_cast_fp16 = einsum(equation = var_12335_equation_0, values = (var_12181_cast_fp16, var_12301_cast_fp16))[name = tensor("op_12335_cast_fp16")]; + tensor var_12337_equation_0 = const()[name = tensor("op_12337_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12337_cast_fp16 = einsum(equation = var_12337_equation_0, values = (var_12185_cast_fp16, var_12302_cast_fp16))[name = tensor("op_12337_cast_fp16")]; + tensor var_12339_equation_0 = const()[name = tensor("op_12339_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12339_cast_fp16 = einsum(equation = var_12339_equation_0, values = (var_12189_cast_fp16, var_12303_cast_fp16))[name = tensor("op_12339_cast_fp16")]; + tensor var_12341_equation_0 = const()[name = tensor("op_12341_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12341_cast_fp16 = einsum(equation = var_12341_equation_0, values = (var_12193_cast_fp16, var_12304_cast_fp16))[name = tensor("op_12341_cast_fp16")]; + tensor var_12343_equation_0 = const()[name = tensor("op_12343_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12343_cast_fp16 = einsum(equation = var_12343_equation_0, values = (var_12197_cast_fp16, var_12305_cast_fp16))[name = tensor("op_12343_cast_fp16")]; + tensor var_12345_equation_0 = const()[name = tensor("op_12345_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12345_cast_fp16 = einsum(equation = var_12345_equation_0, values = (var_12201_cast_fp16, var_12306_cast_fp16))[name = tensor("op_12345_cast_fp16")]; + tensor var_12347_equation_0 = const()[name = tensor("op_12347_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12347_cast_fp16 = einsum(equation = var_12347_equation_0, values = (var_12205_cast_fp16, var_12307_cast_fp16))[name = tensor("op_12347_cast_fp16")]; + tensor input_229_interleave_0 = const()[name = tensor("input_229_interleave_0"), val = tensor(false)]; + tensor input_229_cast_fp16 = concat(axis = var_2624, interleave = input_229_interleave_0, values = (var_12309_cast_fp16, var_12311_cast_fp16, var_12313_cast_fp16, var_12315_cast_fp16, var_12317_cast_fp16, var_12319_cast_fp16, var_12321_cast_fp16, var_12323_cast_fp16, var_12325_cast_fp16, var_12327_cast_fp16, var_12329_cast_fp16, var_12331_cast_fp16, var_12333_cast_fp16, var_12335_cast_fp16, var_12337_cast_fp16, var_12339_cast_fp16, var_12341_cast_fp16, var_12343_cast_fp16, var_12345_cast_fp16, var_12347_cast_fp16))[name = tensor("input_229_cast_fp16")]; + tensor var_12357_pad_type_0 = const()[name = tensor("op_12357_pad_type_0"), val = tensor("valid")]; + tensor var_12357_strides_0 = const()[name = tensor("op_12357_strides_0"), val = tensor([1, 1])]; + tensor var_12357_pad_0 = const()[name = tensor("op_12357_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12357_dilations_0 = const()[name = tensor("op_12357_dilations_0"), val = tensor([1, 1])]; + tensor var_12357_groups_0 = const()[name = tensor("op_12357_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364940096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366168960))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366169152)))]; + tensor var_12357_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_12357_dilations_0, groups = var_12357_groups_0, pad = var_12357_pad_0, pad_type = var_12357_pad_type_0, strides = var_12357_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = tensor("op_12357_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = var_12357_cast_fp16, y = inputs_85_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; + tensor hidden_states_139_axes_0 = const()[name = tensor("hidden_states_139_axes_0"), val = tensor([1])]; + tensor hidden_states_139_gamma_0_to_fp16 = const()[name = tensor("hidden_states_139_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366171776)))]; + tensor hidden_states_139_beta_0_to_fp16 = const()[name = tensor("hidden_states_139_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366174400)))]; + tensor var_12367_to_fp16 = const()[name = tensor("op_12367_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_139_cast_fp16 = layer_norm(axes = hidden_states_139_axes_0, beta = hidden_states_139_beta_0_to_fp16, epsilon = var_12367_to_fp16, gamma = hidden_states_139_gamma_0_to_fp16, x = inputs_87_cast_fp16)[name = tensor("hidden_states_139_cast_fp16")]; + tensor q_59_pad_type_0 = const()[name = tensor("q_59_pad_type_0"), val = tensor("valid")]; + tensor q_59_strides_0 = const()[name = tensor("q_59_strides_0"), val = tensor([1, 1])]; + tensor q_59_pad_0 = const()[name = tensor("q_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_59_dilations_0 = const()[name = tensor("q_59_dilations_0"), val = tensor([1, 1])]; + tensor q_59_groups_0 = const()[name = tensor("q_59_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366177024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367405888))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_59_cast_fp16 = conv(dilations = q_59_dilations_0, groups = q_59_groups_0, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = q_59_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_139_cast_fp16)[name = tensor("q_59_cast_fp16")]; + tensor k_117_pad_type_0 = const()[name = tensor("k_117_pad_type_0"), val = tensor("valid")]; + tensor k_117_strides_0 = const()[name = tensor("k_117_strides_0"), val = tensor([1, 1])]; + tensor k_117_pad_0 = const()[name = tensor("k_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_117_dilations_0 = const()[name = tensor("k_117_dilations_0"), val = tensor([1, 1])]; + tensor k_117_groups_0 = const()[name = tensor("k_117_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367406080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(369372224))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_117_cast_fp16 = conv(dilations = k_117_dilations_0, groups = k_117_groups_0, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = k_117_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_117_cast_fp16")]; + tensor v_59_pad_type_0 = const()[name = tensor("v_59_pad_type_0"), val = tensor("valid")]; + tensor v_59_strides_0 = const()[name = tensor("v_59_strides_0"), val = tensor([1, 1])]; + tensor v_59_pad_0 = const()[name = tensor("v_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_59_dilations_0 = const()[name = tensor("v_59_dilations_0"), val = tensor([1, 1])]; + tensor v_59_groups_0 = const()[name = tensor("v_59_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(369372416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371338560))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_59_cast_fp16 = conv(dilations = v_59_dilations_0, groups = v_59_groups_0, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = v_59_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_59_cast_fp16")]; + tensor var_12400_begin_0 = const()[name = tensor("op_12400_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12400_end_0 = const()[name = tensor("op_12400_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12400_end_mask_0 = const()[name = tensor("op_12400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12400_cast_fp16 = slice_by_index(begin = var_12400_begin_0, end = var_12400_end_0, end_mask = var_12400_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12400_cast_fp16")]; + tensor var_12404_begin_0 = const()[name = tensor("op_12404_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12404_end_0 = const()[name = tensor("op_12404_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12404_end_mask_0 = const()[name = tensor("op_12404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12404_cast_fp16 = slice_by_index(begin = var_12404_begin_0, end = var_12404_end_0, end_mask = var_12404_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12404_cast_fp16")]; + tensor var_12408_begin_0 = const()[name = tensor("op_12408_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12408_end_0 = const()[name = tensor("op_12408_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12408_end_mask_0 = const()[name = tensor("op_12408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12408_cast_fp16 = slice_by_index(begin = var_12408_begin_0, end = var_12408_end_0, end_mask = var_12408_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12408_cast_fp16")]; + tensor var_12412_begin_0 = const()[name = tensor("op_12412_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12412_end_0 = const()[name = tensor("op_12412_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12412_end_mask_0 = const()[name = tensor("op_12412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12412_cast_fp16 = slice_by_index(begin = var_12412_begin_0, end = var_12412_end_0, end_mask = var_12412_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12412_cast_fp16")]; + tensor var_12416_begin_0 = const()[name = tensor("op_12416_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12416_end_0 = const()[name = tensor("op_12416_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12416_end_mask_0 = const()[name = tensor("op_12416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12416_cast_fp16 = slice_by_index(begin = var_12416_begin_0, end = var_12416_end_0, end_mask = var_12416_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12416_cast_fp16")]; + tensor var_12420_begin_0 = const()[name = tensor("op_12420_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12420_end_0 = const()[name = tensor("op_12420_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12420_end_mask_0 = const()[name = tensor("op_12420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12420_cast_fp16 = slice_by_index(begin = var_12420_begin_0, end = var_12420_end_0, end_mask = var_12420_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12420_cast_fp16")]; + tensor var_12424_begin_0 = const()[name = tensor("op_12424_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12424_end_0 = const()[name = tensor("op_12424_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12424_end_mask_0 = const()[name = tensor("op_12424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12424_cast_fp16 = slice_by_index(begin = var_12424_begin_0, end = var_12424_end_0, end_mask = var_12424_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12424_cast_fp16")]; + tensor var_12428_begin_0 = const()[name = tensor("op_12428_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12428_end_0 = const()[name = tensor("op_12428_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12428_end_mask_0 = const()[name = tensor("op_12428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12428_cast_fp16 = slice_by_index(begin = var_12428_begin_0, end = var_12428_end_0, end_mask = var_12428_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12428_cast_fp16")]; + tensor var_12432_begin_0 = const()[name = tensor("op_12432_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12432_end_0 = const()[name = tensor("op_12432_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12432_end_mask_0 = const()[name = tensor("op_12432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12432_cast_fp16 = slice_by_index(begin = var_12432_begin_0, end = var_12432_end_0, end_mask = var_12432_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12432_cast_fp16")]; + tensor var_12436_begin_0 = const()[name = tensor("op_12436_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12436_end_0 = const()[name = tensor("op_12436_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12436_end_mask_0 = const()[name = tensor("op_12436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12436_cast_fp16 = slice_by_index(begin = var_12436_begin_0, end = var_12436_end_0, end_mask = var_12436_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12436_cast_fp16")]; + tensor var_12440_begin_0 = const()[name = tensor("op_12440_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12440_end_0 = const()[name = tensor("op_12440_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12440_end_mask_0 = const()[name = tensor("op_12440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12440_cast_fp16 = slice_by_index(begin = var_12440_begin_0, end = var_12440_end_0, end_mask = var_12440_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12440_cast_fp16")]; + tensor var_12444_begin_0 = const()[name = tensor("op_12444_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12444_end_0 = const()[name = tensor("op_12444_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12444_end_mask_0 = const()[name = tensor("op_12444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12444_cast_fp16 = slice_by_index(begin = var_12444_begin_0, end = var_12444_end_0, end_mask = var_12444_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12444_cast_fp16")]; + tensor var_12448_begin_0 = const()[name = tensor("op_12448_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12448_end_0 = const()[name = tensor("op_12448_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12448_end_mask_0 = const()[name = tensor("op_12448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12448_cast_fp16 = slice_by_index(begin = var_12448_begin_0, end = var_12448_end_0, end_mask = var_12448_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12448_cast_fp16")]; + tensor var_12452_begin_0 = const()[name = tensor("op_12452_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12452_end_0 = const()[name = tensor("op_12452_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12452_end_mask_0 = const()[name = tensor("op_12452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12452_cast_fp16 = slice_by_index(begin = var_12452_begin_0, end = var_12452_end_0, end_mask = var_12452_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12452_cast_fp16")]; + tensor var_12456_begin_0 = const()[name = tensor("op_12456_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12456_end_0 = const()[name = tensor("op_12456_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12456_end_mask_0 = const()[name = tensor("op_12456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12456_cast_fp16 = slice_by_index(begin = var_12456_begin_0, end = var_12456_end_0, end_mask = var_12456_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12456_cast_fp16")]; + tensor var_12460_begin_0 = const()[name = tensor("op_12460_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12460_end_0 = const()[name = tensor("op_12460_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12460_end_mask_0 = const()[name = tensor("op_12460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12460_cast_fp16 = slice_by_index(begin = var_12460_begin_0, end = var_12460_end_0, end_mask = var_12460_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12460_cast_fp16")]; + tensor var_12464_begin_0 = const()[name = tensor("op_12464_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12464_end_0 = const()[name = tensor("op_12464_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12464_end_mask_0 = const()[name = tensor("op_12464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12464_cast_fp16 = slice_by_index(begin = var_12464_begin_0, end = var_12464_end_0, end_mask = var_12464_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12464_cast_fp16")]; + tensor var_12468_begin_0 = const()[name = tensor("op_12468_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12468_end_0 = const()[name = tensor("op_12468_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12468_end_mask_0 = const()[name = tensor("op_12468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12468_cast_fp16 = slice_by_index(begin = var_12468_begin_0, end = var_12468_end_0, end_mask = var_12468_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12468_cast_fp16")]; + tensor var_12472_begin_0 = const()[name = tensor("op_12472_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12472_end_0 = const()[name = tensor("op_12472_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12472_end_mask_0 = const()[name = tensor("op_12472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12472_cast_fp16 = slice_by_index(begin = var_12472_begin_0, end = var_12472_end_0, end_mask = var_12472_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12472_cast_fp16")]; + tensor var_12476_begin_0 = const()[name = tensor("op_12476_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12476_end_0 = const()[name = tensor("op_12476_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12476_end_mask_0 = const()[name = tensor("op_12476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12476_cast_fp16 = slice_by_index(begin = var_12476_begin_0, end = var_12476_end_0, end_mask = var_12476_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12476_cast_fp16")]; + tensor k_119_perm_0 = const()[name = tensor("k_119_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12483_begin_0 = const()[name = tensor("op_12483_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12483_end_0 = const()[name = tensor("op_12483_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_12483_end_mask_0 = const()[name = tensor("op_12483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_119_cast_fp16 = transpose(perm = k_119_perm_0, x = k_117_cast_fp16)[name = tensor("transpose_38")]; + tensor var_12483_cast_fp16 = slice_by_index(begin = var_12483_begin_0, end = var_12483_end_0, end_mask = var_12483_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12483_cast_fp16")]; + tensor var_12487_begin_0 = const()[name = tensor("op_12487_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12487_end_0 = const()[name = tensor("op_12487_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_12487_end_mask_0 = const()[name = tensor("op_12487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12487_cast_fp16 = slice_by_index(begin = var_12487_begin_0, end = var_12487_end_0, end_mask = var_12487_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12487_cast_fp16")]; + tensor var_12491_begin_0 = const()[name = tensor("op_12491_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12491_end_0 = const()[name = tensor("op_12491_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_12491_end_mask_0 = const()[name = tensor("op_12491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12491_cast_fp16 = slice_by_index(begin = var_12491_begin_0, end = var_12491_end_0, end_mask = var_12491_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12491_cast_fp16")]; + tensor var_12495_begin_0 = const()[name = tensor("op_12495_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12495_end_0 = const()[name = tensor("op_12495_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_12495_end_mask_0 = const()[name = tensor("op_12495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12495_cast_fp16 = slice_by_index(begin = var_12495_begin_0, end = var_12495_end_0, end_mask = var_12495_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12495_cast_fp16")]; + tensor var_12499_begin_0 = const()[name = tensor("op_12499_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12499_end_0 = const()[name = tensor("op_12499_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_12499_end_mask_0 = const()[name = tensor("op_12499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12499_cast_fp16 = slice_by_index(begin = var_12499_begin_0, end = var_12499_end_0, end_mask = var_12499_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12499_cast_fp16")]; + tensor var_12503_begin_0 = const()[name = tensor("op_12503_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12503_end_0 = const()[name = tensor("op_12503_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_12503_end_mask_0 = const()[name = tensor("op_12503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12503_cast_fp16 = slice_by_index(begin = var_12503_begin_0, end = var_12503_end_0, end_mask = var_12503_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12503_cast_fp16")]; + tensor var_12507_begin_0 = const()[name = tensor("op_12507_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12507_end_0 = const()[name = tensor("op_12507_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_12507_end_mask_0 = const()[name = tensor("op_12507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12507_cast_fp16 = slice_by_index(begin = var_12507_begin_0, end = var_12507_end_0, end_mask = var_12507_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12507_cast_fp16")]; + tensor var_12511_begin_0 = const()[name = tensor("op_12511_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12511_end_0 = const()[name = tensor("op_12511_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_12511_end_mask_0 = const()[name = tensor("op_12511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12511_cast_fp16 = slice_by_index(begin = var_12511_begin_0, end = var_12511_end_0, end_mask = var_12511_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12511_cast_fp16")]; + tensor var_12515_begin_0 = const()[name = tensor("op_12515_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12515_end_0 = const()[name = tensor("op_12515_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_12515_end_mask_0 = const()[name = tensor("op_12515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12515_cast_fp16 = slice_by_index(begin = var_12515_begin_0, end = var_12515_end_0, end_mask = var_12515_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12515_cast_fp16")]; + tensor var_12519_begin_0 = const()[name = tensor("op_12519_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12519_end_0 = const()[name = tensor("op_12519_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_12519_end_mask_0 = const()[name = tensor("op_12519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12519_cast_fp16 = slice_by_index(begin = var_12519_begin_0, end = var_12519_end_0, end_mask = var_12519_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12519_cast_fp16")]; + tensor var_12523_begin_0 = const()[name = tensor("op_12523_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_12523_end_0 = const()[name = tensor("op_12523_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_12523_end_mask_0 = const()[name = tensor("op_12523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12523_cast_fp16 = slice_by_index(begin = var_12523_begin_0, end = var_12523_end_0, end_mask = var_12523_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12523_cast_fp16")]; + tensor var_12527_begin_0 = const()[name = tensor("op_12527_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_12527_end_0 = const()[name = tensor("op_12527_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_12527_end_mask_0 = const()[name = tensor("op_12527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12527_cast_fp16 = slice_by_index(begin = var_12527_begin_0, end = var_12527_end_0, end_mask = var_12527_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12527_cast_fp16")]; + tensor var_12531_begin_0 = const()[name = tensor("op_12531_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_12531_end_0 = const()[name = tensor("op_12531_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_12531_end_mask_0 = const()[name = tensor("op_12531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12531_cast_fp16 = slice_by_index(begin = var_12531_begin_0, end = var_12531_end_0, end_mask = var_12531_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12531_cast_fp16")]; + tensor var_12535_begin_0 = const()[name = tensor("op_12535_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_12535_end_0 = const()[name = tensor("op_12535_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_12535_end_mask_0 = const()[name = tensor("op_12535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12535_cast_fp16 = slice_by_index(begin = var_12535_begin_0, end = var_12535_end_0, end_mask = var_12535_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12535_cast_fp16")]; + tensor var_12539_begin_0 = const()[name = tensor("op_12539_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_12539_end_0 = const()[name = tensor("op_12539_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_12539_end_mask_0 = const()[name = tensor("op_12539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12539_cast_fp16 = slice_by_index(begin = var_12539_begin_0, end = var_12539_end_0, end_mask = var_12539_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12539_cast_fp16")]; + tensor var_12543_begin_0 = const()[name = tensor("op_12543_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_12543_end_0 = const()[name = tensor("op_12543_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_12543_end_mask_0 = const()[name = tensor("op_12543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12543_cast_fp16 = slice_by_index(begin = var_12543_begin_0, end = var_12543_end_0, end_mask = var_12543_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12543_cast_fp16")]; + tensor var_12547_begin_0 = const()[name = tensor("op_12547_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12547_end_0 = const()[name = tensor("op_12547_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_12547_end_mask_0 = const()[name = tensor("op_12547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12547_cast_fp16 = slice_by_index(begin = var_12547_begin_0, end = var_12547_end_0, end_mask = var_12547_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12547_cast_fp16")]; + tensor var_12551_begin_0 = const()[name = tensor("op_12551_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_12551_end_0 = const()[name = tensor("op_12551_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_12551_end_mask_0 = const()[name = tensor("op_12551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12551_cast_fp16 = slice_by_index(begin = var_12551_begin_0, end = var_12551_end_0, end_mask = var_12551_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12551_cast_fp16")]; + tensor var_12555_begin_0 = const()[name = tensor("op_12555_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_12555_end_0 = const()[name = tensor("op_12555_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_12555_end_mask_0 = const()[name = tensor("op_12555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12555_cast_fp16 = slice_by_index(begin = var_12555_begin_0, end = var_12555_end_0, end_mask = var_12555_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12555_cast_fp16")]; + tensor var_12559_begin_0 = const()[name = tensor("op_12559_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_12559_end_0 = const()[name = tensor("op_12559_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_12559_end_mask_0 = const()[name = tensor("op_12559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12559_cast_fp16 = slice_by_index(begin = var_12559_begin_0, end = var_12559_end_0, end_mask = var_12559_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12559_cast_fp16")]; + tensor var_12561_begin_0 = const()[name = tensor("op_12561_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12561_end_0 = const()[name = tensor("op_12561_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_12561_end_mask_0 = const()[name = tensor("op_12561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12561_cast_fp16 = slice_by_index(begin = var_12561_begin_0, end = var_12561_end_0, end_mask = var_12561_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12561_cast_fp16")]; + tensor var_12565_begin_0 = const()[name = tensor("op_12565_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12565_end_0 = const()[name = tensor("op_12565_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_12565_end_mask_0 = const()[name = tensor("op_12565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12565_cast_fp16 = slice_by_index(begin = var_12565_begin_0, end = var_12565_end_0, end_mask = var_12565_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12565_cast_fp16")]; + tensor var_12569_begin_0 = const()[name = tensor("op_12569_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12569_end_0 = const()[name = tensor("op_12569_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_12569_end_mask_0 = const()[name = tensor("op_12569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12569_cast_fp16 = slice_by_index(begin = var_12569_begin_0, end = var_12569_end_0, end_mask = var_12569_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12569_cast_fp16")]; + tensor var_12573_begin_0 = const()[name = tensor("op_12573_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12573_end_0 = const()[name = tensor("op_12573_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_12573_end_mask_0 = const()[name = tensor("op_12573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12573_cast_fp16 = slice_by_index(begin = var_12573_begin_0, end = var_12573_end_0, end_mask = var_12573_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12573_cast_fp16")]; + tensor var_12577_begin_0 = const()[name = tensor("op_12577_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12577_end_0 = const()[name = tensor("op_12577_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_12577_end_mask_0 = const()[name = tensor("op_12577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12577_cast_fp16 = slice_by_index(begin = var_12577_begin_0, end = var_12577_end_0, end_mask = var_12577_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12577_cast_fp16")]; + tensor var_12581_begin_0 = const()[name = tensor("op_12581_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12581_end_0 = const()[name = tensor("op_12581_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_12581_end_mask_0 = const()[name = tensor("op_12581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12581_cast_fp16 = slice_by_index(begin = var_12581_begin_0, end = var_12581_end_0, end_mask = var_12581_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12581_cast_fp16")]; + tensor var_12585_begin_0 = const()[name = tensor("op_12585_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12585_end_0 = const()[name = tensor("op_12585_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_12585_end_mask_0 = const()[name = tensor("op_12585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12585_cast_fp16 = slice_by_index(begin = var_12585_begin_0, end = var_12585_end_0, end_mask = var_12585_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12585_cast_fp16")]; + tensor var_12589_begin_0 = const()[name = tensor("op_12589_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12589_end_0 = const()[name = tensor("op_12589_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_12589_end_mask_0 = const()[name = tensor("op_12589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12589_cast_fp16 = slice_by_index(begin = var_12589_begin_0, end = var_12589_end_0, end_mask = var_12589_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12589_cast_fp16")]; + tensor var_12593_begin_0 = const()[name = tensor("op_12593_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12593_end_0 = const()[name = tensor("op_12593_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_12593_end_mask_0 = const()[name = tensor("op_12593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12593_cast_fp16 = slice_by_index(begin = var_12593_begin_0, end = var_12593_end_0, end_mask = var_12593_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12593_cast_fp16")]; + tensor var_12597_begin_0 = const()[name = tensor("op_12597_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12597_end_0 = const()[name = tensor("op_12597_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_12597_end_mask_0 = const()[name = tensor("op_12597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12597_cast_fp16 = slice_by_index(begin = var_12597_begin_0, end = var_12597_end_0, end_mask = var_12597_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12597_cast_fp16")]; + tensor var_12601_begin_0 = const()[name = tensor("op_12601_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12601_end_0 = const()[name = tensor("op_12601_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_12601_end_mask_0 = const()[name = tensor("op_12601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12601_cast_fp16 = slice_by_index(begin = var_12601_begin_0, end = var_12601_end_0, end_mask = var_12601_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12601_cast_fp16")]; + tensor var_12605_begin_0 = const()[name = tensor("op_12605_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12605_end_0 = const()[name = tensor("op_12605_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_12605_end_mask_0 = const()[name = tensor("op_12605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12605_cast_fp16 = slice_by_index(begin = var_12605_begin_0, end = var_12605_end_0, end_mask = var_12605_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12605_cast_fp16")]; + tensor var_12609_begin_0 = const()[name = tensor("op_12609_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12609_end_0 = const()[name = tensor("op_12609_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_12609_end_mask_0 = const()[name = tensor("op_12609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12609_cast_fp16 = slice_by_index(begin = var_12609_begin_0, end = var_12609_end_0, end_mask = var_12609_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12609_cast_fp16")]; + tensor var_12613_begin_0 = const()[name = tensor("op_12613_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12613_end_0 = const()[name = tensor("op_12613_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_12613_end_mask_0 = const()[name = tensor("op_12613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12613_cast_fp16 = slice_by_index(begin = var_12613_begin_0, end = var_12613_end_0, end_mask = var_12613_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12613_cast_fp16")]; + tensor var_12617_begin_0 = const()[name = tensor("op_12617_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12617_end_0 = const()[name = tensor("op_12617_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_12617_end_mask_0 = const()[name = tensor("op_12617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12617_cast_fp16 = slice_by_index(begin = var_12617_begin_0, end = var_12617_end_0, end_mask = var_12617_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12617_cast_fp16")]; + tensor var_12621_begin_0 = const()[name = tensor("op_12621_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12621_end_0 = const()[name = tensor("op_12621_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_12621_end_mask_0 = const()[name = tensor("op_12621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12621_cast_fp16 = slice_by_index(begin = var_12621_begin_0, end = var_12621_end_0, end_mask = var_12621_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12621_cast_fp16")]; + tensor var_12625_begin_0 = const()[name = tensor("op_12625_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12625_end_0 = const()[name = tensor("op_12625_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_12625_end_mask_0 = const()[name = tensor("op_12625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12625_cast_fp16 = slice_by_index(begin = var_12625_begin_0, end = var_12625_end_0, end_mask = var_12625_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12625_cast_fp16")]; + tensor var_12629_begin_0 = const()[name = tensor("op_12629_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12629_end_0 = const()[name = tensor("op_12629_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_12629_end_mask_0 = const()[name = tensor("op_12629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12629_cast_fp16 = slice_by_index(begin = var_12629_begin_0, end = var_12629_end_0, end_mask = var_12629_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12629_cast_fp16")]; + tensor var_12633_begin_0 = const()[name = tensor("op_12633_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12633_end_0 = const()[name = tensor("op_12633_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_12633_end_mask_0 = const()[name = tensor("op_12633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12633_cast_fp16 = slice_by_index(begin = var_12633_begin_0, end = var_12633_end_0, end_mask = var_12633_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12633_cast_fp16")]; + tensor var_12637_begin_0 = const()[name = tensor("op_12637_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12637_end_0 = const()[name = tensor("op_12637_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_12637_end_mask_0 = const()[name = tensor("op_12637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12637_cast_fp16 = slice_by_index(begin = var_12637_begin_0, end = var_12637_end_0, end_mask = var_12637_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12637_cast_fp16")]; + tensor var_12641_equation_0 = const()[name = tensor("op_12641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12641_cast_fp16 = einsum(equation = var_12641_equation_0, values = (var_12483_cast_fp16, var_12400_cast_fp16))[name = tensor("op_12641_cast_fp16")]; + tensor var_12642_to_fp16 = const()[name = tensor("op_12642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1001_cast_fp16 = mul(x = var_12641_cast_fp16, y = var_12642_to_fp16)[name = tensor("aw_1001_cast_fp16")]; + tensor var_12645_equation_0 = const()[name = tensor("op_12645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12645_cast_fp16 = einsum(equation = var_12645_equation_0, values = (var_12487_cast_fp16, var_12404_cast_fp16))[name = tensor("op_12645_cast_fp16")]; + tensor var_12646_to_fp16 = const()[name = tensor("op_12646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1003_cast_fp16 = mul(x = var_12645_cast_fp16, y = var_12646_to_fp16)[name = tensor("aw_1003_cast_fp16")]; + tensor var_12649_equation_0 = const()[name = tensor("op_12649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12649_cast_fp16 = einsum(equation = var_12649_equation_0, values = (var_12491_cast_fp16, var_12408_cast_fp16))[name = tensor("op_12649_cast_fp16")]; + tensor var_12650_to_fp16 = const()[name = tensor("op_12650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1005_cast_fp16 = mul(x = var_12649_cast_fp16, y = var_12650_to_fp16)[name = tensor("aw_1005_cast_fp16")]; + tensor var_12653_equation_0 = const()[name = tensor("op_12653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12653_cast_fp16 = einsum(equation = var_12653_equation_0, values = (var_12495_cast_fp16, var_12412_cast_fp16))[name = tensor("op_12653_cast_fp16")]; + tensor var_12654_to_fp16 = const()[name = tensor("op_12654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1007_cast_fp16 = mul(x = var_12653_cast_fp16, y = var_12654_to_fp16)[name = tensor("aw_1007_cast_fp16")]; + tensor var_12657_equation_0 = const()[name = tensor("op_12657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12657_cast_fp16 = einsum(equation = var_12657_equation_0, values = (var_12499_cast_fp16, var_12416_cast_fp16))[name = tensor("op_12657_cast_fp16")]; + tensor var_12658_to_fp16 = const()[name = tensor("op_12658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1009_cast_fp16 = mul(x = var_12657_cast_fp16, y = var_12658_to_fp16)[name = tensor("aw_1009_cast_fp16")]; + tensor var_12661_equation_0 = const()[name = tensor("op_12661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12661_cast_fp16 = einsum(equation = var_12661_equation_0, values = (var_12503_cast_fp16, var_12420_cast_fp16))[name = tensor("op_12661_cast_fp16")]; + tensor var_12662_to_fp16 = const()[name = tensor("op_12662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1011_cast_fp16 = mul(x = var_12661_cast_fp16, y = var_12662_to_fp16)[name = tensor("aw_1011_cast_fp16")]; + tensor var_12665_equation_0 = const()[name = tensor("op_12665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12665_cast_fp16 = einsum(equation = var_12665_equation_0, values = (var_12507_cast_fp16, var_12424_cast_fp16))[name = tensor("op_12665_cast_fp16")]; + tensor var_12666_to_fp16 = const()[name = tensor("op_12666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1013_cast_fp16 = mul(x = var_12665_cast_fp16, y = var_12666_to_fp16)[name = tensor("aw_1013_cast_fp16")]; + tensor var_12669_equation_0 = const()[name = tensor("op_12669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12669_cast_fp16 = einsum(equation = var_12669_equation_0, values = (var_12511_cast_fp16, var_12428_cast_fp16))[name = tensor("op_12669_cast_fp16")]; + tensor var_12670_to_fp16 = const()[name = tensor("op_12670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1015_cast_fp16 = mul(x = var_12669_cast_fp16, y = var_12670_to_fp16)[name = tensor("aw_1015_cast_fp16")]; + tensor var_12673_equation_0 = const()[name = tensor("op_12673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12673_cast_fp16 = einsum(equation = var_12673_equation_0, values = (var_12515_cast_fp16, var_12432_cast_fp16))[name = tensor("op_12673_cast_fp16")]; + tensor var_12674_to_fp16 = const()[name = tensor("op_12674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1017_cast_fp16 = mul(x = var_12673_cast_fp16, y = var_12674_to_fp16)[name = tensor("aw_1017_cast_fp16")]; + tensor var_12677_equation_0 = const()[name = tensor("op_12677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12677_cast_fp16 = einsum(equation = var_12677_equation_0, values = (var_12519_cast_fp16, var_12436_cast_fp16))[name = tensor("op_12677_cast_fp16")]; + tensor var_12678_to_fp16 = const()[name = tensor("op_12678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1019_cast_fp16 = mul(x = var_12677_cast_fp16, y = var_12678_to_fp16)[name = tensor("aw_1019_cast_fp16")]; + tensor var_12681_equation_0 = const()[name = tensor("op_12681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12681_cast_fp16 = einsum(equation = var_12681_equation_0, values = (var_12523_cast_fp16, var_12440_cast_fp16))[name = tensor("op_12681_cast_fp16")]; + tensor var_12682_to_fp16 = const()[name = tensor("op_12682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1021_cast_fp16 = mul(x = var_12681_cast_fp16, y = var_12682_to_fp16)[name = tensor("aw_1021_cast_fp16")]; + tensor var_12685_equation_0 = const()[name = tensor("op_12685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12685_cast_fp16 = einsum(equation = var_12685_equation_0, values = (var_12527_cast_fp16, var_12444_cast_fp16))[name = tensor("op_12685_cast_fp16")]; + tensor var_12686_to_fp16 = const()[name = tensor("op_12686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1023_cast_fp16 = mul(x = var_12685_cast_fp16, y = var_12686_to_fp16)[name = tensor("aw_1023_cast_fp16")]; + tensor var_12689_equation_0 = const()[name = tensor("op_12689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12689_cast_fp16 = einsum(equation = var_12689_equation_0, values = (var_12531_cast_fp16, var_12448_cast_fp16))[name = tensor("op_12689_cast_fp16")]; + tensor var_12690_to_fp16 = const()[name = tensor("op_12690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1025_cast_fp16 = mul(x = var_12689_cast_fp16, y = var_12690_to_fp16)[name = tensor("aw_1025_cast_fp16")]; + tensor var_12693_equation_0 = const()[name = tensor("op_12693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12693_cast_fp16 = einsum(equation = var_12693_equation_0, values = (var_12535_cast_fp16, var_12452_cast_fp16))[name = tensor("op_12693_cast_fp16")]; + tensor var_12694_to_fp16 = const()[name = tensor("op_12694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1027_cast_fp16 = mul(x = var_12693_cast_fp16, y = var_12694_to_fp16)[name = tensor("aw_1027_cast_fp16")]; + tensor var_12697_equation_0 = const()[name = tensor("op_12697_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12697_cast_fp16 = einsum(equation = var_12697_equation_0, values = (var_12539_cast_fp16, var_12456_cast_fp16))[name = tensor("op_12697_cast_fp16")]; + tensor var_12698_to_fp16 = const()[name = tensor("op_12698_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1029_cast_fp16 = mul(x = var_12697_cast_fp16, y = var_12698_to_fp16)[name = tensor("aw_1029_cast_fp16")]; + tensor var_12701_equation_0 = const()[name = tensor("op_12701_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12701_cast_fp16 = einsum(equation = var_12701_equation_0, values = (var_12543_cast_fp16, var_12460_cast_fp16))[name = tensor("op_12701_cast_fp16")]; + tensor var_12702_to_fp16 = const()[name = tensor("op_12702_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1031_cast_fp16 = mul(x = var_12701_cast_fp16, y = var_12702_to_fp16)[name = tensor("aw_1031_cast_fp16")]; + tensor var_12705_equation_0 = const()[name = tensor("op_12705_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12705_cast_fp16 = einsum(equation = var_12705_equation_0, values = (var_12547_cast_fp16, var_12464_cast_fp16))[name = tensor("op_12705_cast_fp16")]; + tensor var_12706_to_fp16 = const()[name = tensor("op_12706_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1033_cast_fp16 = mul(x = var_12705_cast_fp16, y = var_12706_to_fp16)[name = tensor("aw_1033_cast_fp16")]; + tensor var_12709_equation_0 = const()[name = tensor("op_12709_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12709_cast_fp16 = einsum(equation = var_12709_equation_0, values = (var_12551_cast_fp16, var_12468_cast_fp16))[name = tensor("op_12709_cast_fp16")]; + tensor var_12710_to_fp16 = const()[name = tensor("op_12710_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1035_cast_fp16 = mul(x = var_12709_cast_fp16, y = var_12710_to_fp16)[name = tensor("aw_1035_cast_fp16")]; + tensor var_12713_equation_0 = const()[name = tensor("op_12713_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12713_cast_fp16 = einsum(equation = var_12713_equation_0, values = (var_12555_cast_fp16, var_12472_cast_fp16))[name = tensor("op_12713_cast_fp16")]; + tensor var_12714_to_fp16 = const()[name = tensor("op_12714_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1037_cast_fp16 = mul(x = var_12713_cast_fp16, y = var_12714_to_fp16)[name = tensor("aw_1037_cast_fp16")]; + tensor var_12717_equation_0 = const()[name = tensor("op_12717_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12717_cast_fp16 = einsum(equation = var_12717_equation_0, values = (var_12559_cast_fp16, var_12476_cast_fp16))[name = tensor("op_12717_cast_fp16")]; + tensor var_12718_to_fp16 = const()[name = tensor("op_12718_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1039_cast_fp16 = mul(x = var_12717_cast_fp16, y = var_12718_to_fp16)[name = tensor("aw_1039_cast_fp16")]; + tensor var_12720_cast_fp16 = softmax(axis = var_2624, x = aw_1001_cast_fp16)[name = tensor("op_12720_cast_fp16")]; + tensor var_12721_cast_fp16 = softmax(axis = var_2624, x = aw_1003_cast_fp16)[name = tensor("op_12721_cast_fp16")]; + tensor var_12722_cast_fp16 = softmax(axis = var_2624, x = aw_1005_cast_fp16)[name = tensor("op_12722_cast_fp16")]; + tensor var_12723_cast_fp16 = softmax(axis = var_2624, x = aw_1007_cast_fp16)[name = tensor("op_12723_cast_fp16")]; + tensor var_12724_cast_fp16 = softmax(axis = var_2624, x = aw_1009_cast_fp16)[name = tensor("op_12724_cast_fp16")]; + tensor var_12725_cast_fp16 = softmax(axis = var_2624, x = aw_1011_cast_fp16)[name = tensor("op_12725_cast_fp16")]; + tensor var_12726_cast_fp16 = softmax(axis = var_2624, x = aw_1013_cast_fp16)[name = tensor("op_12726_cast_fp16")]; + tensor var_12727_cast_fp16 = softmax(axis = var_2624, x = aw_1015_cast_fp16)[name = tensor("op_12727_cast_fp16")]; + tensor var_12728_cast_fp16 = softmax(axis = var_2624, x = aw_1017_cast_fp16)[name = tensor("op_12728_cast_fp16")]; + tensor var_12729_cast_fp16 = softmax(axis = var_2624, x = aw_1019_cast_fp16)[name = tensor("op_12729_cast_fp16")]; + tensor var_12730_cast_fp16 = softmax(axis = var_2624, x = aw_1021_cast_fp16)[name = tensor("op_12730_cast_fp16")]; + tensor var_12731_cast_fp16 = softmax(axis = var_2624, x = aw_1023_cast_fp16)[name = tensor("op_12731_cast_fp16")]; + tensor var_12732_cast_fp16 = softmax(axis = var_2624, x = aw_1025_cast_fp16)[name = tensor("op_12732_cast_fp16")]; + tensor var_12733_cast_fp16 = softmax(axis = var_2624, x = aw_1027_cast_fp16)[name = tensor("op_12733_cast_fp16")]; + tensor var_12734_cast_fp16 = softmax(axis = var_2624, x = aw_1029_cast_fp16)[name = tensor("op_12734_cast_fp16")]; + tensor var_12735_cast_fp16 = softmax(axis = var_2624, x = aw_1031_cast_fp16)[name = tensor("op_12735_cast_fp16")]; + tensor var_12736_cast_fp16 = softmax(axis = var_2624, x = aw_1033_cast_fp16)[name = tensor("op_12736_cast_fp16")]; + tensor var_12737_cast_fp16 = softmax(axis = var_2624, x = aw_1035_cast_fp16)[name = tensor("op_12737_cast_fp16")]; + tensor var_12738_cast_fp16 = softmax(axis = var_2624, x = aw_1037_cast_fp16)[name = tensor("op_12738_cast_fp16")]; + tensor var_12739_cast_fp16 = softmax(axis = var_2624, x = aw_1039_cast_fp16)[name = tensor("op_12739_cast_fp16")]; + tensor var_12741_equation_0 = const()[name = tensor("op_12741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12741_cast_fp16 = einsum(equation = var_12741_equation_0, values = (var_12561_cast_fp16, var_12720_cast_fp16))[name = tensor("op_12741_cast_fp16")]; + tensor var_12743_equation_0 = const()[name = tensor("op_12743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12743_cast_fp16 = einsum(equation = var_12743_equation_0, values = (var_12565_cast_fp16, var_12721_cast_fp16))[name = tensor("op_12743_cast_fp16")]; + tensor var_12745_equation_0 = const()[name = tensor("op_12745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12745_cast_fp16 = einsum(equation = var_12745_equation_0, values = (var_12569_cast_fp16, var_12722_cast_fp16))[name = tensor("op_12745_cast_fp16")]; + tensor var_12747_equation_0 = const()[name = tensor("op_12747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12747_cast_fp16 = einsum(equation = var_12747_equation_0, values = (var_12573_cast_fp16, var_12723_cast_fp16))[name = tensor("op_12747_cast_fp16")]; + tensor var_12749_equation_0 = const()[name = tensor("op_12749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12749_cast_fp16 = einsum(equation = var_12749_equation_0, values = (var_12577_cast_fp16, var_12724_cast_fp16))[name = tensor("op_12749_cast_fp16")]; + tensor var_12751_equation_0 = const()[name = tensor("op_12751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12751_cast_fp16 = einsum(equation = var_12751_equation_0, values = (var_12581_cast_fp16, var_12725_cast_fp16))[name = tensor("op_12751_cast_fp16")]; + tensor var_12753_equation_0 = const()[name = tensor("op_12753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12753_cast_fp16 = einsum(equation = var_12753_equation_0, values = (var_12585_cast_fp16, var_12726_cast_fp16))[name = tensor("op_12753_cast_fp16")]; + tensor var_12755_equation_0 = const()[name = tensor("op_12755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12755_cast_fp16 = einsum(equation = var_12755_equation_0, values = (var_12589_cast_fp16, var_12727_cast_fp16))[name = tensor("op_12755_cast_fp16")]; + tensor var_12757_equation_0 = const()[name = tensor("op_12757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12757_cast_fp16 = einsum(equation = var_12757_equation_0, values = (var_12593_cast_fp16, var_12728_cast_fp16))[name = tensor("op_12757_cast_fp16")]; + tensor var_12759_equation_0 = const()[name = tensor("op_12759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12759_cast_fp16 = einsum(equation = var_12759_equation_0, values = (var_12597_cast_fp16, var_12729_cast_fp16))[name = tensor("op_12759_cast_fp16")]; + tensor var_12761_equation_0 = const()[name = tensor("op_12761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12761_cast_fp16 = einsum(equation = var_12761_equation_0, values = (var_12601_cast_fp16, var_12730_cast_fp16))[name = tensor("op_12761_cast_fp16")]; + tensor var_12763_equation_0 = const()[name = tensor("op_12763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12763_cast_fp16 = einsum(equation = var_12763_equation_0, values = (var_12605_cast_fp16, var_12731_cast_fp16))[name = tensor("op_12763_cast_fp16")]; + tensor var_12765_equation_0 = const()[name = tensor("op_12765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12765_cast_fp16 = einsum(equation = var_12765_equation_0, values = (var_12609_cast_fp16, var_12732_cast_fp16))[name = tensor("op_12765_cast_fp16")]; + tensor var_12767_equation_0 = const()[name = tensor("op_12767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12767_cast_fp16 = einsum(equation = var_12767_equation_0, values = (var_12613_cast_fp16, var_12733_cast_fp16))[name = tensor("op_12767_cast_fp16")]; + tensor var_12769_equation_0 = const()[name = tensor("op_12769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12769_cast_fp16 = einsum(equation = var_12769_equation_0, values = (var_12617_cast_fp16, var_12734_cast_fp16))[name = tensor("op_12769_cast_fp16")]; + tensor var_12771_equation_0 = const()[name = tensor("op_12771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12771_cast_fp16 = einsum(equation = var_12771_equation_0, values = (var_12621_cast_fp16, var_12735_cast_fp16))[name = tensor("op_12771_cast_fp16")]; + tensor var_12773_equation_0 = const()[name = tensor("op_12773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12773_cast_fp16 = einsum(equation = var_12773_equation_0, values = (var_12625_cast_fp16, var_12736_cast_fp16))[name = tensor("op_12773_cast_fp16")]; + tensor var_12775_equation_0 = const()[name = tensor("op_12775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12775_cast_fp16 = einsum(equation = var_12775_equation_0, values = (var_12629_cast_fp16, var_12737_cast_fp16))[name = tensor("op_12775_cast_fp16")]; + tensor var_12777_equation_0 = const()[name = tensor("op_12777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12777_cast_fp16 = einsum(equation = var_12777_equation_0, values = (var_12633_cast_fp16, var_12738_cast_fp16))[name = tensor("op_12777_cast_fp16")]; + tensor var_12779_equation_0 = const()[name = tensor("op_12779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12779_cast_fp16 = einsum(equation = var_12779_equation_0, values = (var_12637_cast_fp16, var_12739_cast_fp16))[name = tensor("op_12779_cast_fp16")]; + tensor input_231_interleave_0 = const()[name = tensor("input_231_interleave_0"), val = tensor(false)]; + tensor input_231_cast_fp16 = concat(axis = var_2624, interleave = input_231_interleave_0, values = (var_12741_cast_fp16, var_12743_cast_fp16, var_12745_cast_fp16, var_12747_cast_fp16, var_12749_cast_fp16, var_12751_cast_fp16, var_12753_cast_fp16, var_12755_cast_fp16, var_12757_cast_fp16, var_12759_cast_fp16, var_12761_cast_fp16, var_12763_cast_fp16, var_12765_cast_fp16, var_12767_cast_fp16, var_12769_cast_fp16, var_12771_cast_fp16, var_12773_cast_fp16, var_12775_cast_fp16, var_12777_cast_fp16, var_12779_cast_fp16))[name = tensor("input_231_cast_fp16")]; + tensor var_12789_pad_type_0 = const()[name = tensor("op_12789_pad_type_0"), val = tensor("valid")]; + tensor var_12789_strides_0 = const()[name = tensor("op_12789_strides_0"), val = tensor([1, 1])]; + tensor var_12789_pad_0 = const()[name = tensor("op_12789_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12789_dilations_0 = const()[name = tensor("op_12789_dilations_0"), val = tensor([1, 1])]; + tensor var_12789_groups_0 = const()[name = tensor("op_12789_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371338752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372567616))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372567808)))]; + tensor var_12789_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_12789_dilations_0, groups = var_12789_groups_0, pad = var_12789_pad_0, pad_type = var_12789_pad_type_0, strides = var_12789_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_231_cast_fp16)[name = tensor("op_12789_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = var_12789_cast_fp16, y = inputs_87_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; + tensor input_233_axes_0 = const()[name = tensor("input_233_axes_0"), val = tensor([1])]; + tensor input_233_gamma_0_to_fp16 = const()[name = tensor("input_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372570432)))]; + tensor input_233_beta_0_to_fp16 = const()[name = tensor("input_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372573056)))]; + tensor var_12799_to_fp16 = const()[name = tensor("op_12799_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_233_cast_fp16 = layer_norm(axes = input_233_axes_0, beta = input_233_beta_0_to_fp16, epsilon = var_12799_to_fp16, gamma = input_233_gamma_0_to_fp16, x = inputs_89_cast_fp16)[name = tensor("input_233_cast_fp16")]; + tensor var_12819_pad_type_0 = const()[name = tensor("op_12819_pad_type_0"), val = tensor("valid")]; + tensor var_12819_strides_0 = const()[name = tensor("op_12819_strides_0"), val = tensor([1, 1])]; + tensor var_12819_pad_0 = const()[name = tensor("op_12819_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12819_dilations_0 = const()[name = tensor("op_12819_dilations_0"), val = tensor([1, 1])]; + tensor var_12819_groups_0 = const()[name = tensor("op_12819_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372575680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382406144))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382406336)))]; + tensor var_12819_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_12819_dilations_0, groups = var_12819_groups_0, pad = var_12819_pad_0, pad_type = var_12819_pad_type_0, strides = var_12819_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = tensor("op_12819_cast_fp16")]; + tensor var_12820_split_sizes_0 = const()[name = tensor("op_12820_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_12820_axis_0 = const()[name = tensor("op_12820_axis_0"), val = tensor(1)]; + tensor var_12820_cast_fp16_0, tensor var_12820_cast_fp16_1 = split(axis = var_12820_axis_0, split_sizes = var_12820_split_sizes_0, x = var_12819_cast_fp16)[name = tensor("op_12820_cast_fp16")]; + tensor var_12822_mode_0 = const()[name = tensor("op_12822_mode_0"), val = tensor("EXACT")]; + tensor var_12822_cast_fp16 = gelu(mode = var_12822_mode_0, x = var_12820_cast_fp16_1)[name = tensor("op_12822_cast_fp16")]; + tensor input_235_cast_fp16 = mul(x = var_12820_cast_fp16_0, y = var_12822_cast_fp16)[name = tensor("input_235_cast_fp16")]; + tensor var_12830_pad_type_0 = const()[name = tensor("op_12830_pad_type_0"), val = tensor("valid")]; + tensor var_12830_strides_0 = const()[name = tensor("op_12830_strides_0"), val = tensor([1, 1])]; + tensor var_12830_pad_0 = const()[name = tensor("op_12830_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12830_dilations_0 = const()[name = tensor("op_12830_dilations_0"), val = tensor([1, 1])]; + tensor var_12830_groups_0 = const()[name = tensor("op_12830_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382426880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387342144))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387342336)))]; + tensor var_12830_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_12830_dilations_0, groups = var_12830_groups_0, pad = var_12830_pad_0, pad_type = var_12830_pad_type_0, strides = var_12830_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_235_cast_fp16)[name = tensor("op_12830_cast_fp16")]; + tensor inputs_91_cast_fp16 = add(x = var_12830_cast_fp16, y = inputs_89_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; + tensor hidden_states_143_axes_0 = const()[name = tensor("hidden_states_143_axes_0"), val = tensor([1])]; + tensor hidden_states_143_gamma_0_to_fp16 = const()[name = tensor("hidden_states_143_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387344960)))]; + tensor hidden_states_143_beta_0_to_fp16 = const()[name = tensor("hidden_states_143_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387347584)))]; + tensor var_12846_to_fp16 = const()[name = tensor("op_12846_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_143_cast_fp16 = layer_norm(axes = hidden_states_143_axes_0, beta = hidden_states_143_beta_0_to_fp16, epsilon = var_12846_to_fp16, gamma = hidden_states_143_gamma_0_to_fp16, x = inputs_91_cast_fp16)[name = tensor("hidden_states_143_cast_fp16")]; + tensor q_61_pad_type_0 = const()[name = tensor("q_61_pad_type_0"), val = tensor("valid")]; + tensor q_61_strides_0 = const()[name = tensor("q_61_strides_0"), val = tensor([1, 1])]; + tensor q_61_pad_0 = const()[name = tensor("q_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_61_dilations_0 = const()[name = tensor("q_61_dilations_0"), val = tensor([1, 1])]; + tensor q_61_groups_0 = const()[name = tensor("q_61_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387350208))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388579072))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_61_cast_fp16 = conv(dilations = q_61_dilations_0, groups = q_61_groups_0, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = q_61_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_143_cast_fp16)[name = tensor("q_61_cast_fp16")]; + tensor k_121_pad_type_0 = const()[name = tensor("k_121_pad_type_0"), val = tensor("valid")]; + tensor k_121_strides_0 = const()[name = tensor("k_121_strides_0"), val = tensor([1, 1])]; + tensor k_121_pad_0 = const()[name = tensor("k_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_121_dilations_0 = const()[name = tensor("k_121_dilations_0"), val = tensor([1, 1])]; + tensor k_121_groups_0 = const()[name = tensor("k_121_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388579264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389808128))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_121_cast_fp16 = conv(dilations = k_121_dilations_0, groups = k_121_groups_0, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = k_121_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_143_cast_fp16)[name = tensor("k_121_cast_fp16")]; + tensor v_61_pad_type_0 = const()[name = tensor("v_61_pad_type_0"), val = tensor("valid")]; + tensor v_61_strides_0 = const()[name = tensor("v_61_strides_0"), val = tensor([1, 1])]; + tensor v_61_pad_0 = const()[name = tensor("v_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_61_dilations_0 = const()[name = tensor("v_61_dilations_0"), val = tensor([1, 1])]; + tensor v_61_groups_0 = const()[name = tensor("v_61_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389808320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391037184))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_61_cast_fp16 = conv(dilations = v_61_dilations_0, groups = v_61_groups_0, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = v_61_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_143_cast_fp16)[name = tensor("v_61_cast_fp16")]; + tensor var_12879_begin_0 = const()[name = tensor("op_12879_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12879_end_0 = const()[name = tensor("op_12879_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12879_end_mask_0 = const()[name = tensor("op_12879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12879_cast_fp16 = slice_by_index(begin = var_12879_begin_0, end = var_12879_end_0, end_mask = var_12879_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12879_cast_fp16")]; + tensor var_12883_begin_0 = const()[name = tensor("op_12883_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12883_end_0 = const()[name = tensor("op_12883_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12883_end_mask_0 = const()[name = tensor("op_12883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12883_cast_fp16 = slice_by_index(begin = var_12883_begin_0, end = var_12883_end_0, end_mask = var_12883_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12883_cast_fp16")]; + tensor var_12887_begin_0 = const()[name = tensor("op_12887_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12887_end_0 = const()[name = tensor("op_12887_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12887_end_mask_0 = const()[name = tensor("op_12887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12887_cast_fp16 = slice_by_index(begin = var_12887_begin_0, end = var_12887_end_0, end_mask = var_12887_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12887_cast_fp16")]; + tensor var_12891_begin_0 = const()[name = tensor("op_12891_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12891_end_0 = const()[name = tensor("op_12891_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12891_end_mask_0 = const()[name = tensor("op_12891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12891_cast_fp16 = slice_by_index(begin = var_12891_begin_0, end = var_12891_end_0, end_mask = var_12891_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12891_cast_fp16")]; + tensor var_12895_begin_0 = const()[name = tensor("op_12895_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12895_end_0 = const()[name = tensor("op_12895_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12895_end_mask_0 = const()[name = tensor("op_12895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12895_cast_fp16 = slice_by_index(begin = var_12895_begin_0, end = var_12895_end_0, end_mask = var_12895_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12895_cast_fp16")]; + tensor var_12899_begin_0 = const()[name = tensor("op_12899_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12899_end_0 = const()[name = tensor("op_12899_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12899_end_mask_0 = const()[name = tensor("op_12899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12899_cast_fp16 = slice_by_index(begin = var_12899_begin_0, end = var_12899_end_0, end_mask = var_12899_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12899_cast_fp16")]; + tensor var_12903_begin_0 = const()[name = tensor("op_12903_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12903_end_0 = const()[name = tensor("op_12903_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12903_end_mask_0 = const()[name = tensor("op_12903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12903_cast_fp16 = slice_by_index(begin = var_12903_begin_0, end = var_12903_end_0, end_mask = var_12903_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12903_cast_fp16")]; + tensor var_12907_begin_0 = const()[name = tensor("op_12907_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12907_end_0 = const()[name = tensor("op_12907_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12907_end_mask_0 = const()[name = tensor("op_12907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12907_cast_fp16 = slice_by_index(begin = var_12907_begin_0, end = var_12907_end_0, end_mask = var_12907_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12907_cast_fp16")]; + tensor var_12911_begin_0 = const()[name = tensor("op_12911_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12911_end_0 = const()[name = tensor("op_12911_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12911_end_mask_0 = const()[name = tensor("op_12911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12911_cast_fp16 = slice_by_index(begin = var_12911_begin_0, end = var_12911_end_0, end_mask = var_12911_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12911_cast_fp16")]; + tensor var_12915_begin_0 = const()[name = tensor("op_12915_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12915_end_0 = const()[name = tensor("op_12915_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12915_end_mask_0 = const()[name = tensor("op_12915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12915_cast_fp16 = slice_by_index(begin = var_12915_begin_0, end = var_12915_end_0, end_mask = var_12915_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12915_cast_fp16")]; + tensor var_12919_begin_0 = const()[name = tensor("op_12919_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12919_end_0 = const()[name = tensor("op_12919_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12919_end_mask_0 = const()[name = tensor("op_12919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12919_cast_fp16 = slice_by_index(begin = var_12919_begin_0, end = var_12919_end_0, end_mask = var_12919_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12919_cast_fp16")]; + tensor var_12923_begin_0 = const()[name = tensor("op_12923_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12923_end_0 = const()[name = tensor("op_12923_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12923_end_mask_0 = const()[name = tensor("op_12923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12923_cast_fp16 = slice_by_index(begin = var_12923_begin_0, end = var_12923_end_0, end_mask = var_12923_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12923_cast_fp16")]; + tensor var_12927_begin_0 = const()[name = tensor("op_12927_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12927_end_0 = const()[name = tensor("op_12927_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12927_end_mask_0 = const()[name = tensor("op_12927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12927_cast_fp16 = slice_by_index(begin = var_12927_begin_0, end = var_12927_end_0, end_mask = var_12927_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12927_cast_fp16")]; + tensor var_12931_begin_0 = const()[name = tensor("op_12931_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12931_end_0 = const()[name = tensor("op_12931_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12931_end_mask_0 = const()[name = tensor("op_12931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12931_cast_fp16 = slice_by_index(begin = var_12931_begin_0, end = var_12931_end_0, end_mask = var_12931_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12931_cast_fp16")]; + tensor var_12935_begin_0 = const()[name = tensor("op_12935_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12935_end_0 = const()[name = tensor("op_12935_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12935_end_mask_0 = const()[name = tensor("op_12935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12935_cast_fp16 = slice_by_index(begin = var_12935_begin_0, end = var_12935_end_0, end_mask = var_12935_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12935_cast_fp16")]; + tensor var_12939_begin_0 = const()[name = tensor("op_12939_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12939_end_0 = const()[name = tensor("op_12939_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12939_end_mask_0 = const()[name = tensor("op_12939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12939_cast_fp16 = slice_by_index(begin = var_12939_begin_0, end = var_12939_end_0, end_mask = var_12939_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12939_cast_fp16")]; + tensor var_12943_begin_0 = const()[name = tensor("op_12943_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12943_end_0 = const()[name = tensor("op_12943_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12943_end_mask_0 = const()[name = tensor("op_12943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12943_cast_fp16 = slice_by_index(begin = var_12943_begin_0, end = var_12943_end_0, end_mask = var_12943_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12943_cast_fp16")]; + tensor var_12947_begin_0 = const()[name = tensor("op_12947_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12947_end_0 = const()[name = tensor("op_12947_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12947_end_mask_0 = const()[name = tensor("op_12947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12947_cast_fp16 = slice_by_index(begin = var_12947_begin_0, end = var_12947_end_0, end_mask = var_12947_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12947_cast_fp16")]; + tensor var_12951_begin_0 = const()[name = tensor("op_12951_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12951_end_0 = const()[name = tensor("op_12951_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12951_end_mask_0 = const()[name = tensor("op_12951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12951_cast_fp16 = slice_by_index(begin = var_12951_begin_0, end = var_12951_end_0, end_mask = var_12951_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12951_cast_fp16")]; + tensor var_12955_begin_0 = const()[name = tensor("op_12955_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12955_end_0 = const()[name = tensor("op_12955_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12955_end_mask_0 = const()[name = tensor("op_12955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12955_cast_fp16 = slice_by_index(begin = var_12955_begin_0, end = var_12955_end_0, end_mask = var_12955_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12955_cast_fp16")]; + tensor k_123_perm_0 = const()[name = tensor("k_123_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12962_begin_0 = const()[name = tensor("op_12962_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12962_end_0 = const()[name = tensor("op_12962_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_12962_end_mask_0 = const()[name = tensor("op_12962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_123_cast_fp16 = transpose(perm = k_123_perm_0, x = k_121_cast_fp16)[name = tensor("transpose_37")]; + tensor var_12962_cast_fp16 = slice_by_index(begin = var_12962_begin_0, end = var_12962_end_0, end_mask = var_12962_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12962_cast_fp16")]; + tensor var_12966_begin_0 = const()[name = tensor("op_12966_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12966_end_0 = const()[name = tensor("op_12966_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_12966_end_mask_0 = const()[name = tensor("op_12966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12966_cast_fp16 = slice_by_index(begin = var_12966_begin_0, end = var_12966_end_0, end_mask = var_12966_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12966_cast_fp16")]; + tensor var_12970_begin_0 = const()[name = tensor("op_12970_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12970_end_0 = const()[name = tensor("op_12970_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_12970_end_mask_0 = const()[name = tensor("op_12970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12970_cast_fp16 = slice_by_index(begin = var_12970_begin_0, end = var_12970_end_0, end_mask = var_12970_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12970_cast_fp16")]; + tensor var_12974_begin_0 = const()[name = tensor("op_12974_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12974_end_0 = const()[name = tensor("op_12974_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_12974_end_mask_0 = const()[name = tensor("op_12974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12974_cast_fp16 = slice_by_index(begin = var_12974_begin_0, end = var_12974_end_0, end_mask = var_12974_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12974_cast_fp16")]; + tensor var_12978_begin_0 = const()[name = tensor("op_12978_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12978_end_0 = const()[name = tensor("op_12978_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_12978_end_mask_0 = const()[name = tensor("op_12978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12978_cast_fp16 = slice_by_index(begin = var_12978_begin_0, end = var_12978_end_0, end_mask = var_12978_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12978_cast_fp16")]; + tensor var_12982_begin_0 = const()[name = tensor("op_12982_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12982_end_0 = const()[name = tensor("op_12982_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_12982_end_mask_0 = const()[name = tensor("op_12982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12982_cast_fp16 = slice_by_index(begin = var_12982_begin_0, end = var_12982_end_0, end_mask = var_12982_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12982_cast_fp16")]; + tensor var_12986_begin_0 = const()[name = tensor("op_12986_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12986_end_0 = const()[name = tensor("op_12986_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_12986_end_mask_0 = const()[name = tensor("op_12986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12986_cast_fp16 = slice_by_index(begin = var_12986_begin_0, end = var_12986_end_0, end_mask = var_12986_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12986_cast_fp16")]; + tensor var_12990_begin_0 = const()[name = tensor("op_12990_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12990_end_0 = const()[name = tensor("op_12990_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_12990_end_mask_0 = const()[name = tensor("op_12990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12990_cast_fp16 = slice_by_index(begin = var_12990_begin_0, end = var_12990_end_0, end_mask = var_12990_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12990_cast_fp16")]; + tensor var_12994_begin_0 = const()[name = tensor("op_12994_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12994_end_0 = const()[name = tensor("op_12994_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_12994_end_mask_0 = const()[name = tensor("op_12994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12994_cast_fp16 = slice_by_index(begin = var_12994_begin_0, end = var_12994_end_0, end_mask = var_12994_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12994_cast_fp16")]; + tensor var_12998_begin_0 = const()[name = tensor("op_12998_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12998_end_0 = const()[name = tensor("op_12998_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_12998_end_mask_0 = const()[name = tensor("op_12998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12998_cast_fp16 = slice_by_index(begin = var_12998_begin_0, end = var_12998_end_0, end_mask = var_12998_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12998_cast_fp16")]; + tensor var_13002_begin_0 = const()[name = tensor("op_13002_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13002_end_0 = const()[name = tensor("op_13002_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_13002_end_mask_0 = const()[name = tensor("op_13002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13002_cast_fp16 = slice_by_index(begin = var_13002_begin_0, end = var_13002_end_0, end_mask = var_13002_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13002_cast_fp16")]; + tensor var_13006_begin_0 = const()[name = tensor("op_13006_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13006_end_0 = const()[name = tensor("op_13006_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_13006_end_mask_0 = const()[name = tensor("op_13006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13006_cast_fp16 = slice_by_index(begin = var_13006_begin_0, end = var_13006_end_0, end_mask = var_13006_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13006_cast_fp16")]; + tensor var_13010_begin_0 = const()[name = tensor("op_13010_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13010_end_0 = const()[name = tensor("op_13010_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_13010_end_mask_0 = const()[name = tensor("op_13010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13010_cast_fp16 = slice_by_index(begin = var_13010_begin_0, end = var_13010_end_0, end_mask = var_13010_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13010_cast_fp16")]; + tensor var_13014_begin_0 = const()[name = tensor("op_13014_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13014_end_0 = const()[name = tensor("op_13014_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_13014_end_mask_0 = const()[name = tensor("op_13014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13014_cast_fp16 = slice_by_index(begin = var_13014_begin_0, end = var_13014_end_0, end_mask = var_13014_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13014_cast_fp16")]; + tensor var_13018_begin_0 = const()[name = tensor("op_13018_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13018_end_0 = const()[name = tensor("op_13018_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_13018_end_mask_0 = const()[name = tensor("op_13018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13018_cast_fp16 = slice_by_index(begin = var_13018_begin_0, end = var_13018_end_0, end_mask = var_13018_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13018_cast_fp16")]; + tensor var_13022_begin_0 = const()[name = tensor("op_13022_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13022_end_0 = const()[name = tensor("op_13022_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13022_end_mask_0 = const()[name = tensor("op_13022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13022_cast_fp16 = slice_by_index(begin = var_13022_begin_0, end = var_13022_end_0, end_mask = var_13022_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13022_cast_fp16")]; + tensor var_13026_begin_0 = const()[name = tensor("op_13026_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13026_end_0 = const()[name = tensor("op_13026_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_13026_end_mask_0 = const()[name = tensor("op_13026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13026_cast_fp16 = slice_by_index(begin = var_13026_begin_0, end = var_13026_end_0, end_mask = var_13026_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13026_cast_fp16")]; + tensor var_13030_begin_0 = const()[name = tensor("op_13030_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13030_end_0 = const()[name = tensor("op_13030_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_13030_end_mask_0 = const()[name = tensor("op_13030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13030_cast_fp16 = slice_by_index(begin = var_13030_begin_0, end = var_13030_end_0, end_mask = var_13030_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13030_cast_fp16")]; + tensor var_13034_begin_0 = const()[name = tensor("op_13034_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13034_end_0 = const()[name = tensor("op_13034_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_13034_end_mask_0 = const()[name = tensor("op_13034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13034_cast_fp16 = slice_by_index(begin = var_13034_begin_0, end = var_13034_end_0, end_mask = var_13034_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13034_cast_fp16")]; + tensor var_13038_begin_0 = const()[name = tensor("op_13038_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13038_end_0 = const()[name = tensor("op_13038_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_13038_end_mask_0 = const()[name = tensor("op_13038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13038_cast_fp16 = slice_by_index(begin = var_13038_begin_0, end = var_13038_end_0, end_mask = var_13038_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13038_cast_fp16")]; + tensor var_13040_begin_0 = const()[name = tensor("op_13040_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13040_end_0 = const()[name = tensor("op_13040_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13040_end_mask_0 = const()[name = tensor("op_13040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13040_cast_fp16 = slice_by_index(begin = var_13040_begin_0, end = var_13040_end_0, end_mask = var_13040_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13040_cast_fp16")]; + tensor var_13044_begin_0 = const()[name = tensor("op_13044_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13044_end_0 = const()[name = tensor("op_13044_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13044_end_mask_0 = const()[name = tensor("op_13044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13044_cast_fp16 = slice_by_index(begin = var_13044_begin_0, end = var_13044_end_0, end_mask = var_13044_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13044_cast_fp16")]; + tensor var_13048_begin_0 = const()[name = tensor("op_13048_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13048_end_0 = const()[name = tensor("op_13048_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13048_end_mask_0 = const()[name = tensor("op_13048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13048_cast_fp16 = slice_by_index(begin = var_13048_begin_0, end = var_13048_end_0, end_mask = var_13048_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13048_cast_fp16")]; + tensor var_13052_begin_0 = const()[name = tensor("op_13052_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13052_end_0 = const()[name = tensor("op_13052_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13052_end_mask_0 = const()[name = tensor("op_13052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13052_cast_fp16 = slice_by_index(begin = var_13052_begin_0, end = var_13052_end_0, end_mask = var_13052_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13052_cast_fp16")]; + tensor var_13056_begin_0 = const()[name = tensor("op_13056_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13056_end_0 = const()[name = tensor("op_13056_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13056_end_mask_0 = const()[name = tensor("op_13056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13056_cast_fp16 = slice_by_index(begin = var_13056_begin_0, end = var_13056_end_0, end_mask = var_13056_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13056_cast_fp16")]; + tensor var_13060_begin_0 = const()[name = tensor("op_13060_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13060_end_0 = const()[name = tensor("op_13060_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13060_end_mask_0 = const()[name = tensor("op_13060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13060_cast_fp16 = slice_by_index(begin = var_13060_begin_0, end = var_13060_end_0, end_mask = var_13060_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13060_cast_fp16")]; + tensor var_13064_begin_0 = const()[name = tensor("op_13064_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13064_end_0 = const()[name = tensor("op_13064_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13064_end_mask_0 = const()[name = tensor("op_13064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13064_cast_fp16 = slice_by_index(begin = var_13064_begin_0, end = var_13064_end_0, end_mask = var_13064_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13064_cast_fp16")]; + tensor var_13068_begin_0 = const()[name = tensor("op_13068_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13068_end_0 = const()[name = tensor("op_13068_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13068_end_mask_0 = const()[name = tensor("op_13068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13068_cast_fp16 = slice_by_index(begin = var_13068_begin_0, end = var_13068_end_0, end_mask = var_13068_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13068_cast_fp16")]; + tensor var_13072_begin_0 = const()[name = tensor("op_13072_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13072_end_0 = const()[name = tensor("op_13072_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13072_end_mask_0 = const()[name = tensor("op_13072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13072_cast_fp16 = slice_by_index(begin = var_13072_begin_0, end = var_13072_end_0, end_mask = var_13072_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13072_cast_fp16")]; + tensor var_13076_begin_0 = const()[name = tensor("op_13076_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13076_end_0 = const()[name = tensor("op_13076_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13076_end_mask_0 = const()[name = tensor("op_13076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13076_cast_fp16 = slice_by_index(begin = var_13076_begin_0, end = var_13076_end_0, end_mask = var_13076_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13076_cast_fp16")]; + tensor var_13080_begin_0 = const()[name = tensor("op_13080_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13080_end_0 = const()[name = tensor("op_13080_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13080_end_mask_0 = const()[name = tensor("op_13080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13080_cast_fp16 = slice_by_index(begin = var_13080_begin_0, end = var_13080_end_0, end_mask = var_13080_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13080_cast_fp16")]; + tensor var_13084_begin_0 = const()[name = tensor("op_13084_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13084_end_0 = const()[name = tensor("op_13084_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13084_end_mask_0 = const()[name = tensor("op_13084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13084_cast_fp16 = slice_by_index(begin = var_13084_begin_0, end = var_13084_end_0, end_mask = var_13084_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13084_cast_fp16")]; + tensor var_13088_begin_0 = const()[name = tensor("op_13088_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13088_end_0 = const()[name = tensor("op_13088_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13088_end_mask_0 = const()[name = tensor("op_13088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13088_cast_fp16 = slice_by_index(begin = var_13088_begin_0, end = var_13088_end_0, end_mask = var_13088_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13088_cast_fp16")]; + tensor var_13092_begin_0 = const()[name = tensor("op_13092_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13092_end_0 = const()[name = tensor("op_13092_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_13092_end_mask_0 = const()[name = tensor("op_13092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13092_cast_fp16 = slice_by_index(begin = var_13092_begin_0, end = var_13092_end_0, end_mask = var_13092_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13092_cast_fp16")]; + tensor var_13096_begin_0 = const()[name = tensor("op_13096_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13096_end_0 = const()[name = tensor("op_13096_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_13096_end_mask_0 = const()[name = tensor("op_13096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13096_cast_fp16 = slice_by_index(begin = var_13096_begin_0, end = var_13096_end_0, end_mask = var_13096_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13096_cast_fp16")]; + tensor var_13100_begin_0 = const()[name = tensor("op_13100_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13100_end_0 = const()[name = tensor("op_13100_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13100_end_mask_0 = const()[name = tensor("op_13100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13100_cast_fp16 = slice_by_index(begin = var_13100_begin_0, end = var_13100_end_0, end_mask = var_13100_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13100_cast_fp16")]; + tensor var_13104_begin_0 = const()[name = tensor("op_13104_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13104_end_0 = const()[name = tensor("op_13104_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_13104_end_mask_0 = const()[name = tensor("op_13104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13104_cast_fp16 = slice_by_index(begin = var_13104_begin_0, end = var_13104_end_0, end_mask = var_13104_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13104_cast_fp16")]; + tensor var_13108_begin_0 = const()[name = tensor("op_13108_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13108_end_0 = const()[name = tensor("op_13108_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_13108_end_mask_0 = const()[name = tensor("op_13108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13108_cast_fp16 = slice_by_index(begin = var_13108_begin_0, end = var_13108_end_0, end_mask = var_13108_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13108_cast_fp16")]; + tensor var_13112_begin_0 = const()[name = tensor("op_13112_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13112_end_0 = const()[name = tensor("op_13112_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_13112_end_mask_0 = const()[name = tensor("op_13112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13112_cast_fp16 = slice_by_index(begin = var_13112_begin_0, end = var_13112_end_0, end_mask = var_13112_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13112_cast_fp16")]; + tensor var_13116_begin_0 = const()[name = tensor("op_13116_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13116_end_0 = const()[name = tensor("op_13116_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_13116_end_mask_0 = const()[name = tensor("op_13116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13116_cast_fp16 = slice_by_index(begin = var_13116_begin_0, end = var_13116_end_0, end_mask = var_13116_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13116_cast_fp16")]; + tensor var_13120_equation_0 = const()[name = tensor("op_13120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13120_cast_fp16 = einsum(equation = var_13120_equation_0, values = (var_12962_cast_fp16, var_12879_cast_fp16))[name = tensor("op_13120_cast_fp16")]; + tensor var_13121_to_fp16 = const()[name = tensor("op_13121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1041_cast_fp16 = mul(x = var_13120_cast_fp16, y = var_13121_to_fp16)[name = tensor("aw_1041_cast_fp16")]; + tensor var_13124_equation_0 = const()[name = tensor("op_13124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13124_cast_fp16 = einsum(equation = var_13124_equation_0, values = (var_12966_cast_fp16, var_12883_cast_fp16))[name = tensor("op_13124_cast_fp16")]; + tensor var_13125_to_fp16 = const()[name = tensor("op_13125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1043_cast_fp16 = mul(x = var_13124_cast_fp16, y = var_13125_to_fp16)[name = tensor("aw_1043_cast_fp16")]; + tensor var_13128_equation_0 = const()[name = tensor("op_13128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13128_cast_fp16 = einsum(equation = var_13128_equation_0, values = (var_12970_cast_fp16, var_12887_cast_fp16))[name = tensor("op_13128_cast_fp16")]; + tensor var_13129_to_fp16 = const()[name = tensor("op_13129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1045_cast_fp16 = mul(x = var_13128_cast_fp16, y = var_13129_to_fp16)[name = tensor("aw_1045_cast_fp16")]; + tensor var_13132_equation_0 = const()[name = tensor("op_13132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13132_cast_fp16 = einsum(equation = var_13132_equation_0, values = (var_12974_cast_fp16, var_12891_cast_fp16))[name = tensor("op_13132_cast_fp16")]; + tensor var_13133_to_fp16 = const()[name = tensor("op_13133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1047_cast_fp16 = mul(x = var_13132_cast_fp16, y = var_13133_to_fp16)[name = tensor("aw_1047_cast_fp16")]; + tensor var_13136_equation_0 = const()[name = tensor("op_13136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13136_cast_fp16 = einsum(equation = var_13136_equation_0, values = (var_12978_cast_fp16, var_12895_cast_fp16))[name = tensor("op_13136_cast_fp16")]; + tensor var_13137_to_fp16 = const()[name = tensor("op_13137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1049_cast_fp16 = mul(x = var_13136_cast_fp16, y = var_13137_to_fp16)[name = tensor("aw_1049_cast_fp16")]; + tensor var_13140_equation_0 = const()[name = tensor("op_13140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13140_cast_fp16 = einsum(equation = var_13140_equation_0, values = (var_12982_cast_fp16, var_12899_cast_fp16))[name = tensor("op_13140_cast_fp16")]; + tensor var_13141_to_fp16 = const()[name = tensor("op_13141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1051_cast_fp16 = mul(x = var_13140_cast_fp16, y = var_13141_to_fp16)[name = tensor("aw_1051_cast_fp16")]; + tensor var_13144_equation_0 = const()[name = tensor("op_13144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13144_cast_fp16 = einsum(equation = var_13144_equation_0, values = (var_12986_cast_fp16, var_12903_cast_fp16))[name = tensor("op_13144_cast_fp16")]; + tensor var_13145_to_fp16 = const()[name = tensor("op_13145_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1053_cast_fp16 = mul(x = var_13144_cast_fp16, y = var_13145_to_fp16)[name = tensor("aw_1053_cast_fp16")]; + tensor var_13148_equation_0 = const()[name = tensor("op_13148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13148_cast_fp16 = einsum(equation = var_13148_equation_0, values = (var_12990_cast_fp16, var_12907_cast_fp16))[name = tensor("op_13148_cast_fp16")]; + tensor var_13149_to_fp16 = const()[name = tensor("op_13149_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1055_cast_fp16 = mul(x = var_13148_cast_fp16, y = var_13149_to_fp16)[name = tensor("aw_1055_cast_fp16")]; + tensor var_13152_equation_0 = const()[name = tensor("op_13152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13152_cast_fp16 = einsum(equation = var_13152_equation_0, values = (var_12994_cast_fp16, var_12911_cast_fp16))[name = tensor("op_13152_cast_fp16")]; + tensor var_13153_to_fp16 = const()[name = tensor("op_13153_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1057_cast_fp16 = mul(x = var_13152_cast_fp16, y = var_13153_to_fp16)[name = tensor("aw_1057_cast_fp16")]; + tensor var_13156_equation_0 = const()[name = tensor("op_13156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13156_cast_fp16 = einsum(equation = var_13156_equation_0, values = (var_12998_cast_fp16, var_12915_cast_fp16))[name = tensor("op_13156_cast_fp16")]; + tensor var_13157_to_fp16 = const()[name = tensor("op_13157_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1059_cast_fp16 = mul(x = var_13156_cast_fp16, y = var_13157_to_fp16)[name = tensor("aw_1059_cast_fp16")]; + tensor var_13160_equation_0 = const()[name = tensor("op_13160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13160_cast_fp16 = einsum(equation = var_13160_equation_0, values = (var_13002_cast_fp16, var_12919_cast_fp16))[name = tensor("op_13160_cast_fp16")]; + tensor var_13161_to_fp16 = const()[name = tensor("op_13161_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1061_cast_fp16 = mul(x = var_13160_cast_fp16, y = var_13161_to_fp16)[name = tensor("aw_1061_cast_fp16")]; + tensor var_13164_equation_0 = const()[name = tensor("op_13164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13164_cast_fp16 = einsum(equation = var_13164_equation_0, values = (var_13006_cast_fp16, var_12923_cast_fp16))[name = tensor("op_13164_cast_fp16")]; + tensor var_13165_to_fp16 = const()[name = tensor("op_13165_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1063_cast_fp16 = mul(x = var_13164_cast_fp16, y = var_13165_to_fp16)[name = tensor("aw_1063_cast_fp16")]; + tensor var_13168_equation_0 = const()[name = tensor("op_13168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13168_cast_fp16 = einsum(equation = var_13168_equation_0, values = (var_13010_cast_fp16, var_12927_cast_fp16))[name = tensor("op_13168_cast_fp16")]; + tensor var_13169_to_fp16 = const()[name = tensor("op_13169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1065_cast_fp16 = mul(x = var_13168_cast_fp16, y = var_13169_to_fp16)[name = tensor("aw_1065_cast_fp16")]; + tensor var_13172_equation_0 = const()[name = tensor("op_13172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13172_cast_fp16 = einsum(equation = var_13172_equation_0, values = (var_13014_cast_fp16, var_12931_cast_fp16))[name = tensor("op_13172_cast_fp16")]; + tensor var_13173_to_fp16 = const()[name = tensor("op_13173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1067_cast_fp16 = mul(x = var_13172_cast_fp16, y = var_13173_to_fp16)[name = tensor("aw_1067_cast_fp16")]; + tensor var_13176_equation_0 = const()[name = tensor("op_13176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13176_cast_fp16 = einsum(equation = var_13176_equation_0, values = (var_13018_cast_fp16, var_12935_cast_fp16))[name = tensor("op_13176_cast_fp16")]; + tensor var_13177_to_fp16 = const()[name = tensor("op_13177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1069_cast_fp16 = mul(x = var_13176_cast_fp16, y = var_13177_to_fp16)[name = tensor("aw_1069_cast_fp16")]; + tensor var_13180_equation_0 = const()[name = tensor("op_13180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13180_cast_fp16 = einsum(equation = var_13180_equation_0, values = (var_13022_cast_fp16, var_12939_cast_fp16))[name = tensor("op_13180_cast_fp16")]; + tensor var_13181_to_fp16 = const()[name = tensor("op_13181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1071_cast_fp16 = mul(x = var_13180_cast_fp16, y = var_13181_to_fp16)[name = tensor("aw_1071_cast_fp16")]; + tensor var_13184_equation_0 = const()[name = tensor("op_13184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13184_cast_fp16 = einsum(equation = var_13184_equation_0, values = (var_13026_cast_fp16, var_12943_cast_fp16))[name = tensor("op_13184_cast_fp16")]; + tensor var_13185_to_fp16 = const()[name = tensor("op_13185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1073_cast_fp16 = mul(x = var_13184_cast_fp16, y = var_13185_to_fp16)[name = tensor("aw_1073_cast_fp16")]; + tensor var_13188_equation_0 = const()[name = tensor("op_13188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13188_cast_fp16 = einsum(equation = var_13188_equation_0, values = (var_13030_cast_fp16, var_12947_cast_fp16))[name = tensor("op_13188_cast_fp16")]; + tensor var_13189_to_fp16 = const()[name = tensor("op_13189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1075_cast_fp16 = mul(x = var_13188_cast_fp16, y = var_13189_to_fp16)[name = tensor("aw_1075_cast_fp16")]; + tensor var_13192_equation_0 = const()[name = tensor("op_13192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13192_cast_fp16 = einsum(equation = var_13192_equation_0, values = (var_13034_cast_fp16, var_12951_cast_fp16))[name = tensor("op_13192_cast_fp16")]; + tensor var_13193_to_fp16 = const()[name = tensor("op_13193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1077_cast_fp16 = mul(x = var_13192_cast_fp16, y = var_13193_to_fp16)[name = tensor("aw_1077_cast_fp16")]; + tensor var_13196_equation_0 = const()[name = tensor("op_13196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13196_cast_fp16 = einsum(equation = var_13196_equation_0, values = (var_13038_cast_fp16, var_12955_cast_fp16))[name = tensor("op_13196_cast_fp16")]; + tensor var_13197_to_fp16 = const()[name = tensor("op_13197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1079_cast_fp16 = mul(x = var_13196_cast_fp16, y = var_13197_to_fp16)[name = tensor("aw_1079_cast_fp16")]; + tensor var_13199_cast_fp16 = softmax(axis = var_2624, x = aw_1041_cast_fp16)[name = tensor("op_13199_cast_fp16")]; + tensor var_13200_cast_fp16 = softmax(axis = var_2624, x = aw_1043_cast_fp16)[name = tensor("op_13200_cast_fp16")]; + tensor var_13201_cast_fp16 = softmax(axis = var_2624, x = aw_1045_cast_fp16)[name = tensor("op_13201_cast_fp16")]; + tensor var_13202_cast_fp16 = softmax(axis = var_2624, x = aw_1047_cast_fp16)[name = tensor("op_13202_cast_fp16")]; + tensor var_13203_cast_fp16 = softmax(axis = var_2624, x = aw_1049_cast_fp16)[name = tensor("op_13203_cast_fp16")]; + tensor var_13204_cast_fp16 = softmax(axis = var_2624, x = aw_1051_cast_fp16)[name = tensor("op_13204_cast_fp16")]; + tensor var_13205_cast_fp16 = softmax(axis = var_2624, x = aw_1053_cast_fp16)[name = tensor("op_13205_cast_fp16")]; + tensor var_13206_cast_fp16 = softmax(axis = var_2624, x = aw_1055_cast_fp16)[name = tensor("op_13206_cast_fp16")]; + tensor var_13207_cast_fp16 = softmax(axis = var_2624, x = aw_1057_cast_fp16)[name = tensor("op_13207_cast_fp16")]; + tensor var_13208_cast_fp16 = softmax(axis = var_2624, x = aw_1059_cast_fp16)[name = tensor("op_13208_cast_fp16")]; + tensor var_13209_cast_fp16 = softmax(axis = var_2624, x = aw_1061_cast_fp16)[name = tensor("op_13209_cast_fp16")]; + tensor var_13210_cast_fp16 = softmax(axis = var_2624, x = aw_1063_cast_fp16)[name = tensor("op_13210_cast_fp16")]; + tensor var_13211_cast_fp16 = softmax(axis = var_2624, x = aw_1065_cast_fp16)[name = tensor("op_13211_cast_fp16")]; + tensor var_13212_cast_fp16 = softmax(axis = var_2624, x = aw_1067_cast_fp16)[name = tensor("op_13212_cast_fp16")]; + tensor var_13213_cast_fp16 = softmax(axis = var_2624, x = aw_1069_cast_fp16)[name = tensor("op_13213_cast_fp16")]; + tensor var_13214_cast_fp16 = softmax(axis = var_2624, x = aw_1071_cast_fp16)[name = tensor("op_13214_cast_fp16")]; + tensor var_13215_cast_fp16 = softmax(axis = var_2624, x = aw_1073_cast_fp16)[name = tensor("op_13215_cast_fp16")]; + tensor var_13216_cast_fp16 = softmax(axis = var_2624, x = aw_1075_cast_fp16)[name = tensor("op_13216_cast_fp16")]; + tensor var_13217_cast_fp16 = softmax(axis = var_2624, x = aw_1077_cast_fp16)[name = tensor("op_13217_cast_fp16")]; + tensor var_13218_cast_fp16 = softmax(axis = var_2624, x = aw_1079_cast_fp16)[name = tensor("op_13218_cast_fp16")]; + tensor var_13220_equation_0 = const()[name = tensor("op_13220_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13220_cast_fp16 = einsum(equation = var_13220_equation_0, values = (var_13040_cast_fp16, var_13199_cast_fp16))[name = tensor("op_13220_cast_fp16")]; + tensor var_13222_equation_0 = const()[name = tensor("op_13222_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13222_cast_fp16 = einsum(equation = var_13222_equation_0, values = (var_13044_cast_fp16, var_13200_cast_fp16))[name = tensor("op_13222_cast_fp16")]; + tensor var_13224_equation_0 = const()[name = tensor("op_13224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13224_cast_fp16 = einsum(equation = var_13224_equation_0, values = (var_13048_cast_fp16, var_13201_cast_fp16))[name = tensor("op_13224_cast_fp16")]; + tensor var_13226_equation_0 = const()[name = tensor("op_13226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13226_cast_fp16 = einsum(equation = var_13226_equation_0, values = (var_13052_cast_fp16, var_13202_cast_fp16))[name = tensor("op_13226_cast_fp16")]; + tensor var_13228_equation_0 = const()[name = tensor("op_13228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13228_cast_fp16 = einsum(equation = var_13228_equation_0, values = (var_13056_cast_fp16, var_13203_cast_fp16))[name = tensor("op_13228_cast_fp16")]; + tensor var_13230_equation_0 = const()[name = tensor("op_13230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13230_cast_fp16 = einsum(equation = var_13230_equation_0, values = (var_13060_cast_fp16, var_13204_cast_fp16))[name = tensor("op_13230_cast_fp16")]; + tensor var_13232_equation_0 = const()[name = tensor("op_13232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13232_cast_fp16 = einsum(equation = var_13232_equation_0, values = (var_13064_cast_fp16, var_13205_cast_fp16))[name = tensor("op_13232_cast_fp16")]; + tensor var_13234_equation_0 = const()[name = tensor("op_13234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13234_cast_fp16 = einsum(equation = var_13234_equation_0, values = (var_13068_cast_fp16, var_13206_cast_fp16))[name = tensor("op_13234_cast_fp16")]; + tensor var_13236_equation_0 = const()[name = tensor("op_13236_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13236_cast_fp16 = einsum(equation = var_13236_equation_0, values = (var_13072_cast_fp16, var_13207_cast_fp16))[name = tensor("op_13236_cast_fp16")]; + tensor var_13238_equation_0 = const()[name = tensor("op_13238_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13238_cast_fp16 = einsum(equation = var_13238_equation_0, values = (var_13076_cast_fp16, var_13208_cast_fp16))[name = tensor("op_13238_cast_fp16")]; + tensor var_13240_equation_0 = const()[name = tensor("op_13240_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13240_cast_fp16 = einsum(equation = var_13240_equation_0, values = (var_13080_cast_fp16, var_13209_cast_fp16))[name = tensor("op_13240_cast_fp16")]; + tensor var_13242_equation_0 = const()[name = tensor("op_13242_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13242_cast_fp16 = einsum(equation = var_13242_equation_0, values = (var_13084_cast_fp16, var_13210_cast_fp16))[name = tensor("op_13242_cast_fp16")]; + tensor var_13244_equation_0 = const()[name = tensor("op_13244_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13244_cast_fp16 = einsum(equation = var_13244_equation_0, values = (var_13088_cast_fp16, var_13211_cast_fp16))[name = tensor("op_13244_cast_fp16")]; + tensor var_13246_equation_0 = const()[name = tensor("op_13246_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13246_cast_fp16 = einsum(equation = var_13246_equation_0, values = (var_13092_cast_fp16, var_13212_cast_fp16))[name = tensor("op_13246_cast_fp16")]; + tensor var_13248_equation_0 = const()[name = tensor("op_13248_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13248_cast_fp16 = einsum(equation = var_13248_equation_0, values = (var_13096_cast_fp16, var_13213_cast_fp16))[name = tensor("op_13248_cast_fp16")]; + tensor var_13250_equation_0 = const()[name = tensor("op_13250_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13250_cast_fp16 = einsum(equation = var_13250_equation_0, values = (var_13100_cast_fp16, var_13214_cast_fp16))[name = tensor("op_13250_cast_fp16")]; + tensor var_13252_equation_0 = const()[name = tensor("op_13252_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13252_cast_fp16 = einsum(equation = var_13252_equation_0, values = (var_13104_cast_fp16, var_13215_cast_fp16))[name = tensor("op_13252_cast_fp16")]; + tensor var_13254_equation_0 = const()[name = tensor("op_13254_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13254_cast_fp16 = einsum(equation = var_13254_equation_0, values = (var_13108_cast_fp16, var_13216_cast_fp16))[name = tensor("op_13254_cast_fp16")]; + tensor var_13256_equation_0 = const()[name = tensor("op_13256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13256_cast_fp16 = einsum(equation = var_13256_equation_0, values = (var_13112_cast_fp16, var_13217_cast_fp16))[name = tensor("op_13256_cast_fp16")]; + tensor var_13258_equation_0 = const()[name = tensor("op_13258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13258_cast_fp16 = einsum(equation = var_13258_equation_0, values = (var_13116_cast_fp16, var_13218_cast_fp16))[name = tensor("op_13258_cast_fp16")]; + tensor input_237_interleave_0 = const()[name = tensor("input_237_interleave_0"), val = tensor(false)]; + tensor input_237_cast_fp16 = concat(axis = var_2624, interleave = input_237_interleave_0, values = (var_13220_cast_fp16, var_13222_cast_fp16, var_13224_cast_fp16, var_13226_cast_fp16, var_13228_cast_fp16, var_13230_cast_fp16, var_13232_cast_fp16, var_13234_cast_fp16, var_13236_cast_fp16, var_13238_cast_fp16, var_13240_cast_fp16, var_13242_cast_fp16, var_13244_cast_fp16, var_13246_cast_fp16, var_13248_cast_fp16, var_13250_cast_fp16, var_13252_cast_fp16, var_13254_cast_fp16, var_13256_cast_fp16, var_13258_cast_fp16))[name = tensor("input_237_cast_fp16")]; + tensor var_13268_pad_type_0 = const()[name = tensor("op_13268_pad_type_0"), val = tensor("valid")]; + tensor var_13268_strides_0 = const()[name = tensor("op_13268_strides_0"), val = tensor([1, 1])]; + tensor var_13268_pad_0 = const()[name = tensor("op_13268_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13268_dilations_0 = const()[name = tensor("op_13268_dilations_0"), val = tensor([1, 1])]; + tensor var_13268_groups_0 = const()[name = tensor("op_13268_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391037376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392266240))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392266432)))]; + tensor var_13268_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_13268_dilations_0, groups = var_13268_groups_0, pad = var_13268_pad_0, pad_type = var_13268_pad_type_0, strides = var_13268_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_237_cast_fp16)[name = tensor("op_13268_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = var_13268_cast_fp16, y = inputs_91_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; + tensor hidden_states_145_axes_0 = const()[name = tensor("hidden_states_145_axes_0"), val = tensor([1])]; + tensor hidden_states_145_gamma_0_to_fp16 = const()[name = tensor("hidden_states_145_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392269056)))]; + tensor hidden_states_145_beta_0_to_fp16 = const()[name = tensor("hidden_states_145_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392271680)))]; + tensor var_13278_to_fp16 = const()[name = tensor("op_13278_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_145_cast_fp16 = layer_norm(axes = hidden_states_145_axes_0, beta = hidden_states_145_beta_0_to_fp16, epsilon = var_13278_to_fp16, gamma = hidden_states_145_gamma_0_to_fp16, x = inputs_93_cast_fp16)[name = tensor("hidden_states_145_cast_fp16")]; + tensor q_63_pad_type_0 = const()[name = tensor("q_63_pad_type_0"), val = tensor("valid")]; + tensor q_63_strides_0 = const()[name = tensor("q_63_strides_0"), val = tensor([1, 1])]; + tensor q_63_pad_0 = const()[name = tensor("q_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_63_dilations_0 = const()[name = tensor("q_63_dilations_0"), val = tensor([1, 1])]; + tensor q_63_groups_0 = const()[name = tensor("q_63_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392274304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(393503168))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_63_cast_fp16 = conv(dilations = q_63_dilations_0, groups = q_63_groups_0, pad = q_63_pad_0, pad_type = q_63_pad_type_0, strides = q_63_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_145_cast_fp16)[name = tensor("q_63_cast_fp16")]; + tensor k_125_pad_type_0 = const()[name = tensor("k_125_pad_type_0"), val = tensor("valid")]; + tensor k_125_strides_0 = const()[name = tensor("k_125_strides_0"), val = tensor([1, 1])]; + tensor k_125_pad_0 = const()[name = tensor("k_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_125_dilations_0 = const()[name = tensor("k_125_dilations_0"), val = tensor([1, 1])]; + tensor k_125_groups_0 = const()[name = tensor("k_125_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(393503360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395469504))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_125_cast_fp16 = conv(dilations = k_125_dilations_0, groups = k_125_groups_0, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = k_125_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_125_cast_fp16")]; + tensor v_63_pad_type_0 = const()[name = tensor("v_63_pad_type_0"), val = tensor("valid")]; + tensor v_63_strides_0 = const()[name = tensor("v_63_strides_0"), val = tensor([1, 1])]; + tensor v_63_pad_0 = const()[name = tensor("v_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_63_dilations_0 = const()[name = tensor("v_63_dilations_0"), val = tensor([1, 1])]; + tensor v_63_groups_0 = const()[name = tensor("v_63_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395469696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(397435840))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_63_cast_fp16 = conv(dilations = v_63_dilations_0, groups = v_63_groups_0, pad = v_63_pad_0, pad_type = v_63_pad_type_0, strides = v_63_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_63_cast_fp16")]; + tensor var_13311_begin_0 = const()[name = tensor("op_13311_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13311_end_0 = const()[name = tensor("op_13311_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13311_end_mask_0 = const()[name = tensor("op_13311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13311_cast_fp16 = slice_by_index(begin = var_13311_begin_0, end = var_13311_end_0, end_mask = var_13311_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13311_cast_fp16")]; + tensor var_13315_begin_0 = const()[name = tensor("op_13315_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13315_end_0 = const()[name = tensor("op_13315_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13315_end_mask_0 = const()[name = tensor("op_13315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13315_cast_fp16 = slice_by_index(begin = var_13315_begin_0, end = var_13315_end_0, end_mask = var_13315_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13315_cast_fp16")]; + tensor var_13319_begin_0 = const()[name = tensor("op_13319_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13319_end_0 = const()[name = tensor("op_13319_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13319_end_mask_0 = const()[name = tensor("op_13319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13319_cast_fp16 = slice_by_index(begin = var_13319_begin_0, end = var_13319_end_0, end_mask = var_13319_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13319_cast_fp16")]; + tensor var_13323_begin_0 = const()[name = tensor("op_13323_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13323_end_0 = const()[name = tensor("op_13323_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13323_end_mask_0 = const()[name = tensor("op_13323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13323_cast_fp16 = slice_by_index(begin = var_13323_begin_0, end = var_13323_end_0, end_mask = var_13323_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13323_cast_fp16")]; + tensor var_13327_begin_0 = const()[name = tensor("op_13327_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13327_end_0 = const()[name = tensor("op_13327_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13327_end_mask_0 = const()[name = tensor("op_13327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13327_cast_fp16 = slice_by_index(begin = var_13327_begin_0, end = var_13327_end_0, end_mask = var_13327_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13327_cast_fp16")]; + tensor var_13331_begin_0 = const()[name = tensor("op_13331_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13331_end_0 = const()[name = tensor("op_13331_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13331_end_mask_0 = const()[name = tensor("op_13331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13331_cast_fp16 = slice_by_index(begin = var_13331_begin_0, end = var_13331_end_0, end_mask = var_13331_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13331_cast_fp16")]; + tensor var_13335_begin_0 = const()[name = tensor("op_13335_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13335_end_0 = const()[name = tensor("op_13335_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13335_end_mask_0 = const()[name = tensor("op_13335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13335_cast_fp16 = slice_by_index(begin = var_13335_begin_0, end = var_13335_end_0, end_mask = var_13335_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13335_cast_fp16")]; + tensor var_13339_begin_0 = const()[name = tensor("op_13339_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13339_end_0 = const()[name = tensor("op_13339_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13339_end_mask_0 = const()[name = tensor("op_13339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13339_cast_fp16 = slice_by_index(begin = var_13339_begin_0, end = var_13339_end_0, end_mask = var_13339_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13339_cast_fp16")]; + tensor var_13343_begin_0 = const()[name = tensor("op_13343_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13343_end_0 = const()[name = tensor("op_13343_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13343_end_mask_0 = const()[name = tensor("op_13343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13343_cast_fp16 = slice_by_index(begin = var_13343_begin_0, end = var_13343_end_0, end_mask = var_13343_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13343_cast_fp16")]; + tensor var_13347_begin_0 = const()[name = tensor("op_13347_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13347_end_0 = const()[name = tensor("op_13347_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13347_end_mask_0 = const()[name = tensor("op_13347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13347_cast_fp16 = slice_by_index(begin = var_13347_begin_0, end = var_13347_end_0, end_mask = var_13347_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13347_cast_fp16")]; + tensor var_13351_begin_0 = const()[name = tensor("op_13351_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13351_end_0 = const()[name = tensor("op_13351_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13351_end_mask_0 = const()[name = tensor("op_13351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13351_cast_fp16 = slice_by_index(begin = var_13351_begin_0, end = var_13351_end_0, end_mask = var_13351_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13351_cast_fp16")]; + tensor var_13355_begin_0 = const()[name = tensor("op_13355_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13355_end_0 = const()[name = tensor("op_13355_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13355_end_mask_0 = const()[name = tensor("op_13355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13355_cast_fp16 = slice_by_index(begin = var_13355_begin_0, end = var_13355_end_0, end_mask = var_13355_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13355_cast_fp16")]; + tensor var_13359_begin_0 = const()[name = tensor("op_13359_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13359_end_0 = const()[name = tensor("op_13359_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13359_end_mask_0 = const()[name = tensor("op_13359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13359_cast_fp16 = slice_by_index(begin = var_13359_begin_0, end = var_13359_end_0, end_mask = var_13359_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13359_cast_fp16")]; + tensor var_13363_begin_0 = const()[name = tensor("op_13363_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13363_end_0 = const()[name = tensor("op_13363_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_13363_end_mask_0 = const()[name = tensor("op_13363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13363_cast_fp16 = slice_by_index(begin = var_13363_begin_0, end = var_13363_end_0, end_mask = var_13363_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13363_cast_fp16")]; + tensor var_13367_begin_0 = const()[name = tensor("op_13367_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13367_end_0 = const()[name = tensor("op_13367_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_13367_end_mask_0 = const()[name = tensor("op_13367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13367_cast_fp16 = slice_by_index(begin = var_13367_begin_0, end = var_13367_end_0, end_mask = var_13367_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13367_cast_fp16")]; + tensor var_13371_begin_0 = const()[name = tensor("op_13371_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13371_end_0 = const()[name = tensor("op_13371_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13371_end_mask_0 = const()[name = tensor("op_13371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13371_cast_fp16 = slice_by_index(begin = var_13371_begin_0, end = var_13371_end_0, end_mask = var_13371_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13371_cast_fp16")]; + tensor var_13375_begin_0 = const()[name = tensor("op_13375_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13375_end_0 = const()[name = tensor("op_13375_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_13375_end_mask_0 = const()[name = tensor("op_13375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13375_cast_fp16 = slice_by_index(begin = var_13375_begin_0, end = var_13375_end_0, end_mask = var_13375_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13375_cast_fp16")]; + tensor var_13379_begin_0 = const()[name = tensor("op_13379_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13379_end_0 = const()[name = tensor("op_13379_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_13379_end_mask_0 = const()[name = tensor("op_13379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13379_cast_fp16 = slice_by_index(begin = var_13379_begin_0, end = var_13379_end_0, end_mask = var_13379_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13379_cast_fp16")]; + tensor var_13383_begin_0 = const()[name = tensor("op_13383_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13383_end_0 = const()[name = tensor("op_13383_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_13383_end_mask_0 = const()[name = tensor("op_13383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13383_cast_fp16 = slice_by_index(begin = var_13383_begin_0, end = var_13383_end_0, end_mask = var_13383_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13383_cast_fp16")]; + tensor var_13387_begin_0 = const()[name = tensor("op_13387_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13387_end_0 = const()[name = tensor("op_13387_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_13387_end_mask_0 = const()[name = tensor("op_13387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13387_cast_fp16 = slice_by_index(begin = var_13387_begin_0, end = var_13387_end_0, end_mask = var_13387_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13387_cast_fp16")]; + tensor k_127_perm_0 = const()[name = tensor("k_127_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13394_begin_0 = const()[name = tensor("op_13394_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13394_end_0 = const()[name = tensor("op_13394_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_13394_end_mask_0 = const()[name = tensor("op_13394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_127_cast_fp16 = transpose(perm = k_127_perm_0, x = k_125_cast_fp16)[name = tensor("transpose_36")]; + tensor var_13394_cast_fp16 = slice_by_index(begin = var_13394_begin_0, end = var_13394_end_0, end_mask = var_13394_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13394_cast_fp16")]; + tensor var_13398_begin_0 = const()[name = tensor("op_13398_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13398_end_0 = const()[name = tensor("op_13398_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_13398_end_mask_0 = const()[name = tensor("op_13398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13398_cast_fp16 = slice_by_index(begin = var_13398_begin_0, end = var_13398_end_0, end_mask = var_13398_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13398_cast_fp16")]; + tensor var_13402_begin_0 = const()[name = tensor("op_13402_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13402_end_0 = const()[name = tensor("op_13402_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_13402_end_mask_0 = const()[name = tensor("op_13402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13402_cast_fp16 = slice_by_index(begin = var_13402_begin_0, end = var_13402_end_0, end_mask = var_13402_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13402_cast_fp16")]; + tensor var_13406_begin_0 = const()[name = tensor("op_13406_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13406_end_0 = const()[name = tensor("op_13406_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_13406_end_mask_0 = const()[name = tensor("op_13406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13406_cast_fp16 = slice_by_index(begin = var_13406_begin_0, end = var_13406_end_0, end_mask = var_13406_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13406_cast_fp16")]; + tensor var_13410_begin_0 = const()[name = tensor("op_13410_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13410_end_0 = const()[name = tensor("op_13410_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_13410_end_mask_0 = const()[name = tensor("op_13410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13410_cast_fp16 = slice_by_index(begin = var_13410_begin_0, end = var_13410_end_0, end_mask = var_13410_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13410_cast_fp16")]; + tensor var_13414_begin_0 = const()[name = tensor("op_13414_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_13414_end_0 = const()[name = tensor("op_13414_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_13414_end_mask_0 = const()[name = tensor("op_13414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13414_cast_fp16 = slice_by_index(begin = var_13414_begin_0, end = var_13414_end_0, end_mask = var_13414_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13414_cast_fp16")]; + tensor var_13418_begin_0 = const()[name = tensor("op_13418_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_13418_end_0 = const()[name = tensor("op_13418_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_13418_end_mask_0 = const()[name = tensor("op_13418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13418_cast_fp16 = slice_by_index(begin = var_13418_begin_0, end = var_13418_end_0, end_mask = var_13418_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13418_cast_fp16")]; + tensor var_13422_begin_0 = const()[name = tensor("op_13422_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_13422_end_0 = const()[name = tensor("op_13422_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_13422_end_mask_0 = const()[name = tensor("op_13422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13422_cast_fp16 = slice_by_index(begin = var_13422_begin_0, end = var_13422_end_0, end_mask = var_13422_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13422_cast_fp16")]; + tensor var_13426_begin_0 = const()[name = tensor("op_13426_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13426_end_0 = const()[name = tensor("op_13426_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_13426_end_mask_0 = const()[name = tensor("op_13426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13426_cast_fp16 = slice_by_index(begin = var_13426_begin_0, end = var_13426_end_0, end_mask = var_13426_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13426_cast_fp16")]; + tensor var_13430_begin_0 = const()[name = tensor("op_13430_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_13430_end_0 = const()[name = tensor("op_13430_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_13430_end_mask_0 = const()[name = tensor("op_13430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13430_cast_fp16 = slice_by_index(begin = var_13430_begin_0, end = var_13430_end_0, end_mask = var_13430_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13430_cast_fp16")]; + tensor var_13434_begin_0 = const()[name = tensor("op_13434_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13434_end_0 = const()[name = tensor("op_13434_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_13434_end_mask_0 = const()[name = tensor("op_13434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13434_cast_fp16 = slice_by_index(begin = var_13434_begin_0, end = var_13434_end_0, end_mask = var_13434_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13434_cast_fp16")]; + tensor var_13438_begin_0 = const()[name = tensor("op_13438_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13438_end_0 = const()[name = tensor("op_13438_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_13438_end_mask_0 = const()[name = tensor("op_13438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13438_cast_fp16 = slice_by_index(begin = var_13438_begin_0, end = var_13438_end_0, end_mask = var_13438_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13438_cast_fp16")]; + tensor var_13442_begin_0 = const()[name = tensor("op_13442_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13442_end_0 = const()[name = tensor("op_13442_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_13442_end_mask_0 = const()[name = tensor("op_13442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13442_cast_fp16 = slice_by_index(begin = var_13442_begin_0, end = var_13442_end_0, end_mask = var_13442_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13442_cast_fp16")]; + tensor var_13446_begin_0 = const()[name = tensor("op_13446_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13446_end_0 = const()[name = tensor("op_13446_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_13446_end_mask_0 = const()[name = tensor("op_13446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13446_cast_fp16 = slice_by_index(begin = var_13446_begin_0, end = var_13446_end_0, end_mask = var_13446_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13446_cast_fp16")]; + tensor var_13450_begin_0 = const()[name = tensor("op_13450_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13450_end_0 = const()[name = tensor("op_13450_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_13450_end_mask_0 = const()[name = tensor("op_13450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13450_cast_fp16 = slice_by_index(begin = var_13450_begin_0, end = var_13450_end_0, end_mask = var_13450_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13450_cast_fp16")]; + tensor var_13454_begin_0 = const()[name = tensor("op_13454_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13454_end_0 = const()[name = tensor("op_13454_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_13454_end_mask_0 = const()[name = tensor("op_13454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13454_cast_fp16 = slice_by_index(begin = var_13454_begin_0, end = var_13454_end_0, end_mask = var_13454_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13454_cast_fp16")]; + tensor var_13458_begin_0 = const()[name = tensor("op_13458_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13458_end_0 = const()[name = tensor("op_13458_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_13458_end_mask_0 = const()[name = tensor("op_13458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13458_cast_fp16 = slice_by_index(begin = var_13458_begin_0, end = var_13458_end_0, end_mask = var_13458_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13458_cast_fp16")]; + tensor var_13462_begin_0 = const()[name = tensor("op_13462_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13462_end_0 = const()[name = tensor("op_13462_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_13462_end_mask_0 = const()[name = tensor("op_13462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13462_cast_fp16 = slice_by_index(begin = var_13462_begin_0, end = var_13462_end_0, end_mask = var_13462_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13462_cast_fp16")]; + tensor var_13466_begin_0 = const()[name = tensor("op_13466_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13466_end_0 = const()[name = tensor("op_13466_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_13466_end_mask_0 = const()[name = tensor("op_13466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13466_cast_fp16 = slice_by_index(begin = var_13466_begin_0, end = var_13466_end_0, end_mask = var_13466_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13466_cast_fp16")]; + tensor var_13470_begin_0 = const()[name = tensor("op_13470_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13470_end_0 = const()[name = tensor("op_13470_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_13470_end_mask_0 = const()[name = tensor("op_13470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13470_cast_fp16 = slice_by_index(begin = var_13470_begin_0, end = var_13470_end_0, end_mask = var_13470_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13470_cast_fp16")]; + tensor var_13472_begin_0 = const()[name = tensor("op_13472_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13472_end_0 = const()[name = tensor("op_13472_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_13472_end_mask_0 = const()[name = tensor("op_13472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13472_cast_fp16 = slice_by_index(begin = var_13472_begin_0, end = var_13472_end_0, end_mask = var_13472_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13472_cast_fp16")]; + tensor var_13476_begin_0 = const()[name = tensor("op_13476_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13476_end_0 = const()[name = tensor("op_13476_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_13476_end_mask_0 = const()[name = tensor("op_13476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13476_cast_fp16 = slice_by_index(begin = var_13476_begin_0, end = var_13476_end_0, end_mask = var_13476_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13476_cast_fp16")]; + tensor var_13480_begin_0 = const()[name = tensor("op_13480_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13480_end_0 = const()[name = tensor("op_13480_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_13480_end_mask_0 = const()[name = tensor("op_13480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13480_cast_fp16 = slice_by_index(begin = var_13480_begin_0, end = var_13480_end_0, end_mask = var_13480_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13480_cast_fp16")]; + tensor var_13484_begin_0 = const()[name = tensor("op_13484_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13484_end_0 = const()[name = tensor("op_13484_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_13484_end_mask_0 = const()[name = tensor("op_13484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13484_cast_fp16 = slice_by_index(begin = var_13484_begin_0, end = var_13484_end_0, end_mask = var_13484_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13484_cast_fp16")]; + tensor var_13488_begin_0 = const()[name = tensor("op_13488_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13488_end_0 = const()[name = tensor("op_13488_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_13488_end_mask_0 = const()[name = tensor("op_13488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13488_cast_fp16 = slice_by_index(begin = var_13488_begin_0, end = var_13488_end_0, end_mask = var_13488_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13488_cast_fp16")]; + tensor var_13492_begin_0 = const()[name = tensor("op_13492_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13492_end_0 = const()[name = tensor("op_13492_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_13492_end_mask_0 = const()[name = tensor("op_13492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13492_cast_fp16 = slice_by_index(begin = var_13492_begin_0, end = var_13492_end_0, end_mask = var_13492_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13492_cast_fp16")]; + tensor var_13496_begin_0 = const()[name = tensor("op_13496_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13496_end_0 = const()[name = tensor("op_13496_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_13496_end_mask_0 = const()[name = tensor("op_13496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13496_cast_fp16 = slice_by_index(begin = var_13496_begin_0, end = var_13496_end_0, end_mask = var_13496_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13496_cast_fp16")]; + tensor var_13500_begin_0 = const()[name = tensor("op_13500_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13500_end_0 = const()[name = tensor("op_13500_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_13500_end_mask_0 = const()[name = tensor("op_13500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13500_cast_fp16 = slice_by_index(begin = var_13500_begin_0, end = var_13500_end_0, end_mask = var_13500_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13500_cast_fp16")]; + tensor var_13504_begin_0 = const()[name = tensor("op_13504_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13504_end_0 = const()[name = tensor("op_13504_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_13504_end_mask_0 = const()[name = tensor("op_13504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13504_cast_fp16 = slice_by_index(begin = var_13504_begin_0, end = var_13504_end_0, end_mask = var_13504_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13504_cast_fp16")]; + tensor var_13508_begin_0 = const()[name = tensor("op_13508_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13508_end_0 = const()[name = tensor("op_13508_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_13508_end_mask_0 = const()[name = tensor("op_13508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13508_cast_fp16 = slice_by_index(begin = var_13508_begin_0, end = var_13508_end_0, end_mask = var_13508_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13508_cast_fp16")]; + tensor var_13512_begin_0 = const()[name = tensor("op_13512_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13512_end_0 = const()[name = tensor("op_13512_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_13512_end_mask_0 = const()[name = tensor("op_13512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13512_cast_fp16 = slice_by_index(begin = var_13512_begin_0, end = var_13512_end_0, end_mask = var_13512_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13512_cast_fp16")]; + tensor var_13516_begin_0 = const()[name = tensor("op_13516_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13516_end_0 = const()[name = tensor("op_13516_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_13516_end_mask_0 = const()[name = tensor("op_13516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13516_cast_fp16 = slice_by_index(begin = var_13516_begin_0, end = var_13516_end_0, end_mask = var_13516_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13516_cast_fp16")]; + tensor var_13520_begin_0 = const()[name = tensor("op_13520_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13520_end_0 = const()[name = tensor("op_13520_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_13520_end_mask_0 = const()[name = tensor("op_13520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13520_cast_fp16 = slice_by_index(begin = var_13520_begin_0, end = var_13520_end_0, end_mask = var_13520_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13520_cast_fp16")]; + tensor var_13524_begin_0 = const()[name = tensor("op_13524_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13524_end_0 = const()[name = tensor("op_13524_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_13524_end_mask_0 = const()[name = tensor("op_13524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13524_cast_fp16 = slice_by_index(begin = var_13524_begin_0, end = var_13524_end_0, end_mask = var_13524_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13524_cast_fp16")]; + tensor var_13528_begin_0 = const()[name = tensor("op_13528_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13528_end_0 = const()[name = tensor("op_13528_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_13528_end_mask_0 = const()[name = tensor("op_13528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13528_cast_fp16 = slice_by_index(begin = var_13528_begin_0, end = var_13528_end_0, end_mask = var_13528_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13528_cast_fp16")]; + tensor var_13532_begin_0 = const()[name = tensor("op_13532_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13532_end_0 = const()[name = tensor("op_13532_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_13532_end_mask_0 = const()[name = tensor("op_13532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13532_cast_fp16 = slice_by_index(begin = var_13532_begin_0, end = var_13532_end_0, end_mask = var_13532_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13532_cast_fp16")]; + tensor var_13536_begin_0 = const()[name = tensor("op_13536_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13536_end_0 = const()[name = tensor("op_13536_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_13536_end_mask_0 = const()[name = tensor("op_13536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13536_cast_fp16 = slice_by_index(begin = var_13536_begin_0, end = var_13536_end_0, end_mask = var_13536_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13536_cast_fp16")]; + tensor var_13540_begin_0 = const()[name = tensor("op_13540_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13540_end_0 = const()[name = tensor("op_13540_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_13540_end_mask_0 = const()[name = tensor("op_13540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13540_cast_fp16 = slice_by_index(begin = var_13540_begin_0, end = var_13540_end_0, end_mask = var_13540_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13540_cast_fp16")]; + tensor var_13544_begin_0 = const()[name = tensor("op_13544_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13544_end_0 = const()[name = tensor("op_13544_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_13544_end_mask_0 = const()[name = tensor("op_13544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13544_cast_fp16 = slice_by_index(begin = var_13544_begin_0, end = var_13544_end_0, end_mask = var_13544_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13544_cast_fp16")]; + tensor var_13548_begin_0 = const()[name = tensor("op_13548_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13548_end_0 = const()[name = tensor("op_13548_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_13548_end_mask_0 = const()[name = tensor("op_13548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13548_cast_fp16 = slice_by_index(begin = var_13548_begin_0, end = var_13548_end_0, end_mask = var_13548_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13548_cast_fp16")]; + tensor var_13552_equation_0 = const()[name = tensor("op_13552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13552_cast_fp16 = einsum(equation = var_13552_equation_0, values = (var_13394_cast_fp16, var_13311_cast_fp16))[name = tensor("op_13552_cast_fp16")]; + tensor var_13553_to_fp16 = const()[name = tensor("op_13553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1081_cast_fp16 = mul(x = var_13552_cast_fp16, y = var_13553_to_fp16)[name = tensor("aw_1081_cast_fp16")]; + tensor var_13556_equation_0 = const()[name = tensor("op_13556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13556_cast_fp16 = einsum(equation = var_13556_equation_0, values = (var_13398_cast_fp16, var_13315_cast_fp16))[name = tensor("op_13556_cast_fp16")]; + tensor var_13557_to_fp16 = const()[name = tensor("op_13557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1083_cast_fp16 = mul(x = var_13556_cast_fp16, y = var_13557_to_fp16)[name = tensor("aw_1083_cast_fp16")]; + tensor var_13560_equation_0 = const()[name = tensor("op_13560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13560_cast_fp16 = einsum(equation = var_13560_equation_0, values = (var_13402_cast_fp16, var_13319_cast_fp16))[name = tensor("op_13560_cast_fp16")]; + tensor var_13561_to_fp16 = const()[name = tensor("op_13561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1085_cast_fp16 = mul(x = var_13560_cast_fp16, y = var_13561_to_fp16)[name = tensor("aw_1085_cast_fp16")]; + tensor var_13564_equation_0 = const()[name = tensor("op_13564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13564_cast_fp16 = einsum(equation = var_13564_equation_0, values = (var_13406_cast_fp16, var_13323_cast_fp16))[name = tensor("op_13564_cast_fp16")]; + tensor var_13565_to_fp16 = const()[name = tensor("op_13565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1087_cast_fp16 = mul(x = var_13564_cast_fp16, y = var_13565_to_fp16)[name = tensor("aw_1087_cast_fp16")]; + tensor var_13568_equation_0 = const()[name = tensor("op_13568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13568_cast_fp16 = einsum(equation = var_13568_equation_0, values = (var_13410_cast_fp16, var_13327_cast_fp16))[name = tensor("op_13568_cast_fp16")]; + tensor var_13569_to_fp16 = const()[name = tensor("op_13569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1089_cast_fp16 = mul(x = var_13568_cast_fp16, y = var_13569_to_fp16)[name = tensor("aw_1089_cast_fp16")]; + tensor var_13572_equation_0 = const()[name = tensor("op_13572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13572_cast_fp16 = einsum(equation = var_13572_equation_0, values = (var_13414_cast_fp16, var_13331_cast_fp16))[name = tensor("op_13572_cast_fp16")]; + tensor var_13573_to_fp16 = const()[name = tensor("op_13573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1091_cast_fp16 = mul(x = var_13572_cast_fp16, y = var_13573_to_fp16)[name = tensor("aw_1091_cast_fp16")]; + tensor var_13576_equation_0 = const()[name = tensor("op_13576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13576_cast_fp16 = einsum(equation = var_13576_equation_0, values = (var_13418_cast_fp16, var_13335_cast_fp16))[name = tensor("op_13576_cast_fp16")]; + tensor var_13577_to_fp16 = const()[name = tensor("op_13577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1093_cast_fp16 = mul(x = var_13576_cast_fp16, y = var_13577_to_fp16)[name = tensor("aw_1093_cast_fp16")]; + tensor var_13580_equation_0 = const()[name = tensor("op_13580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13580_cast_fp16 = einsum(equation = var_13580_equation_0, values = (var_13422_cast_fp16, var_13339_cast_fp16))[name = tensor("op_13580_cast_fp16")]; + tensor var_13581_to_fp16 = const()[name = tensor("op_13581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1095_cast_fp16 = mul(x = var_13580_cast_fp16, y = var_13581_to_fp16)[name = tensor("aw_1095_cast_fp16")]; + tensor var_13584_equation_0 = const()[name = tensor("op_13584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13584_cast_fp16 = einsum(equation = var_13584_equation_0, values = (var_13426_cast_fp16, var_13343_cast_fp16))[name = tensor("op_13584_cast_fp16")]; + tensor var_13585_to_fp16 = const()[name = tensor("op_13585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1097_cast_fp16 = mul(x = var_13584_cast_fp16, y = var_13585_to_fp16)[name = tensor("aw_1097_cast_fp16")]; + tensor var_13588_equation_0 = const()[name = tensor("op_13588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13588_cast_fp16 = einsum(equation = var_13588_equation_0, values = (var_13430_cast_fp16, var_13347_cast_fp16))[name = tensor("op_13588_cast_fp16")]; + tensor var_13589_to_fp16 = const()[name = tensor("op_13589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1099_cast_fp16 = mul(x = var_13588_cast_fp16, y = var_13589_to_fp16)[name = tensor("aw_1099_cast_fp16")]; + tensor var_13592_equation_0 = const()[name = tensor("op_13592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13592_cast_fp16 = einsum(equation = var_13592_equation_0, values = (var_13434_cast_fp16, var_13351_cast_fp16))[name = tensor("op_13592_cast_fp16")]; + tensor var_13593_to_fp16 = const()[name = tensor("op_13593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1101_cast_fp16 = mul(x = var_13592_cast_fp16, y = var_13593_to_fp16)[name = tensor("aw_1101_cast_fp16")]; + tensor var_13596_equation_0 = const()[name = tensor("op_13596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13596_cast_fp16 = einsum(equation = var_13596_equation_0, values = (var_13438_cast_fp16, var_13355_cast_fp16))[name = tensor("op_13596_cast_fp16")]; + tensor var_13597_to_fp16 = const()[name = tensor("op_13597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1103_cast_fp16 = mul(x = var_13596_cast_fp16, y = var_13597_to_fp16)[name = tensor("aw_1103_cast_fp16")]; + tensor var_13600_equation_0 = const()[name = tensor("op_13600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13600_cast_fp16 = einsum(equation = var_13600_equation_0, values = (var_13442_cast_fp16, var_13359_cast_fp16))[name = tensor("op_13600_cast_fp16")]; + tensor var_13601_to_fp16 = const()[name = tensor("op_13601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1105_cast_fp16 = mul(x = var_13600_cast_fp16, y = var_13601_to_fp16)[name = tensor("aw_1105_cast_fp16")]; + tensor var_13604_equation_0 = const()[name = tensor("op_13604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13604_cast_fp16 = einsum(equation = var_13604_equation_0, values = (var_13446_cast_fp16, var_13363_cast_fp16))[name = tensor("op_13604_cast_fp16")]; + tensor var_13605_to_fp16 = const()[name = tensor("op_13605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1107_cast_fp16 = mul(x = var_13604_cast_fp16, y = var_13605_to_fp16)[name = tensor("aw_1107_cast_fp16")]; + tensor var_13608_equation_0 = const()[name = tensor("op_13608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13608_cast_fp16 = einsum(equation = var_13608_equation_0, values = (var_13450_cast_fp16, var_13367_cast_fp16))[name = tensor("op_13608_cast_fp16")]; + tensor var_13609_to_fp16 = const()[name = tensor("op_13609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1109_cast_fp16 = mul(x = var_13608_cast_fp16, y = var_13609_to_fp16)[name = tensor("aw_1109_cast_fp16")]; + tensor var_13612_equation_0 = const()[name = tensor("op_13612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13612_cast_fp16 = einsum(equation = var_13612_equation_0, values = (var_13454_cast_fp16, var_13371_cast_fp16))[name = tensor("op_13612_cast_fp16")]; + tensor var_13613_to_fp16 = const()[name = tensor("op_13613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1111_cast_fp16 = mul(x = var_13612_cast_fp16, y = var_13613_to_fp16)[name = tensor("aw_1111_cast_fp16")]; + tensor var_13616_equation_0 = const()[name = tensor("op_13616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13616_cast_fp16 = einsum(equation = var_13616_equation_0, values = (var_13458_cast_fp16, var_13375_cast_fp16))[name = tensor("op_13616_cast_fp16")]; + tensor var_13617_to_fp16 = const()[name = tensor("op_13617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1113_cast_fp16 = mul(x = var_13616_cast_fp16, y = var_13617_to_fp16)[name = tensor("aw_1113_cast_fp16")]; + tensor var_13620_equation_0 = const()[name = tensor("op_13620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13620_cast_fp16 = einsum(equation = var_13620_equation_0, values = (var_13462_cast_fp16, var_13379_cast_fp16))[name = tensor("op_13620_cast_fp16")]; + tensor var_13621_to_fp16 = const()[name = tensor("op_13621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1115_cast_fp16 = mul(x = var_13620_cast_fp16, y = var_13621_to_fp16)[name = tensor("aw_1115_cast_fp16")]; + tensor var_13624_equation_0 = const()[name = tensor("op_13624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13624_cast_fp16 = einsum(equation = var_13624_equation_0, values = (var_13466_cast_fp16, var_13383_cast_fp16))[name = tensor("op_13624_cast_fp16")]; + tensor var_13625_to_fp16 = const()[name = tensor("op_13625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1117_cast_fp16 = mul(x = var_13624_cast_fp16, y = var_13625_to_fp16)[name = tensor("aw_1117_cast_fp16")]; + tensor var_13628_equation_0 = const()[name = tensor("op_13628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13628_cast_fp16 = einsum(equation = var_13628_equation_0, values = (var_13470_cast_fp16, var_13387_cast_fp16))[name = tensor("op_13628_cast_fp16")]; + tensor var_13629_to_fp16 = const()[name = tensor("op_13629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1119_cast_fp16 = mul(x = var_13628_cast_fp16, y = var_13629_to_fp16)[name = tensor("aw_1119_cast_fp16")]; + tensor var_13631_cast_fp16 = softmax(axis = var_2624, x = aw_1081_cast_fp16)[name = tensor("op_13631_cast_fp16")]; + tensor var_13632_cast_fp16 = softmax(axis = var_2624, x = aw_1083_cast_fp16)[name = tensor("op_13632_cast_fp16")]; + tensor var_13633_cast_fp16 = softmax(axis = var_2624, x = aw_1085_cast_fp16)[name = tensor("op_13633_cast_fp16")]; + tensor var_13634_cast_fp16 = softmax(axis = var_2624, x = aw_1087_cast_fp16)[name = tensor("op_13634_cast_fp16")]; + tensor var_13635_cast_fp16 = softmax(axis = var_2624, x = aw_1089_cast_fp16)[name = tensor("op_13635_cast_fp16")]; + tensor var_13636_cast_fp16 = softmax(axis = var_2624, x = aw_1091_cast_fp16)[name = tensor("op_13636_cast_fp16")]; + tensor var_13637_cast_fp16 = softmax(axis = var_2624, x = aw_1093_cast_fp16)[name = tensor("op_13637_cast_fp16")]; + tensor var_13638_cast_fp16 = softmax(axis = var_2624, x = aw_1095_cast_fp16)[name = tensor("op_13638_cast_fp16")]; + tensor var_13639_cast_fp16 = softmax(axis = var_2624, x = aw_1097_cast_fp16)[name = tensor("op_13639_cast_fp16")]; + tensor var_13640_cast_fp16 = softmax(axis = var_2624, x = aw_1099_cast_fp16)[name = tensor("op_13640_cast_fp16")]; + tensor var_13641_cast_fp16 = softmax(axis = var_2624, x = aw_1101_cast_fp16)[name = tensor("op_13641_cast_fp16")]; + tensor var_13642_cast_fp16 = softmax(axis = var_2624, x = aw_1103_cast_fp16)[name = tensor("op_13642_cast_fp16")]; + tensor var_13643_cast_fp16 = softmax(axis = var_2624, x = aw_1105_cast_fp16)[name = tensor("op_13643_cast_fp16")]; + tensor var_13644_cast_fp16 = softmax(axis = var_2624, x = aw_1107_cast_fp16)[name = tensor("op_13644_cast_fp16")]; + tensor var_13645_cast_fp16 = softmax(axis = var_2624, x = aw_1109_cast_fp16)[name = tensor("op_13645_cast_fp16")]; + tensor var_13646_cast_fp16 = softmax(axis = var_2624, x = aw_1111_cast_fp16)[name = tensor("op_13646_cast_fp16")]; + tensor var_13647_cast_fp16 = softmax(axis = var_2624, x = aw_1113_cast_fp16)[name = tensor("op_13647_cast_fp16")]; + tensor var_13648_cast_fp16 = softmax(axis = var_2624, x = aw_1115_cast_fp16)[name = tensor("op_13648_cast_fp16")]; + tensor var_13649_cast_fp16 = softmax(axis = var_2624, x = aw_1117_cast_fp16)[name = tensor("op_13649_cast_fp16")]; + tensor var_13650_cast_fp16 = softmax(axis = var_2624, x = aw_1119_cast_fp16)[name = tensor("op_13650_cast_fp16")]; + tensor var_13652_equation_0 = const()[name = tensor("op_13652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13652_cast_fp16 = einsum(equation = var_13652_equation_0, values = (var_13472_cast_fp16, var_13631_cast_fp16))[name = tensor("op_13652_cast_fp16")]; + tensor var_13654_equation_0 = const()[name = tensor("op_13654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13654_cast_fp16 = einsum(equation = var_13654_equation_0, values = (var_13476_cast_fp16, var_13632_cast_fp16))[name = tensor("op_13654_cast_fp16")]; + tensor var_13656_equation_0 = const()[name = tensor("op_13656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13656_cast_fp16 = einsum(equation = var_13656_equation_0, values = (var_13480_cast_fp16, var_13633_cast_fp16))[name = tensor("op_13656_cast_fp16")]; + tensor var_13658_equation_0 = const()[name = tensor("op_13658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13658_cast_fp16 = einsum(equation = var_13658_equation_0, values = (var_13484_cast_fp16, var_13634_cast_fp16))[name = tensor("op_13658_cast_fp16")]; + tensor var_13660_equation_0 = const()[name = tensor("op_13660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13660_cast_fp16 = einsum(equation = var_13660_equation_0, values = (var_13488_cast_fp16, var_13635_cast_fp16))[name = tensor("op_13660_cast_fp16")]; + tensor var_13662_equation_0 = const()[name = tensor("op_13662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13662_cast_fp16 = einsum(equation = var_13662_equation_0, values = (var_13492_cast_fp16, var_13636_cast_fp16))[name = tensor("op_13662_cast_fp16")]; + tensor var_13664_equation_0 = const()[name = tensor("op_13664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13664_cast_fp16 = einsum(equation = var_13664_equation_0, values = (var_13496_cast_fp16, var_13637_cast_fp16))[name = tensor("op_13664_cast_fp16")]; + tensor var_13666_equation_0 = const()[name = tensor("op_13666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13666_cast_fp16 = einsum(equation = var_13666_equation_0, values = (var_13500_cast_fp16, var_13638_cast_fp16))[name = tensor("op_13666_cast_fp16")]; + tensor var_13668_equation_0 = const()[name = tensor("op_13668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13668_cast_fp16 = einsum(equation = var_13668_equation_0, values = (var_13504_cast_fp16, var_13639_cast_fp16))[name = tensor("op_13668_cast_fp16")]; + tensor var_13670_equation_0 = const()[name = tensor("op_13670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13670_cast_fp16 = einsum(equation = var_13670_equation_0, values = (var_13508_cast_fp16, var_13640_cast_fp16))[name = tensor("op_13670_cast_fp16")]; + tensor var_13672_equation_0 = const()[name = tensor("op_13672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13672_cast_fp16 = einsum(equation = var_13672_equation_0, values = (var_13512_cast_fp16, var_13641_cast_fp16))[name = tensor("op_13672_cast_fp16")]; + tensor var_13674_equation_0 = const()[name = tensor("op_13674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13674_cast_fp16 = einsum(equation = var_13674_equation_0, values = (var_13516_cast_fp16, var_13642_cast_fp16))[name = tensor("op_13674_cast_fp16")]; + tensor var_13676_equation_0 = const()[name = tensor("op_13676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13676_cast_fp16 = einsum(equation = var_13676_equation_0, values = (var_13520_cast_fp16, var_13643_cast_fp16))[name = tensor("op_13676_cast_fp16")]; + tensor var_13678_equation_0 = const()[name = tensor("op_13678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13678_cast_fp16 = einsum(equation = var_13678_equation_0, values = (var_13524_cast_fp16, var_13644_cast_fp16))[name = tensor("op_13678_cast_fp16")]; + tensor var_13680_equation_0 = const()[name = tensor("op_13680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13680_cast_fp16 = einsum(equation = var_13680_equation_0, values = (var_13528_cast_fp16, var_13645_cast_fp16))[name = tensor("op_13680_cast_fp16")]; + tensor var_13682_equation_0 = const()[name = tensor("op_13682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13682_cast_fp16 = einsum(equation = var_13682_equation_0, values = (var_13532_cast_fp16, var_13646_cast_fp16))[name = tensor("op_13682_cast_fp16")]; + tensor var_13684_equation_0 = const()[name = tensor("op_13684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13684_cast_fp16 = einsum(equation = var_13684_equation_0, values = (var_13536_cast_fp16, var_13647_cast_fp16))[name = tensor("op_13684_cast_fp16")]; + tensor var_13686_equation_0 = const()[name = tensor("op_13686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13686_cast_fp16 = einsum(equation = var_13686_equation_0, values = (var_13540_cast_fp16, var_13648_cast_fp16))[name = tensor("op_13686_cast_fp16")]; + tensor var_13688_equation_0 = const()[name = tensor("op_13688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13688_cast_fp16 = einsum(equation = var_13688_equation_0, values = (var_13544_cast_fp16, var_13649_cast_fp16))[name = tensor("op_13688_cast_fp16")]; + tensor var_13690_equation_0 = const()[name = tensor("op_13690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13690_cast_fp16 = einsum(equation = var_13690_equation_0, values = (var_13548_cast_fp16, var_13650_cast_fp16))[name = tensor("op_13690_cast_fp16")]; + tensor input_239_interleave_0 = const()[name = tensor("input_239_interleave_0"), val = tensor(false)]; + tensor input_239_cast_fp16 = concat(axis = var_2624, interleave = input_239_interleave_0, values = (var_13652_cast_fp16, var_13654_cast_fp16, var_13656_cast_fp16, var_13658_cast_fp16, var_13660_cast_fp16, var_13662_cast_fp16, var_13664_cast_fp16, var_13666_cast_fp16, var_13668_cast_fp16, var_13670_cast_fp16, var_13672_cast_fp16, var_13674_cast_fp16, var_13676_cast_fp16, var_13678_cast_fp16, var_13680_cast_fp16, var_13682_cast_fp16, var_13684_cast_fp16, var_13686_cast_fp16, var_13688_cast_fp16, var_13690_cast_fp16))[name = tensor("input_239_cast_fp16")]; + tensor var_13700_pad_type_0 = const()[name = tensor("op_13700_pad_type_0"), val = tensor("valid")]; + tensor var_13700_strides_0 = const()[name = tensor("op_13700_strides_0"), val = tensor([1, 1])]; + tensor var_13700_pad_0 = const()[name = tensor("op_13700_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13700_dilations_0 = const()[name = tensor("op_13700_dilations_0"), val = tensor([1, 1])]; + tensor var_13700_groups_0 = const()[name = tensor("op_13700_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(397436032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398664896))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398665088)))]; + tensor var_13700_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_13700_dilations_0, groups = var_13700_groups_0, pad = var_13700_pad_0, pad_type = var_13700_pad_type_0, strides = var_13700_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_239_cast_fp16)[name = tensor("op_13700_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = var_13700_cast_fp16, y = inputs_93_cast_fp16)[name = tensor("inputs_95_cast_fp16")]; + tensor input_241_axes_0 = const()[name = tensor("input_241_axes_0"), val = tensor([1])]; + tensor input_241_gamma_0_to_fp16 = const()[name = tensor("input_241_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398667712)))]; + tensor input_241_beta_0_to_fp16 = const()[name = tensor("input_241_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398670336)))]; + tensor var_13710_to_fp16 = const()[name = tensor("op_13710_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_241_cast_fp16 = layer_norm(axes = input_241_axes_0, beta = input_241_beta_0_to_fp16, epsilon = var_13710_to_fp16, gamma = input_241_gamma_0_to_fp16, x = inputs_95_cast_fp16)[name = tensor("input_241_cast_fp16")]; + tensor var_13730_pad_type_0 = const()[name = tensor("op_13730_pad_type_0"), val = tensor("valid")]; + tensor var_13730_strides_0 = const()[name = tensor("op_13730_strides_0"), val = tensor([1, 1])]; + tensor var_13730_pad_0 = const()[name = tensor("op_13730_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13730_dilations_0 = const()[name = tensor("op_13730_dilations_0"), val = tensor([1, 1])]; + tensor var_13730_groups_0 = const()[name = tensor("op_13730_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(398672960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408503424))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408503616)))]; + tensor var_13730_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_13730_dilations_0, groups = var_13730_groups_0, pad = var_13730_pad_0, pad_type = var_13730_pad_type_0, strides = var_13730_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_241_cast_fp16)[name = tensor("op_13730_cast_fp16")]; + tensor var_13731_split_sizes_0 = const()[name = tensor("op_13731_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_13731_axis_0 = const()[name = tensor("op_13731_axis_0"), val = tensor(1)]; + tensor var_13731_cast_fp16_0, tensor var_13731_cast_fp16_1 = split(axis = var_13731_axis_0, split_sizes = var_13731_split_sizes_0, x = var_13730_cast_fp16)[name = tensor("op_13731_cast_fp16")]; + tensor var_13733_mode_0 = const()[name = tensor("op_13733_mode_0"), val = tensor("EXACT")]; + tensor var_13733_cast_fp16 = gelu(mode = var_13733_mode_0, x = var_13731_cast_fp16_1)[name = tensor("op_13733_cast_fp16")]; + tensor input_243_cast_fp16 = mul(x = var_13731_cast_fp16_0, y = var_13733_cast_fp16)[name = tensor("input_243_cast_fp16")]; + tensor var_13741_pad_type_0 = const()[name = tensor("op_13741_pad_type_0"), val = tensor("valid")]; + tensor var_13741_strides_0 = const()[name = tensor("op_13741_strides_0"), val = tensor([1, 1])]; + tensor var_13741_pad_0 = const()[name = tensor("op_13741_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13741_dilations_0 = const()[name = tensor("op_13741_dilations_0"), val = tensor([1, 1])]; + tensor var_13741_groups_0 = const()[name = tensor("op_13741_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408524160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(413439424))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(413439616)))]; + tensor var_13741_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_13741_dilations_0, groups = var_13741_groups_0, pad = var_13741_pad_0, pad_type = var_13741_pad_type_0, strides = var_13741_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_243_cast_fp16)[name = tensor("op_13741_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = var_13741_cast_fp16, y = inputs_95_cast_fp16)[name = tensor("inputs_97_cast_fp16")]; + tensor hidden_states_149_axes_0 = const()[name = tensor("hidden_states_149_axes_0"), val = tensor([1])]; + tensor hidden_states_149_gamma_0_to_fp16 = const()[name = tensor("hidden_states_149_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(413442240)))]; + tensor hidden_states_149_beta_0_to_fp16 = const()[name = tensor("hidden_states_149_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(413444864)))]; + tensor var_13757_to_fp16 = const()[name = tensor("op_13757_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_149_cast_fp16 = layer_norm(axes = hidden_states_149_axes_0, beta = hidden_states_149_beta_0_to_fp16, epsilon = var_13757_to_fp16, gamma = hidden_states_149_gamma_0_to_fp16, x = inputs_97_cast_fp16)[name = tensor("hidden_states_149_cast_fp16")]; + tensor q_65_pad_type_0 = const()[name = tensor("q_65_pad_type_0"), val = tensor("valid")]; + tensor q_65_strides_0 = const()[name = tensor("q_65_strides_0"), val = tensor([1, 1])]; + tensor q_65_pad_0 = const()[name = tensor("q_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_65_dilations_0 = const()[name = tensor("q_65_dilations_0"), val = tensor([1, 1])]; + tensor q_65_groups_0 = const()[name = tensor("q_65_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(413447488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414676352))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_65_cast_fp16 = conv(dilations = q_65_dilations_0, groups = q_65_groups_0, pad = q_65_pad_0, pad_type = q_65_pad_type_0, strides = q_65_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_149_cast_fp16)[name = tensor("q_65_cast_fp16")]; + tensor k_129_pad_type_0 = const()[name = tensor("k_129_pad_type_0"), val = tensor("valid")]; + tensor k_129_strides_0 = const()[name = tensor("k_129_strides_0"), val = tensor([1, 1])]; + tensor k_129_pad_0 = const()[name = tensor("k_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_129_dilations_0 = const()[name = tensor("k_129_dilations_0"), val = tensor([1, 1])]; + tensor k_129_groups_0 = const()[name = tensor("k_129_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414676544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415905408))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_129_cast_fp16 = conv(dilations = k_129_dilations_0, groups = k_129_groups_0, pad = k_129_pad_0, pad_type = k_129_pad_type_0, strides = k_129_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_149_cast_fp16)[name = tensor("k_129_cast_fp16")]; + tensor v_65_pad_type_0 = const()[name = tensor("v_65_pad_type_0"), val = tensor("valid")]; + tensor v_65_strides_0 = const()[name = tensor("v_65_strides_0"), val = tensor([1, 1])]; + tensor v_65_pad_0 = const()[name = tensor("v_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_65_dilations_0 = const()[name = tensor("v_65_dilations_0"), val = tensor([1, 1])]; + tensor v_65_groups_0 = const()[name = tensor("v_65_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415905600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417134464))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_65_cast_fp16 = conv(dilations = v_65_dilations_0, groups = v_65_groups_0, pad = v_65_pad_0, pad_type = v_65_pad_type_0, strides = v_65_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_149_cast_fp16)[name = tensor("v_65_cast_fp16")]; + tensor var_13790_begin_0 = const()[name = tensor("op_13790_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13790_end_0 = const()[name = tensor("op_13790_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13790_end_mask_0 = const()[name = tensor("op_13790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13790_cast_fp16 = slice_by_index(begin = var_13790_begin_0, end = var_13790_end_0, end_mask = var_13790_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13790_cast_fp16")]; + tensor var_13794_begin_0 = const()[name = tensor("op_13794_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13794_end_0 = const()[name = tensor("op_13794_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13794_end_mask_0 = const()[name = tensor("op_13794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13794_cast_fp16 = slice_by_index(begin = var_13794_begin_0, end = var_13794_end_0, end_mask = var_13794_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13794_cast_fp16")]; + tensor var_13798_begin_0 = const()[name = tensor("op_13798_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13798_end_0 = const()[name = tensor("op_13798_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13798_end_mask_0 = const()[name = tensor("op_13798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13798_cast_fp16 = slice_by_index(begin = var_13798_begin_0, end = var_13798_end_0, end_mask = var_13798_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13798_cast_fp16")]; + tensor var_13802_begin_0 = const()[name = tensor("op_13802_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13802_end_0 = const()[name = tensor("op_13802_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13802_end_mask_0 = const()[name = tensor("op_13802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13802_cast_fp16 = slice_by_index(begin = var_13802_begin_0, end = var_13802_end_0, end_mask = var_13802_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13802_cast_fp16")]; + tensor var_13806_begin_0 = const()[name = tensor("op_13806_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13806_end_0 = const()[name = tensor("op_13806_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13806_end_mask_0 = const()[name = tensor("op_13806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13806_cast_fp16 = slice_by_index(begin = var_13806_begin_0, end = var_13806_end_0, end_mask = var_13806_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13806_cast_fp16")]; + tensor var_13810_begin_0 = const()[name = tensor("op_13810_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13810_end_0 = const()[name = tensor("op_13810_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13810_end_mask_0 = const()[name = tensor("op_13810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13810_cast_fp16 = slice_by_index(begin = var_13810_begin_0, end = var_13810_end_0, end_mask = var_13810_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13810_cast_fp16")]; + tensor var_13814_begin_0 = const()[name = tensor("op_13814_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13814_end_0 = const()[name = tensor("op_13814_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13814_end_mask_0 = const()[name = tensor("op_13814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13814_cast_fp16 = slice_by_index(begin = var_13814_begin_0, end = var_13814_end_0, end_mask = var_13814_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13814_cast_fp16")]; + tensor var_13818_begin_0 = const()[name = tensor("op_13818_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13818_end_0 = const()[name = tensor("op_13818_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13818_end_mask_0 = const()[name = tensor("op_13818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13818_cast_fp16 = slice_by_index(begin = var_13818_begin_0, end = var_13818_end_0, end_mask = var_13818_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13818_cast_fp16")]; + tensor var_13822_begin_0 = const()[name = tensor("op_13822_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13822_end_0 = const()[name = tensor("op_13822_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13822_end_mask_0 = const()[name = tensor("op_13822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13822_cast_fp16 = slice_by_index(begin = var_13822_begin_0, end = var_13822_end_0, end_mask = var_13822_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13822_cast_fp16")]; + tensor var_13826_begin_0 = const()[name = tensor("op_13826_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13826_end_0 = const()[name = tensor("op_13826_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13826_end_mask_0 = const()[name = tensor("op_13826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13826_cast_fp16 = slice_by_index(begin = var_13826_begin_0, end = var_13826_end_0, end_mask = var_13826_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13826_cast_fp16")]; + tensor var_13830_begin_0 = const()[name = tensor("op_13830_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13830_end_0 = const()[name = tensor("op_13830_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13830_end_mask_0 = const()[name = tensor("op_13830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13830_cast_fp16 = slice_by_index(begin = var_13830_begin_0, end = var_13830_end_0, end_mask = var_13830_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13830_cast_fp16")]; + tensor var_13834_begin_0 = const()[name = tensor("op_13834_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13834_end_0 = const()[name = tensor("op_13834_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13834_end_mask_0 = const()[name = tensor("op_13834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13834_cast_fp16 = slice_by_index(begin = var_13834_begin_0, end = var_13834_end_0, end_mask = var_13834_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13834_cast_fp16")]; + tensor var_13838_begin_0 = const()[name = tensor("op_13838_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13838_end_0 = const()[name = tensor("op_13838_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13838_end_mask_0 = const()[name = tensor("op_13838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13838_cast_fp16 = slice_by_index(begin = var_13838_begin_0, end = var_13838_end_0, end_mask = var_13838_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13838_cast_fp16")]; + tensor var_13842_begin_0 = const()[name = tensor("op_13842_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13842_end_0 = const()[name = tensor("op_13842_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_13842_end_mask_0 = const()[name = tensor("op_13842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13842_cast_fp16 = slice_by_index(begin = var_13842_begin_0, end = var_13842_end_0, end_mask = var_13842_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13842_cast_fp16")]; + tensor var_13846_begin_0 = const()[name = tensor("op_13846_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13846_end_0 = const()[name = tensor("op_13846_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_13846_end_mask_0 = const()[name = tensor("op_13846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13846_cast_fp16 = slice_by_index(begin = var_13846_begin_0, end = var_13846_end_0, end_mask = var_13846_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13846_cast_fp16")]; + tensor var_13850_begin_0 = const()[name = tensor("op_13850_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13850_end_0 = const()[name = tensor("op_13850_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13850_end_mask_0 = const()[name = tensor("op_13850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13850_cast_fp16 = slice_by_index(begin = var_13850_begin_0, end = var_13850_end_0, end_mask = var_13850_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13850_cast_fp16")]; + tensor var_13854_begin_0 = const()[name = tensor("op_13854_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13854_end_0 = const()[name = tensor("op_13854_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_13854_end_mask_0 = const()[name = tensor("op_13854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13854_cast_fp16 = slice_by_index(begin = var_13854_begin_0, end = var_13854_end_0, end_mask = var_13854_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13854_cast_fp16")]; + tensor var_13858_begin_0 = const()[name = tensor("op_13858_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13858_end_0 = const()[name = tensor("op_13858_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_13858_end_mask_0 = const()[name = tensor("op_13858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13858_cast_fp16 = slice_by_index(begin = var_13858_begin_0, end = var_13858_end_0, end_mask = var_13858_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13858_cast_fp16")]; + tensor var_13862_begin_0 = const()[name = tensor("op_13862_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13862_end_0 = const()[name = tensor("op_13862_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_13862_end_mask_0 = const()[name = tensor("op_13862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13862_cast_fp16 = slice_by_index(begin = var_13862_begin_0, end = var_13862_end_0, end_mask = var_13862_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13862_cast_fp16")]; + tensor var_13866_begin_0 = const()[name = tensor("op_13866_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13866_end_0 = const()[name = tensor("op_13866_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_13866_end_mask_0 = const()[name = tensor("op_13866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13866_cast_fp16 = slice_by_index(begin = var_13866_begin_0, end = var_13866_end_0, end_mask = var_13866_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13866_cast_fp16")]; + tensor k_131_perm_0 = const()[name = tensor("k_131_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13873_begin_0 = const()[name = tensor("op_13873_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13873_end_0 = const()[name = tensor("op_13873_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_13873_end_mask_0 = const()[name = tensor("op_13873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_131_cast_fp16 = transpose(perm = k_131_perm_0, x = k_129_cast_fp16)[name = tensor("transpose_35")]; + tensor var_13873_cast_fp16 = slice_by_index(begin = var_13873_begin_0, end = var_13873_end_0, end_mask = var_13873_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13873_cast_fp16")]; + tensor var_13877_begin_0 = const()[name = tensor("op_13877_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13877_end_0 = const()[name = tensor("op_13877_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_13877_end_mask_0 = const()[name = tensor("op_13877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13877_cast_fp16 = slice_by_index(begin = var_13877_begin_0, end = var_13877_end_0, end_mask = var_13877_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13877_cast_fp16")]; + tensor var_13881_begin_0 = const()[name = tensor("op_13881_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13881_end_0 = const()[name = tensor("op_13881_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_13881_end_mask_0 = const()[name = tensor("op_13881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13881_cast_fp16 = slice_by_index(begin = var_13881_begin_0, end = var_13881_end_0, end_mask = var_13881_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13881_cast_fp16")]; + tensor var_13885_begin_0 = const()[name = tensor("op_13885_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13885_end_0 = const()[name = tensor("op_13885_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_13885_end_mask_0 = const()[name = tensor("op_13885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13885_cast_fp16 = slice_by_index(begin = var_13885_begin_0, end = var_13885_end_0, end_mask = var_13885_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13885_cast_fp16")]; + tensor var_13889_begin_0 = const()[name = tensor("op_13889_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13889_end_0 = const()[name = tensor("op_13889_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_13889_end_mask_0 = const()[name = tensor("op_13889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13889_cast_fp16 = slice_by_index(begin = var_13889_begin_0, end = var_13889_end_0, end_mask = var_13889_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13889_cast_fp16")]; + tensor var_13893_begin_0 = const()[name = tensor("op_13893_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_13893_end_0 = const()[name = tensor("op_13893_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_13893_end_mask_0 = const()[name = tensor("op_13893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13893_cast_fp16 = slice_by_index(begin = var_13893_begin_0, end = var_13893_end_0, end_mask = var_13893_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13893_cast_fp16")]; + tensor var_13897_begin_0 = const()[name = tensor("op_13897_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_13897_end_0 = const()[name = tensor("op_13897_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_13897_end_mask_0 = const()[name = tensor("op_13897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13897_cast_fp16 = slice_by_index(begin = var_13897_begin_0, end = var_13897_end_0, end_mask = var_13897_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13897_cast_fp16")]; + tensor var_13901_begin_0 = const()[name = tensor("op_13901_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_13901_end_0 = const()[name = tensor("op_13901_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_13901_end_mask_0 = const()[name = tensor("op_13901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13901_cast_fp16 = slice_by_index(begin = var_13901_begin_0, end = var_13901_end_0, end_mask = var_13901_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13901_cast_fp16")]; + tensor var_13905_begin_0 = const()[name = tensor("op_13905_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13905_end_0 = const()[name = tensor("op_13905_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_13905_end_mask_0 = const()[name = tensor("op_13905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13905_cast_fp16 = slice_by_index(begin = var_13905_begin_0, end = var_13905_end_0, end_mask = var_13905_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13905_cast_fp16")]; + tensor var_13909_begin_0 = const()[name = tensor("op_13909_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_13909_end_0 = const()[name = tensor("op_13909_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_13909_end_mask_0 = const()[name = tensor("op_13909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13909_cast_fp16 = slice_by_index(begin = var_13909_begin_0, end = var_13909_end_0, end_mask = var_13909_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13909_cast_fp16")]; + tensor var_13913_begin_0 = const()[name = tensor("op_13913_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13913_end_0 = const()[name = tensor("op_13913_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_13913_end_mask_0 = const()[name = tensor("op_13913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13913_cast_fp16 = slice_by_index(begin = var_13913_begin_0, end = var_13913_end_0, end_mask = var_13913_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13913_cast_fp16")]; + tensor var_13917_begin_0 = const()[name = tensor("op_13917_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13917_end_0 = const()[name = tensor("op_13917_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_13917_end_mask_0 = const()[name = tensor("op_13917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13917_cast_fp16 = slice_by_index(begin = var_13917_begin_0, end = var_13917_end_0, end_mask = var_13917_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13917_cast_fp16")]; + tensor var_13921_begin_0 = const()[name = tensor("op_13921_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13921_end_0 = const()[name = tensor("op_13921_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_13921_end_mask_0 = const()[name = tensor("op_13921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13921_cast_fp16 = slice_by_index(begin = var_13921_begin_0, end = var_13921_end_0, end_mask = var_13921_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13921_cast_fp16")]; + tensor var_13925_begin_0 = const()[name = tensor("op_13925_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13925_end_0 = const()[name = tensor("op_13925_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_13925_end_mask_0 = const()[name = tensor("op_13925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13925_cast_fp16 = slice_by_index(begin = var_13925_begin_0, end = var_13925_end_0, end_mask = var_13925_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13925_cast_fp16")]; + tensor var_13929_begin_0 = const()[name = tensor("op_13929_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13929_end_0 = const()[name = tensor("op_13929_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_13929_end_mask_0 = const()[name = tensor("op_13929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13929_cast_fp16 = slice_by_index(begin = var_13929_begin_0, end = var_13929_end_0, end_mask = var_13929_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13929_cast_fp16")]; + tensor var_13933_begin_0 = const()[name = tensor("op_13933_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13933_end_0 = const()[name = tensor("op_13933_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13933_end_mask_0 = const()[name = tensor("op_13933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13933_cast_fp16 = slice_by_index(begin = var_13933_begin_0, end = var_13933_end_0, end_mask = var_13933_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13933_cast_fp16")]; + tensor var_13937_begin_0 = const()[name = tensor("op_13937_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13937_end_0 = const()[name = tensor("op_13937_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_13937_end_mask_0 = const()[name = tensor("op_13937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13937_cast_fp16 = slice_by_index(begin = var_13937_begin_0, end = var_13937_end_0, end_mask = var_13937_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13937_cast_fp16")]; + tensor var_13941_begin_0 = const()[name = tensor("op_13941_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13941_end_0 = const()[name = tensor("op_13941_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_13941_end_mask_0 = const()[name = tensor("op_13941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13941_cast_fp16 = slice_by_index(begin = var_13941_begin_0, end = var_13941_end_0, end_mask = var_13941_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13941_cast_fp16")]; + tensor var_13945_begin_0 = const()[name = tensor("op_13945_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13945_end_0 = const()[name = tensor("op_13945_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_13945_end_mask_0 = const()[name = tensor("op_13945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13945_cast_fp16 = slice_by_index(begin = var_13945_begin_0, end = var_13945_end_0, end_mask = var_13945_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13945_cast_fp16")]; + tensor var_13949_begin_0 = const()[name = tensor("op_13949_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13949_end_0 = const()[name = tensor("op_13949_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_13949_end_mask_0 = const()[name = tensor("op_13949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13949_cast_fp16 = slice_by_index(begin = var_13949_begin_0, end = var_13949_end_0, end_mask = var_13949_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13949_cast_fp16")]; + tensor var_13951_begin_0 = const()[name = tensor("op_13951_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13951_end_0 = const()[name = tensor("op_13951_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13951_end_mask_0 = const()[name = tensor("op_13951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13951_cast_fp16 = slice_by_index(begin = var_13951_begin_0, end = var_13951_end_0, end_mask = var_13951_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13951_cast_fp16")]; + tensor var_13955_begin_0 = const()[name = tensor("op_13955_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13955_end_0 = const()[name = tensor("op_13955_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13955_end_mask_0 = const()[name = tensor("op_13955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13955_cast_fp16 = slice_by_index(begin = var_13955_begin_0, end = var_13955_end_0, end_mask = var_13955_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13955_cast_fp16")]; + tensor var_13959_begin_0 = const()[name = tensor("op_13959_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13959_end_0 = const()[name = tensor("op_13959_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13959_end_mask_0 = const()[name = tensor("op_13959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13959_cast_fp16 = slice_by_index(begin = var_13959_begin_0, end = var_13959_end_0, end_mask = var_13959_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13959_cast_fp16")]; + tensor var_13963_begin_0 = const()[name = tensor("op_13963_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13963_end_0 = const()[name = tensor("op_13963_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13963_end_mask_0 = const()[name = tensor("op_13963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13963_cast_fp16 = slice_by_index(begin = var_13963_begin_0, end = var_13963_end_0, end_mask = var_13963_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13963_cast_fp16")]; + tensor var_13967_begin_0 = const()[name = tensor("op_13967_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13967_end_0 = const()[name = tensor("op_13967_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13967_end_mask_0 = const()[name = tensor("op_13967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13967_cast_fp16 = slice_by_index(begin = var_13967_begin_0, end = var_13967_end_0, end_mask = var_13967_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13967_cast_fp16")]; + tensor var_13971_begin_0 = const()[name = tensor("op_13971_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13971_end_0 = const()[name = tensor("op_13971_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13971_end_mask_0 = const()[name = tensor("op_13971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13971_cast_fp16 = slice_by_index(begin = var_13971_begin_0, end = var_13971_end_0, end_mask = var_13971_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13971_cast_fp16")]; + tensor var_13975_begin_0 = const()[name = tensor("op_13975_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13975_end_0 = const()[name = tensor("op_13975_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13975_end_mask_0 = const()[name = tensor("op_13975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13975_cast_fp16 = slice_by_index(begin = var_13975_begin_0, end = var_13975_end_0, end_mask = var_13975_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13975_cast_fp16")]; + tensor var_13979_begin_0 = const()[name = tensor("op_13979_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13979_end_0 = const()[name = tensor("op_13979_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13979_end_mask_0 = const()[name = tensor("op_13979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13979_cast_fp16 = slice_by_index(begin = var_13979_begin_0, end = var_13979_end_0, end_mask = var_13979_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13979_cast_fp16")]; + tensor var_13983_begin_0 = const()[name = tensor("op_13983_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13983_end_0 = const()[name = tensor("op_13983_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13983_end_mask_0 = const()[name = tensor("op_13983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13983_cast_fp16 = slice_by_index(begin = var_13983_begin_0, end = var_13983_end_0, end_mask = var_13983_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13983_cast_fp16")]; + tensor var_13987_begin_0 = const()[name = tensor("op_13987_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13987_end_0 = const()[name = tensor("op_13987_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13987_end_mask_0 = const()[name = tensor("op_13987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13987_cast_fp16 = slice_by_index(begin = var_13987_begin_0, end = var_13987_end_0, end_mask = var_13987_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13987_cast_fp16")]; + tensor var_13991_begin_0 = const()[name = tensor("op_13991_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13991_end_0 = const()[name = tensor("op_13991_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13991_end_mask_0 = const()[name = tensor("op_13991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13991_cast_fp16 = slice_by_index(begin = var_13991_begin_0, end = var_13991_end_0, end_mask = var_13991_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13991_cast_fp16")]; + tensor var_13995_begin_0 = const()[name = tensor("op_13995_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13995_end_0 = const()[name = tensor("op_13995_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13995_end_mask_0 = const()[name = tensor("op_13995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13995_cast_fp16 = slice_by_index(begin = var_13995_begin_0, end = var_13995_end_0, end_mask = var_13995_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13995_cast_fp16")]; + tensor var_13999_begin_0 = const()[name = tensor("op_13999_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13999_end_0 = const()[name = tensor("op_13999_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13999_end_mask_0 = const()[name = tensor("op_13999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13999_cast_fp16 = slice_by_index(begin = var_13999_begin_0, end = var_13999_end_0, end_mask = var_13999_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13999_cast_fp16")]; + tensor var_14003_begin_0 = const()[name = tensor("op_14003_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14003_end_0 = const()[name = tensor("op_14003_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14003_end_mask_0 = const()[name = tensor("op_14003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14003_cast_fp16 = slice_by_index(begin = var_14003_begin_0, end = var_14003_end_0, end_mask = var_14003_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14003_cast_fp16")]; + tensor var_14007_begin_0 = const()[name = tensor("op_14007_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14007_end_0 = const()[name = tensor("op_14007_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14007_end_mask_0 = const()[name = tensor("op_14007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14007_cast_fp16 = slice_by_index(begin = var_14007_begin_0, end = var_14007_end_0, end_mask = var_14007_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14007_cast_fp16")]; + tensor var_14011_begin_0 = const()[name = tensor("op_14011_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14011_end_0 = const()[name = tensor("op_14011_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14011_end_mask_0 = const()[name = tensor("op_14011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14011_cast_fp16 = slice_by_index(begin = var_14011_begin_0, end = var_14011_end_0, end_mask = var_14011_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14011_cast_fp16")]; + tensor var_14015_begin_0 = const()[name = tensor("op_14015_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14015_end_0 = const()[name = tensor("op_14015_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14015_end_mask_0 = const()[name = tensor("op_14015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14015_cast_fp16 = slice_by_index(begin = var_14015_begin_0, end = var_14015_end_0, end_mask = var_14015_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14015_cast_fp16")]; + tensor var_14019_begin_0 = const()[name = tensor("op_14019_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14019_end_0 = const()[name = tensor("op_14019_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14019_end_mask_0 = const()[name = tensor("op_14019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14019_cast_fp16 = slice_by_index(begin = var_14019_begin_0, end = var_14019_end_0, end_mask = var_14019_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14019_cast_fp16")]; + tensor var_14023_begin_0 = const()[name = tensor("op_14023_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14023_end_0 = const()[name = tensor("op_14023_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14023_end_mask_0 = const()[name = tensor("op_14023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14023_cast_fp16 = slice_by_index(begin = var_14023_begin_0, end = var_14023_end_0, end_mask = var_14023_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14023_cast_fp16")]; + tensor var_14027_begin_0 = const()[name = tensor("op_14027_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14027_end_0 = const()[name = tensor("op_14027_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14027_end_mask_0 = const()[name = tensor("op_14027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14027_cast_fp16 = slice_by_index(begin = var_14027_begin_0, end = var_14027_end_0, end_mask = var_14027_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14027_cast_fp16")]; + tensor var_14031_equation_0 = const()[name = tensor("op_14031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14031_cast_fp16 = einsum(equation = var_14031_equation_0, values = (var_13873_cast_fp16, var_13790_cast_fp16))[name = tensor("op_14031_cast_fp16")]; + tensor var_14032_to_fp16 = const()[name = tensor("op_14032_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1121_cast_fp16 = mul(x = var_14031_cast_fp16, y = var_14032_to_fp16)[name = tensor("aw_1121_cast_fp16")]; + tensor var_14035_equation_0 = const()[name = tensor("op_14035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14035_cast_fp16 = einsum(equation = var_14035_equation_0, values = (var_13877_cast_fp16, var_13794_cast_fp16))[name = tensor("op_14035_cast_fp16")]; + tensor var_14036_to_fp16 = const()[name = tensor("op_14036_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1123_cast_fp16 = mul(x = var_14035_cast_fp16, y = var_14036_to_fp16)[name = tensor("aw_1123_cast_fp16")]; + tensor var_14039_equation_0 = const()[name = tensor("op_14039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14039_cast_fp16 = einsum(equation = var_14039_equation_0, values = (var_13881_cast_fp16, var_13798_cast_fp16))[name = tensor("op_14039_cast_fp16")]; + tensor var_14040_to_fp16 = const()[name = tensor("op_14040_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1125_cast_fp16 = mul(x = var_14039_cast_fp16, y = var_14040_to_fp16)[name = tensor("aw_1125_cast_fp16")]; + tensor var_14043_equation_0 = const()[name = tensor("op_14043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14043_cast_fp16 = einsum(equation = var_14043_equation_0, values = (var_13885_cast_fp16, var_13802_cast_fp16))[name = tensor("op_14043_cast_fp16")]; + tensor var_14044_to_fp16 = const()[name = tensor("op_14044_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1127_cast_fp16 = mul(x = var_14043_cast_fp16, y = var_14044_to_fp16)[name = tensor("aw_1127_cast_fp16")]; + tensor var_14047_equation_0 = const()[name = tensor("op_14047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14047_cast_fp16 = einsum(equation = var_14047_equation_0, values = (var_13889_cast_fp16, var_13806_cast_fp16))[name = tensor("op_14047_cast_fp16")]; + tensor var_14048_to_fp16 = const()[name = tensor("op_14048_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1129_cast_fp16 = mul(x = var_14047_cast_fp16, y = var_14048_to_fp16)[name = tensor("aw_1129_cast_fp16")]; + tensor var_14051_equation_0 = const()[name = tensor("op_14051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14051_cast_fp16 = einsum(equation = var_14051_equation_0, values = (var_13893_cast_fp16, var_13810_cast_fp16))[name = tensor("op_14051_cast_fp16")]; + tensor var_14052_to_fp16 = const()[name = tensor("op_14052_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1131_cast_fp16 = mul(x = var_14051_cast_fp16, y = var_14052_to_fp16)[name = tensor("aw_1131_cast_fp16")]; + tensor var_14055_equation_0 = const()[name = tensor("op_14055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14055_cast_fp16 = einsum(equation = var_14055_equation_0, values = (var_13897_cast_fp16, var_13814_cast_fp16))[name = tensor("op_14055_cast_fp16")]; + tensor var_14056_to_fp16 = const()[name = tensor("op_14056_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1133_cast_fp16 = mul(x = var_14055_cast_fp16, y = var_14056_to_fp16)[name = tensor("aw_1133_cast_fp16")]; + tensor var_14059_equation_0 = const()[name = tensor("op_14059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14059_cast_fp16 = einsum(equation = var_14059_equation_0, values = (var_13901_cast_fp16, var_13818_cast_fp16))[name = tensor("op_14059_cast_fp16")]; + tensor var_14060_to_fp16 = const()[name = tensor("op_14060_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1135_cast_fp16 = mul(x = var_14059_cast_fp16, y = var_14060_to_fp16)[name = tensor("aw_1135_cast_fp16")]; + tensor var_14063_equation_0 = const()[name = tensor("op_14063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14063_cast_fp16 = einsum(equation = var_14063_equation_0, values = (var_13905_cast_fp16, var_13822_cast_fp16))[name = tensor("op_14063_cast_fp16")]; + tensor var_14064_to_fp16 = const()[name = tensor("op_14064_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1137_cast_fp16 = mul(x = var_14063_cast_fp16, y = var_14064_to_fp16)[name = tensor("aw_1137_cast_fp16")]; + tensor var_14067_equation_0 = const()[name = tensor("op_14067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14067_cast_fp16 = einsum(equation = var_14067_equation_0, values = (var_13909_cast_fp16, var_13826_cast_fp16))[name = tensor("op_14067_cast_fp16")]; + tensor var_14068_to_fp16 = const()[name = tensor("op_14068_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1139_cast_fp16 = mul(x = var_14067_cast_fp16, y = var_14068_to_fp16)[name = tensor("aw_1139_cast_fp16")]; + tensor var_14071_equation_0 = const()[name = tensor("op_14071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14071_cast_fp16 = einsum(equation = var_14071_equation_0, values = (var_13913_cast_fp16, var_13830_cast_fp16))[name = tensor("op_14071_cast_fp16")]; + tensor var_14072_to_fp16 = const()[name = tensor("op_14072_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1141_cast_fp16 = mul(x = var_14071_cast_fp16, y = var_14072_to_fp16)[name = tensor("aw_1141_cast_fp16")]; + tensor var_14075_equation_0 = const()[name = tensor("op_14075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14075_cast_fp16 = einsum(equation = var_14075_equation_0, values = (var_13917_cast_fp16, var_13834_cast_fp16))[name = tensor("op_14075_cast_fp16")]; + tensor var_14076_to_fp16 = const()[name = tensor("op_14076_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1143_cast_fp16 = mul(x = var_14075_cast_fp16, y = var_14076_to_fp16)[name = tensor("aw_1143_cast_fp16")]; + tensor var_14079_equation_0 = const()[name = tensor("op_14079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14079_cast_fp16 = einsum(equation = var_14079_equation_0, values = (var_13921_cast_fp16, var_13838_cast_fp16))[name = tensor("op_14079_cast_fp16")]; + tensor var_14080_to_fp16 = const()[name = tensor("op_14080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1145_cast_fp16 = mul(x = var_14079_cast_fp16, y = var_14080_to_fp16)[name = tensor("aw_1145_cast_fp16")]; + tensor var_14083_equation_0 = const()[name = tensor("op_14083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14083_cast_fp16 = einsum(equation = var_14083_equation_0, values = (var_13925_cast_fp16, var_13842_cast_fp16))[name = tensor("op_14083_cast_fp16")]; + tensor var_14084_to_fp16 = const()[name = tensor("op_14084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1147_cast_fp16 = mul(x = var_14083_cast_fp16, y = var_14084_to_fp16)[name = tensor("aw_1147_cast_fp16")]; + tensor var_14087_equation_0 = const()[name = tensor("op_14087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14087_cast_fp16 = einsum(equation = var_14087_equation_0, values = (var_13929_cast_fp16, var_13846_cast_fp16))[name = tensor("op_14087_cast_fp16")]; + tensor var_14088_to_fp16 = const()[name = tensor("op_14088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1149_cast_fp16 = mul(x = var_14087_cast_fp16, y = var_14088_to_fp16)[name = tensor("aw_1149_cast_fp16")]; + tensor var_14091_equation_0 = const()[name = tensor("op_14091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14091_cast_fp16 = einsum(equation = var_14091_equation_0, values = (var_13933_cast_fp16, var_13850_cast_fp16))[name = tensor("op_14091_cast_fp16")]; + tensor var_14092_to_fp16 = const()[name = tensor("op_14092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1151_cast_fp16 = mul(x = var_14091_cast_fp16, y = var_14092_to_fp16)[name = tensor("aw_1151_cast_fp16")]; + tensor var_14095_equation_0 = const()[name = tensor("op_14095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14095_cast_fp16 = einsum(equation = var_14095_equation_0, values = (var_13937_cast_fp16, var_13854_cast_fp16))[name = tensor("op_14095_cast_fp16")]; + tensor var_14096_to_fp16 = const()[name = tensor("op_14096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1153_cast_fp16 = mul(x = var_14095_cast_fp16, y = var_14096_to_fp16)[name = tensor("aw_1153_cast_fp16")]; + tensor var_14099_equation_0 = const()[name = tensor("op_14099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14099_cast_fp16 = einsum(equation = var_14099_equation_0, values = (var_13941_cast_fp16, var_13858_cast_fp16))[name = tensor("op_14099_cast_fp16")]; + tensor var_14100_to_fp16 = const()[name = tensor("op_14100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1155_cast_fp16 = mul(x = var_14099_cast_fp16, y = var_14100_to_fp16)[name = tensor("aw_1155_cast_fp16")]; + tensor var_14103_equation_0 = const()[name = tensor("op_14103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14103_cast_fp16 = einsum(equation = var_14103_equation_0, values = (var_13945_cast_fp16, var_13862_cast_fp16))[name = tensor("op_14103_cast_fp16")]; + tensor var_14104_to_fp16 = const()[name = tensor("op_14104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1157_cast_fp16 = mul(x = var_14103_cast_fp16, y = var_14104_to_fp16)[name = tensor("aw_1157_cast_fp16")]; + tensor var_14107_equation_0 = const()[name = tensor("op_14107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14107_cast_fp16 = einsum(equation = var_14107_equation_0, values = (var_13949_cast_fp16, var_13866_cast_fp16))[name = tensor("op_14107_cast_fp16")]; + tensor var_14108_to_fp16 = const()[name = tensor("op_14108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1159_cast_fp16 = mul(x = var_14107_cast_fp16, y = var_14108_to_fp16)[name = tensor("aw_1159_cast_fp16")]; + tensor var_14110_cast_fp16 = softmax(axis = var_2624, x = aw_1121_cast_fp16)[name = tensor("op_14110_cast_fp16")]; + tensor var_14111_cast_fp16 = softmax(axis = var_2624, x = aw_1123_cast_fp16)[name = tensor("op_14111_cast_fp16")]; + tensor var_14112_cast_fp16 = softmax(axis = var_2624, x = aw_1125_cast_fp16)[name = tensor("op_14112_cast_fp16")]; + tensor var_14113_cast_fp16 = softmax(axis = var_2624, x = aw_1127_cast_fp16)[name = tensor("op_14113_cast_fp16")]; + tensor var_14114_cast_fp16 = softmax(axis = var_2624, x = aw_1129_cast_fp16)[name = tensor("op_14114_cast_fp16")]; + tensor var_14115_cast_fp16 = softmax(axis = var_2624, x = aw_1131_cast_fp16)[name = tensor("op_14115_cast_fp16")]; + tensor var_14116_cast_fp16 = softmax(axis = var_2624, x = aw_1133_cast_fp16)[name = tensor("op_14116_cast_fp16")]; + tensor var_14117_cast_fp16 = softmax(axis = var_2624, x = aw_1135_cast_fp16)[name = tensor("op_14117_cast_fp16")]; + tensor var_14118_cast_fp16 = softmax(axis = var_2624, x = aw_1137_cast_fp16)[name = tensor("op_14118_cast_fp16")]; + tensor var_14119_cast_fp16 = softmax(axis = var_2624, x = aw_1139_cast_fp16)[name = tensor("op_14119_cast_fp16")]; + tensor var_14120_cast_fp16 = softmax(axis = var_2624, x = aw_1141_cast_fp16)[name = tensor("op_14120_cast_fp16")]; + tensor var_14121_cast_fp16 = softmax(axis = var_2624, x = aw_1143_cast_fp16)[name = tensor("op_14121_cast_fp16")]; + tensor var_14122_cast_fp16 = softmax(axis = var_2624, x = aw_1145_cast_fp16)[name = tensor("op_14122_cast_fp16")]; + tensor var_14123_cast_fp16 = softmax(axis = var_2624, x = aw_1147_cast_fp16)[name = tensor("op_14123_cast_fp16")]; + tensor var_14124_cast_fp16 = softmax(axis = var_2624, x = aw_1149_cast_fp16)[name = tensor("op_14124_cast_fp16")]; + tensor var_14125_cast_fp16 = softmax(axis = var_2624, x = aw_1151_cast_fp16)[name = tensor("op_14125_cast_fp16")]; + tensor var_14126_cast_fp16 = softmax(axis = var_2624, x = aw_1153_cast_fp16)[name = tensor("op_14126_cast_fp16")]; + tensor var_14127_cast_fp16 = softmax(axis = var_2624, x = aw_1155_cast_fp16)[name = tensor("op_14127_cast_fp16")]; + tensor var_14128_cast_fp16 = softmax(axis = var_2624, x = aw_1157_cast_fp16)[name = tensor("op_14128_cast_fp16")]; + tensor var_14129_cast_fp16 = softmax(axis = var_2624, x = aw_1159_cast_fp16)[name = tensor("op_14129_cast_fp16")]; + tensor var_14131_equation_0 = const()[name = tensor("op_14131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14131_cast_fp16 = einsum(equation = var_14131_equation_0, values = (var_13951_cast_fp16, var_14110_cast_fp16))[name = tensor("op_14131_cast_fp16")]; + tensor var_14133_equation_0 = const()[name = tensor("op_14133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14133_cast_fp16 = einsum(equation = var_14133_equation_0, values = (var_13955_cast_fp16, var_14111_cast_fp16))[name = tensor("op_14133_cast_fp16")]; + tensor var_14135_equation_0 = const()[name = tensor("op_14135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14135_cast_fp16 = einsum(equation = var_14135_equation_0, values = (var_13959_cast_fp16, var_14112_cast_fp16))[name = tensor("op_14135_cast_fp16")]; + tensor var_14137_equation_0 = const()[name = tensor("op_14137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14137_cast_fp16 = einsum(equation = var_14137_equation_0, values = (var_13963_cast_fp16, var_14113_cast_fp16))[name = tensor("op_14137_cast_fp16")]; + tensor var_14139_equation_0 = const()[name = tensor("op_14139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14139_cast_fp16 = einsum(equation = var_14139_equation_0, values = (var_13967_cast_fp16, var_14114_cast_fp16))[name = tensor("op_14139_cast_fp16")]; + tensor var_14141_equation_0 = const()[name = tensor("op_14141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14141_cast_fp16 = einsum(equation = var_14141_equation_0, values = (var_13971_cast_fp16, var_14115_cast_fp16))[name = tensor("op_14141_cast_fp16")]; + tensor var_14143_equation_0 = const()[name = tensor("op_14143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14143_cast_fp16 = einsum(equation = var_14143_equation_0, values = (var_13975_cast_fp16, var_14116_cast_fp16))[name = tensor("op_14143_cast_fp16")]; + tensor var_14145_equation_0 = const()[name = tensor("op_14145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14145_cast_fp16 = einsum(equation = var_14145_equation_0, values = (var_13979_cast_fp16, var_14117_cast_fp16))[name = tensor("op_14145_cast_fp16")]; + tensor var_14147_equation_0 = const()[name = tensor("op_14147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14147_cast_fp16 = einsum(equation = var_14147_equation_0, values = (var_13983_cast_fp16, var_14118_cast_fp16))[name = tensor("op_14147_cast_fp16")]; + tensor var_14149_equation_0 = const()[name = tensor("op_14149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14149_cast_fp16 = einsum(equation = var_14149_equation_0, values = (var_13987_cast_fp16, var_14119_cast_fp16))[name = tensor("op_14149_cast_fp16")]; + tensor var_14151_equation_0 = const()[name = tensor("op_14151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14151_cast_fp16 = einsum(equation = var_14151_equation_0, values = (var_13991_cast_fp16, var_14120_cast_fp16))[name = tensor("op_14151_cast_fp16")]; + tensor var_14153_equation_0 = const()[name = tensor("op_14153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14153_cast_fp16 = einsum(equation = var_14153_equation_0, values = (var_13995_cast_fp16, var_14121_cast_fp16))[name = tensor("op_14153_cast_fp16")]; + tensor var_14155_equation_0 = const()[name = tensor("op_14155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14155_cast_fp16 = einsum(equation = var_14155_equation_0, values = (var_13999_cast_fp16, var_14122_cast_fp16))[name = tensor("op_14155_cast_fp16")]; + tensor var_14157_equation_0 = const()[name = tensor("op_14157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14157_cast_fp16 = einsum(equation = var_14157_equation_0, values = (var_14003_cast_fp16, var_14123_cast_fp16))[name = tensor("op_14157_cast_fp16")]; + tensor var_14159_equation_0 = const()[name = tensor("op_14159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14159_cast_fp16 = einsum(equation = var_14159_equation_0, values = (var_14007_cast_fp16, var_14124_cast_fp16))[name = tensor("op_14159_cast_fp16")]; + tensor var_14161_equation_0 = const()[name = tensor("op_14161_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14161_cast_fp16 = einsum(equation = var_14161_equation_0, values = (var_14011_cast_fp16, var_14125_cast_fp16))[name = tensor("op_14161_cast_fp16")]; + tensor var_14163_equation_0 = const()[name = tensor("op_14163_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14163_cast_fp16 = einsum(equation = var_14163_equation_0, values = (var_14015_cast_fp16, var_14126_cast_fp16))[name = tensor("op_14163_cast_fp16")]; + tensor var_14165_equation_0 = const()[name = tensor("op_14165_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14165_cast_fp16 = einsum(equation = var_14165_equation_0, values = (var_14019_cast_fp16, var_14127_cast_fp16))[name = tensor("op_14165_cast_fp16")]; + tensor var_14167_equation_0 = const()[name = tensor("op_14167_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14167_cast_fp16 = einsum(equation = var_14167_equation_0, values = (var_14023_cast_fp16, var_14128_cast_fp16))[name = tensor("op_14167_cast_fp16")]; + tensor var_14169_equation_0 = const()[name = tensor("op_14169_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14169_cast_fp16 = einsum(equation = var_14169_equation_0, values = (var_14027_cast_fp16, var_14129_cast_fp16))[name = tensor("op_14169_cast_fp16")]; + tensor input_245_interleave_0 = const()[name = tensor("input_245_interleave_0"), val = tensor(false)]; + tensor input_245_cast_fp16 = concat(axis = var_2624, interleave = input_245_interleave_0, values = (var_14131_cast_fp16, var_14133_cast_fp16, var_14135_cast_fp16, var_14137_cast_fp16, var_14139_cast_fp16, var_14141_cast_fp16, var_14143_cast_fp16, var_14145_cast_fp16, var_14147_cast_fp16, var_14149_cast_fp16, var_14151_cast_fp16, var_14153_cast_fp16, var_14155_cast_fp16, var_14157_cast_fp16, var_14159_cast_fp16, var_14161_cast_fp16, var_14163_cast_fp16, var_14165_cast_fp16, var_14167_cast_fp16, var_14169_cast_fp16))[name = tensor("input_245_cast_fp16")]; + tensor var_14179_pad_type_0 = const()[name = tensor("op_14179_pad_type_0"), val = tensor("valid")]; + tensor var_14179_strides_0 = const()[name = tensor("op_14179_strides_0"), val = tensor([1, 1])]; + tensor var_14179_pad_0 = const()[name = tensor("op_14179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14179_dilations_0 = const()[name = tensor("op_14179_dilations_0"), val = tensor([1, 1])]; + tensor var_14179_groups_0 = const()[name = tensor("op_14179_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417134656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418363520))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418363712)))]; + tensor var_14179_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_14179_dilations_0, groups = var_14179_groups_0, pad = var_14179_pad_0, pad_type = var_14179_pad_type_0, strides = var_14179_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized, x = input_245_cast_fp16)[name = tensor("op_14179_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = var_14179_cast_fp16, y = inputs_97_cast_fp16)[name = tensor("inputs_99_cast_fp16")]; + tensor hidden_states_151_axes_0 = const()[name = tensor("hidden_states_151_axes_0"), val = tensor([1])]; + tensor hidden_states_151_gamma_0_to_fp16 = const()[name = tensor("hidden_states_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418366336)))]; + tensor hidden_states_151_beta_0_to_fp16 = const()[name = tensor("hidden_states_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418368960)))]; + tensor var_14189_to_fp16 = const()[name = tensor("op_14189_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_151_cast_fp16 = layer_norm(axes = hidden_states_151_axes_0, beta = hidden_states_151_beta_0_to_fp16, epsilon = var_14189_to_fp16, gamma = hidden_states_151_gamma_0_to_fp16, x = inputs_99_cast_fp16)[name = tensor("hidden_states_151_cast_fp16")]; + tensor q_67_pad_type_0 = const()[name = tensor("q_67_pad_type_0"), val = tensor("valid")]; + tensor q_67_strides_0 = const()[name = tensor("q_67_strides_0"), val = tensor([1, 1])]; + tensor q_67_pad_0 = const()[name = tensor("q_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_67_dilations_0 = const()[name = tensor("q_67_dilations_0"), val = tensor([1, 1])]; + tensor q_67_groups_0 = const()[name = tensor("q_67_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418371584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419600448))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_67_cast_fp16 = conv(dilations = q_67_dilations_0, groups = q_67_groups_0, pad = q_67_pad_0, pad_type = q_67_pad_type_0, strides = q_67_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_151_cast_fp16)[name = tensor("q_67_cast_fp16")]; + tensor k_133_pad_type_0 = const()[name = tensor("k_133_pad_type_0"), val = tensor("valid")]; + tensor k_133_strides_0 = const()[name = tensor("k_133_strides_0"), val = tensor([1, 1])]; + tensor k_133_pad_0 = const()[name = tensor("k_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_133_dilations_0 = const()[name = tensor("k_133_dilations_0"), val = tensor([1, 1])]; + tensor k_133_groups_0 = const()[name = tensor("k_133_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419600640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421566784))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_133_cast_fp16 = conv(dilations = k_133_dilations_0, groups = k_133_groups_0, pad = k_133_pad_0, pad_type = k_133_pad_type_0, strides = k_133_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_133_cast_fp16")]; + tensor v_67_pad_type_0 = const()[name = tensor("v_67_pad_type_0"), val = tensor("valid")]; + tensor v_67_strides_0 = const()[name = tensor("v_67_strides_0"), val = tensor([1, 1])]; + tensor v_67_pad_0 = const()[name = tensor("v_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_67_dilations_0 = const()[name = tensor("v_67_dilations_0"), val = tensor([1, 1])]; + tensor v_67_groups_0 = const()[name = tensor("v_67_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421566976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423533120))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_67_cast_fp16 = conv(dilations = v_67_dilations_0, groups = v_67_groups_0, pad = v_67_pad_0, pad_type = v_67_pad_type_0, strides = v_67_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_67_cast_fp16")]; + tensor var_14222_begin_0 = const()[name = tensor("op_14222_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14222_end_0 = const()[name = tensor("op_14222_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14222_end_mask_0 = const()[name = tensor("op_14222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14222_cast_fp16 = slice_by_index(begin = var_14222_begin_0, end = var_14222_end_0, end_mask = var_14222_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14222_cast_fp16")]; + tensor var_14226_begin_0 = const()[name = tensor("op_14226_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14226_end_0 = const()[name = tensor("op_14226_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_14226_end_mask_0 = const()[name = tensor("op_14226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14226_cast_fp16 = slice_by_index(begin = var_14226_begin_0, end = var_14226_end_0, end_mask = var_14226_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14226_cast_fp16")]; + tensor var_14230_begin_0 = const()[name = tensor("op_14230_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14230_end_0 = const()[name = tensor("op_14230_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_14230_end_mask_0 = const()[name = tensor("op_14230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14230_cast_fp16 = slice_by_index(begin = var_14230_begin_0, end = var_14230_end_0, end_mask = var_14230_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14230_cast_fp16")]; + tensor var_14234_begin_0 = const()[name = tensor("op_14234_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14234_end_0 = const()[name = tensor("op_14234_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_14234_end_mask_0 = const()[name = tensor("op_14234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14234_cast_fp16 = slice_by_index(begin = var_14234_begin_0, end = var_14234_end_0, end_mask = var_14234_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14234_cast_fp16")]; + tensor var_14238_begin_0 = const()[name = tensor("op_14238_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14238_end_0 = const()[name = tensor("op_14238_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_14238_end_mask_0 = const()[name = tensor("op_14238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14238_cast_fp16 = slice_by_index(begin = var_14238_begin_0, end = var_14238_end_0, end_mask = var_14238_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14238_cast_fp16")]; + tensor var_14242_begin_0 = const()[name = tensor("op_14242_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14242_end_0 = const()[name = tensor("op_14242_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_14242_end_mask_0 = const()[name = tensor("op_14242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14242_cast_fp16 = slice_by_index(begin = var_14242_begin_0, end = var_14242_end_0, end_mask = var_14242_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14242_cast_fp16")]; + tensor var_14246_begin_0 = const()[name = tensor("op_14246_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14246_end_0 = const()[name = tensor("op_14246_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_14246_end_mask_0 = const()[name = tensor("op_14246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14246_cast_fp16 = slice_by_index(begin = var_14246_begin_0, end = var_14246_end_0, end_mask = var_14246_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14246_cast_fp16")]; + tensor var_14250_begin_0 = const()[name = tensor("op_14250_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14250_end_0 = const()[name = tensor("op_14250_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_14250_end_mask_0 = const()[name = tensor("op_14250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14250_cast_fp16 = slice_by_index(begin = var_14250_begin_0, end = var_14250_end_0, end_mask = var_14250_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14250_cast_fp16")]; + tensor var_14254_begin_0 = const()[name = tensor("op_14254_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14254_end_0 = const()[name = tensor("op_14254_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_14254_end_mask_0 = const()[name = tensor("op_14254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14254_cast_fp16 = slice_by_index(begin = var_14254_begin_0, end = var_14254_end_0, end_mask = var_14254_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14254_cast_fp16")]; + tensor var_14258_begin_0 = const()[name = tensor("op_14258_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14258_end_0 = const()[name = tensor("op_14258_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_14258_end_mask_0 = const()[name = tensor("op_14258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14258_cast_fp16 = slice_by_index(begin = var_14258_begin_0, end = var_14258_end_0, end_mask = var_14258_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14258_cast_fp16")]; + tensor var_14262_begin_0 = const()[name = tensor("op_14262_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14262_end_0 = const()[name = tensor("op_14262_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_14262_end_mask_0 = const()[name = tensor("op_14262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14262_cast_fp16 = slice_by_index(begin = var_14262_begin_0, end = var_14262_end_0, end_mask = var_14262_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14262_cast_fp16")]; + tensor var_14266_begin_0 = const()[name = tensor("op_14266_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14266_end_0 = const()[name = tensor("op_14266_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_14266_end_mask_0 = const()[name = tensor("op_14266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14266_cast_fp16 = slice_by_index(begin = var_14266_begin_0, end = var_14266_end_0, end_mask = var_14266_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14266_cast_fp16")]; + tensor var_14270_begin_0 = const()[name = tensor("op_14270_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14270_end_0 = const()[name = tensor("op_14270_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_14270_end_mask_0 = const()[name = tensor("op_14270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14270_cast_fp16 = slice_by_index(begin = var_14270_begin_0, end = var_14270_end_0, end_mask = var_14270_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14270_cast_fp16")]; + tensor var_14274_begin_0 = const()[name = tensor("op_14274_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14274_end_0 = const()[name = tensor("op_14274_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14274_end_mask_0 = const()[name = tensor("op_14274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14274_cast_fp16 = slice_by_index(begin = var_14274_begin_0, end = var_14274_end_0, end_mask = var_14274_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14274_cast_fp16")]; + tensor var_14278_begin_0 = const()[name = tensor("op_14278_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14278_end_0 = const()[name = tensor("op_14278_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14278_end_mask_0 = const()[name = tensor("op_14278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14278_cast_fp16 = slice_by_index(begin = var_14278_begin_0, end = var_14278_end_0, end_mask = var_14278_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14278_cast_fp16")]; + tensor var_14282_begin_0 = const()[name = tensor("op_14282_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14282_end_0 = const()[name = tensor("op_14282_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14282_end_mask_0 = const()[name = tensor("op_14282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14282_cast_fp16 = slice_by_index(begin = var_14282_begin_0, end = var_14282_end_0, end_mask = var_14282_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14282_cast_fp16")]; + tensor var_14286_begin_0 = const()[name = tensor("op_14286_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14286_end_0 = const()[name = tensor("op_14286_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14286_end_mask_0 = const()[name = tensor("op_14286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14286_cast_fp16 = slice_by_index(begin = var_14286_begin_0, end = var_14286_end_0, end_mask = var_14286_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14286_cast_fp16")]; + tensor var_14290_begin_0 = const()[name = tensor("op_14290_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14290_end_0 = const()[name = tensor("op_14290_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14290_end_mask_0 = const()[name = tensor("op_14290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14290_cast_fp16 = slice_by_index(begin = var_14290_begin_0, end = var_14290_end_0, end_mask = var_14290_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14290_cast_fp16")]; + tensor var_14294_begin_0 = const()[name = tensor("op_14294_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14294_end_0 = const()[name = tensor("op_14294_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14294_end_mask_0 = const()[name = tensor("op_14294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14294_cast_fp16 = slice_by_index(begin = var_14294_begin_0, end = var_14294_end_0, end_mask = var_14294_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14294_cast_fp16")]; + tensor var_14298_begin_0 = const()[name = tensor("op_14298_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14298_end_0 = const()[name = tensor("op_14298_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14298_end_mask_0 = const()[name = tensor("op_14298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14298_cast_fp16 = slice_by_index(begin = var_14298_begin_0, end = var_14298_end_0, end_mask = var_14298_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14298_cast_fp16")]; + tensor k_135_perm_0 = const()[name = tensor("k_135_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14305_begin_0 = const()[name = tensor("op_14305_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14305_end_0 = const()[name = tensor("op_14305_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_14305_end_mask_0 = const()[name = tensor("op_14305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_135_cast_fp16 = transpose(perm = k_135_perm_0, x = k_133_cast_fp16)[name = tensor("transpose_34")]; + tensor var_14305_cast_fp16 = slice_by_index(begin = var_14305_begin_0, end = var_14305_end_0, end_mask = var_14305_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14305_cast_fp16")]; + tensor var_14309_begin_0 = const()[name = tensor("op_14309_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14309_end_0 = const()[name = tensor("op_14309_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_14309_end_mask_0 = const()[name = tensor("op_14309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14309_cast_fp16 = slice_by_index(begin = var_14309_begin_0, end = var_14309_end_0, end_mask = var_14309_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14309_cast_fp16")]; + tensor var_14313_begin_0 = const()[name = tensor("op_14313_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14313_end_0 = const()[name = tensor("op_14313_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_14313_end_mask_0 = const()[name = tensor("op_14313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14313_cast_fp16 = slice_by_index(begin = var_14313_begin_0, end = var_14313_end_0, end_mask = var_14313_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14313_cast_fp16")]; + tensor var_14317_begin_0 = const()[name = tensor("op_14317_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14317_end_0 = const()[name = tensor("op_14317_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_14317_end_mask_0 = const()[name = tensor("op_14317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14317_cast_fp16 = slice_by_index(begin = var_14317_begin_0, end = var_14317_end_0, end_mask = var_14317_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14317_cast_fp16")]; + tensor var_14321_begin_0 = const()[name = tensor("op_14321_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14321_end_0 = const()[name = tensor("op_14321_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_14321_end_mask_0 = const()[name = tensor("op_14321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14321_cast_fp16 = slice_by_index(begin = var_14321_begin_0, end = var_14321_end_0, end_mask = var_14321_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14321_cast_fp16")]; + tensor var_14325_begin_0 = const()[name = tensor("op_14325_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_14325_end_0 = const()[name = tensor("op_14325_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_14325_end_mask_0 = const()[name = tensor("op_14325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14325_cast_fp16 = slice_by_index(begin = var_14325_begin_0, end = var_14325_end_0, end_mask = var_14325_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14325_cast_fp16")]; + tensor var_14329_begin_0 = const()[name = tensor("op_14329_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_14329_end_0 = const()[name = tensor("op_14329_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_14329_end_mask_0 = const()[name = tensor("op_14329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14329_cast_fp16 = slice_by_index(begin = var_14329_begin_0, end = var_14329_end_0, end_mask = var_14329_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14329_cast_fp16")]; + tensor var_14333_begin_0 = const()[name = tensor("op_14333_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_14333_end_0 = const()[name = tensor("op_14333_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_14333_end_mask_0 = const()[name = tensor("op_14333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14333_cast_fp16 = slice_by_index(begin = var_14333_begin_0, end = var_14333_end_0, end_mask = var_14333_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14333_cast_fp16")]; + tensor var_14337_begin_0 = const()[name = tensor("op_14337_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14337_end_0 = const()[name = tensor("op_14337_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_14337_end_mask_0 = const()[name = tensor("op_14337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14337_cast_fp16 = slice_by_index(begin = var_14337_begin_0, end = var_14337_end_0, end_mask = var_14337_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14337_cast_fp16")]; + tensor var_14341_begin_0 = const()[name = tensor("op_14341_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_14341_end_0 = const()[name = tensor("op_14341_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_14341_end_mask_0 = const()[name = tensor("op_14341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14341_cast_fp16 = slice_by_index(begin = var_14341_begin_0, end = var_14341_end_0, end_mask = var_14341_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14341_cast_fp16")]; + tensor var_14345_begin_0 = const()[name = tensor("op_14345_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_14345_end_0 = const()[name = tensor("op_14345_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_14345_end_mask_0 = const()[name = tensor("op_14345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14345_cast_fp16 = slice_by_index(begin = var_14345_begin_0, end = var_14345_end_0, end_mask = var_14345_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14345_cast_fp16")]; + tensor var_14349_begin_0 = const()[name = tensor("op_14349_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_14349_end_0 = const()[name = tensor("op_14349_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_14349_end_mask_0 = const()[name = tensor("op_14349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14349_cast_fp16 = slice_by_index(begin = var_14349_begin_0, end = var_14349_end_0, end_mask = var_14349_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14349_cast_fp16")]; + tensor var_14353_begin_0 = const()[name = tensor("op_14353_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_14353_end_0 = const()[name = tensor("op_14353_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_14353_end_mask_0 = const()[name = tensor("op_14353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14353_cast_fp16 = slice_by_index(begin = var_14353_begin_0, end = var_14353_end_0, end_mask = var_14353_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14353_cast_fp16")]; + tensor var_14357_begin_0 = const()[name = tensor("op_14357_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_14357_end_0 = const()[name = tensor("op_14357_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_14357_end_mask_0 = const()[name = tensor("op_14357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14357_cast_fp16 = slice_by_index(begin = var_14357_begin_0, end = var_14357_end_0, end_mask = var_14357_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14357_cast_fp16")]; + tensor var_14361_begin_0 = const()[name = tensor("op_14361_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_14361_end_0 = const()[name = tensor("op_14361_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_14361_end_mask_0 = const()[name = tensor("op_14361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14361_cast_fp16 = slice_by_index(begin = var_14361_begin_0, end = var_14361_end_0, end_mask = var_14361_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14361_cast_fp16")]; + tensor var_14365_begin_0 = const()[name = tensor("op_14365_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_14365_end_0 = const()[name = tensor("op_14365_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_14365_end_mask_0 = const()[name = tensor("op_14365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14365_cast_fp16 = slice_by_index(begin = var_14365_begin_0, end = var_14365_end_0, end_mask = var_14365_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14365_cast_fp16")]; + tensor var_14369_begin_0 = const()[name = tensor("op_14369_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14369_end_0 = const()[name = tensor("op_14369_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_14369_end_mask_0 = const()[name = tensor("op_14369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14369_cast_fp16 = slice_by_index(begin = var_14369_begin_0, end = var_14369_end_0, end_mask = var_14369_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14369_cast_fp16")]; + tensor var_14373_begin_0 = const()[name = tensor("op_14373_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_14373_end_0 = const()[name = tensor("op_14373_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_14373_end_mask_0 = const()[name = tensor("op_14373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14373_cast_fp16 = slice_by_index(begin = var_14373_begin_0, end = var_14373_end_0, end_mask = var_14373_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14373_cast_fp16")]; + tensor var_14377_begin_0 = const()[name = tensor("op_14377_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_14377_end_0 = const()[name = tensor("op_14377_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_14377_end_mask_0 = const()[name = tensor("op_14377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14377_cast_fp16 = slice_by_index(begin = var_14377_begin_0, end = var_14377_end_0, end_mask = var_14377_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14377_cast_fp16")]; + tensor var_14381_begin_0 = const()[name = tensor("op_14381_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_14381_end_0 = const()[name = tensor("op_14381_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_14381_end_mask_0 = const()[name = tensor("op_14381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14381_cast_fp16 = slice_by_index(begin = var_14381_begin_0, end = var_14381_end_0, end_mask = var_14381_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14381_cast_fp16")]; + tensor var_14383_begin_0 = const()[name = tensor("op_14383_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14383_end_0 = const()[name = tensor("op_14383_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_14383_end_mask_0 = const()[name = tensor("op_14383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14383_cast_fp16 = slice_by_index(begin = var_14383_begin_0, end = var_14383_end_0, end_mask = var_14383_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14383_cast_fp16")]; + tensor var_14387_begin_0 = const()[name = tensor("op_14387_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14387_end_0 = const()[name = tensor("op_14387_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_14387_end_mask_0 = const()[name = tensor("op_14387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14387_cast_fp16 = slice_by_index(begin = var_14387_begin_0, end = var_14387_end_0, end_mask = var_14387_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14387_cast_fp16")]; + tensor var_14391_begin_0 = const()[name = tensor("op_14391_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14391_end_0 = const()[name = tensor("op_14391_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_14391_end_mask_0 = const()[name = tensor("op_14391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14391_cast_fp16 = slice_by_index(begin = var_14391_begin_0, end = var_14391_end_0, end_mask = var_14391_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14391_cast_fp16")]; + tensor var_14395_begin_0 = const()[name = tensor("op_14395_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14395_end_0 = const()[name = tensor("op_14395_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_14395_end_mask_0 = const()[name = tensor("op_14395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14395_cast_fp16 = slice_by_index(begin = var_14395_begin_0, end = var_14395_end_0, end_mask = var_14395_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14395_cast_fp16")]; + tensor var_14399_begin_0 = const()[name = tensor("op_14399_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14399_end_0 = const()[name = tensor("op_14399_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_14399_end_mask_0 = const()[name = tensor("op_14399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14399_cast_fp16 = slice_by_index(begin = var_14399_begin_0, end = var_14399_end_0, end_mask = var_14399_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14399_cast_fp16")]; + tensor var_14403_begin_0 = const()[name = tensor("op_14403_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14403_end_0 = const()[name = tensor("op_14403_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_14403_end_mask_0 = const()[name = tensor("op_14403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14403_cast_fp16 = slice_by_index(begin = var_14403_begin_0, end = var_14403_end_0, end_mask = var_14403_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14403_cast_fp16")]; + tensor var_14407_begin_0 = const()[name = tensor("op_14407_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14407_end_0 = const()[name = tensor("op_14407_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_14407_end_mask_0 = const()[name = tensor("op_14407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14407_cast_fp16 = slice_by_index(begin = var_14407_begin_0, end = var_14407_end_0, end_mask = var_14407_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14407_cast_fp16")]; + tensor var_14411_begin_0 = const()[name = tensor("op_14411_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14411_end_0 = const()[name = tensor("op_14411_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_14411_end_mask_0 = const()[name = tensor("op_14411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14411_cast_fp16 = slice_by_index(begin = var_14411_begin_0, end = var_14411_end_0, end_mask = var_14411_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14411_cast_fp16")]; + tensor var_14415_begin_0 = const()[name = tensor("op_14415_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14415_end_0 = const()[name = tensor("op_14415_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_14415_end_mask_0 = const()[name = tensor("op_14415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14415_cast_fp16 = slice_by_index(begin = var_14415_begin_0, end = var_14415_end_0, end_mask = var_14415_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14415_cast_fp16")]; + tensor var_14419_begin_0 = const()[name = tensor("op_14419_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14419_end_0 = const()[name = tensor("op_14419_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_14419_end_mask_0 = const()[name = tensor("op_14419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14419_cast_fp16 = slice_by_index(begin = var_14419_begin_0, end = var_14419_end_0, end_mask = var_14419_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14419_cast_fp16")]; + tensor var_14423_begin_0 = const()[name = tensor("op_14423_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14423_end_0 = const()[name = tensor("op_14423_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_14423_end_mask_0 = const()[name = tensor("op_14423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14423_cast_fp16 = slice_by_index(begin = var_14423_begin_0, end = var_14423_end_0, end_mask = var_14423_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14423_cast_fp16")]; + tensor var_14427_begin_0 = const()[name = tensor("op_14427_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14427_end_0 = const()[name = tensor("op_14427_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_14427_end_mask_0 = const()[name = tensor("op_14427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14427_cast_fp16 = slice_by_index(begin = var_14427_begin_0, end = var_14427_end_0, end_mask = var_14427_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14427_cast_fp16")]; + tensor var_14431_begin_0 = const()[name = tensor("op_14431_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14431_end_0 = const()[name = tensor("op_14431_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_14431_end_mask_0 = const()[name = tensor("op_14431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14431_cast_fp16 = slice_by_index(begin = var_14431_begin_0, end = var_14431_end_0, end_mask = var_14431_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14431_cast_fp16")]; + tensor var_14435_begin_0 = const()[name = tensor("op_14435_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14435_end_0 = const()[name = tensor("op_14435_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_14435_end_mask_0 = const()[name = tensor("op_14435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14435_cast_fp16 = slice_by_index(begin = var_14435_begin_0, end = var_14435_end_0, end_mask = var_14435_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14435_cast_fp16")]; + tensor var_14439_begin_0 = const()[name = tensor("op_14439_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14439_end_0 = const()[name = tensor("op_14439_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_14439_end_mask_0 = const()[name = tensor("op_14439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14439_cast_fp16 = slice_by_index(begin = var_14439_begin_0, end = var_14439_end_0, end_mask = var_14439_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14439_cast_fp16")]; + tensor var_14443_begin_0 = const()[name = tensor("op_14443_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14443_end_0 = const()[name = tensor("op_14443_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_14443_end_mask_0 = const()[name = tensor("op_14443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14443_cast_fp16 = slice_by_index(begin = var_14443_begin_0, end = var_14443_end_0, end_mask = var_14443_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14443_cast_fp16")]; + tensor var_14447_begin_0 = const()[name = tensor("op_14447_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14447_end_0 = const()[name = tensor("op_14447_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_14447_end_mask_0 = const()[name = tensor("op_14447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14447_cast_fp16 = slice_by_index(begin = var_14447_begin_0, end = var_14447_end_0, end_mask = var_14447_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14447_cast_fp16")]; + tensor var_14451_begin_0 = const()[name = tensor("op_14451_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14451_end_0 = const()[name = tensor("op_14451_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_14451_end_mask_0 = const()[name = tensor("op_14451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14451_cast_fp16 = slice_by_index(begin = var_14451_begin_0, end = var_14451_end_0, end_mask = var_14451_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14451_cast_fp16")]; + tensor var_14455_begin_0 = const()[name = tensor("op_14455_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14455_end_0 = const()[name = tensor("op_14455_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_14455_end_mask_0 = const()[name = tensor("op_14455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14455_cast_fp16 = slice_by_index(begin = var_14455_begin_0, end = var_14455_end_0, end_mask = var_14455_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14455_cast_fp16")]; + tensor var_14459_begin_0 = const()[name = tensor("op_14459_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14459_end_0 = const()[name = tensor("op_14459_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_14459_end_mask_0 = const()[name = tensor("op_14459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14459_cast_fp16 = slice_by_index(begin = var_14459_begin_0, end = var_14459_end_0, end_mask = var_14459_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14459_cast_fp16")]; + tensor var_14463_equation_0 = const()[name = tensor("op_14463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14463_cast_fp16 = einsum(equation = var_14463_equation_0, values = (var_14305_cast_fp16, var_14222_cast_fp16))[name = tensor("op_14463_cast_fp16")]; + tensor var_14464_to_fp16 = const()[name = tensor("op_14464_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1161_cast_fp16 = mul(x = var_14463_cast_fp16, y = var_14464_to_fp16)[name = tensor("aw_1161_cast_fp16")]; + tensor var_14467_equation_0 = const()[name = tensor("op_14467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14467_cast_fp16 = einsum(equation = var_14467_equation_0, values = (var_14309_cast_fp16, var_14226_cast_fp16))[name = tensor("op_14467_cast_fp16")]; + tensor var_14468_to_fp16 = const()[name = tensor("op_14468_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1163_cast_fp16 = mul(x = var_14467_cast_fp16, y = var_14468_to_fp16)[name = tensor("aw_1163_cast_fp16")]; + tensor var_14471_equation_0 = const()[name = tensor("op_14471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14471_cast_fp16 = einsum(equation = var_14471_equation_0, values = (var_14313_cast_fp16, var_14230_cast_fp16))[name = tensor("op_14471_cast_fp16")]; + tensor var_14472_to_fp16 = const()[name = tensor("op_14472_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1165_cast_fp16 = mul(x = var_14471_cast_fp16, y = var_14472_to_fp16)[name = tensor("aw_1165_cast_fp16")]; + tensor var_14475_equation_0 = const()[name = tensor("op_14475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14475_cast_fp16 = einsum(equation = var_14475_equation_0, values = (var_14317_cast_fp16, var_14234_cast_fp16))[name = tensor("op_14475_cast_fp16")]; + tensor var_14476_to_fp16 = const()[name = tensor("op_14476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1167_cast_fp16 = mul(x = var_14475_cast_fp16, y = var_14476_to_fp16)[name = tensor("aw_1167_cast_fp16")]; + tensor var_14479_equation_0 = const()[name = tensor("op_14479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14479_cast_fp16 = einsum(equation = var_14479_equation_0, values = (var_14321_cast_fp16, var_14238_cast_fp16))[name = tensor("op_14479_cast_fp16")]; + tensor var_14480_to_fp16 = const()[name = tensor("op_14480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1169_cast_fp16 = mul(x = var_14479_cast_fp16, y = var_14480_to_fp16)[name = tensor("aw_1169_cast_fp16")]; + tensor var_14483_equation_0 = const()[name = tensor("op_14483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14483_cast_fp16 = einsum(equation = var_14483_equation_0, values = (var_14325_cast_fp16, var_14242_cast_fp16))[name = tensor("op_14483_cast_fp16")]; + tensor var_14484_to_fp16 = const()[name = tensor("op_14484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1171_cast_fp16 = mul(x = var_14483_cast_fp16, y = var_14484_to_fp16)[name = tensor("aw_1171_cast_fp16")]; + tensor var_14487_equation_0 = const()[name = tensor("op_14487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14487_cast_fp16 = einsum(equation = var_14487_equation_0, values = (var_14329_cast_fp16, var_14246_cast_fp16))[name = tensor("op_14487_cast_fp16")]; + tensor var_14488_to_fp16 = const()[name = tensor("op_14488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1173_cast_fp16 = mul(x = var_14487_cast_fp16, y = var_14488_to_fp16)[name = tensor("aw_1173_cast_fp16")]; + tensor var_14491_equation_0 = const()[name = tensor("op_14491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14491_cast_fp16 = einsum(equation = var_14491_equation_0, values = (var_14333_cast_fp16, var_14250_cast_fp16))[name = tensor("op_14491_cast_fp16")]; + tensor var_14492_to_fp16 = const()[name = tensor("op_14492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1175_cast_fp16 = mul(x = var_14491_cast_fp16, y = var_14492_to_fp16)[name = tensor("aw_1175_cast_fp16")]; + tensor var_14495_equation_0 = const()[name = tensor("op_14495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14495_cast_fp16 = einsum(equation = var_14495_equation_0, values = (var_14337_cast_fp16, var_14254_cast_fp16))[name = tensor("op_14495_cast_fp16")]; + tensor var_14496_to_fp16 = const()[name = tensor("op_14496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1177_cast_fp16 = mul(x = var_14495_cast_fp16, y = var_14496_to_fp16)[name = tensor("aw_1177_cast_fp16")]; + tensor var_14499_equation_0 = const()[name = tensor("op_14499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14499_cast_fp16 = einsum(equation = var_14499_equation_0, values = (var_14341_cast_fp16, var_14258_cast_fp16))[name = tensor("op_14499_cast_fp16")]; + tensor var_14500_to_fp16 = const()[name = tensor("op_14500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1179_cast_fp16 = mul(x = var_14499_cast_fp16, y = var_14500_to_fp16)[name = tensor("aw_1179_cast_fp16")]; + tensor var_14503_equation_0 = const()[name = tensor("op_14503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14503_cast_fp16 = einsum(equation = var_14503_equation_0, values = (var_14345_cast_fp16, var_14262_cast_fp16))[name = tensor("op_14503_cast_fp16")]; + tensor var_14504_to_fp16 = const()[name = tensor("op_14504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1181_cast_fp16 = mul(x = var_14503_cast_fp16, y = var_14504_to_fp16)[name = tensor("aw_1181_cast_fp16")]; + tensor var_14507_equation_0 = const()[name = tensor("op_14507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14507_cast_fp16 = einsum(equation = var_14507_equation_0, values = (var_14349_cast_fp16, var_14266_cast_fp16))[name = tensor("op_14507_cast_fp16")]; + tensor var_14508_to_fp16 = const()[name = tensor("op_14508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1183_cast_fp16 = mul(x = var_14507_cast_fp16, y = var_14508_to_fp16)[name = tensor("aw_1183_cast_fp16")]; + tensor var_14511_equation_0 = const()[name = tensor("op_14511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14511_cast_fp16 = einsum(equation = var_14511_equation_0, values = (var_14353_cast_fp16, var_14270_cast_fp16))[name = tensor("op_14511_cast_fp16")]; + tensor var_14512_to_fp16 = const()[name = tensor("op_14512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1185_cast_fp16 = mul(x = var_14511_cast_fp16, y = var_14512_to_fp16)[name = tensor("aw_1185_cast_fp16")]; + tensor var_14515_equation_0 = const()[name = tensor("op_14515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14515_cast_fp16 = einsum(equation = var_14515_equation_0, values = (var_14357_cast_fp16, var_14274_cast_fp16))[name = tensor("op_14515_cast_fp16")]; + tensor var_14516_to_fp16 = const()[name = tensor("op_14516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1187_cast_fp16 = mul(x = var_14515_cast_fp16, y = var_14516_to_fp16)[name = tensor("aw_1187_cast_fp16")]; + tensor var_14519_equation_0 = const()[name = tensor("op_14519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14519_cast_fp16 = einsum(equation = var_14519_equation_0, values = (var_14361_cast_fp16, var_14278_cast_fp16))[name = tensor("op_14519_cast_fp16")]; + tensor var_14520_to_fp16 = const()[name = tensor("op_14520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1189_cast_fp16 = mul(x = var_14519_cast_fp16, y = var_14520_to_fp16)[name = tensor("aw_1189_cast_fp16")]; + tensor var_14523_equation_0 = const()[name = tensor("op_14523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14523_cast_fp16 = einsum(equation = var_14523_equation_0, values = (var_14365_cast_fp16, var_14282_cast_fp16))[name = tensor("op_14523_cast_fp16")]; + tensor var_14524_to_fp16 = const()[name = tensor("op_14524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1191_cast_fp16 = mul(x = var_14523_cast_fp16, y = var_14524_to_fp16)[name = tensor("aw_1191_cast_fp16")]; + tensor var_14527_equation_0 = const()[name = tensor("op_14527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14527_cast_fp16 = einsum(equation = var_14527_equation_0, values = (var_14369_cast_fp16, var_14286_cast_fp16))[name = tensor("op_14527_cast_fp16")]; + tensor var_14528_to_fp16 = const()[name = tensor("op_14528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1193_cast_fp16 = mul(x = var_14527_cast_fp16, y = var_14528_to_fp16)[name = tensor("aw_1193_cast_fp16")]; + tensor var_14531_equation_0 = const()[name = tensor("op_14531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14531_cast_fp16 = einsum(equation = var_14531_equation_0, values = (var_14373_cast_fp16, var_14290_cast_fp16))[name = tensor("op_14531_cast_fp16")]; + tensor var_14532_to_fp16 = const()[name = tensor("op_14532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1195_cast_fp16 = mul(x = var_14531_cast_fp16, y = var_14532_to_fp16)[name = tensor("aw_1195_cast_fp16")]; + tensor var_14535_equation_0 = const()[name = tensor("op_14535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14535_cast_fp16 = einsum(equation = var_14535_equation_0, values = (var_14377_cast_fp16, var_14294_cast_fp16))[name = tensor("op_14535_cast_fp16")]; + tensor var_14536_to_fp16 = const()[name = tensor("op_14536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1197_cast_fp16 = mul(x = var_14535_cast_fp16, y = var_14536_to_fp16)[name = tensor("aw_1197_cast_fp16")]; + tensor var_14539_equation_0 = const()[name = tensor("op_14539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14539_cast_fp16 = einsum(equation = var_14539_equation_0, values = (var_14381_cast_fp16, var_14298_cast_fp16))[name = tensor("op_14539_cast_fp16")]; + tensor var_14540_to_fp16 = const()[name = tensor("op_14540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1199_cast_fp16 = mul(x = var_14539_cast_fp16, y = var_14540_to_fp16)[name = tensor("aw_1199_cast_fp16")]; + tensor var_14542_cast_fp16 = softmax(axis = var_2624, x = aw_1161_cast_fp16)[name = tensor("op_14542_cast_fp16")]; + tensor var_14543_cast_fp16 = softmax(axis = var_2624, x = aw_1163_cast_fp16)[name = tensor("op_14543_cast_fp16")]; + tensor var_14544_cast_fp16 = softmax(axis = var_2624, x = aw_1165_cast_fp16)[name = tensor("op_14544_cast_fp16")]; + tensor var_14545_cast_fp16 = softmax(axis = var_2624, x = aw_1167_cast_fp16)[name = tensor("op_14545_cast_fp16")]; + tensor var_14546_cast_fp16 = softmax(axis = var_2624, x = aw_1169_cast_fp16)[name = tensor("op_14546_cast_fp16")]; + tensor var_14547_cast_fp16 = softmax(axis = var_2624, x = aw_1171_cast_fp16)[name = tensor("op_14547_cast_fp16")]; + tensor var_14548_cast_fp16 = softmax(axis = var_2624, x = aw_1173_cast_fp16)[name = tensor("op_14548_cast_fp16")]; + tensor var_14549_cast_fp16 = softmax(axis = var_2624, x = aw_1175_cast_fp16)[name = tensor("op_14549_cast_fp16")]; + tensor var_14550_cast_fp16 = softmax(axis = var_2624, x = aw_1177_cast_fp16)[name = tensor("op_14550_cast_fp16")]; + tensor var_14551_cast_fp16 = softmax(axis = var_2624, x = aw_1179_cast_fp16)[name = tensor("op_14551_cast_fp16")]; + tensor var_14552_cast_fp16 = softmax(axis = var_2624, x = aw_1181_cast_fp16)[name = tensor("op_14552_cast_fp16")]; + tensor var_14553_cast_fp16 = softmax(axis = var_2624, x = aw_1183_cast_fp16)[name = tensor("op_14553_cast_fp16")]; + tensor var_14554_cast_fp16 = softmax(axis = var_2624, x = aw_1185_cast_fp16)[name = tensor("op_14554_cast_fp16")]; + tensor var_14555_cast_fp16 = softmax(axis = var_2624, x = aw_1187_cast_fp16)[name = tensor("op_14555_cast_fp16")]; + tensor var_14556_cast_fp16 = softmax(axis = var_2624, x = aw_1189_cast_fp16)[name = tensor("op_14556_cast_fp16")]; + tensor var_14557_cast_fp16 = softmax(axis = var_2624, x = aw_1191_cast_fp16)[name = tensor("op_14557_cast_fp16")]; + tensor var_14558_cast_fp16 = softmax(axis = var_2624, x = aw_1193_cast_fp16)[name = tensor("op_14558_cast_fp16")]; + tensor var_14559_cast_fp16 = softmax(axis = var_2624, x = aw_1195_cast_fp16)[name = tensor("op_14559_cast_fp16")]; + tensor var_14560_cast_fp16 = softmax(axis = var_2624, x = aw_1197_cast_fp16)[name = tensor("op_14560_cast_fp16")]; + tensor var_14561_cast_fp16 = softmax(axis = var_2624, x = aw_1199_cast_fp16)[name = tensor("op_14561_cast_fp16")]; + tensor var_14563_equation_0 = const()[name = tensor("op_14563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14563_cast_fp16 = einsum(equation = var_14563_equation_0, values = (var_14383_cast_fp16, var_14542_cast_fp16))[name = tensor("op_14563_cast_fp16")]; + tensor var_14565_equation_0 = const()[name = tensor("op_14565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14565_cast_fp16 = einsum(equation = var_14565_equation_0, values = (var_14387_cast_fp16, var_14543_cast_fp16))[name = tensor("op_14565_cast_fp16")]; + tensor var_14567_equation_0 = const()[name = tensor("op_14567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14567_cast_fp16 = einsum(equation = var_14567_equation_0, values = (var_14391_cast_fp16, var_14544_cast_fp16))[name = tensor("op_14567_cast_fp16")]; + tensor var_14569_equation_0 = const()[name = tensor("op_14569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14569_cast_fp16 = einsum(equation = var_14569_equation_0, values = (var_14395_cast_fp16, var_14545_cast_fp16))[name = tensor("op_14569_cast_fp16")]; + tensor var_14571_equation_0 = const()[name = tensor("op_14571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14571_cast_fp16 = einsum(equation = var_14571_equation_0, values = (var_14399_cast_fp16, var_14546_cast_fp16))[name = tensor("op_14571_cast_fp16")]; + tensor var_14573_equation_0 = const()[name = tensor("op_14573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14573_cast_fp16 = einsum(equation = var_14573_equation_0, values = (var_14403_cast_fp16, var_14547_cast_fp16))[name = tensor("op_14573_cast_fp16")]; + tensor var_14575_equation_0 = const()[name = tensor("op_14575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14575_cast_fp16 = einsum(equation = var_14575_equation_0, values = (var_14407_cast_fp16, var_14548_cast_fp16))[name = tensor("op_14575_cast_fp16")]; + tensor var_14577_equation_0 = const()[name = tensor("op_14577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14577_cast_fp16 = einsum(equation = var_14577_equation_0, values = (var_14411_cast_fp16, var_14549_cast_fp16))[name = tensor("op_14577_cast_fp16")]; + tensor var_14579_equation_0 = const()[name = tensor("op_14579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14579_cast_fp16 = einsum(equation = var_14579_equation_0, values = (var_14415_cast_fp16, var_14550_cast_fp16))[name = tensor("op_14579_cast_fp16")]; + tensor var_14581_equation_0 = const()[name = tensor("op_14581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14581_cast_fp16 = einsum(equation = var_14581_equation_0, values = (var_14419_cast_fp16, var_14551_cast_fp16))[name = tensor("op_14581_cast_fp16")]; + tensor var_14583_equation_0 = const()[name = tensor("op_14583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14583_cast_fp16 = einsum(equation = var_14583_equation_0, values = (var_14423_cast_fp16, var_14552_cast_fp16))[name = tensor("op_14583_cast_fp16")]; + tensor var_14585_equation_0 = const()[name = tensor("op_14585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14585_cast_fp16 = einsum(equation = var_14585_equation_0, values = (var_14427_cast_fp16, var_14553_cast_fp16))[name = tensor("op_14585_cast_fp16")]; + tensor var_14587_equation_0 = const()[name = tensor("op_14587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14587_cast_fp16 = einsum(equation = var_14587_equation_0, values = (var_14431_cast_fp16, var_14554_cast_fp16))[name = tensor("op_14587_cast_fp16")]; + tensor var_14589_equation_0 = const()[name = tensor("op_14589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14589_cast_fp16 = einsum(equation = var_14589_equation_0, values = (var_14435_cast_fp16, var_14555_cast_fp16))[name = tensor("op_14589_cast_fp16")]; + tensor var_14591_equation_0 = const()[name = tensor("op_14591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14591_cast_fp16 = einsum(equation = var_14591_equation_0, values = (var_14439_cast_fp16, var_14556_cast_fp16))[name = tensor("op_14591_cast_fp16")]; + tensor var_14593_equation_0 = const()[name = tensor("op_14593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14593_cast_fp16 = einsum(equation = var_14593_equation_0, values = (var_14443_cast_fp16, var_14557_cast_fp16))[name = tensor("op_14593_cast_fp16")]; + tensor var_14595_equation_0 = const()[name = tensor("op_14595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14595_cast_fp16 = einsum(equation = var_14595_equation_0, values = (var_14447_cast_fp16, var_14558_cast_fp16))[name = tensor("op_14595_cast_fp16")]; + tensor var_14597_equation_0 = const()[name = tensor("op_14597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14597_cast_fp16 = einsum(equation = var_14597_equation_0, values = (var_14451_cast_fp16, var_14559_cast_fp16))[name = tensor("op_14597_cast_fp16")]; + tensor var_14599_equation_0 = const()[name = tensor("op_14599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14599_cast_fp16 = einsum(equation = var_14599_equation_0, values = (var_14455_cast_fp16, var_14560_cast_fp16))[name = tensor("op_14599_cast_fp16")]; + tensor var_14601_equation_0 = const()[name = tensor("op_14601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14601_cast_fp16 = einsum(equation = var_14601_equation_0, values = (var_14459_cast_fp16, var_14561_cast_fp16))[name = tensor("op_14601_cast_fp16")]; + tensor input_247_interleave_0 = const()[name = tensor("input_247_interleave_0"), val = tensor(false)]; + tensor input_247_cast_fp16 = concat(axis = var_2624, interleave = input_247_interleave_0, values = (var_14563_cast_fp16, var_14565_cast_fp16, var_14567_cast_fp16, var_14569_cast_fp16, var_14571_cast_fp16, var_14573_cast_fp16, var_14575_cast_fp16, var_14577_cast_fp16, var_14579_cast_fp16, var_14581_cast_fp16, var_14583_cast_fp16, var_14585_cast_fp16, var_14587_cast_fp16, var_14589_cast_fp16, var_14591_cast_fp16, var_14593_cast_fp16, var_14595_cast_fp16, var_14597_cast_fp16, var_14599_cast_fp16, var_14601_cast_fp16))[name = tensor("input_247_cast_fp16")]; + tensor var_14611_pad_type_0 = const()[name = tensor("op_14611_pad_type_0"), val = tensor("valid")]; + tensor var_14611_strides_0 = const()[name = tensor("op_14611_strides_0"), val = tensor([1, 1])]; + tensor var_14611_pad_0 = const()[name = tensor("op_14611_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14611_dilations_0 = const()[name = tensor("op_14611_dilations_0"), val = tensor([1, 1])]; + tensor var_14611_groups_0 = const()[name = tensor("op_14611_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423533312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(424762176))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(424762368)))]; + tensor var_14611_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_14611_dilations_0, groups = var_14611_groups_0, pad = var_14611_pad_0, pad_type = var_14611_pad_type_0, strides = var_14611_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized, x = input_247_cast_fp16)[name = tensor("op_14611_cast_fp16")]; + tensor inputs_101_cast_fp16 = add(x = var_14611_cast_fp16, y = inputs_99_cast_fp16)[name = tensor("inputs_101_cast_fp16")]; + tensor input_249_axes_0 = const()[name = tensor("input_249_axes_0"), val = tensor([1])]; + tensor input_249_gamma_0_to_fp16 = const()[name = tensor("input_249_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(424764992)))]; + tensor input_249_beta_0_to_fp16 = const()[name = tensor("input_249_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(424767616)))]; + tensor var_14621_to_fp16 = const()[name = tensor("op_14621_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_249_cast_fp16 = layer_norm(axes = input_249_axes_0, beta = input_249_beta_0_to_fp16, epsilon = var_14621_to_fp16, gamma = input_249_gamma_0_to_fp16, x = inputs_101_cast_fp16)[name = tensor("input_249_cast_fp16")]; + tensor var_14641_pad_type_0 = const()[name = tensor("op_14641_pad_type_0"), val = tensor("valid")]; + tensor var_14641_strides_0 = const()[name = tensor("op_14641_strides_0"), val = tensor([1, 1])]; + tensor var_14641_pad_0 = const()[name = tensor("op_14641_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14641_dilations_0 = const()[name = tensor("op_14641_dilations_0"), val = tensor([1, 1])]; + tensor var_14641_groups_0 = const()[name = tensor("op_14641_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(424770240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434600704))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434600896)))]; + tensor var_14641_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_14641_dilations_0, groups = var_14641_groups_0, pad = var_14641_pad_0, pad_type = var_14641_pad_type_0, strides = var_14641_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = tensor("op_14641_cast_fp16")]; + tensor var_14642_split_sizes_0 = const()[name = tensor("op_14642_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_14642_axis_0 = const()[name = tensor("op_14642_axis_0"), val = tensor(1)]; + tensor var_14642_cast_fp16_0, tensor var_14642_cast_fp16_1 = split(axis = var_14642_axis_0, split_sizes = var_14642_split_sizes_0, x = var_14641_cast_fp16)[name = tensor("op_14642_cast_fp16")]; + tensor var_14644_mode_0 = const()[name = tensor("op_14644_mode_0"), val = tensor("EXACT")]; + tensor var_14644_cast_fp16 = gelu(mode = var_14644_mode_0, x = var_14642_cast_fp16_1)[name = tensor("op_14644_cast_fp16")]; + tensor input_251_cast_fp16 = mul(x = var_14642_cast_fp16_0, y = var_14644_cast_fp16)[name = tensor("input_251_cast_fp16")]; + tensor var_14652_pad_type_0 = const()[name = tensor("op_14652_pad_type_0"), val = tensor("valid")]; + tensor var_14652_strides_0 = const()[name = tensor("op_14652_strides_0"), val = tensor([1, 1])]; + tensor var_14652_pad_0 = const()[name = tensor("op_14652_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14652_dilations_0 = const()[name = tensor("op_14652_dilations_0"), val = tensor([1, 1])]; + tensor var_14652_groups_0 = const()[name = tensor("op_14652_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434621440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(439536704))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(439536896)))]; + tensor var_14652_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_14652_dilations_0, groups = var_14652_groups_0, pad = var_14652_pad_0, pad_type = var_14652_pad_type_0, strides = var_14652_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized, x = input_251_cast_fp16)[name = tensor("op_14652_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = var_14652_cast_fp16, y = inputs_101_cast_fp16)[name = tensor("inputs_103_cast_fp16")]; + tensor hidden_states_155_axes_0 = const()[name = tensor("hidden_states_155_axes_0"), val = tensor([1])]; + tensor hidden_states_155_gamma_0_to_fp16 = const()[name = tensor("hidden_states_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(439539520)))]; + tensor hidden_states_155_beta_0_to_fp16 = const()[name = tensor("hidden_states_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(439542144)))]; + tensor var_14668_to_fp16 = const()[name = tensor("op_14668_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_155_cast_fp16 = layer_norm(axes = hidden_states_155_axes_0, beta = hidden_states_155_beta_0_to_fp16, epsilon = var_14668_to_fp16, gamma = hidden_states_155_gamma_0_to_fp16, x = inputs_103_cast_fp16)[name = tensor("hidden_states_155_cast_fp16")]; + tensor q_69_pad_type_0 = const()[name = tensor("q_69_pad_type_0"), val = tensor("valid")]; + tensor q_69_strides_0 = const()[name = tensor("q_69_strides_0"), val = tensor([1, 1])]; + tensor q_69_pad_0 = const()[name = tensor("q_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_69_dilations_0 = const()[name = tensor("q_69_dilations_0"), val = tensor([1, 1])]; + tensor q_69_groups_0 = const()[name = tensor("q_69_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(439544768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440773632))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_69_cast_fp16 = conv(dilations = q_69_dilations_0, groups = q_69_groups_0, pad = q_69_pad_0, pad_type = q_69_pad_type_0, strides = q_69_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_155_cast_fp16)[name = tensor("q_69_cast_fp16")]; + tensor k_137_pad_type_0 = const()[name = tensor("k_137_pad_type_0"), val = tensor("valid")]; + tensor k_137_strides_0 = const()[name = tensor("k_137_strides_0"), val = tensor([1, 1])]; + tensor k_137_pad_0 = const()[name = tensor("k_137_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_137_dilations_0 = const()[name = tensor("k_137_dilations_0"), val = tensor([1, 1])]; + tensor k_137_groups_0 = const()[name = tensor("k_137_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440773824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442002688))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_137_cast_fp16 = conv(dilations = k_137_dilations_0, groups = k_137_groups_0, pad = k_137_pad_0, pad_type = k_137_pad_type_0, strides = k_137_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_155_cast_fp16)[name = tensor("k_137_cast_fp16")]; + tensor v_69_pad_type_0 = const()[name = tensor("v_69_pad_type_0"), val = tensor("valid")]; + tensor v_69_strides_0 = const()[name = tensor("v_69_strides_0"), val = tensor([1, 1])]; + tensor v_69_pad_0 = const()[name = tensor("v_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_69_dilations_0 = const()[name = tensor("v_69_dilations_0"), val = tensor([1, 1])]; + tensor v_69_groups_0 = const()[name = tensor("v_69_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442002880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(443231744))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_69_cast_fp16 = conv(dilations = v_69_dilations_0, groups = v_69_groups_0, pad = v_69_pad_0, pad_type = v_69_pad_type_0, strides = v_69_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_155_cast_fp16)[name = tensor("v_69_cast_fp16")]; + tensor var_14701_begin_0 = const()[name = tensor("op_14701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14701_end_0 = const()[name = tensor("op_14701_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14701_end_mask_0 = const()[name = tensor("op_14701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14701_cast_fp16 = slice_by_index(begin = var_14701_begin_0, end = var_14701_end_0, end_mask = var_14701_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14701_cast_fp16")]; + tensor var_14705_begin_0 = const()[name = tensor("op_14705_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14705_end_0 = const()[name = tensor("op_14705_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_14705_end_mask_0 = const()[name = tensor("op_14705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14705_cast_fp16 = slice_by_index(begin = var_14705_begin_0, end = var_14705_end_0, end_mask = var_14705_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14705_cast_fp16")]; + tensor var_14709_begin_0 = const()[name = tensor("op_14709_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14709_end_0 = const()[name = tensor("op_14709_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_14709_end_mask_0 = const()[name = tensor("op_14709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14709_cast_fp16 = slice_by_index(begin = var_14709_begin_0, end = var_14709_end_0, end_mask = var_14709_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14709_cast_fp16")]; + tensor var_14713_begin_0 = const()[name = tensor("op_14713_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14713_end_0 = const()[name = tensor("op_14713_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_14713_end_mask_0 = const()[name = tensor("op_14713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14713_cast_fp16 = slice_by_index(begin = var_14713_begin_0, end = var_14713_end_0, end_mask = var_14713_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14713_cast_fp16")]; + tensor var_14717_begin_0 = const()[name = tensor("op_14717_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14717_end_0 = const()[name = tensor("op_14717_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_14717_end_mask_0 = const()[name = tensor("op_14717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14717_cast_fp16 = slice_by_index(begin = var_14717_begin_0, end = var_14717_end_0, end_mask = var_14717_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14717_cast_fp16")]; + tensor var_14721_begin_0 = const()[name = tensor("op_14721_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14721_end_0 = const()[name = tensor("op_14721_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_14721_end_mask_0 = const()[name = tensor("op_14721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14721_cast_fp16 = slice_by_index(begin = var_14721_begin_0, end = var_14721_end_0, end_mask = var_14721_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14721_cast_fp16")]; + tensor var_14725_begin_0 = const()[name = tensor("op_14725_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14725_end_0 = const()[name = tensor("op_14725_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_14725_end_mask_0 = const()[name = tensor("op_14725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14725_cast_fp16 = slice_by_index(begin = var_14725_begin_0, end = var_14725_end_0, end_mask = var_14725_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14725_cast_fp16")]; + tensor var_14729_begin_0 = const()[name = tensor("op_14729_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14729_end_0 = const()[name = tensor("op_14729_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_14729_end_mask_0 = const()[name = tensor("op_14729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14729_cast_fp16 = slice_by_index(begin = var_14729_begin_0, end = var_14729_end_0, end_mask = var_14729_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14729_cast_fp16")]; + tensor var_14733_begin_0 = const()[name = tensor("op_14733_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14733_end_0 = const()[name = tensor("op_14733_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_14733_end_mask_0 = const()[name = tensor("op_14733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14733_cast_fp16 = slice_by_index(begin = var_14733_begin_0, end = var_14733_end_0, end_mask = var_14733_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14733_cast_fp16")]; + tensor var_14737_begin_0 = const()[name = tensor("op_14737_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14737_end_0 = const()[name = tensor("op_14737_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_14737_end_mask_0 = const()[name = tensor("op_14737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14737_cast_fp16 = slice_by_index(begin = var_14737_begin_0, end = var_14737_end_0, end_mask = var_14737_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14737_cast_fp16")]; + tensor var_14741_begin_0 = const()[name = tensor("op_14741_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14741_end_0 = const()[name = tensor("op_14741_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_14741_end_mask_0 = const()[name = tensor("op_14741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14741_cast_fp16 = slice_by_index(begin = var_14741_begin_0, end = var_14741_end_0, end_mask = var_14741_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14741_cast_fp16")]; + tensor var_14745_begin_0 = const()[name = tensor("op_14745_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14745_end_0 = const()[name = tensor("op_14745_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_14745_end_mask_0 = const()[name = tensor("op_14745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14745_cast_fp16 = slice_by_index(begin = var_14745_begin_0, end = var_14745_end_0, end_mask = var_14745_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14745_cast_fp16")]; + tensor var_14749_begin_0 = const()[name = tensor("op_14749_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14749_end_0 = const()[name = tensor("op_14749_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_14749_end_mask_0 = const()[name = tensor("op_14749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14749_cast_fp16 = slice_by_index(begin = var_14749_begin_0, end = var_14749_end_0, end_mask = var_14749_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14749_cast_fp16")]; + tensor var_14753_begin_0 = const()[name = tensor("op_14753_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14753_end_0 = const()[name = tensor("op_14753_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14753_end_mask_0 = const()[name = tensor("op_14753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14753_cast_fp16 = slice_by_index(begin = var_14753_begin_0, end = var_14753_end_0, end_mask = var_14753_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14753_cast_fp16")]; + tensor var_14757_begin_0 = const()[name = tensor("op_14757_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14757_end_0 = const()[name = tensor("op_14757_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14757_end_mask_0 = const()[name = tensor("op_14757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14757_cast_fp16 = slice_by_index(begin = var_14757_begin_0, end = var_14757_end_0, end_mask = var_14757_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14757_cast_fp16")]; + tensor var_14761_begin_0 = const()[name = tensor("op_14761_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14761_end_0 = const()[name = tensor("op_14761_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14761_end_mask_0 = const()[name = tensor("op_14761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14761_cast_fp16 = slice_by_index(begin = var_14761_begin_0, end = var_14761_end_0, end_mask = var_14761_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14761_cast_fp16")]; + tensor var_14765_begin_0 = const()[name = tensor("op_14765_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14765_end_0 = const()[name = tensor("op_14765_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14765_end_mask_0 = const()[name = tensor("op_14765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14765_cast_fp16 = slice_by_index(begin = var_14765_begin_0, end = var_14765_end_0, end_mask = var_14765_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14765_cast_fp16")]; + tensor var_14769_begin_0 = const()[name = tensor("op_14769_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14769_end_0 = const()[name = tensor("op_14769_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14769_end_mask_0 = const()[name = tensor("op_14769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14769_cast_fp16 = slice_by_index(begin = var_14769_begin_0, end = var_14769_end_0, end_mask = var_14769_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14769_cast_fp16")]; + tensor var_14773_begin_0 = const()[name = tensor("op_14773_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14773_end_0 = const()[name = tensor("op_14773_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14773_end_mask_0 = const()[name = tensor("op_14773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14773_cast_fp16 = slice_by_index(begin = var_14773_begin_0, end = var_14773_end_0, end_mask = var_14773_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14773_cast_fp16")]; + tensor var_14777_begin_0 = const()[name = tensor("op_14777_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14777_end_0 = const()[name = tensor("op_14777_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14777_end_mask_0 = const()[name = tensor("op_14777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14777_cast_fp16 = slice_by_index(begin = var_14777_begin_0, end = var_14777_end_0, end_mask = var_14777_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14777_cast_fp16")]; + tensor k_139_perm_0 = const()[name = tensor("k_139_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14784_begin_0 = const()[name = tensor("op_14784_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14784_end_0 = const()[name = tensor("op_14784_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_14784_end_mask_0 = const()[name = tensor("op_14784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_139_cast_fp16 = transpose(perm = k_139_perm_0, x = k_137_cast_fp16)[name = tensor("transpose_33")]; + tensor var_14784_cast_fp16 = slice_by_index(begin = var_14784_begin_0, end = var_14784_end_0, end_mask = var_14784_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14784_cast_fp16")]; + tensor var_14788_begin_0 = const()[name = tensor("op_14788_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14788_end_0 = const()[name = tensor("op_14788_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_14788_end_mask_0 = const()[name = tensor("op_14788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14788_cast_fp16 = slice_by_index(begin = var_14788_begin_0, end = var_14788_end_0, end_mask = var_14788_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14788_cast_fp16")]; + tensor var_14792_begin_0 = const()[name = tensor("op_14792_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14792_end_0 = const()[name = tensor("op_14792_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_14792_end_mask_0 = const()[name = tensor("op_14792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14792_cast_fp16 = slice_by_index(begin = var_14792_begin_0, end = var_14792_end_0, end_mask = var_14792_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14792_cast_fp16")]; + tensor var_14796_begin_0 = const()[name = tensor("op_14796_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14796_end_0 = const()[name = tensor("op_14796_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_14796_end_mask_0 = const()[name = tensor("op_14796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14796_cast_fp16 = slice_by_index(begin = var_14796_begin_0, end = var_14796_end_0, end_mask = var_14796_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14796_cast_fp16")]; + tensor var_14800_begin_0 = const()[name = tensor("op_14800_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14800_end_0 = const()[name = tensor("op_14800_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_14800_end_mask_0 = const()[name = tensor("op_14800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14800_cast_fp16 = slice_by_index(begin = var_14800_begin_0, end = var_14800_end_0, end_mask = var_14800_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14800_cast_fp16")]; + tensor var_14804_begin_0 = const()[name = tensor("op_14804_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_14804_end_0 = const()[name = tensor("op_14804_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_14804_end_mask_0 = const()[name = tensor("op_14804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14804_cast_fp16 = slice_by_index(begin = var_14804_begin_0, end = var_14804_end_0, end_mask = var_14804_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14804_cast_fp16")]; + tensor var_14808_begin_0 = const()[name = tensor("op_14808_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_14808_end_0 = const()[name = tensor("op_14808_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_14808_end_mask_0 = const()[name = tensor("op_14808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14808_cast_fp16 = slice_by_index(begin = var_14808_begin_0, end = var_14808_end_0, end_mask = var_14808_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14808_cast_fp16")]; + tensor var_14812_begin_0 = const()[name = tensor("op_14812_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_14812_end_0 = const()[name = tensor("op_14812_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_14812_end_mask_0 = const()[name = tensor("op_14812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14812_cast_fp16 = slice_by_index(begin = var_14812_begin_0, end = var_14812_end_0, end_mask = var_14812_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14812_cast_fp16")]; + tensor var_14816_begin_0 = const()[name = tensor("op_14816_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14816_end_0 = const()[name = tensor("op_14816_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_14816_end_mask_0 = const()[name = tensor("op_14816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14816_cast_fp16 = slice_by_index(begin = var_14816_begin_0, end = var_14816_end_0, end_mask = var_14816_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14816_cast_fp16")]; + tensor var_14820_begin_0 = const()[name = tensor("op_14820_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_14820_end_0 = const()[name = tensor("op_14820_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_14820_end_mask_0 = const()[name = tensor("op_14820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14820_cast_fp16 = slice_by_index(begin = var_14820_begin_0, end = var_14820_end_0, end_mask = var_14820_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14820_cast_fp16")]; + tensor var_14824_begin_0 = const()[name = tensor("op_14824_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_14824_end_0 = const()[name = tensor("op_14824_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_14824_end_mask_0 = const()[name = tensor("op_14824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14824_cast_fp16 = slice_by_index(begin = var_14824_begin_0, end = var_14824_end_0, end_mask = var_14824_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14824_cast_fp16")]; + tensor var_14828_begin_0 = const()[name = tensor("op_14828_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_14828_end_0 = const()[name = tensor("op_14828_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_14828_end_mask_0 = const()[name = tensor("op_14828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14828_cast_fp16 = slice_by_index(begin = var_14828_begin_0, end = var_14828_end_0, end_mask = var_14828_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14828_cast_fp16")]; + tensor var_14832_begin_0 = const()[name = tensor("op_14832_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_14832_end_0 = const()[name = tensor("op_14832_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_14832_end_mask_0 = const()[name = tensor("op_14832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14832_cast_fp16 = slice_by_index(begin = var_14832_begin_0, end = var_14832_end_0, end_mask = var_14832_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14832_cast_fp16")]; + tensor var_14836_begin_0 = const()[name = tensor("op_14836_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_14836_end_0 = const()[name = tensor("op_14836_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_14836_end_mask_0 = const()[name = tensor("op_14836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14836_cast_fp16 = slice_by_index(begin = var_14836_begin_0, end = var_14836_end_0, end_mask = var_14836_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14836_cast_fp16")]; + tensor var_14840_begin_0 = const()[name = tensor("op_14840_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_14840_end_0 = const()[name = tensor("op_14840_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_14840_end_mask_0 = const()[name = tensor("op_14840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14840_cast_fp16 = slice_by_index(begin = var_14840_begin_0, end = var_14840_end_0, end_mask = var_14840_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14840_cast_fp16")]; + tensor var_14844_begin_0 = const()[name = tensor("op_14844_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_14844_end_0 = const()[name = tensor("op_14844_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14844_end_mask_0 = const()[name = tensor("op_14844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14844_cast_fp16 = slice_by_index(begin = var_14844_begin_0, end = var_14844_end_0, end_mask = var_14844_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14844_cast_fp16")]; + tensor var_14848_begin_0 = const()[name = tensor("op_14848_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14848_end_0 = const()[name = tensor("op_14848_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_14848_end_mask_0 = const()[name = tensor("op_14848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14848_cast_fp16 = slice_by_index(begin = var_14848_begin_0, end = var_14848_end_0, end_mask = var_14848_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14848_cast_fp16")]; + tensor var_14852_begin_0 = const()[name = tensor("op_14852_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_14852_end_0 = const()[name = tensor("op_14852_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_14852_end_mask_0 = const()[name = tensor("op_14852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14852_cast_fp16 = slice_by_index(begin = var_14852_begin_0, end = var_14852_end_0, end_mask = var_14852_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14852_cast_fp16")]; + tensor var_14856_begin_0 = const()[name = tensor("op_14856_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_14856_end_0 = const()[name = tensor("op_14856_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_14856_end_mask_0 = const()[name = tensor("op_14856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14856_cast_fp16 = slice_by_index(begin = var_14856_begin_0, end = var_14856_end_0, end_mask = var_14856_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14856_cast_fp16")]; + tensor var_14860_begin_0 = const()[name = tensor("op_14860_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_14860_end_0 = const()[name = tensor("op_14860_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_14860_end_mask_0 = const()[name = tensor("op_14860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14860_cast_fp16 = slice_by_index(begin = var_14860_begin_0, end = var_14860_end_0, end_mask = var_14860_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14860_cast_fp16")]; + tensor var_14862_begin_0 = const()[name = tensor("op_14862_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14862_end_0 = const()[name = tensor("op_14862_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14862_end_mask_0 = const()[name = tensor("op_14862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14862_cast_fp16 = slice_by_index(begin = var_14862_begin_0, end = var_14862_end_0, end_mask = var_14862_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14862_cast_fp16")]; + tensor var_14866_begin_0 = const()[name = tensor("op_14866_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14866_end_0 = const()[name = tensor("op_14866_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_14866_end_mask_0 = const()[name = tensor("op_14866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14866_cast_fp16 = slice_by_index(begin = var_14866_begin_0, end = var_14866_end_0, end_mask = var_14866_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14866_cast_fp16")]; + tensor var_14870_begin_0 = const()[name = tensor("op_14870_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14870_end_0 = const()[name = tensor("op_14870_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_14870_end_mask_0 = const()[name = tensor("op_14870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14870_cast_fp16 = slice_by_index(begin = var_14870_begin_0, end = var_14870_end_0, end_mask = var_14870_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14870_cast_fp16")]; + tensor var_14874_begin_0 = const()[name = tensor("op_14874_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14874_end_0 = const()[name = tensor("op_14874_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_14874_end_mask_0 = const()[name = tensor("op_14874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14874_cast_fp16 = slice_by_index(begin = var_14874_begin_0, end = var_14874_end_0, end_mask = var_14874_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14874_cast_fp16")]; + tensor var_14878_begin_0 = const()[name = tensor("op_14878_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14878_end_0 = const()[name = tensor("op_14878_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_14878_end_mask_0 = const()[name = tensor("op_14878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14878_cast_fp16 = slice_by_index(begin = var_14878_begin_0, end = var_14878_end_0, end_mask = var_14878_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14878_cast_fp16")]; + tensor var_14882_begin_0 = const()[name = tensor("op_14882_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14882_end_0 = const()[name = tensor("op_14882_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_14882_end_mask_0 = const()[name = tensor("op_14882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14882_cast_fp16 = slice_by_index(begin = var_14882_begin_0, end = var_14882_end_0, end_mask = var_14882_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14882_cast_fp16")]; + tensor var_14886_begin_0 = const()[name = tensor("op_14886_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14886_end_0 = const()[name = tensor("op_14886_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_14886_end_mask_0 = const()[name = tensor("op_14886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14886_cast_fp16 = slice_by_index(begin = var_14886_begin_0, end = var_14886_end_0, end_mask = var_14886_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14886_cast_fp16")]; + tensor var_14890_begin_0 = const()[name = tensor("op_14890_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14890_end_0 = const()[name = tensor("op_14890_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_14890_end_mask_0 = const()[name = tensor("op_14890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14890_cast_fp16 = slice_by_index(begin = var_14890_begin_0, end = var_14890_end_0, end_mask = var_14890_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14890_cast_fp16")]; + tensor var_14894_begin_0 = const()[name = tensor("op_14894_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14894_end_0 = const()[name = tensor("op_14894_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_14894_end_mask_0 = const()[name = tensor("op_14894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14894_cast_fp16 = slice_by_index(begin = var_14894_begin_0, end = var_14894_end_0, end_mask = var_14894_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14894_cast_fp16")]; + tensor var_14898_begin_0 = const()[name = tensor("op_14898_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14898_end_0 = const()[name = tensor("op_14898_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_14898_end_mask_0 = const()[name = tensor("op_14898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14898_cast_fp16 = slice_by_index(begin = var_14898_begin_0, end = var_14898_end_0, end_mask = var_14898_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14898_cast_fp16")]; + tensor var_14902_begin_0 = const()[name = tensor("op_14902_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14902_end_0 = const()[name = tensor("op_14902_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_14902_end_mask_0 = const()[name = tensor("op_14902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14902_cast_fp16 = slice_by_index(begin = var_14902_begin_0, end = var_14902_end_0, end_mask = var_14902_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14902_cast_fp16")]; + tensor var_14906_begin_0 = const()[name = tensor("op_14906_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14906_end_0 = const()[name = tensor("op_14906_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_14906_end_mask_0 = const()[name = tensor("op_14906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14906_cast_fp16 = slice_by_index(begin = var_14906_begin_0, end = var_14906_end_0, end_mask = var_14906_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14906_cast_fp16")]; + tensor var_14910_begin_0 = const()[name = tensor("op_14910_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14910_end_0 = const()[name = tensor("op_14910_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_14910_end_mask_0 = const()[name = tensor("op_14910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14910_cast_fp16 = slice_by_index(begin = var_14910_begin_0, end = var_14910_end_0, end_mask = var_14910_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14910_cast_fp16")]; + tensor var_14914_begin_0 = const()[name = tensor("op_14914_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14914_end_0 = const()[name = tensor("op_14914_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14914_end_mask_0 = const()[name = tensor("op_14914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14914_cast_fp16 = slice_by_index(begin = var_14914_begin_0, end = var_14914_end_0, end_mask = var_14914_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14914_cast_fp16")]; + tensor var_14918_begin_0 = const()[name = tensor("op_14918_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14918_end_0 = const()[name = tensor("op_14918_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14918_end_mask_0 = const()[name = tensor("op_14918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14918_cast_fp16 = slice_by_index(begin = var_14918_begin_0, end = var_14918_end_0, end_mask = var_14918_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14918_cast_fp16")]; + tensor var_14922_begin_0 = const()[name = tensor("op_14922_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14922_end_0 = const()[name = tensor("op_14922_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14922_end_mask_0 = const()[name = tensor("op_14922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14922_cast_fp16 = slice_by_index(begin = var_14922_begin_0, end = var_14922_end_0, end_mask = var_14922_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14922_cast_fp16")]; + tensor var_14926_begin_0 = const()[name = tensor("op_14926_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14926_end_0 = const()[name = tensor("op_14926_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14926_end_mask_0 = const()[name = tensor("op_14926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14926_cast_fp16 = slice_by_index(begin = var_14926_begin_0, end = var_14926_end_0, end_mask = var_14926_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14926_cast_fp16")]; + tensor var_14930_begin_0 = const()[name = tensor("op_14930_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14930_end_0 = const()[name = tensor("op_14930_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14930_end_mask_0 = const()[name = tensor("op_14930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14930_cast_fp16 = slice_by_index(begin = var_14930_begin_0, end = var_14930_end_0, end_mask = var_14930_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14930_cast_fp16")]; + tensor var_14934_begin_0 = const()[name = tensor("op_14934_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14934_end_0 = const()[name = tensor("op_14934_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14934_end_mask_0 = const()[name = tensor("op_14934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14934_cast_fp16 = slice_by_index(begin = var_14934_begin_0, end = var_14934_end_0, end_mask = var_14934_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14934_cast_fp16")]; + tensor var_14938_begin_0 = const()[name = tensor("op_14938_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14938_end_0 = const()[name = tensor("op_14938_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14938_end_mask_0 = const()[name = tensor("op_14938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14938_cast_fp16 = slice_by_index(begin = var_14938_begin_0, end = var_14938_end_0, end_mask = var_14938_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14938_cast_fp16")]; + tensor var_14942_equation_0 = const()[name = tensor("op_14942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14942_cast_fp16 = einsum(equation = var_14942_equation_0, values = (var_14784_cast_fp16, var_14701_cast_fp16))[name = tensor("op_14942_cast_fp16")]; + tensor var_14943_to_fp16 = const()[name = tensor("op_14943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1201_cast_fp16 = mul(x = var_14942_cast_fp16, y = var_14943_to_fp16)[name = tensor("aw_1201_cast_fp16")]; + tensor var_14946_equation_0 = const()[name = tensor("op_14946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14946_cast_fp16 = einsum(equation = var_14946_equation_0, values = (var_14788_cast_fp16, var_14705_cast_fp16))[name = tensor("op_14946_cast_fp16")]; + tensor var_14947_to_fp16 = const()[name = tensor("op_14947_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1203_cast_fp16 = mul(x = var_14946_cast_fp16, y = var_14947_to_fp16)[name = tensor("aw_1203_cast_fp16")]; + tensor var_14950_equation_0 = const()[name = tensor("op_14950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14950_cast_fp16 = einsum(equation = var_14950_equation_0, values = (var_14792_cast_fp16, var_14709_cast_fp16))[name = tensor("op_14950_cast_fp16")]; + tensor var_14951_to_fp16 = const()[name = tensor("op_14951_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1205_cast_fp16 = mul(x = var_14950_cast_fp16, y = var_14951_to_fp16)[name = tensor("aw_1205_cast_fp16")]; + tensor var_14954_equation_0 = const()[name = tensor("op_14954_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14954_cast_fp16 = einsum(equation = var_14954_equation_0, values = (var_14796_cast_fp16, var_14713_cast_fp16))[name = tensor("op_14954_cast_fp16")]; + tensor var_14955_to_fp16 = const()[name = tensor("op_14955_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1207_cast_fp16 = mul(x = var_14954_cast_fp16, y = var_14955_to_fp16)[name = tensor("aw_1207_cast_fp16")]; + tensor var_14958_equation_0 = const()[name = tensor("op_14958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14958_cast_fp16 = einsum(equation = var_14958_equation_0, values = (var_14800_cast_fp16, var_14717_cast_fp16))[name = tensor("op_14958_cast_fp16")]; + tensor var_14959_to_fp16 = const()[name = tensor("op_14959_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1209_cast_fp16 = mul(x = var_14958_cast_fp16, y = var_14959_to_fp16)[name = tensor("aw_1209_cast_fp16")]; + tensor var_14962_equation_0 = const()[name = tensor("op_14962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14962_cast_fp16 = einsum(equation = var_14962_equation_0, values = (var_14804_cast_fp16, var_14721_cast_fp16))[name = tensor("op_14962_cast_fp16")]; + tensor var_14963_to_fp16 = const()[name = tensor("op_14963_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1211_cast_fp16 = mul(x = var_14962_cast_fp16, y = var_14963_to_fp16)[name = tensor("aw_1211_cast_fp16")]; + tensor var_14966_equation_0 = const()[name = tensor("op_14966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14966_cast_fp16 = einsum(equation = var_14966_equation_0, values = (var_14808_cast_fp16, var_14725_cast_fp16))[name = tensor("op_14966_cast_fp16")]; + tensor var_14967_to_fp16 = const()[name = tensor("op_14967_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1213_cast_fp16 = mul(x = var_14966_cast_fp16, y = var_14967_to_fp16)[name = tensor("aw_1213_cast_fp16")]; + tensor var_14970_equation_0 = const()[name = tensor("op_14970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14970_cast_fp16 = einsum(equation = var_14970_equation_0, values = (var_14812_cast_fp16, var_14729_cast_fp16))[name = tensor("op_14970_cast_fp16")]; + tensor var_14971_to_fp16 = const()[name = tensor("op_14971_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1215_cast_fp16 = mul(x = var_14970_cast_fp16, y = var_14971_to_fp16)[name = tensor("aw_1215_cast_fp16")]; + tensor var_14974_equation_0 = const()[name = tensor("op_14974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14974_cast_fp16 = einsum(equation = var_14974_equation_0, values = (var_14816_cast_fp16, var_14733_cast_fp16))[name = tensor("op_14974_cast_fp16")]; + tensor var_14975_to_fp16 = const()[name = tensor("op_14975_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1217_cast_fp16 = mul(x = var_14974_cast_fp16, y = var_14975_to_fp16)[name = tensor("aw_1217_cast_fp16")]; + tensor var_14978_equation_0 = const()[name = tensor("op_14978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14978_cast_fp16 = einsum(equation = var_14978_equation_0, values = (var_14820_cast_fp16, var_14737_cast_fp16))[name = tensor("op_14978_cast_fp16")]; + tensor var_14979_to_fp16 = const()[name = tensor("op_14979_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1219_cast_fp16 = mul(x = var_14978_cast_fp16, y = var_14979_to_fp16)[name = tensor("aw_1219_cast_fp16")]; + tensor var_14982_equation_0 = const()[name = tensor("op_14982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14982_cast_fp16 = einsum(equation = var_14982_equation_0, values = (var_14824_cast_fp16, var_14741_cast_fp16))[name = tensor("op_14982_cast_fp16")]; + tensor var_14983_to_fp16 = const()[name = tensor("op_14983_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1221_cast_fp16 = mul(x = var_14982_cast_fp16, y = var_14983_to_fp16)[name = tensor("aw_1221_cast_fp16")]; + tensor var_14986_equation_0 = const()[name = tensor("op_14986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14986_cast_fp16 = einsum(equation = var_14986_equation_0, values = (var_14828_cast_fp16, var_14745_cast_fp16))[name = tensor("op_14986_cast_fp16")]; + tensor var_14987_to_fp16 = const()[name = tensor("op_14987_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1223_cast_fp16 = mul(x = var_14986_cast_fp16, y = var_14987_to_fp16)[name = tensor("aw_1223_cast_fp16")]; + tensor var_14990_equation_0 = const()[name = tensor("op_14990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14990_cast_fp16 = einsum(equation = var_14990_equation_0, values = (var_14832_cast_fp16, var_14749_cast_fp16))[name = tensor("op_14990_cast_fp16")]; + tensor var_14991_to_fp16 = const()[name = tensor("op_14991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1225_cast_fp16 = mul(x = var_14990_cast_fp16, y = var_14991_to_fp16)[name = tensor("aw_1225_cast_fp16")]; + tensor var_14994_equation_0 = const()[name = tensor("op_14994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14994_cast_fp16 = einsum(equation = var_14994_equation_0, values = (var_14836_cast_fp16, var_14753_cast_fp16))[name = tensor("op_14994_cast_fp16")]; + tensor var_14995_to_fp16 = const()[name = tensor("op_14995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1227_cast_fp16 = mul(x = var_14994_cast_fp16, y = var_14995_to_fp16)[name = tensor("aw_1227_cast_fp16")]; + tensor var_14998_equation_0 = const()[name = tensor("op_14998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14998_cast_fp16 = einsum(equation = var_14998_equation_0, values = (var_14840_cast_fp16, var_14757_cast_fp16))[name = tensor("op_14998_cast_fp16")]; + tensor var_14999_to_fp16 = const()[name = tensor("op_14999_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1229_cast_fp16 = mul(x = var_14998_cast_fp16, y = var_14999_to_fp16)[name = tensor("aw_1229_cast_fp16")]; + tensor var_15002_equation_0 = const()[name = tensor("op_15002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15002_cast_fp16 = einsum(equation = var_15002_equation_0, values = (var_14844_cast_fp16, var_14761_cast_fp16))[name = tensor("op_15002_cast_fp16")]; + tensor var_15003_to_fp16 = const()[name = tensor("op_15003_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1231_cast_fp16 = mul(x = var_15002_cast_fp16, y = var_15003_to_fp16)[name = tensor("aw_1231_cast_fp16")]; + tensor var_15006_equation_0 = const()[name = tensor("op_15006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15006_cast_fp16 = einsum(equation = var_15006_equation_0, values = (var_14848_cast_fp16, var_14765_cast_fp16))[name = tensor("op_15006_cast_fp16")]; + tensor var_15007_to_fp16 = const()[name = tensor("op_15007_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1233_cast_fp16 = mul(x = var_15006_cast_fp16, y = var_15007_to_fp16)[name = tensor("aw_1233_cast_fp16")]; + tensor var_15010_equation_0 = const()[name = tensor("op_15010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15010_cast_fp16 = einsum(equation = var_15010_equation_0, values = (var_14852_cast_fp16, var_14769_cast_fp16))[name = tensor("op_15010_cast_fp16")]; + tensor var_15011_to_fp16 = const()[name = tensor("op_15011_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1235_cast_fp16 = mul(x = var_15010_cast_fp16, y = var_15011_to_fp16)[name = tensor("aw_1235_cast_fp16")]; + tensor var_15014_equation_0 = const()[name = tensor("op_15014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15014_cast_fp16 = einsum(equation = var_15014_equation_0, values = (var_14856_cast_fp16, var_14773_cast_fp16))[name = tensor("op_15014_cast_fp16")]; + tensor var_15015_to_fp16 = const()[name = tensor("op_15015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1237_cast_fp16 = mul(x = var_15014_cast_fp16, y = var_15015_to_fp16)[name = tensor("aw_1237_cast_fp16")]; + tensor var_15018_equation_0 = const()[name = tensor("op_15018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15018_cast_fp16 = einsum(equation = var_15018_equation_0, values = (var_14860_cast_fp16, var_14777_cast_fp16))[name = tensor("op_15018_cast_fp16")]; + tensor var_15019_to_fp16 = const()[name = tensor("op_15019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1239_cast_fp16 = mul(x = var_15018_cast_fp16, y = var_15019_to_fp16)[name = tensor("aw_1239_cast_fp16")]; + tensor var_15021_cast_fp16 = softmax(axis = var_2624, x = aw_1201_cast_fp16)[name = tensor("op_15021_cast_fp16")]; + tensor var_15022_cast_fp16 = softmax(axis = var_2624, x = aw_1203_cast_fp16)[name = tensor("op_15022_cast_fp16")]; + tensor var_15023_cast_fp16 = softmax(axis = var_2624, x = aw_1205_cast_fp16)[name = tensor("op_15023_cast_fp16")]; + tensor var_15024_cast_fp16 = softmax(axis = var_2624, x = aw_1207_cast_fp16)[name = tensor("op_15024_cast_fp16")]; + tensor var_15025_cast_fp16 = softmax(axis = var_2624, x = aw_1209_cast_fp16)[name = tensor("op_15025_cast_fp16")]; + tensor var_15026_cast_fp16 = softmax(axis = var_2624, x = aw_1211_cast_fp16)[name = tensor("op_15026_cast_fp16")]; + tensor var_15027_cast_fp16 = softmax(axis = var_2624, x = aw_1213_cast_fp16)[name = tensor("op_15027_cast_fp16")]; + tensor var_15028_cast_fp16 = softmax(axis = var_2624, x = aw_1215_cast_fp16)[name = tensor("op_15028_cast_fp16")]; + tensor var_15029_cast_fp16 = softmax(axis = var_2624, x = aw_1217_cast_fp16)[name = tensor("op_15029_cast_fp16")]; + tensor var_15030_cast_fp16 = softmax(axis = var_2624, x = aw_1219_cast_fp16)[name = tensor("op_15030_cast_fp16")]; + tensor var_15031_cast_fp16 = softmax(axis = var_2624, x = aw_1221_cast_fp16)[name = tensor("op_15031_cast_fp16")]; + tensor var_15032_cast_fp16 = softmax(axis = var_2624, x = aw_1223_cast_fp16)[name = tensor("op_15032_cast_fp16")]; + tensor var_15033_cast_fp16 = softmax(axis = var_2624, x = aw_1225_cast_fp16)[name = tensor("op_15033_cast_fp16")]; + tensor var_15034_cast_fp16 = softmax(axis = var_2624, x = aw_1227_cast_fp16)[name = tensor("op_15034_cast_fp16")]; + tensor var_15035_cast_fp16 = softmax(axis = var_2624, x = aw_1229_cast_fp16)[name = tensor("op_15035_cast_fp16")]; + tensor var_15036_cast_fp16 = softmax(axis = var_2624, x = aw_1231_cast_fp16)[name = tensor("op_15036_cast_fp16")]; + tensor var_15037_cast_fp16 = softmax(axis = var_2624, x = aw_1233_cast_fp16)[name = tensor("op_15037_cast_fp16")]; + tensor var_15038_cast_fp16 = softmax(axis = var_2624, x = aw_1235_cast_fp16)[name = tensor("op_15038_cast_fp16")]; + tensor var_15039_cast_fp16 = softmax(axis = var_2624, x = aw_1237_cast_fp16)[name = tensor("op_15039_cast_fp16")]; + tensor var_15040_cast_fp16 = softmax(axis = var_2624, x = aw_1239_cast_fp16)[name = tensor("op_15040_cast_fp16")]; + tensor var_15042_equation_0 = const()[name = tensor("op_15042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15042_cast_fp16 = einsum(equation = var_15042_equation_0, values = (var_14862_cast_fp16, var_15021_cast_fp16))[name = tensor("op_15042_cast_fp16")]; + tensor var_15044_equation_0 = const()[name = tensor("op_15044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15044_cast_fp16 = einsum(equation = var_15044_equation_0, values = (var_14866_cast_fp16, var_15022_cast_fp16))[name = tensor("op_15044_cast_fp16")]; + tensor var_15046_equation_0 = const()[name = tensor("op_15046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15046_cast_fp16 = einsum(equation = var_15046_equation_0, values = (var_14870_cast_fp16, var_15023_cast_fp16))[name = tensor("op_15046_cast_fp16")]; + tensor var_15048_equation_0 = const()[name = tensor("op_15048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15048_cast_fp16 = einsum(equation = var_15048_equation_0, values = (var_14874_cast_fp16, var_15024_cast_fp16))[name = tensor("op_15048_cast_fp16")]; + tensor var_15050_equation_0 = const()[name = tensor("op_15050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15050_cast_fp16 = einsum(equation = var_15050_equation_0, values = (var_14878_cast_fp16, var_15025_cast_fp16))[name = tensor("op_15050_cast_fp16")]; + tensor var_15052_equation_0 = const()[name = tensor("op_15052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15052_cast_fp16 = einsum(equation = var_15052_equation_0, values = (var_14882_cast_fp16, var_15026_cast_fp16))[name = tensor("op_15052_cast_fp16")]; + tensor var_15054_equation_0 = const()[name = tensor("op_15054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15054_cast_fp16 = einsum(equation = var_15054_equation_0, values = (var_14886_cast_fp16, var_15027_cast_fp16))[name = tensor("op_15054_cast_fp16")]; + tensor var_15056_equation_0 = const()[name = tensor("op_15056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15056_cast_fp16 = einsum(equation = var_15056_equation_0, values = (var_14890_cast_fp16, var_15028_cast_fp16))[name = tensor("op_15056_cast_fp16")]; + tensor var_15058_equation_0 = const()[name = tensor("op_15058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15058_cast_fp16 = einsum(equation = var_15058_equation_0, values = (var_14894_cast_fp16, var_15029_cast_fp16))[name = tensor("op_15058_cast_fp16")]; + tensor var_15060_equation_0 = const()[name = tensor("op_15060_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15060_cast_fp16 = einsum(equation = var_15060_equation_0, values = (var_14898_cast_fp16, var_15030_cast_fp16))[name = tensor("op_15060_cast_fp16")]; + tensor var_15062_equation_0 = const()[name = tensor("op_15062_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15062_cast_fp16 = einsum(equation = var_15062_equation_0, values = (var_14902_cast_fp16, var_15031_cast_fp16))[name = tensor("op_15062_cast_fp16")]; + tensor var_15064_equation_0 = const()[name = tensor("op_15064_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15064_cast_fp16 = einsum(equation = var_15064_equation_0, values = (var_14906_cast_fp16, var_15032_cast_fp16))[name = tensor("op_15064_cast_fp16")]; + tensor var_15066_equation_0 = const()[name = tensor("op_15066_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15066_cast_fp16 = einsum(equation = var_15066_equation_0, values = (var_14910_cast_fp16, var_15033_cast_fp16))[name = tensor("op_15066_cast_fp16")]; + tensor var_15068_equation_0 = const()[name = tensor("op_15068_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15068_cast_fp16 = einsum(equation = var_15068_equation_0, values = (var_14914_cast_fp16, var_15034_cast_fp16))[name = tensor("op_15068_cast_fp16")]; + tensor var_15070_equation_0 = const()[name = tensor("op_15070_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15070_cast_fp16 = einsum(equation = var_15070_equation_0, values = (var_14918_cast_fp16, var_15035_cast_fp16))[name = tensor("op_15070_cast_fp16")]; + tensor var_15072_equation_0 = const()[name = tensor("op_15072_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15072_cast_fp16 = einsum(equation = var_15072_equation_0, values = (var_14922_cast_fp16, var_15036_cast_fp16))[name = tensor("op_15072_cast_fp16")]; + tensor var_15074_equation_0 = const()[name = tensor("op_15074_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15074_cast_fp16 = einsum(equation = var_15074_equation_0, values = (var_14926_cast_fp16, var_15037_cast_fp16))[name = tensor("op_15074_cast_fp16")]; + tensor var_15076_equation_0 = const()[name = tensor("op_15076_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15076_cast_fp16 = einsum(equation = var_15076_equation_0, values = (var_14930_cast_fp16, var_15038_cast_fp16))[name = tensor("op_15076_cast_fp16")]; + tensor var_15078_equation_0 = const()[name = tensor("op_15078_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15078_cast_fp16 = einsum(equation = var_15078_equation_0, values = (var_14934_cast_fp16, var_15039_cast_fp16))[name = tensor("op_15078_cast_fp16")]; + tensor var_15080_equation_0 = const()[name = tensor("op_15080_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15080_cast_fp16 = einsum(equation = var_15080_equation_0, values = (var_14938_cast_fp16, var_15040_cast_fp16))[name = tensor("op_15080_cast_fp16")]; + tensor input_253_interleave_0 = const()[name = tensor("input_253_interleave_0"), val = tensor(false)]; + tensor input_253_cast_fp16 = concat(axis = var_2624, interleave = input_253_interleave_0, values = (var_15042_cast_fp16, var_15044_cast_fp16, var_15046_cast_fp16, var_15048_cast_fp16, var_15050_cast_fp16, var_15052_cast_fp16, var_15054_cast_fp16, var_15056_cast_fp16, var_15058_cast_fp16, var_15060_cast_fp16, var_15062_cast_fp16, var_15064_cast_fp16, var_15066_cast_fp16, var_15068_cast_fp16, var_15070_cast_fp16, var_15072_cast_fp16, var_15074_cast_fp16, var_15076_cast_fp16, var_15078_cast_fp16, var_15080_cast_fp16))[name = tensor("input_253_cast_fp16")]; + tensor var_15090_pad_type_0 = const()[name = tensor("op_15090_pad_type_0"), val = tensor("valid")]; + tensor var_15090_strides_0 = const()[name = tensor("op_15090_strides_0"), val = tensor([1, 1])]; + tensor var_15090_pad_0 = const()[name = tensor("op_15090_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15090_dilations_0 = const()[name = tensor("op_15090_dilations_0"), val = tensor([1, 1])]; + tensor var_15090_groups_0 = const()[name = tensor("op_15090_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(443231936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444460800))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444460992)))]; + tensor var_15090_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_15090_dilations_0, groups = var_15090_groups_0, pad = var_15090_pad_0, pad_type = var_15090_pad_type_0, strides = var_15090_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized, x = input_253_cast_fp16)[name = tensor("op_15090_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = var_15090_cast_fp16, y = inputs_103_cast_fp16)[name = tensor("inputs_105_cast_fp16")]; + tensor hidden_states_157_axes_0 = const()[name = tensor("hidden_states_157_axes_0"), val = tensor([1])]; + tensor hidden_states_157_gamma_0_to_fp16 = const()[name = tensor("hidden_states_157_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444463616)))]; + tensor hidden_states_157_beta_0_to_fp16 = const()[name = tensor("hidden_states_157_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444466240)))]; + tensor var_15100_to_fp16 = const()[name = tensor("op_15100_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_157_cast_fp16 = layer_norm(axes = hidden_states_157_axes_0, beta = hidden_states_157_beta_0_to_fp16, epsilon = var_15100_to_fp16, gamma = hidden_states_157_gamma_0_to_fp16, x = inputs_105_cast_fp16)[name = tensor("hidden_states_157_cast_fp16")]; + tensor q_71_pad_type_0 = const()[name = tensor("q_71_pad_type_0"), val = tensor("valid")]; + tensor q_71_strides_0 = const()[name = tensor("q_71_strides_0"), val = tensor([1, 1])]; + tensor q_71_pad_0 = const()[name = tensor("q_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_71_dilations_0 = const()[name = tensor("q_71_dilations_0"), val = tensor([1, 1])]; + tensor q_71_groups_0 = const()[name = tensor("q_71_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444468864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445697728))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_71_cast_fp16 = conv(dilations = q_71_dilations_0, groups = q_71_groups_0, pad = q_71_pad_0, pad_type = q_71_pad_type_0, strides = q_71_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_157_cast_fp16)[name = tensor("q_71_cast_fp16")]; + tensor k_141_pad_type_0 = const()[name = tensor("k_141_pad_type_0"), val = tensor("valid")]; + tensor k_141_strides_0 = const()[name = tensor("k_141_strides_0"), val = tensor([1, 1])]; + tensor k_141_pad_0 = const()[name = tensor("k_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_141_dilations_0 = const()[name = tensor("k_141_dilations_0"), val = tensor([1, 1])]; + tensor k_141_groups_0 = const()[name = tensor("k_141_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445697920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447664064))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_141_cast_fp16 = conv(dilations = k_141_dilations_0, groups = k_141_groups_0, pad = k_141_pad_0, pad_type = k_141_pad_type_0, strides = k_141_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_141_cast_fp16")]; + tensor v_71_pad_type_0 = const()[name = tensor("v_71_pad_type_0"), val = tensor("valid")]; + tensor v_71_strides_0 = const()[name = tensor("v_71_strides_0"), val = tensor([1, 1])]; + tensor v_71_pad_0 = const()[name = tensor("v_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_71_dilations_0 = const()[name = tensor("v_71_dilations_0"), val = tensor([1, 1])]; + tensor v_71_groups_0 = const()[name = tensor("v_71_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447664256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449630400))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_71_cast_fp16 = conv(dilations = v_71_dilations_0, groups = v_71_groups_0, pad = v_71_pad_0, pad_type = v_71_pad_type_0, strides = v_71_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_71_cast_fp16")]; + tensor var_15133_begin_0 = const()[name = tensor("op_15133_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15133_end_0 = const()[name = tensor("op_15133_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15133_end_mask_0 = const()[name = tensor("op_15133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15133_cast_fp16 = slice_by_index(begin = var_15133_begin_0, end = var_15133_end_0, end_mask = var_15133_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15133_cast_fp16")]; + tensor var_15137_begin_0 = const()[name = tensor("op_15137_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15137_end_0 = const()[name = tensor("op_15137_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_15137_end_mask_0 = const()[name = tensor("op_15137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15137_cast_fp16 = slice_by_index(begin = var_15137_begin_0, end = var_15137_end_0, end_mask = var_15137_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15137_cast_fp16")]; + tensor var_15141_begin_0 = const()[name = tensor("op_15141_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15141_end_0 = const()[name = tensor("op_15141_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_15141_end_mask_0 = const()[name = tensor("op_15141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15141_cast_fp16 = slice_by_index(begin = var_15141_begin_0, end = var_15141_end_0, end_mask = var_15141_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15141_cast_fp16")]; + tensor var_15145_begin_0 = const()[name = tensor("op_15145_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15145_end_0 = const()[name = tensor("op_15145_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_15145_end_mask_0 = const()[name = tensor("op_15145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15145_cast_fp16 = slice_by_index(begin = var_15145_begin_0, end = var_15145_end_0, end_mask = var_15145_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15145_cast_fp16")]; + tensor var_15149_begin_0 = const()[name = tensor("op_15149_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15149_end_0 = const()[name = tensor("op_15149_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_15149_end_mask_0 = const()[name = tensor("op_15149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15149_cast_fp16 = slice_by_index(begin = var_15149_begin_0, end = var_15149_end_0, end_mask = var_15149_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15149_cast_fp16")]; + tensor var_15153_begin_0 = const()[name = tensor("op_15153_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15153_end_0 = const()[name = tensor("op_15153_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_15153_end_mask_0 = const()[name = tensor("op_15153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15153_cast_fp16 = slice_by_index(begin = var_15153_begin_0, end = var_15153_end_0, end_mask = var_15153_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15153_cast_fp16")]; + tensor var_15157_begin_0 = const()[name = tensor("op_15157_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15157_end_0 = const()[name = tensor("op_15157_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_15157_end_mask_0 = const()[name = tensor("op_15157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15157_cast_fp16 = slice_by_index(begin = var_15157_begin_0, end = var_15157_end_0, end_mask = var_15157_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15157_cast_fp16")]; + tensor var_15161_begin_0 = const()[name = tensor("op_15161_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15161_end_0 = const()[name = tensor("op_15161_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_15161_end_mask_0 = const()[name = tensor("op_15161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15161_cast_fp16 = slice_by_index(begin = var_15161_begin_0, end = var_15161_end_0, end_mask = var_15161_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15161_cast_fp16")]; + tensor var_15165_begin_0 = const()[name = tensor("op_15165_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15165_end_0 = const()[name = tensor("op_15165_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_15165_end_mask_0 = const()[name = tensor("op_15165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15165_cast_fp16 = slice_by_index(begin = var_15165_begin_0, end = var_15165_end_0, end_mask = var_15165_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15165_cast_fp16")]; + tensor var_15169_begin_0 = const()[name = tensor("op_15169_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15169_end_0 = const()[name = tensor("op_15169_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_15169_end_mask_0 = const()[name = tensor("op_15169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15169_cast_fp16 = slice_by_index(begin = var_15169_begin_0, end = var_15169_end_0, end_mask = var_15169_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15169_cast_fp16")]; + tensor var_15173_begin_0 = const()[name = tensor("op_15173_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15173_end_0 = const()[name = tensor("op_15173_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_15173_end_mask_0 = const()[name = tensor("op_15173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15173_cast_fp16 = slice_by_index(begin = var_15173_begin_0, end = var_15173_end_0, end_mask = var_15173_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15173_cast_fp16")]; + tensor var_15177_begin_0 = const()[name = tensor("op_15177_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15177_end_0 = const()[name = tensor("op_15177_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_15177_end_mask_0 = const()[name = tensor("op_15177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15177_cast_fp16 = slice_by_index(begin = var_15177_begin_0, end = var_15177_end_0, end_mask = var_15177_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15177_cast_fp16")]; + tensor var_15181_begin_0 = const()[name = tensor("op_15181_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15181_end_0 = const()[name = tensor("op_15181_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_15181_end_mask_0 = const()[name = tensor("op_15181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15181_cast_fp16 = slice_by_index(begin = var_15181_begin_0, end = var_15181_end_0, end_mask = var_15181_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15181_cast_fp16")]; + tensor var_15185_begin_0 = const()[name = tensor("op_15185_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15185_end_0 = const()[name = tensor("op_15185_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_15185_end_mask_0 = const()[name = tensor("op_15185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15185_cast_fp16 = slice_by_index(begin = var_15185_begin_0, end = var_15185_end_0, end_mask = var_15185_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15185_cast_fp16")]; + tensor var_15189_begin_0 = const()[name = tensor("op_15189_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15189_end_0 = const()[name = tensor("op_15189_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_15189_end_mask_0 = const()[name = tensor("op_15189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15189_cast_fp16 = slice_by_index(begin = var_15189_begin_0, end = var_15189_end_0, end_mask = var_15189_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15189_cast_fp16")]; + tensor var_15193_begin_0 = const()[name = tensor("op_15193_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15193_end_0 = const()[name = tensor("op_15193_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15193_end_mask_0 = const()[name = tensor("op_15193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15193_cast_fp16 = slice_by_index(begin = var_15193_begin_0, end = var_15193_end_0, end_mask = var_15193_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15193_cast_fp16")]; + tensor var_15197_begin_0 = const()[name = tensor("op_15197_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15197_end_0 = const()[name = tensor("op_15197_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_15197_end_mask_0 = const()[name = tensor("op_15197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15197_cast_fp16 = slice_by_index(begin = var_15197_begin_0, end = var_15197_end_0, end_mask = var_15197_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15197_cast_fp16")]; + tensor var_15201_begin_0 = const()[name = tensor("op_15201_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15201_end_0 = const()[name = tensor("op_15201_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_15201_end_mask_0 = const()[name = tensor("op_15201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15201_cast_fp16 = slice_by_index(begin = var_15201_begin_0, end = var_15201_end_0, end_mask = var_15201_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15201_cast_fp16")]; + tensor var_15205_begin_0 = const()[name = tensor("op_15205_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15205_end_0 = const()[name = tensor("op_15205_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_15205_end_mask_0 = const()[name = tensor("op_15205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15205_cast_fp16 = slice_by_index(begin = var_15205_begin_0, end = var_15205_end_0, end_mask = var_15205_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15205_cast_fp16")]; + tensor var_15209_begin_0 = const()[name = tensor("op_15209_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15209_end_0 = const()[name = tensor("op_15209_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_15209_end_mask_0 = const()[name = tensor("op_15209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15209_cast_fp16 = slice_by_index(begin = var_15209_begin_0, end = var_15209_end_0, end_mask = var_15209_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15209_cast_fp16")]; + tensor k_143_perm_0 = const()[name = tensor("k_143_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_15216_begin_0 = const()[name = tensor("op_15216_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15216_end_0 = const()[name = tensor("op_15216_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_15216_end_mask_0 = const()[name = tensor("op_15216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_143_cast_fp16 = transpose(perm = k_143_perm_0, x = k_141_cast_fp16)[name = tensor("transpose_32")]; + tensor var_15216_cast_fp16 = slice_by_index(begin = var_15216_begin_0, end = var_15216_end_0, end_mask = var_15216_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15216_cast_fp16")]; + tensor var_15220_begin_0 = const()[name = tensor("op_15220_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_15220_end_0 = const()[name = tensor("op_15220_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_15220_end_mask_0 = const()[name = tensor("op_15220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15220_cast_fp16 = slice_by_index(begin = var_15220_begin_0, end = var_15220_end_0, end_mask = var_15220_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15220_cast_fp16")]; + tensor var_15224_begin_0 = const()[name = tensor("op_15224_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_15224_end_0 = const()[name = tensor("op_15224_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_15224_end_mask_0 = const()[name = tensor("op_15224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15224_cast_fp16 = slice_by_index(begin = var_15224_begin_0, end = var_15224_end_0, end_mask = var_15224_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15224_cast_fp16")]; + tensor var_15228_begin_0 = const()[name = tensor("op_15228_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_15228_end_0 = const()[name = tensor("op_15228_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_15228_end_mask_0 = const()[name = tensor("op_15228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15228_cast_fp16 = slice_by_index(begin = var_15228_begin_0, end = var_15228_end_0, end_mask = var_15228_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15228_cast_fp16")]; + tensor var_15232_begin_0 = const()[name = tensor("op_15232_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_15232_end_0 = const()[name = tensor("op_15232_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_15232_end_mask_0 = const()[name = tensor("op_15232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15232_cast_fp16 = slice_by_index(begin = var_15232_begin_0, end = var_15232_end_0, end_mask = var_15232_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15232_cast_fp16")]; + tensor var_15236_begin_0 = const()[name = tensor("op_15236_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_15236_end_0 = const()[name = tensor("op_15236_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_15236_end_mask_0 = const()[name = tensor("op_15236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15236_cast_fp16 = slice_by_index(begin = var_15236_begin_0, end = var_15236_end_0, end_mask = var_15236_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15236_cast_fp16")]; + tensor var_15240_begin_0 = const()[name = tensor("op_15240_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_15240_end_0 = const()[name = tensor("op_15240_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_15240_end_mask_0 = const()[name = tensor("op_15240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15240_cast_fp16 = slice_by_index(begin = var_15240_begin_0, end = var_15240_end_0, end_mask = var_15240_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15240_cast_fp16")]; + tensor var_15244_begin_0 = const()[name = tensor("op_15244_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_15244_end_0 = const()[name = tensor("op_15244_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_15244_end_mask_0 = const()[name = tensor("op_15244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15244_cast_fp16 = slice_by_index(begin = var_15244_begin_0, end = var_15244_end_0, end_mask = var_15244_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15244_cast_fp16")]; + tensor var_15248_begin_0 = const()[name = tensor("op_15248_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15248_end_0 = const()[name = tensor("op_15248_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_15248_end_mask_0 = const()[name = tensor("op_15248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15248_cast_fp16 = slice_by_index(begin = var_15248_begin_0, end = var_15248_end_0, end_mask = var_15248_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15248_cast_fp16")]; + tensor var_15252_begin_0 = const()[name = tensor("op_15252_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_15252_end_0 = const()[name = tensor("op_15252_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_15252_end_mask_0 = const()[name = tensor("op_15252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15252_cast_fp16 = slice_by_index(begin = var_15252_begin_0, end = var_15252_end_0, end_mask = var_15252_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15252_cast_fp16")]; + tensor var_15256_begin_0 = const()[name = tensor("op_15256_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_15256_end_0 = const()[name = tensor("op_15256_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_15256_end_mask_0 = const()[name = tensor("op_15256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15256_cast_fp16 = slice_by_index(begin = var_15256_begin_0, end = var_15256_end_0, end_mask = var_15256_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15256_cast_fp16")]; + tensor var_15260_begin_0 = const()[name = tensor("op_15260_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_15260_end_0 = const()[name = tensor("op_15260_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_15260_end_mask_0 = const()[name = tensor("op_15260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15260_cast_fp16 = slice_by_index(begin = var_15260_begin_0, end = var_15260_end_0, end_mask = var_15260_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15260_cast_fp16")]; + tensor var_15264_begin_0 = const()[name = tensor("op_15264_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_15264_end_0 = const()[name = tensor("op_15264_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_15264_end_mask_0 = const()[name = tensor("op_15264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15264_cast_fp16 = slice_by_index(begin = var_15264_begin_0, end = var_15264_end_0, end_mask = var_15264_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15264_cast_fp16")]; + tensor var_15268_begin_0 = const()[name = tensor("op_15268_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_15268_end_0 = const()[name = tensor("op_15268_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_15268_end_mask_0 = const()[name = tensor("op_15268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15268_cast_fp16 = slice_by_index(begin = var_15268_begin_0, end = var_15268_end_0, end_mask = var_15268_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15268_cast_fp16")]; + tensor var_15272_begin_0 = const()[name = tensor("op_15272_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_15272_end_0 = const()[name = tensor("op_15272_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_15272_end_mask_0 = const()[name = tensor("op_15272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15272_cast_fp16 = slice_by_index(begin = var_15272_begin_0, end = var_15272_end_0, end_mask = var_15272_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15272_cast_fp16")]; + tensor var_15276_begin_0 = const()[name = tensor("op_15276_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_15276_end_0 = const()[name = tensor("op_15276_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_15276_end_mask_0 = const()[name = tensor("op_15276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15276_cast_fp16 = slice_by_index(begin = var_15276_begin_0, end = var_15276_end_0, end_mask = var_15276_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15276_cast_fp16")]; + tensor var_15280_begin_0 = const()[name = tensor("op_15280_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15280_end_0 = const()[name = tensor("op_15280_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_15280_end_mask_0 = const()[name = tensor("op_15280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15280_cast_fp16 = slice_by_index(begin = var_15280_begin_0, end = var_15280_end_0, end_mask = var_15280_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15280_cast_fp16")]; + tensor var_15284_begin_0 = const()[name = tensor("op_15284_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_15284_end_0 = const()[name = tensor("op_15284_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_15284_end_mask_0 = const()[name = tensor("op_15284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15284_cast_fp16 = slice_by_index(begin = var_15284_begin_0, end = var_15284_end_0, end_mask = var_15284_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15284_cast_fp16")]; + tensor var_15288_begin_0 = const()[name = tensor("op_15288_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_15288_end_0 = const()[name = tensor("op_15288_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_15288_end_mask_0 = const()[name = tensor("op_15288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15288_cast_fp16 = slice_by_index(begin = var_15288_begin_0, end = var_15288_end_0, end_mask = var_15288_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15288_cast_fp16")]; + tensor var_15292_begin_0 = const()[name = tensor("op_15292_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_15292_end_0 = const()[name = tensor("op_15292_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_15292_end_mask_0 = const()[name = tensor("op_15292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15292_cast_fp16 = slice_by_index(begin = var_15292_begin_0, end = var_15292_end_0, end_mask = var_15292_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15292_cast_fp16")]; + tensor var_15294_begin_0 = const()[name = tensor("op_15294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15294_end_0 = const()[name = tensor("op_15294_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_15294_end_mask_0 = const()[name = tensor("op_15294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15294_cast_fp16 = slice_by_index(begin = var_15294_begin_0, end = var_15294_end_0, end_mask = var_15294_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15294_cast_fp16")]; + tensor var_15298_begin_0 = const()[name = tensor("op_15298_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15298_end_0 = const()[name = tensor("op_15298_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_15298_end_mask_0 = const()[name = tensor("op_15298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15298_cast_fp16 = slice_by_index(begin = var_15298_begin_0, end = var_15298_end_0, end_mask = var_15298_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15298_cast_fp16")]; + tensor var_15302_begin_0 = const()[name = tensor("op_15302_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15302_end_0 = const()[name = tensor("op_15302_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_15302_end_mask_0 = const()[name = tensor("op_15302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15302_cast_fp16 = slice_by_index(begin = var_15302_begin_0, end = var_15302_end_0, end_mask = var_15302_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15302_cast_fp16")]; + tensor var_15306_begin_0 = const()[name = tensor("op_15306_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15306_end_0 = const()[name = tensor("op_15306_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_15306_end_mask_0 = const()[name = tensor("op_15306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15306_cast_fp16 = slice_by_index(begin = var_15306_begin_0, end = var_15306_end_0, end_mask = var_15306_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15306_cast_fp16")]; + tensor var_15310_begin_0 = const()[name = tensor("op_15310_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15310_end_0 = const()[name = tensor("op_15310_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_15310_end_mask_0 = const()[name = tensor("op_15310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15310_cast_fp16 = slice_by_index(begin = var_15310_begin_0, end = var_15310_end_0, end_mask = var_15310_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15310_cast_fp16")]; + tensor var_15314_begin_0 = const()[name = tensor("op_15314_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15314_end_0 = const()[name = tensor("op_15314_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_15314_end_mask_0 = const()[name = tensor("op_15314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15314_cast_fp16 = slice_by_index(begin = var_15314_begin_0, end = var_15314_end_0, end_mask = var_15314_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15314_cast_fp16")]; + tensor var_15318_begin_0 = const()[name = tensor("op_15318_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15318_end_0 = const()[name = tensor("op_15318_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_15318_end_mask_0 = const()[name = tensor("op_15318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15318_cast_fp16 = slice_by_index(begin = var_15318_begin_0, end = var_15318_end_0, end_mask = var_15318_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15318_cast_fp16")]; + tensor var_15322_begin_0 = const()[name = tensor("op_15322_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15322_end_0 = const()[name = tensor("op_15322_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_15322_end_mask_0 = const()[name = tensor("op_15322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15322_cast_fp16 = slice_by_index(begin = var_15322_begin_0, end = var_15322_end_0, end_mask = var_15322_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15322_cast_fp16")]; + tensor var_15326_begin_0 = const()[name = tensor("op_15326_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15326_end_0 = const()[name = tensor("op_15326_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_15326_end_mask_0 = const()[name = tensor("op_15326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15326_cast_fp16 = slice_by_index(begin = var_15326_begin_0, end = var_15326_end_0, end_mask = var_15326_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15326_cast_fp16")]; + tensor var_15330_begin_0 = const()[name = tensor("op_15330_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15330_end_0 = const()[name = tensor("op_15330_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_15330_end_mask_0 = const()[name = tensor("op_15330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15330_cast_fp16 = slice_by_index(begin = var_15330_begin_0, end = var_15330_end_0, end_mask = var_15330_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15330_cast_fp16")]; + tensor var_15334_begin_0 = const()[name = tensor("op_15334_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15334_end_0 = const()[name = tensor("op_15334_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_15334_end_mask_0 = const()[name = tensor("op_15334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15334_cast_fp16 = slice_by_index(begin = var_15334_begin_0, end = var_15334_end_0, end_mask = var_15334_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15334_cast_fp16")]; + tensor var_15338_begin_0 = const()[name = tensor("op_15338_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15338_end_0 = const()[name = tensor("op_15338_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_15338_end_mask_0 = const()[name = tensor("op_15338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15338_cast_fp16 = slice_by_index(begin = var_15338_begin_0, end = var_15338_end_0, end_mask = var_15338_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15338_cast_fp16")]; + tensor var_15342_begin_0 = const()[name = tensor("op_15342_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15342_end_0 = const()[name = tensor("op_15342_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_15342_end_mask_0 = const()[name = tensor("op_15342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15342_cast_fp16 = slice_by_index(begin = var_15342_begin_0, end = var_15342_end_0, end_mask = var_15342_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15342_cast_fp16")]; + tensor var_15346_begin_0 = const()[name = tensor("op_15346_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15346_end_0 = const()[name = tensor("op_15346_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_15346_end_mask_0 = const()[name = tensor("op_15346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15346_cast_fp16 = slice_by_index(begin = var_15346_begin_0, end = var_15346_end_0, end_mask = var_15346_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15346_cast_fp16")]; + tensor var_15350_begin_0 = const()[name = tensor("op_15350_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15350_end_0 = const()[name = tensor("op_15350_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_15350_end_mask_0 = const()[name = tensor("op_15350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15350_cast_fp16 = slice_by_index(begin = var_15350_begin_0, end = var_15350_end_0, end_mask = var_15350_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15350_cast_fp16")]; + tensor var_15354_begin_0 = const()[name = tensor("op_15354_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15354_end_0 = const()[name = tensor("op_15354_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_15354_end_mask_0 = const()[name = tensor("op_15354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15354_cast_fp16 = slice_by_index(begin = var_15354_begin_0, end = var_15354_end_0, end_mask = var_15354_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15354_cast_fp16")]; + tensor var_15358_begin_0 = const()[name = tensor("op_15358_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15358_end_0 = const()[name = tensor("op_15358_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_15358_end_mask_0 = const()[name = tensor("op_15358_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15358_cast_fp16 = slice_by_index(begin = var_15358_begin_0, end = var_15358_end_0, end_mask = var_15358_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15358_cast_fp16")]; + tensor var_15362_begin_0 = const()[name = tensor("op_15362_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15362_end_0 = const()[name = tensor("op_15362_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_15362_end_mask_0 = const()[name = tensor("op_15362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15362_cast_fp16 = slice_by_index(begin = var_15362_begin_0, end = var_15362_end_0, end_mask = var_15362_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15362_cast_fp16")]; + tensor var_15366_begin_0 = const()[name = tensor("op_15366_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15366_end_0 = const()[name = tensor("op_15366_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_15366_end_mask_0 = const()[name = tensor("op_15366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15366_cast_fp16 = slice_by_index(begin = var_15366_begin_0, end = var_15366_end_0, end_mask = var_15366_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15366_cast_fp16")]; + tensor var_15370_begin_0 = const()[name = tensor("op_15370_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15370_end_0 = const()[name = tensor("op_15370_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_15370_end_mask_0 = const()[name = tensor("op_15370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15370_cast_fp16 = slice_by_index(begin = var_15370_begin_0, end = var_15370_end_0, end_mask = var_15370_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15370_cast_fp16")]; + tensor var_15374_equation_0 = const()[name = tensor("op_15374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15374_cast_fp16 = einsum(equation = var_15374_equation_0, values = (var_15216_cast_fp16, var_15133_cast_fp16))[name = tensor("op_15374_cast_fp16")]; + tensor var_15375_to_fp16 = const()[name = tensor("op_15375_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1241_cast_fp16 = mul(x = var_15374_cast_fp16, y = var_15375_to_fp16)[name = tensor("aw_1241_cast_fp16")]; + tensor var_15378_equation_0 = const()[name = tensor("op_15378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15378_cast_fp16 = einsum(equation = var_15378_equation_0, values = (var_15220_cast_fp16, var_15137_cast_fp16))[name = tensor("op_15378_cast_fp16")]; + tensor var_15379_to_fp16 = const()[name = tensor("op_15379_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1243_cast_fp16 = mul(x = var_15378_cast_fp16, y = var_15379_to_fp16)[name = tensor("aw_1243_cast_fp16")]; + tensor var_15382_equation_0 = const()[name = tensor("op_15382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15382_cast_fp16 = einsum(equation = var_15382_equation_0, values = (var_15224_cast_fp16, var_15141_cast_fp16))[name = tensor("op_15382_cast_fp16")]; + tensor var_15383_to_fp16 = const()[name = tensor("op_15383_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1245_cast_fp16 = mul(x = var_15382_cast_fp16, y = var_15383_to_fp16)[name = tensor("aw_1245_cast_fp16")]; + tensor var_15386_equation_0 = const()[name = tensor("op_15386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15386_cast_fp16 = einsum(equation = var_15386_equation_0, values = (var_15228_cast_fp16, var_15145_cast_fp16))[name = tensor("op_15386_cast_fp16")]; + tensor var_15387_to_fp16 = const()[name = tensor("op_15387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1247_cast_fp16 = mul(x = var_15386_cast_fp16, y = var_15387_to_fp16)[name = tensor("aw_1247_cast_fp16")]; + tensor var_15390_equation_0 = const()[name = tensor("op_15390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15390_cast_fp16 = einsum(equation = var_15390_equation_0, values = (var_15232_cast_fp16, var_15149_cast_fp16))[name = tensor("op_15390_cast_fp16")]; + tensor var_15391_to_fp16 = const()[name = tensor("op_15391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1249_cast_fp16 = mul(x = var_15390_cast_fp16, y = var_15391_to_fp16)[name = tensor("aw_1249_cast_fp16")]; + tensor var_15394_equation_0 = const()[name = tensor("op_15394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15394_cast_fp16 = einsum(equation = var_15394_equation_0, values = (var_15236_cast_fp16, var_15153_cast_fp16))[name = tensor("op_15394_cast_fp16")]; + tensor var_15395_to_fp16 = const()[name = tensor("op_15395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1251_cast_fp16 = mul(x = var_15394_cast_fp16, y = var_15395_to_fp16)[name = tensor("aw_1251_cast_fp16")]; + tensor var_15398_equation_0 = const()[name = tensor("op_15398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15398_cast_fp16 = einsum(equation = var_15398_equation_0, values = (var_15240_cast_fp16, var_15157_cast_fp16))[name = tensor("op_15398_cast_fp16")]; + tensor var_15399_to_fp16 = const()[name = tensor("op_15399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1253_cast_fp16 = mul(x = var_15398_cast_fp16, y = var_15399_to_fp16)[name = tensor("aw_1253_cast_fp16")]; + tensor var_15402_equation_0 = const()[name = tensor("op_15402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15402_cast_fp16 = einsum(equation = var_15402_equation_0, values = (var_15244_cast_fp16, var_15161_cast_fp16))[name = tensor("op_15402_cast_fp16")]; + tensor var_15403_to_fp16 = const()[name = tensor("op_15403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1255_cast_fp16 = mul(x = var_15402_cast_fp16, y = var_15403_to_fp16)[name = tensor("aw_1255_cast_fp16")]; + tensor var_15406_equation_0 = const()[name = tensor("op_15406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15406_cast_fp16 = einsum(equation = var_15406_equation_0, values = (var_15248_cast_fp16, var_15165_cast_fp16))[name = tensor("op_15406_cast_fp16")]; + tensor var_15407_to_fp16 = const()[name = tensor("op_15407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1257_cast_fp16 = mul(x = var_15406_cast_fp16, y = var_15407_to_fp16)[name = tensor("aw_1257_cast_fp16")]; + tensor var_15410_equation_0 = const()[name = tensor("op_15410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15410_cast_fp16 = einsum(equation = var_15410_equation_0, values = (var_15252_cast_fp16, var_15169_cast_fp16))[name = tensor("op_15410_cast_fp16")]; + tensor var_15411_to_fp16 = const()[name = tensor("op_15411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1259_cast_fp16 = mul(x = var_15410_cast_fp16, y = var_15411_to_fp16)[name = tensor("aw_1259_cast_fp16")]; + tensor var_15414_equation_0 = const()[name = tensor("op_15414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15414_cast_fp16 = einsum(equation = var_15414_equation_0, values = (var_15256_cast_fp16, var_15173_cast_fp16))[name = tensor("op_15414_cast_fp16")]; + tensor var_15415_to_fp16 = const()[name = tensor("op_15415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1261_cast_fp16 = mul(x = var_15414_cast_fp16, y = var_15415_to_fp16)[name = tensor("aw_1261_cast_fp16")]; + tensor var_15418_equation_0 = const()[name = tensor("op_15418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15418_cast_fp16 = einsum(equation = var_15418_equation_0, values = (var_15260_cast_fp16, var_15177_cast_fp16))[name = tensor("op_15418_cast_fp16")]; + tensor var_15419_to_fp16 = const()[name = tensor("op_15419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1263_cast_fp16 = mul(x = var_15418_cast_fp16, y = var_15419_to_fp16)[name = tensor("aw_1263_cast_fp16")]; + tensor var_15422_equation_0 = const()[name = tensor("op_15422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15422_cast_fp16 = einsum(equation = var_15422_equation_0, values = (var_15264_cast_fp16, var_15181_cast_fp16))[name = tensor("op_15422_cast_fp16")]; + tensor var_15423_to_fp16 = const()[name = tensor("op_15423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1265_cast_fp16 = mul(x = var_15422_cast_fp16, y = var_15423_to_fp16)[name = tensor("aw_1265_cast_fp16")]; + tensor var_15426_equation_0 = const()[name = tensor("op_15426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15426_cast_fp16 = einsum(equation = var_15426_equation_0, values = (var_15268_cast_fp16, var_15185_cast_fp16))[name = tensor("op_15426_cast_fp16")]; + tensor var_15427_to_fp16 = const()[name = tensor("op_15427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1267_cast_fp16 = mul(x = var_15426_cast_fp16, y = var_15427_to_fp16)[name = tensor("aw_1267_cast_fp16")]; + tensor var_15430_equation_0 = const()[name = tensor("op_15430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15430_cast_fp16 = einsum(equation = var_15430_equation_0, values = (var_15272_cast_fp16, var_15189_cast_fp16))[name = tensor("op_15430_cast_fp16")]; + tensor var_15431_to_fp16 = const()[name = tensor("op_15431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1269_cast_fp16 = mul(x = var_15430_cast_fp16, y = var_15431_to_fp16)[name = tensor("aw_1269_cast_fp16")]; + tensor var_15434_equation_0 = const()[name = tensor("op_15434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15434_cast_fp16 = einsum(equation = var_15434_equation_0, values = (var_15276_cast_fp16, var_15193_cast_fp16))[name = tensor("op_15434_cast_fp16")]; + tensor var_15435_to_fp16 = const()[name = tensor("op_15435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1271_cast_fp16 = mul(x = var_15434_cast_fp16, y = var_15435_to_fp16)[name = tensor("aw_1271_cast_fp16")]; + tensor var_15438_equation_0 = const()[name = tensor("op_15438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15438_cast_fp16 = einsum(equation = var_15438_equation_0, values = (var_15280_cast_fp16, var_15197_cast_fp16))[name = tensor("op_15438_cast_fp16")]; + tensor var_15439_to_fp16 = const()[name = tensor("op_15439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1273_cast_fp16 = mul(x = var_15438_cast_fp16, y = var_15439_to_fp16)[name = tensor("aw_1273_cast_fp16")]; + tensor var_15442_equation_0 = const()[name = tensor("op_15442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15442_cast_fp16 = einsum(equation = var_15442_equation_0, values = (var_15284_cast_fp16, var_15201_cast_fp16))[name = tensor("op_15442_cast_fp16")]; + tensor var_15443_to_fp16 = const()[name = tensor("op_15443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1275_cast_fp16 = mul(x = var_15442_cast_fp16, y = var_15443_to_fp16)[name = tensor("aw_1275_cast_fp16")]; + tensor var_15446_equation_0 = const()[name = tensor("op_15446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15446_cast_fp16 = einsum(equation = var_15446_equation_0, values = (var_15288_cast_fp16, var_15205_cast_fp16))[name = tensor("op_15446_cast_fp16")]; + tensor var_15447_to_fp16 = const()[name = tensor("op_15447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1277_cast_fp16 = mul(x = var_15446_cast_fp16, y = var_15447_to_fp16)[name = tensor("aw_1277_cast_fp16")]; + tensor var_15450_equation_0 = const()[name = tensor("op_15450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15450_cast_fp16 = einsum(equation = var_15450_equation_0, values = (var_15292_cast_fp16, var_15209_cast_fp16))[name = tensor("op_15450_cast_fp16")]; + tensor var_15451_to_fp16 = const()[name = tensor("op_15451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1279_cast_fp16 = mul(x = var_15450_cast_fp16, y = var_15451_to_fp16)[name = tensor("aw_1279_cast_fp16")]; + tensor var_15453_cast_fp16 = softmax(axis = var_2624, x = aw_1241_cast_fp16)[name = tensor("op_15453_cast_fp16")]; + tensor var_15454_cast_fp16 = softmax(axis = var_2624, x = aw_1243_cast_fp16)[name = tensor("op_15454_cast_fp16")]; + tensor var_15455_cast_fp16 = softmax(axis = var_2624, x = aw_1245_cast_fp16)[name = tensor("op_15455_cast_fp16")]; + tensor var_15456_cast_fp16 = softmax(axis = var_2624, x = aw_1247_cast_fp16)[name = tensor("op_15456_cast_fp16")]; + tensor var_15457_cast_fp16 = softmax(axis = var_2624, x = aw_1249_cast_fp16)[name = tensor("op_15457_cast_fp16")]; + tensor var_15458_cast_fp16 = softmax(axis = var_2624, x = aw_1251_cast_fp16)[name = tensor("op_15458_cast_fp16")]; + tensor var_15459_cast_fp16 = softmax(axis = var_2624, x = aw_1253_cast_fp16)[name = tensor("op_15459_cast_fp16")]; + tensor var_15460_cast_fp16 = softmax(axis = var_2624, x = aw_1255_cast_fp16)[name = tensor("op_15460_cast_fp16")]; + tensor var_15461_cast_fp16 = softmax(axis = var_2624, x = aw_1257_cast_fp16)[name = tensor("op_15461_cast_fp16")]; + tensor var_15462_cast_fp16 = softmax(axis = var_2624, x = aw_1259_cast_fp16)[name = tensor("op_15462_cast_fp16")]; + tensor var_15463_cast_fp16 = softmax(axis = var_2624, x = aw_1261_cast_fp16)[name = tensor("op_15463_cast_fp16")]; + tensor var_15464_cast_fp16 = softmax(axis = var_2624, x = aw_1263_cast_fp16)[name = tensor("op_15464_cast_fp16")]; + tensor var_15465_cast_fp16 = softmax(axis = var_2624, x = aw_1265_cast_fp16)[name = tensor("op_15465_cast_fp16")]; + tensor var_15466_cast_fp16 = softmax(axis = var_2624, x = aw_1267_cast_fp16)[name = tensor("op_15466_cast_fp16")]; + tensor var_15467_cast_fp16 = softmax(axis = var_2624, x = aw_1269_cast_fp16)[name = tensor("op_15467_cast_fp16")]; + tensor var_15468_cast_fp16 = softmax(axis = var_2624, x = aw_1271_cast_fp16)[name = tensor("op_15468_cast_fp16")]; + tensor var_15469_cast_fp16 = softmax(axis = var_2624, x = aw_1273_cast_fp16)[name = tensor("op_15469_cast_fp16")]; + tensor var_15470_cast_fp16 = softmax(axis = var_2624, x = aw_1275_cast_fp16)[name = tensor("op_15470_cast_fp16")]; + tensor var_15471_cast_fp16 = softmax(axis = var_2624, x = aw_1277_cast_fp16)[name = tensor("op_15471_cast_fp16")]; + tensor var_15472_cast_fp16 = softmax(axis = var_2624, x = aw_1279_cast_fp16)[name = tensor("op_15472_cast_fp16")]; + tensor var_15474_equation_0 = const()[name = tensor("op_15474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15474_cast_fp16 = einsum(equation = var_15474_equation_0, values = (var_15294_cast_fp16, var_15453_cast_fp16))[name = tensor("op_15474_cast_fp16")]; + tensor var_15476_equation_0 = const()[name = tensor("op_15476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15476_cast_fp16 = einsum(equation = var_15476_equation_0, values = (var_15298_cast_fp16, var_15454_cast_fp16))[name = tensor("op_15476_cast_fp16")]; + tensor var_15478_equation_0 = const()[name = tensor("op_15478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15478_cast_fp16 = einsum(equation = var_15478_equation_0, values = (var_15302_cast_fp16, var_15455_cast_fp16))[name = tensor("op_15478_cast_fp16")]; + tensor var_15480_equation_0 = const()[name = tensor("op_15480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15480_cast_fp16 = einsum(equation = var_15480_equation_0, values = (var_15306_cast_fp16, var_15456_cast_fp16))[name = tensor("op_15480_cast_fp16")]; + tensor var_15482_equation_0 = const()[name = tensor("op_15482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15482_cast_fp16 = einsum(equation = var_15482_equation_0, values = (var_15310_cast_fp16, var_15457_cast_fp16))[name = tensor("op_15482_cast_fp16")]; + tensor var_15484_equation_0 = const()[name = tensor("op_15484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15484_cast_fp16 = einsum(equation = var_15484_equation_0, values = (var_15314_cast_fp16, var_15458_cast_fp16))[name = tensor("op_15484_cast_fp16")]; + tensor var_15486_equation_0 = const()[name = tensor("op_15486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15486_cast_fp16 = einsum(equation = var_15486_equation_0, values = (var_15318_cast_fp16, var_15459_cast_fp16))[name = tensor("op_15486_cast_fp16")]; + tensor var_15488_equation_0 = const()[name = tensor("op_15488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15488_cast_fp16 = einsum(equation = var_15488_equation_0, values = (var_15322_cast_fp16, var_15460_cast_fp16))[name = tensor("op_15488_cast_fp16")]; + tensor var_15490_equation_0 = const()[name = tensor("op_15490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15490_cast_fp16 = einsum(equation = var_15490_equation_0, values = (var_15326_cast_fp16, var_15461_cast_fp16))[name = tensor("op_15490_cast_fp16")]; + tensor var_15492_equation_0 = const()[name = tensor("op_15492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15492_cast_fp16 = einsum(equation = var_15492_equation_0, values = (var_15330_cast_fp16, var_15462_cast_fp16))[name = tensor("op_15492_cast_fp16")]; + tensor var_15494_equation_0 = const()[name = tensor("op_15494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15494_cast_fp16 = einsum(equation = var_15494_equation_0, values = (var_15334_cast_fp16, var_15463_cast_fp16))[name = tensor("op_15494_cast_fp16")]; + tensor var_15496_equation_0 = const()[name = tensor("op_15496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15496_cast_fp16 = einsum(equation = var_15496_equation_0, values = (var_15338_cast_fp16, var_15464_cast_fp16))[name = tensor("op_15496_cast_fp16")]; + tensor var_15498_equation_0 = const()[name = tensor("op_15498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15498_cast_fp16 = einsum(equation = var_15498_equation_0, values = (var_15342_cast_fp16, var_15465_cast_fp16))[name = tensor("op_15498_cast_fp16")]; + tensor var_15500_equation_0 = const()[name = tensor("op_15500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15500_cast_fp16 = einsum(equation = var_15500_equation_0, values = (var_15346_cast_fp16, var_15466_cast_fp16))[name = tensor("op_15500_cast_fp16")]; + tensor var_15502_equation_0 = const()[name = tensor("op_15502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15502_cast_fp16 = einsum(equation = var_15502_equation_0, values = (var_15350_cast_fp16, var_15467_cast_fp16))[name = tensor("op_15502_cast_fp16")]; + tensor var_15504_equation_0 = const()[name = tensor("op_15504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15504_cast_fp16 = einsum(equation = var_15504_equation_0, values = (var_15354_cast_fp16, var_15468_cast_fp16))[name = tensor("op_15504_cast_fp16")]; + tensor var_15506_equation_0 = const()[name = tensor("op_15506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15506_cast_fp16 = einsum(equation = var_15506_equation_0, values = (var_15358_cast_fp16, var_15469_cast_fp16))[name = tensor("op_15506_cast_fp16")]; + tensor var_15508_equation_0 = const()[name = tensor("op_15508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15508_cast_fp16 = einsum(equation = var_15508_equation_0, values = (var_15362_cast_fp16, var_15470_cast_fp16))[name = tensor("op_15508_cast_fp16")]; + tensor var_15510_equation_0 = const()[name = tensor("op_15510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15510_cast_fp16 = einsum(equation = var_15510_equation_0, values = (var_15366_cast_fp16, var_15471_cast_fp16))[name = tensor("op_15510_cast_fp16")]; + tensor var_15512_equation_0 = const()[name = tensor("op_15512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15512_cast_fp16 = einsum(equation = var_15512_equation_0, values = (var_15370_cast_fp16, var_15472_cast_fp16))[name = tensor("op_15512_cast_fp16")]; + tensor input_255_interleave_0 = const()[name = tensor("input_255_interleave_0"), val = tensor(false)]; + tensor input_255_cast_fp16 = concat(axis = var_2624, interleave = input_255_interleave_0, values = (var_15474_cast_fp16, var_15476_cast_fp16, var_15478_cast_fp16, var_15480_cast_fp16, var_15482_cast_fp16, var_15484_cast_fp16, var_15486_cast_fp16, var_15488_cast_fp16, var_15490_cast_fp16, var_15492_cast_fp16, var_15494_cast_fp16, var_15496_cast_fp16, var_15498_cast_fp16, var_15500_cast_fp16, var_15502_cast_fp16, var_15504_cast_fp16, var_15506_cast_fp16, var_15508_cast_fp16, var_15510_cast_fp16, var_15512_cast_fp16))[name = tensor("input_255_cast_fp16")]; + tensor var_15522_pad_type_0 = const()[name = tensor("op_15522_pad_type_0"), val = tensor("valid")]; + tensor var_15522_strides_0 = const()[name = tensor("op_15522_strides_0"), val = tensor([1, 1])]; + tensor var_15522_pad_0 = const()[name = tensor("op_15522_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15522_dilations_0 = const()[name = tensor("op_15522_dilations_0"), val = tensor([1, 1])]; + tensor var_15522_groups_0 = const()[name = tensor("op_15522_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449630592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450859456))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450859648)))]; + tensor var_15522_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_15522_dilations_0, groups = var_15522_groups_0, pad = var_15522_pad_0, pad_type = var_15522_pad_type_0, strides = var_15522_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized, x = input_255_cast_fp16)[name = tensor("op_15522_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = var_15522_cast_fp16, y = inputs_105_cast_fp16)[name = tensor("inputs_107_cast_fp16")]; + tensor input_257_axes_0 = const()[name = tensor("input_257_axes_0"), val = tensor([1])]; + tensor input_257_gamma_0_to_fp16 = const()[name = tensor("input_257_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450862272)))]; + tensor input_257_beta_0_to_fp16 = const()[name = tensor("input_257_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450864896)))]; + tensor var_15532_to_fp16 = const()[name = tensor("op_15532_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_257_cast_fp16 = layer_norm(axes = input_257_axes_0, beta = input_257_beta_0_to_fp16, epsilon = var_15532_to_fp16, gamma = input_257_gamma_0_to_fp16, x = inputs_107_cast_fp16)[name = tensor("input_257_cast_fp16")]; + tensor var_15552_pad_type_0 = const()[name = tensor("op_15552_pad_type_0"), val = tensor("valid")]; + tensor var_15552_strides_0 = const()[name = tensor("op_15552_strides_0"), val = tensor([1, 1])]; + tensor var_15552_pad_0 = const()[name = tensor("op_15552_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15552_dilations_0 = const()[name = tensor("op_15552_dilations_0"), val = tensor([1, 1])]; + tensor var_15552_groups_0 = const()[name = tensor("op_15552_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450867520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460697984))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460698176)))]; + tensor var_15552_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_15552_dilations_0, groups = var_15552_groups_0, pad = var_15552_pad_0, pad_type = var_15552_pad_type_0, strides = var_15552_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized, x = input_257_cast_fp16)[name = tensor("op_15552_cast_fp16")]; + tensor var_15553_split_sizes_0 = const()[name = tensor("op_15553_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_15553_axis_0 = const()[name = tensor("op_15553_axis_0"), val = tensor(1)]; + tensor var_15553_cast_fp16_0, tensor var_15553_cast_fp16_1 = split(axis = var_15553_axis_0, split_sizes = var_15553_split_sizes_0, x = var_15552_cast_fp16)[name = tensor("op_15553_cast_fp16")]; + tensor var_15555_mode_0 = const()[name = tensor("op_15555_mode_0"), val = tensor("EXACT")]; + tensor var_15555_cast_fp16 = gelu(mode = var_15555_mode_0, x = var_15553_cast_fp16_1)[name = tensor("op_15555_cast_fp16")]; + tensor input_259_cast_fp16 = mul(x = var_15553_cast_fp16_0, y = var_15555_cast_fp16)[name = tensor("input_259_cast_fp16")]; + tensor var_15563_pad_type_0 = const()[name = tensor("op_15563_pad_type_0"), val = tensor("valid")]; + tensor var_15563_strides_0 = const()[name = tensor("op_15563_strides_0"), val = tensor([1, 1])]; + tensor var_15563_pad_0 = const()[name = tensor("op_15563_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15563_dilations_0 = const()[name = tensor("op_15563_dilations_0"), val = tensor([1, 1])]; + tensor var_15563_groups_0 = const()[name = tensor("op_15563_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460718720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465633984))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465634176)))]; + tensor var_15563_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_15563_dilations_0, groups = var_15563_groups_0, pad = var_15563_pad_0, pad_type = var_15563_pad_type_0, strides = var_15563_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized, x = input_259_cast_fp16)[name = tensor("op_15563_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = var_15563_cast_fp16, y = inputs_107_cast_fp16)[name = tensor("inputs_109_cast_fp16")]; + tensor hidden_states_161_axes_0 = const()[name = tensor("hidden_states_161_axes_0"), val = tensor([1])]; + tensor hidden_states_161_gamma_0_to_fp16 = const()[name = tensor("hidden_states_161_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465636800)))]; + tensor hidden_states_161_beta_0_to_fp16 = const()[name = tensor("hidden_states_161_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465639424)))]; + tensor var_15579_to_fp16 = const()[name = tensor("op_15579_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_161_cast_fp16 = layer_norm(axes = hidden_states_161_axes_0, beta = hidden_states_161_beta_0_to_fp16, epsilon = var_15579_to_fp16, gamma = hidden_states_161_gamma_0_to_fp16, x = inputs_109_cast_fp16)[name = tensor("hidden_states_161_cast_fp16")]; + tensor q_73_pad_type_0 = const()[name = tensor("q_73_pad_type_0"), val = tensor("valid")]; + tensor q_73_strides_0 = const()[name = tensor("q_73_strides_0"), val = tensor([1, 1])]; + tensor q_73_pad_0 = const()[name = tensor("q_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_73_dilations_0 = const()[name = tensor("q_73_dilations_0"), val = tensor([1, 1])]; + tensor q_73_groups_0 = const()[name = tensor("q_73_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465642048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(466870912))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_73_cast_fp16 = conv(dilations = q_73_dilations_0, groups = q_73_groups_0, pad = q_73_pad_0, pad_type = q_73_pad_type_0, strides = q_73_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_161_cast_fp16)[name = tensor("q_73_cast_fp16")]; + tensor k_145_pad_type_0 = const()[name = tensor("k_145_pad_type_0"), val = tensor("valid")]; + tensor k_145_strides_0 = const()[name = tensor("k_145_strides_0"), val = tensor([1, 1])]; + tensor k_145_pad_0 = const()[name = tensor("k_145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_145_dilations_0 = const()[name = tensor("k_145_dilations_0"), val = tensor([1, 1])]; + tensor k_145_groups_0 = const()[name = tensor("k_145_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(466871104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468099968))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_145_cast_fp16 = conv(dilations = k_145_dilations_0, groups = k_145_groups_0, pad = k_145_pad_0, pad_type = k_145_pad_type_0, strides = k_145_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_161_cast_fp16)[name = tensor("k_145_cast_fp16")]; + tensor v_73_pad_type_0 = const()[name = tensor("v_73_pad_type_0"), val = tensor("valid")]; + tensor v_73_strides_0 = const()[name = tensor("v_73_strides_0"), val = tensor([1, 1])]; + tensor v_73_pad_0 = const()[name = tensor("v_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_73_dilations_0 = const()[name = tensor("v_73_dilations_0"), val = tensor([1, 1])]; + tensor v_73_groups_0 = const()[name = tensor("v_73_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468100160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469329024))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_73_cast_fp16 = conv(dilations = v_73_dilations_0, groups = v_73_groups_0, pad = v_73_pad_0, pad_type = v_73_pad_type_0, strides = v_73_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_161_cast_fp16)[name = tensor("v_73_cast_fp16")]; + tensor var_15612_begin_0 = const()[name = tensor("op_15612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15612_end_0 = const()[name = tensor("op_15612_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15612_end_mask_0 = const()[name = tensor("op_15612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15612_cast_fp16 = slice_by_index(begin = var_15612_begin_0, end = var_15612_end_0, end_mask = var_15612_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15612_cast_fp16")]; + tensor var_15616_begin_0 = const()[name = tensor("op_15616_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15616_end_0 = const()[name = tensor("op_15616_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_15616_end_mask_0 = const()[name = tensor("op_15616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15616_cast_fp16 = slice_by_index(begin = var_15616_begin_0, end = var_15616_end_0, end_mask = var_15616_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15616_cast_fp16")]; + tensor var_15620_begin_0 = const()[name = tensor("op_15620_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15620_end_0 = const()[name = tensor("op_15620_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_15620_end_mask_0 = const()[name = tensor("op_15620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15620_cast_fp16 = slice_by_index(begin = var_15620_begin_0, end = var_15620_end_0, end_mask = var_15620_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15620_cast_fp16")]; + tensor var_15624_begin_0 = const()[name = tensor("op_15624_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15624_end_0 = const()[name = tensor("op_15624_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_15624_end_mask_0 = const()[name = tensor("op_15624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15624_cast_fp16 = slice_by_index(begin = var_15624_begin_0, end = var_15624_end_0, end_mask = var_15624_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15624_cast_fp16")]; + tensor var_15628_begin_0 = const()[name = tensor("op_15628_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15628_end_0 = const()[name = tensor("op_15628_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_15628_end_mask_0 = const()[name = tensor("op_15628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15628_cast_fp16 = slice_by_index(begin = var_15628_begin_0, end = var_15628_end_0, end_mask = var_15628_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15628_cast_fp16")]; + tensor var_15632_begin_0 = const()[name = tensor("op_15632_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15632_end_0 = const()[name = tensor("op_15632_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_15632_end_mask_0 = const()[name = tensor("op_15632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15632_cast_fp16 = slice_by_index(begin = var_15632_begin_0, end = var_15632_end_0, end_mask = var_15632_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15632_cast_fp16")]; + tensor var_15636_begin_0 = const()[name = tensor("op_15636_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15636_end_0 = const()[name = tensor("op_15636_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_15636_end_mask_0 = const()[name = tensor("op_15636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15636_cast_fp16 = slice_by_index(begin = var_15636_begin_0, end = var_15636_end_0, end_mask = var_15636_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15636_cast_fp16")]; + tensor var_15640_begin_0 = const()[name = tensor("op_15640_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15640_end_0 = const()[name = tensor("op_15640_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_15640_end_mask_0 = const()[name = tensor("op_15640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15640_cast_fp16 = slice_by_index(begin = var_15640_begin_0, end = var_15640_end_0, end_mask = var_15640_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15640_cast_fp16")]; + tensor var_15644_begin_0 = const()[name = tensor("op_15644_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15644_end_0 = const()[name = tensor("op_15644_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_15644_end_mask_0 = const()[name = tensor("op_15644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15644_cast_fp16 = slice_by_index(begin = var_15644_begin_0, end = var_15644_end_0, end_mask = var_15644_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15644_cast_fp16")]; + tensor var_15648_begin_0 = const()[name = tensor("op_15648_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15648_end_0 = const()[name = tensor("op_15648_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_15648_end_mask_0 = const()[name = tensor("op_15648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15648_cast_fp16 = slice_by_index(begin = var_15648_begin_0, end = var_15648_end_0, end_mask = var_15648_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15648_cast_fp16")]; + tensor var_15652_begin_0 = const()[name = tensor("op_15652_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15652_end_0 = const()[name = tensor("op_15652_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_15652_end_mask_0 = const()[name = tensor("op_15652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15652_cast_fp16 = slice_by_index(begin = var_15652_begin_0, end = var_15652_end_0, end_mask = var_15652_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15652_cast_fp16")]; + tensor var_15656_begin_0 = const()[name = tensor("op_15656_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15656_end_0 = const()[name = tensor("op_15656_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_15656_end_mask_0 = const()[name = tensor("op_15656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15656_cast_fp16 = slice_by_index(begin = var_15656_begin_0, end = var_15656_end_0, end_mask = var_15656_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15656_cast_fp16")]; + tensor var_15660_begin_0 = const()[name = tensor("op_15660_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15660_end_0 = const()[name = tensor("op_15660_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_15660_end_mask_0 = const()[name = tensor("op_15660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15660_cast_fp16 = slice_by_index(begin = var_15660_begin_0, end = var_15660_end_0, end_mask = var_15660_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15660_cast_fp16")]; + tensor var_15664_begin_0 = const()[name = tensor("op_15664_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15664_end_0 = const()[name = tensor("op_15664_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_15664_end_mask_0 = const()[name = tensor("op_15664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15664_cast_fp16 = slice_by_index(begin = var_15664_begin_0, end = var_15664_end_0, end_mask = var_15664_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15664_cast_fp16")]; + tensor var_15668_begin_0 = const()[name = tensor("op_15668_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15668_end_0 = const()[name = tensor("op_15668_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_15668_end_mask_0 = const()[name = tensor("op_15668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15668_cast_fp16 = slice_by_index(begin = var_15668_begin_0, end = var_15668_end_0, end_mask = var_15668_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15668_cast_fp16")]; + tensor var_15672_begin_0 = const()[name = tensor("op_15672_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15672_end_0 = const()[name = tensor("op_15672_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15672_end_mask_0 = const()[name = tensor("op_15672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15672_cast_fp16 = slice_by_index(begin = var_15672_begin_0, end = var_15672_end_0, end_mask = var_15672_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15672_cast_fp16")]; + tensor var_15676_begin_0 = const()[name = tensor("op_15676_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15676_end_0 = const()[name = tensor("op_15676_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_15676_end_mask_0 = const()[name = tensor("op_15676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15676_cast_fp16 = slice_by_index(begin = var_15676_begin_0, end = var_15676_end_0, end_mask = var_15676_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15676_cast_fp16")]; + tensor var_15680_begin_0 = const()[name = tensor("op_15680_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15680_end_0 = const()[name = tensor("op_15680_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_15680_end_mask_0 = const()[name = tensor("op_15680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15680_cast_fp16 = slice_by_index(begin = var_15680_begin_0, end = var_15680_end_0, end_mask = var_15680_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15680_cast_fp16")]; + tensor var_15684_begin_0 = const()[name = tensor("op_15684_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15684_end_0 = const()[name = tensor("op_15684_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_15684_end_mask_0 = const()[name = tensor("op_15684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15684_cast_fp16 = slice_by_index(begin = var_15684_begin_0, end = var_15684_end_0, end_mask = var_15684_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15684_cast_fp16")]; + tensor var_15688_begin_0 = const()[name = tensor("op_15688_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15688_end_0 = const()[name = tensor("op_15688_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_15688_end_mask_0 = const()[name = tensor("op_15688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15688_cast_fp16 = slice_by_index(begin = var_15688_begin_0, end = var_15688_end_0, end_mask = var_15688_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15688_cast_fp16")]; + tensor k_147_perm_0 = const()[name = tensor("k_147_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_15695_begin_0 = const()[name = tensor("op_15695_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15695_end_0 = const()[name = tensor("op_15695_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_15695_end_mask_0 = const()[name = tensor("op_15695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_147_cast_fp16 = transpose(perm = k_147_perm_0, x = k_145_cast_fp16)[name = tensor("transpose_31")]; + tensor var_15695_cast_fp16 = slice_by_index(begin = var_15695_begin_0, end = var_15695_end_0, end_mask = var_15695_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15695_cast_fp16")]; + tensor var_15699_begin_0 = const()[name = tensor("op_15699_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_15699_end_0 = const()[name = tensor("op_15699_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_15699_end_mask_0 = const()[name = tensor("op_15699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15699_cast_fp16 = slice_by_index(begin = var_15699_begin_0, end = var_15699_end_0, end_mask = var_15699_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15699_cast_fp16")]; + tensor var_15703_begin_0 = const()[name = tensor("op_15703_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_15703_end_0 = const()[name = tensor("op_15703_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_15703_end_mask_0 = const()[name = tensor("op_15703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15703_cast_fp16 = slice_by_index(begin = var_15703_begin_0, end = var_15703_end_0, end_mask = var_15703_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15703_cast_fp16")]; + tensor var_15707_begin_0 = const()[name = tensor("op_15707_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_15707_end_0 = const()[name = tensor("op_15707_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_15707_end_mask_0 = const()[name = tensor("op_15707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15707_cast_fp16 = slice_by_index(begin = var_15707_begin_0, end = var_15707_end_0, end_mask = var_15707_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15707_cast_fp16")]; + tensor var_15711_begin_0 = const()[name = tensor("op_15711_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_15711_end_0 = const()[name = tensor("op_15711_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_15711_end_mask_0 = const()[name = tensor("op_15711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15711_cast_fp16 = slice_by_index(begin = var_15711_begin_0, end = var_15711_end_0, end_mask = var_15711_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15711_cast_fp16")]; + tensor var_15715_begin_0 = const()[name = tensor("op_15715_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_15715_end_0 = const()[name = tensor("op_15715_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_15715_end_mask_0 = const()[name = tensor("op_15715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15715_cast_fp16 = slice_by_index(begin = var_15715_begin_0, end = var_15715_end_0, end_mask = var_15715_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15715_cast_fp16")]; + tensor var_15719_begin_0 = const()[name = tensor("op_15719_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_15719_end_0 = const()[name = tensor("op_15719_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_15719_end_mask_0 = const()[name = tensor("op_15719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15719_cast_fp16 = slice_by_index(begin = var_15719_begin_0, end = var_15719_end_0, end_mask = var_15719_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15719_cast_fp16")]; + tensor var_15723_begin_0 = const()[name = tensor("op_15723_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_15723_end_0 = const()[name = tensor("op_15723_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_15723_end_mask_0 = const()[name = tensor("op_15723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15723_cast_fp16 = slice_by_index(begin = var_15723_begin_0, end = var_15723_end_0, end_mask = var_15723_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15723_cast_fp16")]; + tensor var_15727_begin_0 = const()[name = tensor("op_15727_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15727_end_0 = const()[name = tensor("op_15727_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_15727_end_mask_0 = const()[name = tensor("op_15727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15727_cast_fp16 = slice_by_index(begin = var_15727_begin_0, end = var_15727_end_0, end_mask = var_15727_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15727_cast_fp16")]; + tensor var_15731_begin_0 = const()[name = tensor("op_15731_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_15731_end_0 = const()[name = tensor("op_15731_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_15731_end_mask_0 = const()[name = tensor("op_15731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15731_cast_fp16 = slice_by_index(begin = var_15731_begin_0, end = var_15731_end_0, end_mask = var_15731_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15731_cast_fp16")]; + tensor var_15735_begin_0 = const()[name = tensor("op_15735_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_15735_end_0 = const()[name = tensor("op_15735_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_15735_end_mask_0 = const()[name = tensor("op_15735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15735_cast_fp16 = slice_by_index(begin = var_15735_begin_0, end = var_15735_end_0, end_mask = var_15735_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15735_cast_fp16")]; + tensor var_15739_begin_0 = const()[name = tensor("op_15739_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_15739_end_0 = const()[name = tensor("op_15739_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_15739_end_mask_0 = const()[name = tensor("op_15739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15739_cast_fp16 = slice_by_index(begin = var_15739_begin_0, end = var_15739_end_0, end_mask = var_15739_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15739_cast_fp16")]; + tensor var_15743_begin_0 = const()[name = tensor("op_15743_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_15743_end_0 = const()[name = tensor("op_15743_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_15743_end_mask_0 = const()[name = tensor("op_15743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15743_cast_fp16 = slice_by_index(begin = var_15743_begin_0, end = var_15743_end_0, end_mask = var_15743_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15743_cast_fp16")]; + tensor var_15747_begin_0 = const()[name = tensor("op_15747_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_15747_end_0 = const()[name = tensor("op_15747_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_15747_end_mask_0 = const()[name = tensor("op_15747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15747_cast_fp16 = slice_by_index(begin = var_15747_begin_0, end = var_15747_end_0, end_mask = var_15747_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15747_cast_fp16")]; + tensor var_15751_begin_0 = const()[name = tensor("op_15751_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_15751_end_0 = const()[name = tensor("op_15751_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_15751_end_mask_0 = const()[name = tensor("op_15751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15751_cast_fp16 = slice_by_index(begin = var_15751_begin_0, end = var_15751_end_0, end_mask = var_15751_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15751_cast_fp16")]; + tensor var_15755_begin_0 = const()[name = tensor("op_15755_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_15755_end_0 = const()[name = tensor("op_15755_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15755_end_mask_0 = const()[name = tensor("op_15755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15755_cast_fp16 = slice_by_index(begin = var_15755_begin_0, end = var_15755_end_0, end_mask = var_15755_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15755_cast_fp16")]; + tensor var_15759_begin_0 = const()[name = tensor("op_15759_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15759_end_0 = const()[name = tensor("op_15759_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_15759_end_mask_0 = const()[name = tensor("op_15759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15759_cast_fp16 = slice_by_index(begin = var_15759_begin_0, end = var_15759_end_0, end_mask = var_15759_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15759_cast_fp16")]; + tensor var_15763_begin_0 = const()[name = tensor("op_15763_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_15763_end_0 = const()[name = tensor("op_15763_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_15763_end_mask_0 = const()[name = tensor("op_15763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15763_cast_fp16 = slice_by_index(begin = var_15763_begin_0, end = var_15763_end_0, end_mask = var_15763_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15763_cast_fp16")]; + tensor var_15767_begin_0 = const()[name = tensor("op_15767_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_15767_end_0 = const()[name = tensor("op_15767_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_15767_end_mask_0 = const()[name = tensor("op_15767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15767_cast_fp16 = slice_by_index(begin = var_15767_begin_0, end = var_15767_end_0, end_mask = var_15767_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15767_cast_fp16")]; + tensor var_15771_begin_0 = const()[name = tensor("op_15771_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_15771_end_0 = const()[name = tensor("op_15771_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_15771_end_mask_0 = const()[name = tensor("op_15771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15771_cast_fp16 = slice_by_index(begin = var_15771_begin_0, end = var_15771_end_0, end_mask = var_15771_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15771_cast_fp16")]; + tensor var_15773_begin_0 = const()[name = tensor("op_15773_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15773_end_0 = const()[name = tensor("op_15773_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15773_end_mask_0 = const()[name = tensor("op_15773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15773_cast_fp16 = slice_by_index(begin = var_15773_begin_0, end = var_15773_end_0, end_mask = var_15773_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15773_cast_fp16")]; + tensor var_15777_begin_0 = const()[name = tensor("op_15777_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15777_end_0 = const()[name = tensor("op_15777_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_15777_end_mask_0 = const()[name = tensor("op_15777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15777_cast_fp16 = slice_by_index(begin = var_15777_begin_0, end = var_15777_end_0, end_mask = var_15777_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15777_cast_fp16")]; + tensor var_15781_begin_0 = const()[name = tensor("op_15781_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15781_end_0 = const()[name = tensor("op_15781_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_15781_end_mask_0 = const()[name = tensor("op_15781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15781_cast_fp16 = slice_by_index(begin = var_15781_begin_0, end = var_15781_end_0, end_mask = var_15781_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15781_cast_fp16")]; + tensor var_15785_begin_0 = const()[name = tensor("op_15785_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15785_end_0 = const()[name = tensor("op_15785_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_15785_end_mask_0 = const()[name = tensor("op_15785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15785_cast_fp16 = slice_by_index(begin = var_15785_begin_0, end = var_15785_end_0, end_mask = var_15785_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15785_cast_fp16")]; + tensor var_15789_begin_0 = const()[name = tensor("op_15789_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15789_end_0 = const()[name = tensor("op_15789_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_15789_end_mask_0 = const()[name = tensor("op_15789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15789_cast_fp16 = slice_by_index(begin = var_15789_begin_0, end = var_15789_end_0, end_mask = var_15789_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15789_cast_fp16")]; + tensor var_15793_begin_0 = const()[name = tensor("op_15793_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15793_end_0 = const()[name = tensor("op_15793_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_15793_end_mask_0 = const()[name = tensor("op_15793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15793_cast_fp16 = slice_by_index(begin = var_15793_begin_0, end = var_15793_end_0, end_mask = var_15793_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15793_cast_fp16")]; + tensor var_15797_begin_0 = const()[name = tensor("op_15797_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15797_end_0 = const()[name = tensor("op_15797_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_15797_end_mask_0 = const()[name = tensor("op_15797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15797_cast_fp16 = slice_by_index(begin = var_15797_begin_0, end = var_15797_end_0, end_mask = var_15797_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15797_cast_fp16")]; + tensor var_15801_begin_0 = const()[name = tensor("op_15801_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15801_end_0 = const()[name = tensor("op_15801_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_15801_end_mask_0 = const()[name = tensor("op_15801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15801_cast_fp16 = slice_by_index(begin = var_15801_begin_0, end = var_15801_end_0, end_mask = var_15801_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15801_cast_fp16")]; + tensor var_15805_begin_0 = const()[name = tensor("op_15805_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15805_end_0 = const()[name = tensor("op_15805_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_15805_end_mask_0 = const()[name = tensor("op_15805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15805_cast_fp16 = slice_by_index(begin = var_15805_begin_0, end = var_15805_end_0, end_mask = var_15805_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15805_cast_fp16")]; + tensor var_15809_begin_0 = const()[name = tensor("op_15809_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15809_end_0 = const()[name = tensor("op_15809_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_15809_end_mask_0 = const()[name = tensor("op_15809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15809_cast_fp16 = slice_by_index(begin = var_15809_begin_0, end = var_15809_end_0, end_mask = var_15809_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15809_cast_fp16")]; + tensor var_15813_begin_0 = const()[name = tensor("op_15813_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15813_end_0 = const()[name = tensor("op_15813_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_15813_end_mask_0 = const()[name = tensor("op_15813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15813_cast_fp16 = slice_by_index(begin = var_15813_begin_0, end = var_15813_end_0, end_mask = var_15813_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15813_cast_fp16")]; + tensor var_15817_begin_0 = const()[name = tensor("op_15817_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15817_end_0 = const()[name = tensor("op_15817_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_15817_end_mask_0 = const()[name = tensor("op_15817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15817_cast_fp16 = slice_by_index(begin = var_15817_begin_0, end = var_15817_end_0, end_mask = var_15817_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15817_cast_fp16")]; + tensor var_15821_begin_0 = const()[name = tensor("op_15821_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15821_end_0 = const()[name = tensor("op_15821_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_15821_end_mask_0 = const()[name = tensor("op_15821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15821_cast_fp16 = slice_by_index(begin = var_15821_begin_0, end = var_15821_end_0, end_mask = var_15821_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15821_cast_fp16")]; + tensor var_15825_begin_0 = const()[name = tensor("op_15825_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15825_end_0 = const()[name = tensor("op_15825_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_15825_end_mask_0 = const()[name = tensor("op_15825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15825_cast_fp16 = slice_by_index(begin = var_15825_begin_0, end = var_15825_end_0, end_mask = var_15825_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15825_cast_fp16")]; + tensor var_15829_begin_0 = const()[name = tensor("op_15829_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15829_end_0 = const()[name = tensor("op_15829_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_15829_end_mask_0 = const()[name = tensor("op_15829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15829_cast_fp16 = slice_by_index(begin = var_15829_begin_0, end = var_15829_end_0, end_mask = var_15829_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15829_cast_fp16")]; + tensor var_15833_begin_0 = const()[name = tensor("op_15833_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15833_end_0 = const()[name = tensor("op_15833_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15833_end_mask_0 = const()[name = tensor("op_15833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15833_cast_fp16 = slice_by_index(begin = var_15833_begin_0, end = var_15833_end_0, end_mask = var_15833_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15833_cast_fp16")]; + tensor var_15837_begin_0 = const()[name = tensor("op_15837_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15837_end_0 = const()[name = tensor("op_15837_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_15837_end_mask_0 = const()[name = tensor("op_15837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15837_cast_fp16 = slice_by_index(begin = var_15837_begin_0, end = var_15837_end_0, end_mask = var_15837_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15837_cast_fp16")]; + tensor var_15841_begin_0 = const()[name = tensor("op_15841_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15841_end_0 = const()[name = tensor("op_15841_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_15841_end_mask_0 = const()[name = tensor("op_15841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15841_cast_fp16 = slice_by_index(begin = var_15841_begin_0, end = var_15841_end_0, end_mask = var_15841_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15841_cast_fp16")]; + tensor var_15845_begin_0 = const()[name = tensor("op_15845_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15845_end_0 = const()[name = tensor("op_15845_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_15845_end_mask_0 = const()[name = tensor("op_15845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15845_cast_fp16 = slice_by_index(begin = var_15845_begin_0, end = var_15845_end_0, end_mask = var_15845_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15845_cast_fp16")]; + tensor var_15849_begin_0 = const()[name = tensor("op_15849_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15849_end_0 = const()[name = tensor("op_15849_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_15849_end_mask_0 = const()[name = tensor("op_15849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15849_cast_fp16 = slice_by_index(begin = var_15849_begin_0, end = var_15849_end_0, end_mask = var_15849_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15849_cast_fp16")]; + tensor var_15853_equation_0 = const()[name = tensor("op_15853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15853_cast_fp16 = einsum(equation = var_15853_equation_0, values = (var_15695_cast_fp16, var_15612_cast_fp16))[name = tensor("op_15853_cast_fp16")]; + tensor var_15854_to_fp16 = const()[name = tensor("op_15854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1281_cast_fp16 = mul(x = var_15853_cast_fp16, y = var_15854_to_fp16)[name = tensor("aw_1281_cast_fp16")]; + tensor var_15857_equation_0 = const()[name = tensor("op_15857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15857_cast_fp16 = einsum(equation = var_15857_equation_0, values = (var_15699_cast_fp16, var_15616_cast_fp16))[name = tensor("op_15857_cast_fp16")]; + tensor var_15858_to_fp16 = const()[name = tensor("op_15858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1283_cast_fp16 = mul(x = var_15857_cast_fp16, y = var_15858_to_fp16)[name = tensor("aw_1283_cast_fp16")]; + tensor var_15861_equation_0 = const()[name = tensor("op_15861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15861_cast_fp16 = einsum(equation = var_15861_equation_0, values = (var_15703_cast_fp16, var_15620_cast_fp16))[name = tensor("op_15861_cast_fp16")]; + tensor var_15862_to_fp16 = const()[name = tensor("op_15862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1285_cast_fp16 = mul(x = var_15861_cast_fp16, y = var_15862_to_fp16)[name = tensor("aw_1285_cast_fp16")]; + tensor var_15865_equation_0 = const()[name = tensor("op_15865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15865_cast_fp16 = einsum(equation = var_15865_equation_0, values = (var_15707_cast_fp16, var_15624_cast_fp16))[name = tensor("op_15865_cast_fp16")]; + tensor var_15866_to_fp16 = const()[name = tensor("op_15866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1287_cast_fp16 = mul(x = var_15865_cast_fp16, y = var_15866_to_fp16)[name = tensor("aw_1287_cast_fp16")]; + tensor var_15869_equation_0 = const()[name = tensor("op_15869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15869_cast_fp16 = einsum(equation = var_15869_equation_0, values = (var_15711_cast_fp16, var_15628_cast_fp16))[name = tensor("op_15869_cast_fp16")]; + tensor var_15870_to_fp16 = const()[name = tensor("op_15870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1289_cast_fp16 = mul(x = var_15869_cast_fp16, y = var_15870_to_fp16)[name = tensor("aw_1289_cast_fp16")]; + tensor var_15873_equation_0 = const()[name = tensor("op_15873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15873_cast_fp16 = einsum(equation = var_15873_equation_0, values = (var_15715_cast_fp16, var_15632_cast_fp16))[name = tensor("op_15873_cast_fp16")]; + tensor var_15874_to_fp16 = const()[name = tensor("op_15874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1291_cast_fp16 = mul(x = var_15873_cast_fp16, y = var_15874_to_fp16)[name = tensor("aw_1291_cast_fp16")]; + tensor var_15877_equation_0 = const()[name = tensor("op_15877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15877_cast_fp16 = einsum(equation = var_15877_equation_0, values = (var_15719_cast_fp16, var_15636_cast_fp16))[name = tensor("op_15877_cast_fp16")]; + tensor var_15878_to_fp16 = const()[name = tensor("op_15878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1293_cast_fp16 = mul(x = var_15877_cast_fp16, y = var_15878_to_fp16)[name = tensor("aw_1293_cast_fp16")]; + tensor var_15881_equation_0 = const()[name = tensor("op_15881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15881_cast_fp16 = einsum(equation = var_15881_equation_0, values = (var_15723_cast_fp16, var_15640_cast_fp16))[name = tensor("op_15881_cast_fp16")]; + tensor var_15882_to_fp16 = const()[name = tensor("op_15882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1295_cast_fp16 = mul(x = var_15881_cast_fp16, y = var_15882_to_fp16)[name = tensor("aw_1295_cast_fp16")]; + tensor var_15885_equation_0 = const()[name = tensor("op_15885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15885_cast_fp16 = einsum(equation = var_15885_equation_0, values = (var_15727_cast_fp16, var_15644_cast_fp16))[name = tensor("op_15885_cast_fp16")]; + tensor var_15886_to_fp16 = const()[name = tensor("op_15886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1297_cast_fp16 = mul(x = var_15885_cast_fp16, y = var_15886_to_fp16)[name = tensor("aw_1297_cast_fp16")]; + tensor var_15889_equation_0 = const()[name = tensor("op_15889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15889_cast_fp16 = einsum(equation = var_15889_equation_0, values = (var_15731_cast_fp16, var_15648_cast_fp16))[name = tensor("op_15889_cast_fp16")]; + tensor var_15890_to_fp16 = const()[name = tensor("op_15890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1299_cast_fp16 = mul(x = var_15889_cast_fp16, y = var_15890_to_fp16)[name = tensor("aw_1299_cast_fp16")]; + tensor var_15893_equation_0 = const()[name = tensor("op_15893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15893_cast_fp16 = einsum(equation = var_15893_equation_0, values = (var_15735_cast_fp16, var_15652_cast_fp16))[name = tensor("op_15893_cast_fp16")]; + tensor var_15894_to_fp16 = const()[name = tensor("op_15894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1301_cast_fp16 = mul(x = var_15893_cast_fp16, y = var_15894_to_fp16)[name = tensor("aw_1301_cast_fp16")]; + tensor var_15897_equation_0 = const()[name = tensor("op_15897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15897_cast_fp16 = einsum(equation = var_15897_equation_0, values = (var_15739_cast_fp16, var_15656_cast_fp16))[name = tensor("op_15897_cast_fp16")]; + tensor var_15898_to_fp16 = const()[name = tensor("op_15898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1303_cast_fp16 = mul(x = var_15897_cast_fp16, y = var_15898_to_fp16)[name = tensor("aw_1303_cast_fp16")]; + tensor var_15901_equation_0 = const()[name = tensor("op_15901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15901_cast_fp16 = einsum(equation = var_15901_equation_0, values = (var_15743_cast_fp16, var_15660_cast_fp16))[name = tensor("op_15901_cast_fp16")]; + tensor var_15902_to_fp16 = const()[name = tensor("op_15902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1305_cast_fp16 = mul(x = var_15901_cast_fp16, y = var_15902_to_fp16)[name = tensor("aw_1305_cast_fp16")]; + tensor var_15905_equation_0 = const()[name = tensor("op_15905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15905_cast_fp16 = einsum(equation = var_15905_equation_0, values = (var_15747_cast_fp16, var_15664_cast_fp16))[name = tensor("op_15905_cast_fp16")]; + tensor var_15906_to_fp16 = const()[name = tensor("op_15906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1307_cast_fp16 = mul(x = var_15905_cast_fp16, y = var_15906_to_fp16)[name = tensor("aw_1307_cast_fp16")]; + tensor var_15909_equation_0 = const()[name = tensor("op_15909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15909_cast_fp16 = einsum(equation = var_15909_equation_0, values = (var_15751_cast_fp16, var_15668_cast_fp16))[name = tensor("op_15909_cast_fp16")]; + tensor var_15910_to_fp16 = const()[name = tensor("op_15910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1309_cast_fp16 = mul(x = var_15909_cast_fp16, y = var_15910_to_fp16)[name = tensor("aw_1309_cast_fp16")]; + tensor var_15913_equation_0 = const()[name = tensor("op_15913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15913_cast_fp16 = einsum(equation = var_15913_equation_0, values = (var_15755_cast_fp16, var_15672_cast_fp16))[name = tensor("op_15913_cast_fp16")]; + tensor var_15914_to_fp16 = const()[name = tensor("op_15914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1311_cast_fp16 = mul(x = var_15913_cast_fp16, y = var_15914_to_fp16)[name = tensor("aw_1311_cast_fp16")]; + tensor var_15917_equation_0 = const()[name = tensor("op_15917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15917_cast_fp16 = einsum(equation = var_15917_equation_0, values = (var_15759_cast_fp16, var_15676_cast_fp16))[name = tensor("op_15917_cast_fp16")]; + tensor var_15918_to_fp16 = const()[name = tensor("op_15918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1313_cast_fp16 = mul(x = var_15917_cast_fp16, y = var_15918_to_fp16)[name = tensor("aw_1313_cast_fp16")]; + tensor var_15921_equation_0 = const()[name = tensor("op_15921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15921_cast_fp16 = einsum(equation = var_15921_equation_0, values = (var_15763_cast_fp16, var_15680_cast_fp16))[name = tensor("op_15921_cast_fp16")]; + tensor var_15922_to_fp16 = const()[name = tensor("op_15922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1315_cast_fp16 = mul(x = var_15921_cast_fp16, y = var_15922_to_fp16)[name = tensor("aw_1315_cast_fp16")]; + tensor var_15925_equation_0 = const()[name = tensor("op_15925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15925_cast_fp16 = einsum(equation = var_15925_equation_0, values = (var_15767_cast_fp16, var_15684_cast_fp16))[name = tensor("op_15925_cast_fp16")]; + tensor var_15926_to_fp16 = const()[name = tensor("op_15926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1317_cast_fp16 = mul(x = var_15925_cast_fp16, y = var_15926_to_fp16)[name = tensor("aw_1317_cast_fp16")]; + tensor var_15929_equation_0 = const()[name = tensor("op_15929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15929_cast_fp16 = einsum(equation = var_15929_equation_0, values = (var_15771_cast_fp16, var_15688_cast_fp16))[name = tensor("op_15929_cast_fp16")]; + tensor var_15930_to_fp16 = const()[name = tensor("op_15930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1319_cast_fp16 = mul(x = var_15929_cast_fp16, y = var_15930_to_fp16)[name = tensor("aw_1319_cast_fp16")]; + tensor var_15932_cast_fp16 = softmax(axis = var_2624, x = aw_1281_cast_fp16)[name = tensor("op_15932_cast_fp16")]; + tensor var_15933_cast_fp16 = softmax(axis = var_2624, x = aw_1283_cast_fp16)[name = tensor("op_15933_cast_fp16")]; + tensor var_15934_cast_fp16 = softmax(axis = var_2624, x = aw_1285_cast_fp16)[name = tensor("op_15934_cast_fp16")]; + tensor var_15935_cast_fp16 = softmax(axis = var_2624, x = aw_1287_cast_fp16)[name = tensor("op_15935_cast_fp16")]; + tensor var_15936_cast_fp16 = softmax(axis = var_2624, x = aw_1289_cast_fp16)[name = tensor("op_15936_cast_fp16")]; + tensor var_15937_cast_fp16 = softmax(axis = var_2624, x = aw_1291_cast_fp16)[name = tensor("op_15937_cast_fp16")]; + tensor var_15938_cast_fp16 = softmax(axis = var_2624, x = aw_1293_cast_fp16)[name = tensor("op_15938_cast_fp16")]; + tensor var_15939_cast_fp16 = softmax(axis = var_2624, x = aw_1295_cast_fp16)[name = tensor("op_15939_cast_fp16")]; + tensor var_15940_cast_fp16 = softmax(axis = var_2624, x = aw_1297_cast_fp16)[name = tensor("op_15940_cast_fp16")]; + tensor var_15941_cast_fp16 = softmax(axis = var_2624, x = aw_1299_cast_fp16)[name = tensor("op_15941_cast_fp16")]; + tensor var_15942_cast_fp16 = softmax(axis = var_2624, x = aw_1301_cast_fp16)[name = tensor("op_15942_cast_fp16")]; + tensor var_15943_cast_fp16 = softmax(axis = var_2624, x = aw_1303_cast_fp16)[name = tensor("op_15943_cast_fp16")]; + tensor var_15944_cast_fp16 = softmax(axis = var_2624, x = aw_1305_cast_fp16)[name = tensor("op_15944_cast_fp16")]; + tensor var_15945_cast_fp16 = softmax(axis = var_2624, x = aw_1307_cast_fp16)[name = tensor("op_15945_cast_fp16")]; + tensor var_15946_cast_fp16 = softmax(axis = var_2624, x = aw_1309_cast_fp16)[name = tensor("op_15946_cast_fp16")]; + tensor var_15947_cast_fp16 = softmax(axis = var_2624, x = aw_1311_cast_fp16)[name = tensor("op_15947_cast_fp16")]; + tensor var_15948_cast_fp16 = softmax(axis = var_2624, x = aw_1313_cast_fp16)[name = tensor("op_15948_cast_fp16")]; + tensor var_15949_cast_fp16 = softmax(axis = var_2624, x = aw_1315_cast_fp16)[name = tensor("op_15949_cast_fp16")]; + tensor var_15950_cast_fp16 = softmax(axis = var_2624, x = aw_1317_cast_fp16)[name = tensor("op_15950_cast_fp16")]; + tensor var_15951_cast_fp16 = softmax(axis = var_2624, x = aw_1319_cast_fp16)[name = tensor("op_15951_cast_fp16")]; + tensor var_15953_equation_0 = const()[name = tensor("op_15953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15953_cast_fp16 = einsum(equation = var_15953_equation_0, values = (var_15773_cast_fp16, var_15932_cast_fp16))[name = tensor("op_15953_cast_fp16")]; + tensor var_15955_equation_0 = const()[name = tensor("op_15955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15955_cast_fp16 = einsum(equation = var_15955_equation_0, values = (var_15777_cast_fp16, var_15933_cast_fp16))[name = tensor("op_15955_cast_fp16")]; + tensor var_15957_equation_0 = const()[name = tensor("op_15957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15957_cast_fp16 = einsum(equation = var_15957_equation_0, values = (var_15781_cast_fp16, var_15934_cast_fp16))[name = tensor("op_15957_cast_fp16")]; + tensor var_15959_equation_0 = const()[name = tensor("op_15959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15959_cast_fp16 = einsum(equation = var_15959_equation_0, values = (var_15785_cast_fp16, var_15935_cast_fp16))[name = tensor("op_15959_cast_fp16")]; + tensor var_15961_equation_0 = const()[name = tensor("op_15961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15961_cast_fp16 = einsum(equation = var_15961_equation_0, values = (var_15789_cast_fp16, var_15936_cast_fp16))[name = tensor("op_15961_cast_fp16")]; + tensor var_15963_equation_0 = const()[name = tensor("op_15963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15963_cast_fp16 = einsum(equation = var_15963_equation_0, values = (var_15793_cast_fp16, var_15937_cast_fp16))[name = tensor("op_15963_cast_fp16")]; + tensor var_15965_equation_0 = const()[name = tensor("op_15965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15965_cast_fp16 = einsum(equation = var_15965_equation_0, values = (var_15797_cast_fp16, var_15938_cast_fp16))[name = tensor("op_15965_cast_fp16")]; + tensor var_15967_equation_0 = const()[name = tensor("op_15967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15967_cast_fp16 = einsum(equation = var_15967_equation_0, values = (var_15801_cast_fp16, var_15939_cast_fp16))[name = tensor("op_15967_cast_fp16")]; + tensor var_15969_equation_0 = const()[name = tensor("op_15969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15969_cast_fp16 = einsum(equation = var_15969_equation_0, values = (var_15805_cast_fp16, var_15940_cast_fp16))[name = tensor("op_15969_cast_fp16")]; + tensor var_15971_equation_0 = const()[name = tensor("op_15971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15971_cast_fp16 = einsum(equation = var_15971_equation_0, values = (var_15809_cast_fp16, var_15941_cast_fp16))[name = tensor("op_15971_cast_fp16")]; + tensor var_15973_equation_0 = const()[name = tensor("op_15973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15973_cast_fp16 = einsum(equation = var_15973_equation_0, values = (var_15813_cast_fp16, var_15942_cast_fp16))[name = tensor("op_15973_cast_fp16")]; + tensor var_15975_equation_0 = const()[name = tensor("op_15975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15975_cast_fp16 = einsum(equation = var_15975_equation_0, values = (var_15817_cast_fp16, var_15943_cast_fp16))[name = tensor("op_15975_cast_fp16")]; + tensor var_15977_equation_0 = const()[name = tensor("op_15977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15977_cast_fp16 = einsum(equation = var_15977_equation_0, values = (var_15821_cast_fp16, var_15944_cast_fp16))[name = tensor("op_15977_cast_fp16")]; + tensor var_15979_equation_0 = const()[name = tensor("op_15979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15979_cast_fp16 = einsum(equation = var_15979_equation_0, values = (var_15825_cast_fp16, var_15945_cast_fp16))[name = tensor("op_15979_cast_fp16")]; + tensor var_15981_equation_0 = const()[name = tensor("op_15981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15981_cast_fp16 = einsum(equation = var_15981_equation_0, values = (var_15829_cast_fp16, var_15946_cast_fp16))[name = tensor("op_15981_cast_fp16")]; + tensor var_15983_equation_0 = const()[name = tensor("op_15983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15983_cast_fp16 = einsum(equation = var_15983_equation_0, values = (var_15833_cast_fp16, var_15947_cast_fp16))[name = tensor("op_15983_cast_fp16")]; + tensor var_15985_equation_0 = const()[name = tensor("op_15985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15985_cast_fp16 = einsum(equation = var_15985_equation_0, values = (var_15837_cast_fp16, var_15948_cast_fp16))[name = tensor("op_15985_cast_fp16")]; + tensor var_15987_equation_0 = const()[name = tensor("op_15987_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15987_cast_fp16 = einsum(equation = var_15987_equation_0, values = (var_15841_cast_fp16, var_15949_cast_fp16))[name = tensor("op_15987_cast_fp16")]; + tensor var_15989_equation_0 = const()[name = tensor("op_15989_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15989_cast_fp16 = einsum(equation = var_15989_equation_0, values = (var_15845_cast_fp16, var_15950_cast_fp16))[name = tensor("op_15989_cast_fp16")]; + tensor var_15991_equation_0 = const()[name = tensor("op_15991_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15991_cast_fp16 = einsum(equation = var_15991_equation_0, values = (var_15849_cast_fp16, var_15951_cast_fp16))[name = tensor("op_15991_cast_fp16")]; + tensor input_261_interleave_0 = const()[name = tensor("input_261_interleave_0"), val = tensor(false)]; + tensor input_261_cast_fp16 = concat(axis = var_2624, interleave = input_261_interleave_0, values = (var_15953_cast_fp16, var_15955_cast_fp16, var_15957_cast_fp16, var_15959_cast_fp16, var_15961_cast_fp16, var_15963_cast_fp16, var_15965_cast_fp16, var_15967_cast_fp16, var_15969_cast_fp16, var_15971_cast_fp16, var_15973_cast_fp16, var_15975_cast_fp16, var_15977_cast_fp16, var_15979_cast_fp16, var_15981_cast_fp16, var_15983_cast_fp16, var_15985_cast_fp16, var_15987_cast_fp16, var_15989_cast_fp16, var_15991_cast_fp16))[name = tensor("input_261_cast_fp16")]; + tensor var_16001_pad_type_0 = const()[name = tensor("op_16001_pad_type_0"), val = tensor("valid")]; + tensor var_16001_strides_0 = const()[name = tensor("op_16001_strides_0"), val = tensor([1, 1])]; + tensor var_16001_pad_0 = const()[name = tensor("op_16001_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16001_dilations_0 = const()[name = tensor("op_16001_dilations_0"), val = tensor([1, 1])]; + tensor var_16001_groups_0 = const()[name = tensor("op_16001_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469329216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470558080))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470558272)))]; + tensor var_16001_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_16001_dilations_0, groups = var_16001_groups_0, pad = var_16001_pad_0, pad_type = var_16001_pad_type_0, strides = var_16001_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized, x = input_261_cast_fp16)[name = tensor("op_16001_cast_fp16")]; + tensor inputs_111_cast_fp16 = add(x = var_16001_cast_fp16, y = inputs_109_cast_fp16)[name = tensor("inputs_111_cast_fp16")]; + tensor hidden_states_163_axes_0 = const()[name = tensor("hidden_states_163_axes_0"), val = tensor([1])]; + tensor hidden_states_163_gamma_0_to_fp16 = const()[name = tensor("hidden_states_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470560896)))]; + tensor hidden_states_163_beta_0_to_fp16 = const()[name = tensor("hidden_states_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470563520)))]; + tensor var_16011_to_fp16 = const()[name = tensor("op_16011_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_163_cast_fp16 = layer_norm(axes = hidden_states_163_axes_0, beta = hidden_states_163_beta_0_to_fp16, epsilon = var_16011_to_fp16, gamma = hidden_states_163_gamma_0_to_fp16, x = inputs_111_cast_fp16)[name = tensor("hidden_states_163_cast_fp16")]; + tensor q_75_pad_type_0 = const()[name = tensor("q_75_pad_type_0"), val = tensor("valid")]; + tensor q_75_strides_0 = const()[name = tensor("q_75_strides_0"), val = tensor([1, 1])]; + tensor q_75_pad_0 = const()[name = tensor("q_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_75_dilations_0 = const()[name = tensor("q_75_dilations_0"), val = tensor([1, 1])]; + tensor q_75_groups_0 = const()[name = tensor("q_75_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470566144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471795008))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_75_cast_fp16 = conv(dilations = q_75_dilations_0, groups = q_75_groups_0, pad = q_75_pad_0, pad_type = q_75_pad_type_0, strides = q_75_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_163_cast_fp16)[name = tensor("q_75_cast_fp16")]; + tensor k_149_pad_type_0 = const()[name = tensor("k_149_pad_type_0"), val = tensor("valid")]; + tensor k_149_strides_0 = const()[name = tensor("k_149_strides_0"), val = tensor([1, 1])]; + tensor k_149_pad_0 = const()[name = tensor("k_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_149_dilations_0 = const()[name = tensor("k_149_dilations_0"), val = tensor([1, 1])]; + tensor k_149_groups_0 = const()[name = tensor("k_149_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471795200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473761344))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_149_cast_fp16 = conv(dilations = k_149_dilations_0, groups = k_149_groups_0, pad = k_149_pad_0, pad_type = k_149_pad_type_0, strides = k_149_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_149_cast_fp16")]; + tensor v_75_pad_type_0 = const()[name = tensor("v_75_pad_type_0"), val = tensor("valid")]; + tensor v_75_strides_0 = const()[name = tensor("v_75_strides_0"), val = tensor([1, 1])]; + tensor v_75_pad_0 = const()[name = tensor("v_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_75_dilations_0 = const()[name = tensor("v_75_dilations_0"), val = tensor([1, 1])]; + tensor v_75_groups_0 = const()[name = tensor("v_75_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473761536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475727680))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_75_cast_fp16 = conv(dilations = v_75_dilations_0, groups = v_75_groups_0, pad = v_75_pad_0, pad_type = v_75_pad_type_0, strides = v_75_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_75_cast_fp16")]; + tensor var_16044_begin_0 = const()[name = tensor("op_16044_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16044_end_0 = const()[name = tensor("op_16044_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16044_end_mask_0 = const()[name = tensor("op_16044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16044_cast_fp16 = slice_by_index(begin = var_16044_begin_0, end = var_16044_end_0, end_mask = var_16044_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16044_cast_fp16")]; + tensor var_16048_begin_0 = const()[name = tensor("op_16048_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16048_end_0 = const()[name = tensor("op_16048_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16048_end_mask_0 = const()[name = tensor("op_16048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16048_cast_fp16 = slice_by_index(begin = var_16048_begin_0, end = var_16048_end_0, end_mask = var_16048_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16048_cast_fp16")]; + tensor var_16052_begin_0 = const()[name = tensor("op_16052_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16052_end_0 = const()[name = tensor("op_16052_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16052_end_mask_0 = const()[name = tensor("op_16052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16052_cast_fp16 = slice_by_index(begin = var_16052_begin_0, end = var_16052_end_0, end_mask = var_16052_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16052_cast_fp16")]; + tensor var_16056_begin_0 = const()[name = tensor("op_16056_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16056_end_0 = const()[name = tensor("op_16056_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16056_end_mask_0 = const()[name = tensor("op_16056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16056_cast_fp16 = slice_by_index(begin = var_16056_begin_0, end = var_16056_end_0, end_mask = var_16056_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16056_cast_fp16")]; + tensor var_16060_begin_0 = const()[name = tensor("op_16060_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16060_end_0 = const()[name = tensor("op_16060_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16060_end_mask_0 = const()[name = tensor("op_16060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16060_cast_fp16 = slice_by_index(begin = var_16060_begin_0, end = var_16060_end_0, end_mask = var_16060_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16060_cast_fp16")]; + tensor var_16064_begin_0 = const()[name = tensor("op_16064_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16064_end_0 = const()[name = tensor("op_16064_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16064_end_mask_0 = const()[name = tensor("op_16064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16064_cast_fp16 = slice_by_index(begin = var_16064_begin_0, end = var_16064_end_0, end_mask = var_16064_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16064_cast_fp16")]; + tensor var_16068_begin_0 = const()[name = tensor("op_16068_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16068_end_0 = const()[name = tensor("op_16068_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16068_end_mask_0 = const()[name = tensor("op_16068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16068_cast_fp16 = slice_by_index(begin = var_16068_begin_0, end = var_16068_end_0, end_mask = var_16068_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16068_cast_fp16")]; + tensor var_16072_begin_0 = const()[name = tensor("op_16072_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16072_end_0 = const()[name = tensor("op_16072_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16072_end_mask_0 = const()[name = tensor("op_16072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16072_cast_fp16 = slice_by_index(begin = var_16072_begin_0, end = var_16072_end_0, end_mask = var_16072_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16072_cast_fp16")]; + tensor var_16076_begin_0 = const()[name = tensor("op_16076_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16076_end_0 = const()[name = tensor("op_16076_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16076_end_mask_0 = const()[name = tensor("op_16076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16076_cast_fp16 = slice_by_index(begin = var_16076_begin_0, end = var_16076_end_0, end_mask = var_16076_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16076_cast_fp16")]; + tensor var_16080_begin_0 = const()[name = tensor("op_16080_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16080_end_0 = const()[name = tensor("op_16080_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16080_end_mask_0 = const()[name = tensor("op_16080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16080_cast_fp16 = slice_by_index(begin = var_16080_begin_0, end = var_16080_end_0, end_mask = var_16080_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16080_cast_fp16")]; + tensor var_16084_begin_0 = const()[name = tensor("op_16084_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16084_end_0 = const()[name = tensor("op_16084_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16084_end_mask_0 = const()[name = tensor("op_16084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16084_cast_fp16 = slice_by_index(begin = var_16084_begin_0, end = var_16084_end_0, end_mask = var_16084_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16084_cast_fp16")]; + tensor var_16088_begin_0 = const()[name = tensor("op_16088_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16088_end_0 = const()[name = tensor("op_16088_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16088_end_mask_0 = const()[name = tensor("op_16088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16088_cast_fp16 = slice_by_index(begin = var_16088_begin_0, end = var_16088_end_0, end_mask = var_16088_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16088_cast_fp16")]; + tensor var_16092_begin_0 = const()[name = tensor("op_16092_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16092_end_0 = const()[name = tensor("op_16092_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_16092_end_mask_0 = const()[name = tensor("op_16092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16092_cast_fp16 = slice_by_index(begin = var_16092_begin_0, end = var_16092_end_0, end_mask = var_16092_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16092_cast_fp16")]; + tensor var_16096_begin_0 = const()[name = tensor("op_16096_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16096_end_0 = const()[name = tensor("op_16096_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_16096_end_mask_0 = const()[name = tensor("op_16096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16096_cast_fp16 = slice_by_index(begin = var_16096_begin_0, end = var_16096_end_0, end_mask = var_16096_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16096_cast_fp16")]; + tensor var_16100_begin_0 = const()[name = tensor("op_16100_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16100_end_0 = const()[name = tensor("op_16100_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_16100_end_mask_0 = const()[name = tensor("op_16100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16100_cast_fp16 = slice_by_index(begin = var_16100_begin_0, end = var_16100_end_0, end_mask = var_16100_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16100_cast_fp16")]; + tensor var_16104_begin_0 = const()[name = tensor("op_16104_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16104_end_0 = const()[name = tensor("op_16104_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16104_end_mask_0 = const()[name = tensor("op_16104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16104_cast_fp16 = slice_by_index(begin = var_16104_begin_0, end = var_16104_end_0, end_mask = var_16104_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16104_cast_fp16")]; + tensor var_16108_begin_0 = const()[name = tensor("op_16108_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16108_end_0 = const()[name = tensor("op_16108_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_16108_end_mask_0 = const()[name = tensor("op_16108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16108_cast_fp16 = slice_by_index(begin = var_16108_begin_0, end = var_16108_end_0, end_mask = var_16108_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16108_cast_fp16")]; + tensor var_16112_begin_0 = const()[name = tensor("op_16112_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16112_end_0 = const()[name = tensor("op_16112_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_16112_end_mask_0 = const()[name = tensor("op_16112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16112_cast_fp16 = slice_by_index(begin = var_16112_begin_0, end = var_16112_end_0, end_mask = var_16112_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16112_cast_fp16")]; + tensor var_16116_begin_0 = const()[name = tensor("op_16116_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16116_end_0 = const()[name = tensor("op_16116_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_16116_end_mask_0 = const()[name = tensor("op_16116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16116_cast_fp16 = slice_by_index(begin = var_16116_begin_0, end = var_16116_end_0, end_mask = var_16116_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16116_cast_fp16")]; + tensor var_16120_begin_0 = const()[name = tensor("op_16120_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16120_end_0 = const()[name = tensor("op_16120_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_16120_end_mask_0 = const()[name = tensor("op_16120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16120_cast_fp16 = slice_by_index(begin = var_16120_begin_0, end = var_16120_end_0, end_mask = var_16120_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16120_cast_fp16")]; + tensor k_151_perm_0 = const()[name = tensor("k_151_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_16127_begin_0 = const()[name = tensor("op_16127_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16127_end_0 = const()[name = tensor("op_16127_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_16127_end_mask_0 = const()[name = tensor("op_16127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_151_cast_fp16 = transpose(perm = k_151_perm_0, x = k_149_cast_fp16)[name = tensor("transpose_30")]; + tensor var_16127_cast_fp16 = slice_by_index(begin = var_16127_begin_0, end = var_16127_end_0, end_mask = var_16127_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16127_cast_fp16")]; + tensor var_16131_begin_0 = const()[name = tensor("op_16131_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_16131_end_0 = const()[name = tensor("op_16131_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_16131_end_mask_0 = const()[name = tensor("op_16131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16131_cast_fp16 = slice_by_index(begin = var_16131_begin_0, end = var_16131_end_0, end_mask = var_16131_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16131_cast_fp16")]; + tensor var_16135_begin_0 = const()[name = tensor("op_16135_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_16135_end_0 = const()[name = tensor("op_16135_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_16135_end_mask_0 = const()[name = tensor("op_16135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16135_cast_fp16 = slice_by_index(begin = var_16135_begin_0, end = var_16135_end_0, end_mask = var_16135_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16135_cast_fp16")]; + tensor var_16139_begin_0 = const()[name = tensor("op_16139_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_16139_end_0 = const()[name = tensor("op_16139_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_16139_end_mask_0 = const()[name = tensor("op_16139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16139_cast_fp16 = slice_by_index(begin = var_16139_begin_0, end = var_16139_end_0, end_mask = var_16139_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16139_cast_fp16")]; + tensor var_16143_begin_0 = const()[name = tensor("op_16143_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_16143_end_0 = const()[name = tensor("op_16143_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_16143_end_mask_0 = const()[name = tensor("op_16143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16143_cast_fp16 = slice_by_index(begin = var_16143_begin_0, end = var_16143_end_0, end_mask = var_16143_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16143_cast_fp16")]; + tensor var_16147_begin_0 = const()[name = tensor("op_16147_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_16147_end_0 = const()[name = tensor("op_16147_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_16147_end_mask_0 = const()[name = tensor("op_16147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16147_cast_fp16 = slice_by_index(begin = var_16147_begin_0, end = var_16147_end_0, end_mask = var_16147_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16147_cast_fp16")]; + tensor var_16151_begin_0 = const()[name = tensor("op_16151_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_16151_end_0 = const()[name = tensor("op_16151_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_16151_end_mask_0 = const()[name = tensor("op_16151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16151_cast_fp16 = slice_by_index(begin = var_16151_begin_0, end = var_16151_end_0, end_mask = var_16151_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16151_cast_fp16")]; + tensor var_16155_begin_0 = const()[name = tensor("op_16155_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_16155_end_0 = const()[name = tensor("op_16155_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_16155_end_mask_0 = const()[name = tensor("op_16155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16155_cast_fp16 = slice_by_index(begin = var_16155_begin_0, end = var_16155_end_0, end_mask = var_16155_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16155_cast_fp16")]; + tensor var_16159_begin_0 = const()[name = tensor("op_16159_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_16159_end_0 = const()[name = tensor("op_16159_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_16159_end_mask_0 = const()[name = tensor("op_16159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16159_cast_fp16 = slice_by_index(begin = var_16159_begin_0, end = var_16159_end_0, end_mask = var_16159_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16159_cast_fp16")]; + tensor var_16163_begin_0 = const()[name = tensor("op_16163_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_16163_end_0 = const()[name = tensor("op_16163_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_16163_end_mask_0 = const()[name = tensor("op_16163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16163_cast_fp16 = slice_by_index(begin = var_16163_begin_0, end = var_16163_end_0, end_mask = var_16163_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16163_cast_fp16")]; + tensor var_16167_begin_0 = const()[name = tensor("op_16167_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_16167_end_0 = const()[name = tensor("op_16167_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_16167_end_mask_0 = const()[name = tensor("op_16167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16167_cast_fp16 = slice_by_index(begin = var_16167_begin_0, end = var_16167_end_0, end_mask = var_16167_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16167_cast_fp16")]; + tensor var_16171_begin_0 = const()[name = tensor("op_16171_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_16171_end_0 = const()[name = tensor("op_16171_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_16171_end_mask_0 = const()[name = tensor("op_16171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16171_cast_fp16 = slice_by_index(begin = var_16171_begin_0, end = var_16171_end_0, end_mask = var_16171_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16171_cast_fp16")]; + tensor var_16175_begin_0 = const()[name = tensor("op_16175_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_16175_end_0 = const()[name = tensor("op_16175_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_16175_end_mask_0 = const()[name = tensor("op_16175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16175_cast_fp16 = slice_by_index(begin = var_16175_begin_0, end = var_16175_end_0, end_mask = var_16175_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16175_cast_fp16")]; + tensor var_16179_begin_0 = const()[name = tensor("op_16179_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_16179_end_0 = const()[name = tensor("op_16179_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_16179_end_mask_0 = const()[name = tensor("op_16179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16179_cast_fp16 = slice_by_index(begin = var_16179_begin_0, end = var_16179_end_0, end_mask = var_16179_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16179_cast_fp16")]; + tensor var_16183_begin_0 = const()[name = tensor("op_16183_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_16183_end_0 = const()[name = tensor("op_16183_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_16183_end_mask_0 = const()[name = tensor("op_16183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16183_cast_fp16 = slice_by_index(begin = var_16183_begin_0, end = var_16183_end_0, end_mask = var_16183_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16183_cast_fp16")]; + tensor var_16187_begin_0 = const()[name = tensor("op_16187_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_16187_end_0 = const()[name = tensor("op_16187_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_16187_end_mask_0 = const()[name = tensor("op_16187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16187_cast_fp16 = slice_by_index(begin = var_16187_begin_0, end = var_16187_end_0, end_mask = var_16187_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16187_cast_fp16")]; + tensor var_16191_begin_0 = const()[name = tensor("op_16191_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_16191_end_0 = const()[name = tensor("op_16191_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_16191_end_mask_0 = const()[name = tensor("op_16191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16191_cast_fp16 = slice_by_index(begin = var_16191_begin_0, end = var_16191_end_0, end_mask = var_16191_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16191_cast_fp16")]; + tensor var_16195_begin_0 = const()[name = tensor("op_16195_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_16195_end_0 = const()[name = tensor("op_16195_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_16195_end_mask_0 = const()[name = tensor("op_16195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16195_cast_fp16 = slice_by_index(begin = var_16195_begin_0, end = var_16195_end_0, end_mask = var_16195_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16195_cast_fp16")]; + tensor var_16199_begin_0 = const()[name = tensor("op_16199_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_16199_end_0 = const()[name = tensor("op_16199_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_16199_end_mask_0 = const()[name = tensor("op_16199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16199_cast_fp16 = slice_by_index(begin = var_16199_begin_0, end = var_16199_end_0, end_mask = var_16199_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16199_cast_fp16")]; + tensor var_16203_begin_0 = const()[name = tensor("op_16203_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_16203_end_0 = const()[name = tensor("op_16203_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_16203_end_mask_0 = const()[name = tensor("op_16203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16203_cast_fp16 = slice_by_index(begin = var_16203_begin_0, end = var_16203_end_0, end_mask = var_16203_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16203_cast_fp16")]; + tensor var_16205_begin_0 = const()[name = tensor("op_16205_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16205_end_0 = const()[name = tensor("op_16205_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_16205_end_mask_0 = const()[name = tensor("op_16205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16205_cast_fp16 = slice_by_index(begin = var_16205_begin_0, end = var_16205_end_0, end_mask = var_16205_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16205_cast_fp16")]; + tensor var_16209_begin_0 = const()[name = tensor("op_16209_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16209_end_0 = const()[name = tensor("op_16209_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_16209_end_mask_0 = const()[name = tensor("op_16209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16209_cast_fp16 = slice_by_index(begin = var_16209_begin_0, end = var_16209_end_0, end_mask = var_16209_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16209_cast_fp16")]; + tensor var_16213_begin_0 = const()[name = tensor("op_16213_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16213_end_0 = const()[name = tensor("op_16213_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_16213_end_mask_0 = const()[name = tensor("op_16213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16213_cast_fp16 = slice_by_index(begin = var_16213_begin_0, end = var_16213_end_0, end_mask = var_16213_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16213_cast_fp16")]; + tensor var_16217_begin_0 = const()[name = tensor("op_16217_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16217_end_0 = const()[name = tensor("op_16217_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_16217_end_mask_0 = const()[name = tensor("op_16217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16217_cast_fp16 = slice_by_index(begin = var_16217_begin_0, end = var_16217_end_0, end_mask = var_16217_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16217_cast_fp16")]; + tensor var_16221_begin_0 = const()[name = tensor("op_16221_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16221_end_0 = const()[name = tensor("op_16221_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_16221_end_mask_0 = const()[name = tensor("op_16221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16221_cast_fp16 = slice_by_index(begin = var_16221_begin_0, end = var_16221_end_0, end_mask = var_16221_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16221_cast_fp16")]; + tensor var_16225_begin_0 = const()[name = tensor("op_16225_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16225_end_0 = const()[name = tensor("op_16225_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_16225_end_mask_0 = const()[name = tensor("op_16225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16225_cast_fp16 = slice_by_index(begin = var_16225_begin_0, end = var_16225_end_0, end_mask = var_16225_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16225_cast_fp16")]; + tensor var_16229_begin_0 = const()[name = tensor("op_16229_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16229_end_0 = const()[name = tensor("op_16229_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_16229_end_mask_0 = const()[name = tensor("op_16229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16229_cast_fp16 = slice_by_index(begin = var_16229_begin_0, end = var_16229_end_0, end_mask = var_16229_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16229_cast_fp16")]; + tensor var_16233_begin_0 = const()[name = tensor("op_16233_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16233_end_0 = const()[name = tensor("op_16233_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_16233_end_mask_0 = const()[name = tensor("op_16233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16233_cast_fp16 = slice_by_index(begin = var_16233_begin_0, end = var_16233_end_0, end_mask = var_16233_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16233_cast_fp16")]; + tensor var_16237_begin_0 = const()[name = tensor("op_16237_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16237_end_0 = const()[name = tensor("op_16237_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_16237_end_mask_0 = const()[name = tensor("op_16237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16237_cast_fp16 = slice_by_index(begin = var_16237_begin_0, end = var_16237_end_0, end_mask = var_16237_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16237_cast_fp16")]; + tensor var_16241_begin_0 = const()[name = tensor("op_16241_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16241_end_0 = const()[name = tensor("op_16241_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_16241_end_mask_0 = const()[name = tensor("op_16241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16241_cast_fp16 = slice_by_index(begin = var_16241_begin_0, end = var_16241_end_0, end_mask = var_16241_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16241_cast_fp16")]; + tensor var_16245_begin_0 = const()[name = tensor("op_16245_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16245_end_0 = const()[name = tensor("op_16245_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_16245_end_mask_0 = const()[name = tensor("op_16245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16245_cast_fp16 = slice_by_index(begin = var_16245_begin_0, end = var_16245_end_0, end_mask = var_16245_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16245_cast_fp16")]; + tensor var_16249_begin_0 = const()[name = tensor("op_16249_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16249_end_0 = const()[name = tensor("op_16249_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_16249_end_mask_0 = const()[name = tensor("op_16249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16249_cast_fp16 = slice_by_index(begin = var_16249_begin_0, end = var_16249_end_0, end_mask = var_16249_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16249_cast_fp16")]; + tensor var_16253_begin_0 = const()[name = tensor("op_16253_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16253_end_0 = const()[name = tensor("op_16253_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_16253_end_mask_0 = const()[name = tensor("op_16253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16253_cast_fp16 = slice_by_index(begin = var_16253_begin_0, end = var_16253_end_0, end_mask = var_16253_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16253_cast_fp16")]; + tensor var_16257_begin_0 = const()[name = tensor("op_16257_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16257_end_0 = const()[name = tensor("op_16257_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_16257_end_mask_0 = const()[name = tensor("op_16257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16257_cast_fp16 = slice_by_index(begin = var_16257_begin_0, end = var_16257_end_0, end_mask = var_16257_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16257_cast_fp16")]; + tensor var_16261_begin_0 = const()[name = tensor("op_16261_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16261_end_0 = const()[name = tensor("op_16261_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_16261_end_mask_0 = const()[name = tensor("op_16261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16261_cast_fp16 = slice_by_index(begin = var_16261_begin_0, end = var_16261_end_0, end_mask = var_16261_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16261_cast_fp16")]; + tensor var_16265_begin_0 = const()[name = tensor("op_16265_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16265_end_0 = const()[name = tensor("op_16265_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_16265_end_mask_0 = const()[name = tensor("op_16265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16265_cast_fp16 = slice_by_index(begin = var_16265_begin_0, end = var_16265_end_0, end_mask = var_16265_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16265_cast_fp16")]; + tensor var_16269_begin_0 = const()[name = tensor("op_16269_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16269_end_0 = const()[name = tensor("op_16269_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_16269_end_mask_0 = const()[name = tensor("op_16269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16269_cast_fp16 = slice_by_index(begin = var_16269_begin_0, end = var_16269_end_0, end_mask = var_16269_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16269_cast_fp16")]; + tensor var_16273_begin_0 = const()[name = tensor("op_16273_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16273_end_0 = const()[name = tensor("op_16273_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_16273_end_mask_0 = const()[name = tensor("op_16273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16273_cast_fp16 = slice_by_index(begin = var_16273_begin_0, end = var_16273_end_0, end_mask = var_16273_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16273_cast_fp16")]; + tensor var_16277_begin_0 = const()[name = tensor("op_16277_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16277_end_0 = const()[name = tensor("op_16277_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_16277_end_mask_0 = const()[name = tensor("op_16277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16277_cast_fp16 = slice_by_index(begin = var_16277_begin_0, end = var_16277_end_0, end_mask = var_16277_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16277_cast_fp16")]; + tensor var_16281_begin_0 = const()[name = tensor("op_16281_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16281_end_0 = const()[name = tensor("op_16281_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_16281_end_mask_0 = const()[name = tensor("op_16281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16281_cast_fp16 = slice_by_index(begin = var_16281_begin_0, end = var_16281_end_0, end_mask = var_16281_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16281_cast_fp16")]; + tensor var_16285_equation_0 = const()[name = tensor("op_16285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16285_cast_fp16 = einsum(equation = var_16285_equation_0, values = (var_16127_cast_fp16, var_16044_cast_fp16))[name = tensor("op_16285_cast_fp16")]; + tensor var_16286_to_fp16 = const()[name = tensor("op_16286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1321_cast_fp16 = mul(x = var_16285_cast_fp16, y = var_16286_to_fp16)[name = tensor("aw_1321_cast_fp16")]; + tensor var_16289_equation_0 = const()[name = tensor("op_16289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16289_cast_fp16 = einsum(equation = var_16289_equation_0, values = (var_16131_cast_fp16, var_16048_cast_fp16))[name = tensor("op_16289_cast_fp16")]; + tensor var_16290_to_fp16 = const()[name = tensor("op_16290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1323_cast_fp16 = mul(x = var_16289_cast_fp16, y = var_16290_to_fp16)[name = tensor("aw_1323_cast_fp16")]; + tensor var_16293_equation_0 = const()[name = tensor("op_16293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16293_cast_fp16 = einsum(equation = var_16293_equation_0, values = (var_16135_cast_fp16, var_16052_cast_fp16))[name = tensor("op_16293_cast_fp16")]; + tensor var_16294_to_fp16 = const()[name = tensor("op_16294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1325_cast_fp16 = mul(x = var_16293_cast_fp16, y = var_16294_to_fp16)[name = tensor("aw_1325_cast_fp16")]; + tensor var_16297_equation_0 = const()[name = tensor("op_16297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16297_cast_fp16 = einsum(equation = var_16297_equation_0, values = (var_16139_cast_fp16, var_16056_cast_fp16))[name = tensor("op_16297_cast_fp16")]; + tensor var_16298_to_fp16 = const()[name = tensor("op_16298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1327_cast_fp16 = mul(x = var_16297_cast_fp16, y = var_16298_to_fp16)[name = tensor("aw_1327_cast_fp16")]; + tensor var_16301_equation_0 = const()[name = tensor("op_16301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16301_cast_fp16 = einsum(equation = var_16301_equation_0, values = (var_16143_cast_fp16, var_16060_cast_fp16))[name = tensor("op_16301_cast_fp16")]; + tensor var_16302_to_fp16 = const()[name = tensor("op_16302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1329_cast_fp16 = mul(x = var_16301_cast_fp16, y = var_16302_to_fp16)[name = tensor("aw_1329_cast_fp16")]; + tensor var_16305_equation_0 = const()[name = tensor("op_16305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16305_cast_fp16 = einsum(equation = var_16305_equation_0, values = (var_16147_cast_fp16, var_16064_cast_fp16))[name = tensor("op_16305_cast_fp16")]; + tensor var_16306_to_fp16 = const()[name = tensor("op_16306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1331_cast_fp16 = mul(x = var_16305_cast_fp16, y = var_16306_to_fp16)[name = tensor("aw_1331_cast_fp16")]; + tensor var_16309_equation_0 = const()[name = tensor("op_16309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16309_cast_fp16 = einsum(equation = var_16309_equation_0, values = (var_16151_cast_fp16, var_16068_cast_fp16))[name = tensor("op_16309_cast_fp16")]; + tensor var_16310_to_fp16 = const()[name = tensor("op_16310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1333_cast_fp16 = mul(x = var_16309_cast_fp16, y = var_16310_to_fp16)[name = tensor("aw_1333_cast_fp16")]; + tensor var_16313_equation_0 = const()[name = tensor("op_16313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16313_cast_fp16 = einsum(equation = var_16313_equation_0, values = (var_16155_cast_fp16, var_16072_cast_fp16))[name = tensor("op_16313_cast_fp16")]; + tensor var_16314_to_fp16 = const()[name = tensor("op_16314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1335_cast_fp16 = mul(x = var_16313_cast_fp16, y = var_16314_to_fp16)[name = tensor("aw_1335_cast_fp16")]; + tensor var_16317_equation_0 = const()[name = tensor("op_16317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16317_cast_fp16 = einsum(equation = var_16317_equation_0, values = (var_16159_cast_fp16, var_16076_cast_fp16))[name = tensor("op_16317_cast_fp16")]; + tensor var_16318_to_fp16 = const()[name = tensor("op_16318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1337_cast_fp16 = mul(x = var_16317_cast_fp16, y = var_16318_to_fp16)[name = tensor("aw_1337_cast_fp16")]; + tensor var_16321_equation_0 = const()[name = tensor("op_16321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16321_cast_fp16 = einsum(equation = var_16321_equation_0, values = (var_16163_cast_fp16, var_16080_cast_fp16))[name = tensor("op_16321_cast_fp16")]; + tensor var_16322_to_fp16 = const()[name = tensor("op_16322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1339_cast_fp16 = mul(x = var_16321_cast_fp16, y = var_16322_to_fp16)[name = tensor("aw_1339_cast_fp16")]; + tensor var_16325_equation_0 = const()[name = tensor("op_16325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16325_cast_fp16 = einsum(equation = var_16325_equation_0, values = (var_16167_cast_fp16, var_16084_cast_fp16))[name = tensor("op_16325_cast_fp16")]; + tensor var_16326_to_fp16 = const()[name = tensor("op_16326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1341_cast_fp16 = mul(x = var_16325_cast_fp16, y = var_16326_to_fp16)[name = tensor("aw_1341_cast_fp16")]; + tensor var_16329_equation_0 = const()[name = tensor("op_16329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16329_cast_fp16 = einsum(equation = var_16329_equation_0, values = (var_16171_cast_fp16, var_16088_cast_fp16))[name = tensor("op_16329_cast_fp16")]; + tensor var_16330_to_fp16 = const()[name = tensor("op_16330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1343_cast_fp16 = mul(x = var_16329_cast_fp16, y = var_16330_to_fp16)[name = tensor("aw_1343_cast_fp16")]; + tensor var_16333_equation_0 = const()[name = tensor("op_16333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16333_cast_fp16 = einsum(equation = var_16333_equation_0, values = (var_16175_cast_fp16, var_16092_cast_fp16))[name = tensor("op_16333_cast_fp16")]; + tensor var_16334_to_fp16 = const()[name = tensor("op_16334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1345_cast_fp16 = mul(x = var_16333_cast_fp16, y = var_16334_to_fp16)[name = tensor("aw_1345_cast_fp16")]; + tensor var_16337_equation_0 = const()[name = tensor("op_16337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16337_cast_fp16 = einsum(equation = var_16337_equation_0, values = (var_16179_cast_fp16, var_16096_cast_fp16))[name = tensor("op_16337_cast_fp16")]; + tensor var_16338_to_fp16 = const()[name = tensor("op_16338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1347_cast_fp16 = mul(x = var_16337_cast_fp16, y = var_16338_to_fp16)[name = tensor("aw_1347_cast_fp16")]; + tensor var_16341_equation_0 = const()[name = tensor("op_16341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16341_cast_fp16 = einsum(equation = var_16341_equation_0, values = (var_16183_cast_fp16, var_16100_cast_fp16))[name = tensor("op_16341_cast_fp16")]; + tensor var_16342_to_fp16 = const()[name = tensor("op_16342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1349_cast_fp16 = mul(x = var_16341_cast_fp16, y = var_16342_to_fp16)[name = tensor("aw_1349_cast_fp16")]; + tensor var_16345_equation_0 = const()[name = tensor("op_16345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16345_cast_fp16 = einsum(equation = var_16345_equation_0, values = (var_16187_cast_fp16, var_16104_cast_fp16))[name = tensor("op_16345_cast_fp16")]; + tensor var_16346_to_fp16 = const()[name = tensor("op_16346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1351_cast_fp16 = mul(x = var_16345_cast_fp16, y = var_16346_to_fp16)[name = tensor("aw_1351_cast_fp16")]; + tensor var_16349_equation_0 = const()[name = tensor("op_16349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16349_cast_fp16 = einsum(equation = var_16349_equation_0, values = (var_16191_cast_fp16, var_16108_cast_fp16))[name = tensor("op_16349_cast_fp16")]; + tensor var_16350_to_fp16 = const()[name = tensor("op_16350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1353_cast_fp16 = mul(x = var_16349_cast_fp16, y = var_16350_to_fp16)[name = tensor("aw_1353_cast_fp16")]; + tensor var_16353_equation_0 = const()[name = tensor("op_16353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16353_cast_fp16 = einsum(equation = var_16353_equation_0, values = (var_16195_cast_fp16, var_16112_cast_fp16))[name = tensor("op_16353_cast_fp16")]; + tensor var_16354_to_fp16 = const()[name = tensor("op_16354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1355_cast_fp16 = mul(x = var_16353_cast_fp16, y = var_16354_to_fp16)[name = tensor("aw_1355_cast_fp16")]; + tensor var_16357_equation_0 = const()[name = tensor("op_16357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16357_cast_fp16 = einsum(equation = var_16357_equation_0, values = (var_16199_cast_fp16, var_16116_cast_fp16))[name = tensor("op_16357_cast_fp16")]; + tensor var_16358_to_fp16 = const()[name = tensor("op_16358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1357_cast_fp16 = mul(x = var_16357_cast_fp16, y = var_16358_to_fp16)[name = tensor("aw_1357_cast_fp16")]; + tensor var_16361_equation_0 = const()[name = tensor("op_16361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16361_cast_fp16 = einsum(equation = var_16361_equation_0, values = (var_16203_cast_fp16, var_16120_cast_fp16))[name = tensor("op_16361_cast_fp16")]; + tensor var_16362_to_fp16 = const()[name = tensor("op_16362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1359_cast_fp16 = mul(x = var_16361_cast_fp16, y = var_16362_to_fp16)[name = tensor("aw_1359_cast_fp16")]; + tensor var_16364_cast_fp16 = softmax(axis = var_2624, x = aw_1321_cast_fp16)[name = tensor("op_16364_cast_fp16")]; + tensor var_16365_cast_fp16 = softmax(axis = var_2624, x = aw_1323_cast_fp16)[name = tensor("op_16365_cast_fp16")]; + tensor var_16366_cast_fp16 = softmax(axis = var_2624, x = aw_1325_cast_fp16)[name = tensor("op_16366_cast_fp16")]; + tensor var_16367_cast_fp16 = softmax(axis = var_2624, x = aw_1327_cast_fp16)[name = tensor("op_16367_cast_fp16")]; + tensor var_16368_cast_fp16 = softmax(axis = var_2624, x = aw_1329_cast_fp16)[name = tensor("op_16368_cast_fp16")]; + tensor var_16369_cast_fp16 = softmax(axis = var_2624, x = aw_1331_cast_fp16)[name = tensor("op_16369_cast_fp16")]; + tensor var_16370_cast_fp16 = softmax(axis = var_2624, x = aw_1333_cast_fp16)[name = tensor("op_16370_cast_fp16")]; + tensor var_16371_cast_fp16 = softmax(axis = var_2624, x = aw_1335_cast_fp16)[name = tensor("op_16371_cast_fp16")]; + tensor var_16372_cast_fp16 = softmax(axis = var_2624, x = aw_1337_cast_fp16)[name = tensor("op_16372_cast_fp16")]; + tensor var_16373_cast_fp16 = softmax(axis = var_2624, x = aw_1339_cast_fp16)[name = tensor("op_16373_cast_fp16")]; + tensor var_16374_cast_fp16 = softmax(axis = var_2624, x = aw_1341_cast_fp16)[name = tensor("op_16374_cast_fp16")]; + tensor var_16375_cast_fp16 = softmax(axis = var_2624, x = aw_1343_cast_fp16)[name = tensor("op_16375_cast_fp16")]; + tensor var_16376_cast_fp16 = softmax(axis = var_2624, x = aw_1345_cast_fp16)[name = tensor("op_16376_cast_fp16")]; + tensor var_16377_cast_fp16 = softmax(axis = var_2624, x = aw_1347_cast_fp16)[name = tensor("op_16377_cast_fp16")]; + tensor var_16378_cast_fp16 = softmax(axis = var_2624, x = aw_1349_cast_fp16)[name = tensor("op_16378_cast_fp16")]; + tensor var_16379_cast_fp16 = softmax(axis = var_2624, x = aw_1351_cast_fp16)[name = tensor("op_16379_cast_fp16")]; + tensor var_16380_cast_fp16 = softmax(axis = var_2624, x = aw_1353_cast_fp16)[name = tensor("op_16380_cast_fp16")]; + tensor var_16381_cast_fp16 = softmax(axis = var_2624, x = aw_1355_cast_fp16)[name = tensor("op_16381_cast_fp16")]; + tensor var_16382_cast_fp16 = softmax(axis = var_2624, x = aw_1357_cast_fp16)[name = tensor("op_16382_cast_fp16")]; + tensor var_16383_cast_fp16 = softmax(axis = var_2624, x = aw_1359_cast_fp16)[name = tensor("op_16383_cast_fp16")]; + tensor var_16385_equation_0 = const()[name = tensor("op_16385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16385_cast_fp16 = einsum(equation = var_16385_equation_0, values = (var_16205_cast_fp16, var_16364_cast_fp16))[name = tensor("op_16385_cast_fp16")]; + tensor var_16387_equation_0 = const()[name = tensor("op_16387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16387_cast_fp16 = einsum(equation = var_16387_equation_0, values = (var_16209_cast_fp16, var_16365_cast_fp16))[name = tensor("op_16387_cast_fp16")]; + tensor var_16389_equation_0 = const()[name = tensor("op_16389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16389_cast_fp16 = einsum(equation = var_16389_equation_0, values = (var_16213_cast_fp16, var_16366_cast_fp16))[name = tensor("op_16389_cast_fp16")]; + tensor var_16391_equation_0 = const()[name = tensor("op_16391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16391_cast_fp16 = einsum(equation = var_16391_equation_0, values = (var_16217_cast_fp16, var_16367_cast_fp16))[name = tensor("op_16391_cast_fp16")]; + tensor var_16393_equation_0 = const()[name = tensor("op_16393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16393_cast_fp16 = einsum(equation = var_16393_equation_0, values = (var_16221_cast_fp16, var_16368_cast_fp16))[name = tensor("op_16393_cast_fp16")]; + tensor var_16395_equation_0 = const()[name = tensor("op_16395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16395_cast_fp16 = einsum(equation = var_16395_equation_0, values = (var_16225_cast_fp16, var_16369_cast_fp16))[name = tensor("op_16395_cast_fp16")]; + tensor var_16397_equation_0 = const()[name = tensor("op_16397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16397_cast_fp16 = einsum(equation = var_16397_equation_0, values = (var_16229_cast_fp16, var_16370_cast_fp16))[name = tensor("op_16397_cast_fp16")]; + tensor var_16399_equation_0 = const()[name = tensor("op_16399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16399_cast_fp16 = einsum(equation = var_16399_equation_0, values = (var_16233_cast_fp16, var_16371_cast_fp16))[name = tensor("op_16399_cast_fp16")]; + tensor var_16401_equation_0 = const()[name = tensor("op_16401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16401_cast_fp16 = einsum(equation = var_16401_equation_0, values = (var_16237_cast_fp16, var_16372_cast_fp16))[name = tensor("op_16401_cast_fp16")]; + tensor var_16403_equation_0 = const()[name = tensor("op_16403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16403_cast_fp16 = einsum(equation = var_16403_equation_0, values = (var_16241_cast_fp16, var_16373_cast_fp16))[name = tensor("op_16403_cast_fp16")]; + tensor var_16405_equation_0 = const()[name = tensor("op_16405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16405_cast_fp16 = einsum(equation = var_16405_equation_0, values = (var_16245_cast_fp16, var_16374_cast_fp16))[name = tensor("op_16405_cast_fp16")]; + tensor var_16407_equation_0 = const()[name = tensor("op_16407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16407_cast_fp16 = einsum(equation = var_16407_equation_0, values = (var_16249_cast_fp16, var_16375_cast_fp16))[name = tensor("op_16407_cast_fp16")]; + tensor var_16409_equation_0 = const()[name = tensor("op_16409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16409_cast_fp16 = einsum(equation = var_16409_equation_0, values = (var_16253_cast_fp16, var_16376_cast_fp16))[name = tensor("op_16409_cast_fp16")]; + tensor var_16411_equation_0 = const()[name = tensor("op_16411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16411_cast_fp16 = einsum(equation = var_16411_equation_0, values = (var_16257_cast_fp16, var_16377_cast_fp16))[name = tensor("op_16411_cast_fp16")]; + tensor var_16413_equation_0 = const()[name = tensor("op_16413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16413_cast_fp16 = einsum(equation = var_16413_equation_0, values = (var_16261_cast_fp16, var_16378_cast_fp16))[name = tensor("op_16413_cast_fp16")]; + tensor var_16415_equation_0 = const()[name = tensor("op_16415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16415_cast_fp16 = einsum(equation = var_16415_equation_0, values = (var_16265_cast_fp16, var_16379_cast_fp16))[name = tensor("op_16415_cast_fp16")]; + tensor var_16417_equation_0 = const()[name = tensor("op_16417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16417_cast_fp16 = einsum(equation = var_16417_equation_0, values = (var_16269_cast_fp16, var_16380_cast_fp16))[name = tensor("op_16417_cast_fp16")]; + tensor var_16419_equation_0 = const()[name = tensor("op_16419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16419_cast_fp16 = einsum(equation = var_16419_equation_0, values = (var_16273_cast_fp16, var_16381_cast_fp16))[name = tensor("op_16419_cast_fp16")]; + tensor var_16421_equation_0 = const()[name = tensor("op_16421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16421_cast_fp16 = einsum(equation = var_16421_equation_0, values = (var_16277_cast_fp16, var_16382_cast_fp16))[name = tensor("op_16421_cast_fp16")]; + tensor var_16423_equation_0 = const()[name = tensor("op_16423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16423_cast_fp16 = einsum(equation = var_16423_equation_0, values = (var_16281_cast_fp16, var_16383_cast_fp16))[name = tensor("op_16423_cast_fp16")]; + tensor input_263_interleave_0 = const()[name = tensor("input_263_interleave_0"), val = tensor(false)]; + tensor input_263_cast_fp16 = concat(axis = var_2624, interleave = input_263_interleave_0, values = (var_16385_cast_fp16, var_16387_cast_fp16, var_16389_cast_fp16, var_16391_cast_fp16, var_16393_cast_fp16, var_16395_cast_fp16, var_16397_cast_fp16, var_16399_cast_fp16, var_16401_cast_fp16, var_16403_cast_fp16, var_16405_cast_fp16, var_16407_cast_fp16, var_16409_cast_fp16, var_16411_cast_fp16, var_16413_cast_fp16, var_16415_cast_fp16, var_16417_cast_fp16, var_16419_cast_fp16, var_16421_cast_fp16, var_16423_cast_fp16))[name = tensor("input_263_cast_fp16")]; + tensor var_16433_pad_type_0 = const()[name = tensor("op_16433_pad_type_0"), val = tensor("valid")]; + tensor var_16433_strides_0 = const()[name = tensor("op_16433_strides_0"), val = tensor([1, 1])]; + tensor var_16433_pad_0 = const()[name = tensor("op_16433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16433_dilations_0 = const()[name = tensor("op_16433_dilations_0"), val = tensor([1, 1])]; + tensor var_16433_groups_0 = const()[name = tensor("op_16433_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475727872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476956736))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476956928)))]; + tensor var_16433_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_16433_dilations_0, groups = var_16433_groups_0, pad = var_16433_pad_0, pad_type = var_16433_pad_type_0, strides = var_16433_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized, x = input_263_cast_fp16)[name = tensor("op_16433_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = var_16433_cast_fp16, y = inputs_111_cast_fp16)[name = tensor("inputs_113_cast_fp16")]; + tensor input_265_axes_0 = const()[name = tensor("input_265_axes_0"), val = tensor([1])]; + tensor input_265_gamma_0_to_fp16 = const()[name = tensor("input_265_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476959552)))]; + tensor input_265_beta_0_to_fp16 = const()[name = tensor("input_265_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476962176)))]; + tensor var_16443_to_fp16 = const()[name = tensor("op_16443_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_265_cast_fp16 = layer_norm(axes = input_265_axes_0, beta = input_265_beta_0_to_fp16, epsilon = var_16443_to_fp16, gamma = input_265_gamma_0_to_fp16, x = inputs_113_cast_fp16)[name = tensor("input_265_cast_fp16")]; + tensor var_16463_pad_type_0 = const()[name = tensor("op_16463_pad_type_0"), val = tensor("valid")]; + tensor var_16463_strides_0 = const()[name = tensor("op_16463_strides_0"), val = tensor([1, 1])]; + tensor var_16463_pad_0 = const()[name = tensor("op_16463_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16463_dilations_0 = const()[name = tensor("op_16463_dilations_0"), val = tensor([1, 1])]; + tensor var_16463_groups_0 = const()[name = tensor("op_16463_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476964800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486795264))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486795456)))]; + tensor var_16463_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_16463_dilations_0, groups = var_16463_groups_0, pad = var_16463_pad_0, pad_type = var_16463_pad_type_0, strides = var_16463_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized, x = input_265_cast_fp16)[name = tensor("op_16463_cast_fp16")]; + tensor var_16464_split_sizes_0 = const()[name = tensor("op_16464_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_16464_axis_0 = const()[name = tensor("op_16464_axis_0"), val = tensor(1)]; + tensor var_16464_cast_fp16_0, tensor var_16464_cast_fp16_1 = split(axis = var_16464_axis_0, split_sizes = var_16464_split_sizes_0, x = var_16463_cast_fp16)[name = tensor("op_16464_cast_fp16")]; + tensor var_16466_mode_0 = const()[name = tensor("op_16466_mode_0"), val = tensor("EXACT")]; + tensor var_16466_cast_fp16 = gelu(mode = var_16466_mode_0, x = var_16464_cast_fp16_1)[name = tensor("op_16466_cast_fp16")]; + tensor input_267_cast_fp16 = mul(x = var_16464_cast_fp16_0, y = var_16466_cast_fp16)[name = tensor("input_267_cast_fp16")]; + tensor var_16474_pad_type_0 = const()[name = tensor("op_16474_pad_type_0"), val = tensor("valid")]; + tensor var_16474_strides_0 = const()[name = tensor("op_16474_strides_0"), val = tensor([1, 1])]; + tensor var_16474_pad_0 = const()[name = tensor("op_16474_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16474_dilations_0 = const()[name = tensor("op_16474_dilations_0"), val = tensor([1, 1])]; + tensor var_16474_groups_0 = const()[name = tensor("op_16474_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486816000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491731264))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491731456)))]; + tensor var_16474_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_16474_dilations_0, groups = var_16474_groups_0, pad = var_16474_pad_0, pad_type = var_16474_pad_type_0, strides = var_16474_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized, x = input_267_cast_fp16)[name = tensor("op_16474_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = var_16474_cast_fp16, y = inputs_113_cast_fp16)[name = tensor("inputs_115_cast_fp16")]; + tensor hidden_states_167_axes_0 = const()[name = tensor("hidden_states_167_axes_0"), val = tensor([1])]; + tensor hidden_states_167_gamma_0_to_fp16 = const()[name = tensor("hidden_states_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491734080)))]; + tensor hidden_states_167_beta_0_to_fp16 = const()[name = tensor("hidden_states_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491736704)))]; + tensor var_16490_to_fp16 = const()[name = tensor("op_16490_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_167_cast_fp16 = layer_norm(axes = hidden_states_167_axes_0, beta = hidden_states_167_beta_0_to_fp16, epsilon = var_16490_to_fp16, gamma = hidden_states_167_gamma_0_to_fp16, x = inputs_115_cast_fp16)[name = tensor("hidden_states_167_cast_fp16")]; + tensor q_77_pad_type_0 = const()[name = tensor("q_77_pad_type_0"), val = tensor("valid")]; + tensor q_77_strides_0 = const()[name = tensor("q_77_strides_0"), val = tensor([1, 1])]; + tensor q_77_pad_0 = const()[name = tensor("q_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_77_dilations_0 = const()[name = tensor("q_77_dilations_0"), val = tensor([1, 1])]; + tensor q_77_groups_0 = const()[name = tensor("q_77_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491739328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492968192))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_77_cast_fp16 = conv(dilations = q_77_dilations_0, groups = q_77_groups_0, pad = q_77_pad_0, pad_type = q_77_pad_type_0, strides = q_77_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_167_cast_fp16)[name = tensor("q_77_cast_fp16")]; + tensor k_153_pad_type_0 = const()[name = tensor("k_153_pad_type_0"), val = tensor("valid")]; + tensor k_153_strides_0 = const()[name = tensor("k_153_strides_0"), val = tensor([1, 1])]; + tensor k_153_pad_0 = const()[name = tensor("k_153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_153_dilations_0 = const()[name = tensor("k_153_dilations_0"), val = tensor([1, 1])]; + tensor k_153_groups_0 = const()[name = tensor("k_153_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492968384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494197248))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_153_cast_fp16 = conv(dilations = k_153_dilations_0, groups = k_153_groups_0, pad = k_153_pad_0, pad_type = k_153_pad_type_0, strides = k_153_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_167_cast_fp16)[name = tensor("k_153_cast_fp16")]; + tensor v_77_pad_type_0 = const()[name = tensor("v_77_pad_type_0"), val = tensor("valid")]; + tensor v_77_strides_0 = const()[name = tensor("v_77_strides_0"), val = tensor([1, 1])]; + tensor v_77_pad_0 = const()[name = tensor("v_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_77_dilations_0 = const()[name = tensor("v_77_dilations_0"), val = tensor([1, 1])]; + tensor v_77_groups_0 = const()[name = tensor("v_77_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494197440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(495426304))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_77_cast_fp16 = conv(dilations = v_77_dilations_0, groups = v_77_groups_0, pad = v_77_pad_0, pad_type = v_77_pad_type_0, strides = v_77_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_167_cast_fp16)[name = tensor("v_77_cast_fp16")]; + tensor var_16523_begin_0 = const()[name = tensor("op_16523_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16523_end_0 = const()[name = tensor("op_16523_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16523_end_mask_0 = const()[name = tensor("op_16523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16523_cast_fp16 = slice_by_index(begin = var_16523_begin_0, end = var_16523_end_0, end_mask = var_16523_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16523_cast_fp16")]; + tensor var_16527_begin_0 = const()[name = tensor("op_16527_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16527_end_0 = const()[name = tensor("op_16527_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16527_end_mask_0 = const()[name = tensor("op_16527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16527_cast_fp16 = slice_by_index(begin = var_16527_begin_0, end = var_16527_end_0, end_mask = var_16527_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16527_cast_fp16")]; + tensor var_16531_begin_0 = const()[name = tensor("op_16531_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16531_end_0 = const()[name = tensor("op_16531_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16531_end_mask_0 = const()[name = tensor("op_16531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16531_cast_fp16 = slice_by_index(begin = var_16531_begin_0, end = var_16531_end_0, end_mask = var_16531_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16531_cast_fp16")]; + tensor var_16535_begin_0 = const()[name = tensor("op_16535_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16535_end_0 = const()[name = tensor("op_16535_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16535_end_mask_0 = const()[name = tensor("op_16535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16535_cast_fp16 = slice_by_index(begin = var_16535_begin_0, end = var_16535_end_0, end_mask = var_16535_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16535_cast_fp16")]; + tensor var_16539_begin_0 = const()[name = tensor("op_16539_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16539_end_0 = const()[name = tensor("op_16539_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16539_end_mask_0 = const()[name = tensor("op_16539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16539_cast_fp16 = slice_by_index(begin = var_16539_begin_0, end = var_16539_end_0, end_mask = var_16539_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16539_cast_fp16")]; + tensor var_16543_begin_0 = const()[name = tensor("op_16543_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16543_end_0 = const()[name = tensor("op_16543_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16543_end_mask_0 = const()[name = tensor("op_16543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16543_cast_fp16 = slice_by_index(begin = var_16543_begin_0, end = var_16543_end_0, end_mask = var_16543_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16543_cast_fp16")]; + tensor var_16547_begin_0 = const()[name = tensor("op_16547_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16547_end_0 = const()[name = tensor("op_16547_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16547_end_mask_0 = const()[name = tensor("op_16547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16547_cast_fp16 = slice_by_index(begin = var_16547_begin_0, end = var_16547_end_0, end_mask = var_16547_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16547_cast_fp16")]; + tensor var_16551_begin_0 = const()[name = tensor("op_16551_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16551_end_0 = const()[name = tensor("op_16551_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16551_end_mask_0 = const()[name = tensor("op_16551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16551_cast_fp16 = slice_by_index(begin = var_16551_begin_0, end = var_16551_end_0, end_mask = var_16551_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16551_cast_fp16")]; + tensor var_16555_begin_0 = const()[name = tensor("op_16555_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16555_end_0 = const()[name = tensor("op_16555_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16555_end_mask_0 = const()[name = tensor("op_16555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16555_cast_fp16 = slice_by_index(begin = var_16555_begin_0, end = var_16555_end_0, end_mask = var_16555_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16555_cast_fp16")]; + tensor var_16559_begin_0 = const()[name = tensor("op_16559_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16559_end_0 = const()[name = tensor("op_16559_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16559_end_mask_0 = const()[name = tensor("op_16559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16559_cast_fp16 = slice_by_index(begin = var_16559_begin_0, end = var_16559_end_0, end_mask = var_16559_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16559_cast_fp16")]; + tensor var_16563_begin_0 = const()[name = tensor("op_16563_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16563_end_0 = const()[name = tensor("op_16563_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16563_end_mask_0 = const()[name = tensor("op_16563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16563_cast_fp16 = slice_by_index(begin = var_16563_begin_0, end = var_16563_end_0, end_mask = var_16563_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16563_cast_fp16")]; + tensor var_16567_begin_0 = const()[name = tensor("op_16567_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16567_end_0 = const()[name = tensor("op_16567_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16567_end_mask_0 = const()[name = tensor("op_16567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16567_cast_fp16 = slice_by_index(begin = var_16567_begin_0, end = var_16567_end_0, end_mask = var_16567_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16567_cast_fp16")]; + tensor var_16571_begin_0 = const()[name = tensor("op_16571_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16571_end_0 = const()[name = tensor("op_16571_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_16571_end_mask_0 = const()[name = tensor("op_16571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16571_cast_fp16 = slice_by_index(begin = var_16571_begin_0, end = var_16571_end_0, end_mask = var_16571_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16571_cast_fp16")]; + tensor var_16575_begin_0 = const()[name = tensor("op_16575_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16575_end_0 = const()[name = tensor("op_16575_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_16575_end_mask_0 = const()[name = tensor("op_16575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16575_cast_fp16 = slice_by_index(begin = var_16575_begin_0, end = var_16575_end_0, end_mask = var_16575_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16575_cast_fp16")]; + tensor var_16579_begin_0 = const()[name = tensor("op_16579_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16579_end_0 = const()[name = tensor("op_16579_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_16579_end_mask_0 = const()[name = tensor("op_16579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16579_cast_fp16 = slice_by_index(begin = var_16579_begin_0, end = var_16579_end_0, end_mask = var_16579_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16579_cast_fp16")]; + tensor var_16583_begin_0 = const()[name = tensor("op_16583_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16583_end_0 = const()[name = tensor("op_16583_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16583_end_mask_0 = const()[name = tensor("op_16583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16583_cast_fp16 = slice_by_index(begin = var_16583_begin_0, end = var_16583_end_0, end_mask = var_16583_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16583_cast_fp16")]; + tensor var_16587_begin_0 = const()[name = tensor("op_16587_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16587_end_0 = const()[name = tensor("op_16587_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_16587_end_mask_0 = const()[name = tensor("op_16587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16587_cast_fp16 = slice_by_index(begin = var_16587_begin_0, end = var_16587_end_0, end_mask = var_16587_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16587_cast_fp16")]; + tensor var_16591_begin_0 = const()[name = tensor("op_16591_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16591_end_0 = const()[name = tensor("op_16591_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_16591_end_mask_0 = const()[name = tensor("op_16591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16591_cast_fp16 = slice_by_index(begin = var_16591_begin_0, end = var_16591_end_0, end_mask = var_16591_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16591_cast_fp16")]; + tensor var_16595_begin_0 = const()[name = tensor("op_16595_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16595_end_0 = const()[name = tensor("op_16595_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_16595_end_mask_0 = const()[name = tensor("op_16595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16595_cast_fp16 = slice_by_index(begin = var_16595_begin_0, end = var_16595_end_0, end_mask = var_16595_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16595_cast_fp16")]; + tensor var_16599_begin_0 = const()[name = tensor("op_16599_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16599_end_0 = const()[name = tensor("op_16599_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_16599_end_mask_0 = const()[name = tensor("op_16599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16599_cast_fp16 = slice_by_index(begin = var_16599_begin_0, end = var_16599_end_0, end_mask = var_16599_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16599_cast_fp16")]; + tensor k_155_perm_0 = const()[name = tensor("k_155_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_16606_begin_0 = const()[name = tensor("op_16606_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16606_end_0 = const()[name = tensor("op_16606_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_16606_end_mask_0 = const()[name = tensor("op_16606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_155_cast_fp16 = transpose(perm = k_155_perm_0, x = k_153_cast_fp16)[name = tensor("transpose_29")]; + tensor var_16606_cast_fp16 = slice_by_index(begin = var_16606_begin_0, end = var_16606_end_0, end_mask = var_16606_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16606_cast_fp16")]; + tensor var_16610_begin_0 = const()[name = tensor("op_16610_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_16610_end_0 = const()[name = tensor("op_16610_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_16610_end_mask_0 = const()[name = tensor("op_16610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16610_cast_fp16 = slice_by_index(begin = var_16610_begin_0, end = var_16610_end_0, end_mask = var_16610_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16610_cast_fp16")]; + tensor var_16614_begin_0 = const()[name = tensor("op_16614_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_16614_end_0 = const()[name = tensor("op_16614_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_16614_end_mask_0 = const()[name = tensor("op_16614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16614_cast_fp16 = slice_by_index(begin = var_16614_begin_0, end = var_16614_end_0, end_mask = var_16614_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16614_cast_fp16")]; + tensor var_16618_begin_0 = const()[name = tensor("op_16618_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_16618_end_0 = const()[name = tensor("op_16618_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_16618_end_mask_0 = const()[name = tensor("op_16618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16618_cast_fp16 = slice_by_index(begin = var_16618_begin_0, end = var_16618_end_0, end_mask = var_16618_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16618_cast_fp16")]; + tensor var_16622_begin_0 = const()[name = tensor("op_16622_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_16622_end_0 = const()[name = tensor("op_16622_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_16622_end_mask_0 = const()[name = tensor("op_16622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16622_cast_fp16 = slice_by_index(begin = var_16622_begin_0, end = var_16622_end_0, end_mask = var_16622_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16622_cast_fp16")]; + tensor var_16626_begin_0 = const()[name = tensor("op_16626_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_16626_end_0 = const()[name = tensor("op_16626_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_16626_end_mask_0 = const()[name = tensor("op_16626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16626_cast_fp16 = slice_by_index(begin = var_16626_begin_0, end = var_16626_end_0, end_mask = var_16626_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16626_cast_fp16")]; + tensor var_16630_begin_0 = const()[name = tensor("op_16630_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_16630_end_0 = const()[name = tensor("op_16630_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_16630_end_mask_0 = const()[name = tensor("op_16630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16630_cast_fp16 = slice_by_index(begin = var_16630_begin_0, end = var_16630_end_0, end_mask = var_16630_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16630_cast_fp16")]; + tensor var_16634_begin_0 = const()[name = tensor("op_16634_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_16634_end_0 = const()[name = tensor("op_16634_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_16634_end_mask_0 = const()[name = tensor("op_16634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16634_cast_fp16 = slice_by_index(begin = var_16634_begin_0, end = var_16634_end_0, end_mask = var_16634_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16634_cast_fp16")]; + tensor var_16638_begin_0 = const()[name = tensor("op_16638_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_16638_end_0 = const()[name = tensor("op_16638_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_16638_end_mask_0 = const()[name = tensor("op_16638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16638_cast_fp16 = slice_by_index(begin = var_16638_begin_0, end = var_16638_end_0, end_mask = var_16638_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16638_cast_fp16")]; + tensor var_16642_begin_0 = const()[name = tensor("op_16642_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_16642_end_0 = const()[name = tensor("op_16642_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_16642_end_mask_0 = const()[name = tensor("op_16642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16642_cast_fp16 = slice_by_index(begin = var_16642_begin_0, end = var_16642_end_0, end_mask = var_16642_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16642_cast_fp16")]; + tensor var_16646_begin_0 = const()[name = tensor("op_16646_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_16646_end_0 = const()[name = tensor("op_16646_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_16646_end_mask_0 = const()[name = tensor("op_16646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16646_cast_fp16 = slice_by_index(begin = var_16646_begin_0, end = var_16646_end_0, end_mask = var_16646_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16646_cast_fp16")]; + tensor var_16650_begin_0 = const()[name = tensor("op_16650_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_16650_end_0 = const()[name = tensor("op_16650_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_16650_end_mask_0 = const()[name = tensor("op_16650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16650_cast_fp16 = slice_by_index(begin = var_16650_begin_0, end = var_16650_end_0, end_mask = var_16650_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16650_cast_fp16")]; + tensor var_16654_begin_0 = const()[name = tensor("op_16654_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_16654_end_0 = const()[name = tensor("op_16654_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_16654_end_mask_0 = const()[name = tensor("op_16654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16654_cast_fp16 = slice_by_index(begin = var_16654_begin_0, end = var_16654_end_0, end_mask = var_16654_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16654_cast_fp16")]; + tensor var_16658_begin_0 = const()[name = tensor("op_16658_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_16658_end_0 = const()[name = tensor("op_16658_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_16658_end_mask_0 = const()[name = tensor("op_16658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16658_cast_fp16 = slice_by_index(begin = var_16658_begin_0, end = var_16658_end_0, end_mask = var_16658_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16658_cast_fp16")]; + tensor var_16662_begin_0 = const()[name = tensor("op_16662_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_16662_end_0 = const()[name = tensor("op_16662_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_16662_end_mask_0 = const()[name = tensor("op_16662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16662_cast_fp16 = slice_by_index(begin = var_16662_begin_0, end = var_16662_end_0, end_mask = var_16662_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16662_cast_fp16")]; + tensor var_16666_begin_0 = const()[name = tensor("op_16666_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_16666_end_0 = const()[name = tensor("op_16666_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16666_end_mask_0 = const()[name = tensor("op_16666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16666_cast_fp16 = slice_by_index(begin = var_16666_begin_0, end = var_16666_end_0, end_mask = var_16666_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16666_cast_fp16")]; + tensor var_16670_begin_0 = const()[name = tensor("op_16670_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_16670_end_0 = const()[name = tensor("op_16670_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_16670_end_mask_0 = const()[name = tensor("op_16670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16670_cast_fp16 = slice_by_index(begin = var_16670_begin_0, end = var_16670_end_0, end_mask = var_16670_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16670_cast_fp16")]; + tensor var_16674_begin_0 = const()[name = tensor("op_16674_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_16674_end_0 = const()[name = tensor("op_16674_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_16674_end_mask_0 = const()[name = tensor("op_16674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16674_cast_fp16 = slice_by_index(begin = var_16674_begin_0, end = var_16674_end_0, end_mask = var_16674_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16674_cast_fp16")]; + tensor var_16678_begin_0 = const()[name = tensor("op_16678_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_16678_end_0 = const()[name = tensor("op_16678_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_16678_end_mask_0 = const()[name = tensor("op_16678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16678_cast_fp16 = slice_by_index(begin = var_16678_begin_0, end = var_16678_end_0, end_mask = var_16678_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16678_cast_fp16")]; + tensor var_16682_begin_0 = const()[name = tensor("op_16682_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_16682_end_0 = const()[name = tensor("op_16682_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_16682_end_mask_0 = const()[name = tensor("op_16682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16682_cast_fp16 = slice_by_index(begin = var_16682_begin_0, end = var_16682_end_0, end_mask = var_16682_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16682_cast_fp16")]; + tensor var_16684_begin_0 = const()[name = tensor("op_16684_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16684_end_0 = const()[name = tensor("op_16684_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16684_end_mask_0 = const()[name = tensor("op_16684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16684_cast_fp16 = slice_by_index(begin = var_16684_begin_0, end = var_16684_end_0, end_mask = var_16684_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16684_cast_fp16")]; + tensor var_16688_begin_0 = const()[name = tensor("op_16688_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16688_end_0 = const()[name = tensor("op_16688_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16688_end_mask_0 = const()[name = tensor("op_16688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16688_cast_fp16 = slice_by_index(begin = var_16688_begin_0, end = var_16688_end_0, end_mask = var_16688_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16688_cast_fp16")]; + tensor var_16692_begin_0 = const()[name = tensor("op_16692_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16692_end_0 = const()[name = tensor("op_16692_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16692_end_mask_0 = const()[name = tensor("op_16692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16692_cast_fp16 = slice_by_index(begin = var_16692_begin_0, end = var_16692_end_0, end_mask = var_16692_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16692_cast_fp16")]; + tensor var_16696_begin_0 = const()[name = tensor("op_16696_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16696_end_0 = const()[name = tensor("op_16696_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16696_end_mask_0 = const()[name = tensor("op_16696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16696_cast_fp16 = slice_by_index(begin = var_16696_begin_0, end = var_16696_end_0, end_mask = var_16696_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16696_cast_fp16")]; + tensor var_16700_begin_0 = const()[name = tensor("op_16700_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16700_end_0 = const()[name = tensor("op_16700_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16700_end_mask_0 = const()[name = tensor("op_16700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16700_cast_fp16 = slice_by_index(begin = var_16700_begin_0, end = var_16700_end_0, end_mask = var_16700_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16700_cast_fp16")]; + tensor var_16704_begin_0 = const()[name = tensor("op_16704_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16704_end_0 = const()[name = tensor("op_16704_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16704_end_mask_0 = const()[name = tensor("op_16704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16704_cast_fp16 = slice_by_index(begin = var_16704_begin_0, end = var_16704_end_0, end_mask = var_16704_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16704_cast_fp16")]; + tensor var_16708_begin_0 = const()[name = tensor("op_16708_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16708_end_0 = const()[name = tensor("op_16708_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16708_end_mask_0 = const()[name = tensor("op_16708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16708_cast_fp16 = slice_by_index(begin = var_16708_begin_0, end = var_16708_end_0, end_mask = var_16708_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16708_cast_fp16")]; + tensor var_16712_begin_0 = const()[name = tensor("op_16712_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16712_end_0 = const()[name = tensor("op_16712_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16712_end_mask_0 = const()[name = tensor("op_16712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16712_cast_fp16 = slice_by_index(begin = var_16712_begin_0, end = var_16712_end_0, end_mask = var_16712_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16712_cast_fp16")]; + tensor var_16716_begin_0 = const()[name = tensor("op_16716_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16716_end_0 = const()[name = tensor("op_16716_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16716_end_mask_0 = const()[name = tensor("op_16716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16716_cast_fp16 = slice_by_index(begin = var_16716_begin_0, end = var_16716_end_0, end_mask = var_16716_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16716_cast_fp16")]; + tensor var_16720_begin_0 = const()[name = tensor("op_16720_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16720_end_0 = const()[name = tensor("op_16720_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16720_end_mask_0 = const()[name = tensor("op_16720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16720_cast_fp16 = slice_by_index(begin = var_16720_begin_0, end = var_16720_end_0, end_mask = var_16720_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16720_cast_fp16")]; + tensor var_16724_begin_0 = const()[name = tensor("op_16724_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16724_end_0 = const()[name = tensor("op_16724_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16724_end_mask_0 = const()[name = tensor("op_16724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16724_cast_fp16 = slice_by_index(begin = var_16724_begin_0, end = var_16724_end_0, end_mask = var_16724_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16724_cast_fp16")]; + tensor var_16728_begin_0 = const()[name = tensor("op_16728_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16728_end_0 = const()[name = tensor("op_16728_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16728_end_mask_0 = const()[name = tensor("op_16728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16728_cast_fp16 = slice_by_index(begin = var_16728_begin_0, end = var_16728_end_0, end_mask = var_16728_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16728_cast_fp16")]; + tensor var_16732_begin_0 = const()[name = tensor("op_16732_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16732_end_0 = const()[name = tensor("op_16732_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_16732_end_mask_0 = const()[name = tensor("op_16732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16732_cast_fp16 = slice_by_index(begin = var_16732_begin_0, end = var_16732_end_0, end_mask = var_16732_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16732_cast_fp16")]; + tensor var_16736_begin_0 = const()[name = tensor("op_16736_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16736_end_0 = const()[name = tensor("op_16736_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_16736_end_mask_0 = const()[name = tensor("op_16736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16736_cast_fp16 = slice_by_index(begin = var_16736_begin_0, end = var_16736_end_0, end_mask = var_16736_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16736_cast_fp16")]; + tensor var_16740_begin_0 = const()[name = tensor("op_16740_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16740_end_0 = const()[name = tensor("op_16740_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_16740_end_mask_0 = const()[name = tensor("op_16740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16740_cast_fp16 = slice_by_index(begin = var_16740_begin_0, end = var_16740_end_0, end_mask = var_16740_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16740_cast_fp16")]; + tensor var_16744_begin_0 = const()[name = tensor("op_16744_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16744_end_0 = const()[name = tensor("op_16744_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16744_end_mask_0 = const()[name = tensor("op_16744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16744_cast_fp16 = slice_by_index(begin = var_16744_begin_0, end = var_16744_end_0, end_mask = var_16744_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16744_cast_fp16")]; + tensor var_16748_begin_0 = const()[name = tensor("op_16748_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16748_end_0 = const()[name = tensor("op_16748_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_16748_end_mask_0 = const()[name = tensor("op_16748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16748_cast_fp16 = slice_by_index(begin = var_16748_begin_0, end = var_16748_end_0, end_mask = var_16748_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16748_cast_fp16")]; + tensor var_16752_begin_0 = const()[name = tensor("op_16752_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16752_end_0 = const()[name = tensor("op_16752_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_16752_end_mask_0 = const()[name = tensor("op_16752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16752_cast_fp16 = slice_by_index(begin = var_16752_begin_0, end = var_16752_end_0, end_mask = var_16752_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16752_cast_fp16")]; + tensor var_16756_begin_0 = const()[name = tensor("op_16756_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16756_end_0 = const()[name = tensor("op_16756_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_16756_end_mask_0 = const()[name = tensor("op_16756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16756_cast_fp16 = slice_by_index(begin = var_16756_begin_0, end = var_16756_end_0, end_mask = var_16756_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16756_cast_fp16")]; + tensor var_16760_begin_0 = const()[name = tensor("op_16760_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16760_end_0 = const()[name = tensor("op_16760_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_16760_end_mask_0 = const()[name = tensor("op_16760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16760_cast_fp16 = slice_by_index(begin = var_16760_begin_0, end = var_16760_end_0, end_mask = var_16760_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16760_cast_fp16")]; + tensor var_16764_equation_0 = const()[name = tensor("op_16764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16764_cast_fp16 = einsum(equation = var_16764_equation_0, values = (var_16606_cast_fp16, var_16523_cast_fp16))[name = tensor("op_16764_cast_fp16")]; + tensor var_16765_to_fp16 = const()[name = tensor("op_16765_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1361_cast_fp16 = mul(x = var_16764_cast_fp16, y = var_16765_to_fp16)[name = tensor("aw_1361_cast_fp16")]; + tensor var_16768_equation_0 = const()[name = tensor("op_16768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16768_cast_fp16 = einsum(equation = var_16768_equation_0, values = (var_16610_cast_fp16, var_16527_cast_fp16))[name = tensor("op_16768_cast_fp16")]; + tensor var_16769_to_fp16 = const()[name = tensor("op_16769_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1363_cast_fp16 = mul(x = var_16768_cast_fp16, y = var_16769_to_fp16)[name = tensor("aw_1363_cast_fp16")]; + tensor var_16772_equation_0 = const()[name = tensor("op_16772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16772_cast_fp16 = einsum(equation = var_16772_equation_0, values = (var_16614_cast_fp16, var_16531_cast_fp16))[name = tensor("op_16772_cast_fp16")]; + tensor var_16773_to_fp16 = const()[name = tensor("op_16773_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1365_cast_fp16 = mul(x = var_16772_cast_fp16, y = var_16773_to_fp16)[name = tensor("aw_1365_cast_fp16")]; + tensor var_16776_equation_0 = const()[name = tensor("op_16776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16776_cast_fp16 = einsum(equation = var_16776_equation_0, values = (var_16618_cast_fp16, var_16535_cast_fp16))[name = tensor("op_16776_cast_fp16")]; + tensor var_16777_to_fp16 = const()[name = tensor("op_16777_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1367_cast_fp16 = mul(x = var_16776_cast_fp16, y = var_16777_to_fp16)[name = tensor("aw_1367_cast_fp16")]; + tensor var_16780_equation_0 = const()[name = tensor("op_16780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16780_cast_fp16 = einsum(equation = var_16780_equation_0, values = (var_16622_cast_fp16, var_16539_cast_fp16))[name = tensor("op_16780_cast_fp16")]; + tensor var_16781_to_fp16 = const()[name = tensor("op_16781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1369_cast_fp16 = mul(x = var_16780_cast_fp16, y = var_16781_to_fp16)[name = tensor("aw_1369_cast_fp16")]; + tensor var_16784_equation_0 = const()[name = tensor("op_16784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16784_cast_fp16 = einsum(equation = var_16784_equation_0, values = (var_16626_cast_fp16, var_16543_cast_fp16))[name = tensor("op_16784_cast_fp16")]; + tensor var_16785_to_fp16 = const()[name = tensor("op_16785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1371_cast_fp16 = mul(x = var_16784_cast_fp16, y = var_16785_to_fp16)[name = tensor("aw_1371_cast_fp16")]; + tensor var_16788_equation_0 = const()[name = tensor("op_16788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16788_cast_fp16 = einsum(equation = var_16788_equation_0, values = (var_16630_cast_fp16, var_16547_cast_fp16))[name = tensor("op_16788_cast_fp16")]; + tensor var_16789_to_fp16 = const()[name = tensor("op_16789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1373_cast_fp16 = mul(x = var_16788_cast_fp16, y = var_16789_to_fp16)[name = tensor("aw_1373_cast_fp16")]; + tensor var_16792_equation_0 = const()[name = tensor("op_16792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16792_cast_fp16 = einsum(equation = var_16792_equation_0, values = (var_16634_cast_fp16, var_16551_cast_fp16))[name = tensor("op_16792_cast_fp16")]; + tensor var_16793_to_fp16 = const()[name = tensor("op_16793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1375_cast_fp16 = mul(x = var_16792_cast_fp16, y = var_16793_to_fp16)[name = tensor("aw_1375_cast_fp16")]; + tensor var_16796_equation_0 = const()[name = tensor("op_16796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16796_cast_fp16 = einsum(equation = var_16796_equation_0, values = (var_16638_cast_fp16, var_16555_cast_fp16))[name = tensor("op_16796_cast_fp16")]; + tensor var_16797_to_fp16 = const()[name = tensor("op_16797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1377_cast_fp16 = mul(x = var_16796_cast_fp16, y = var_16797_to_fp16)[name = tensor("aw_1377_cast_fp16")]; + tensor var_16800_equation_0 = const()[name = tensor("op_16800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16800_cast_fp16 = einsum(equation = var_16800_equation_0, values = (var_16642_cast_fp16, var_16559_cast_fp16))[name = tensor("op_16800_cast_fp16")]; + tensor var_16801_to_fp16 = const()[name = tensor("op_16801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1379_cast_fp16 = mul(x = var_16800_cast_fp16, y = var_16801_to_fp16)[name = tensor("aw_1379_cast_fp16")]; + tensor var_16804_equation_0 = const()[name = tensor("op_16804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16804_cast_fp16 = einsum(equation = var_16804_equation_0, values = (var_16646_cast_fp16, var_16563_cast_fp16))[name = tensor("op_16804_cast_fp16")]; + tensor var_16805_to_fp16 = const()[name = tensor("op_16805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1381_cast_fp16 = mul(x = var_16804_cast_fp16, y = var_16805_to_fp16)[name = tensor("aw_1381_cast_fp16")]; + tensor var_16808_equation_0 = const()[name = tensor("op_16808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16808_cast_fp16 = einsum(equation = var_16808_equation_0, values = (var_16650_cast_fp16, var_16567_cast_fp16))[name = tensor("op_16808_cast_fp16")]; + tensor var_16809_to_fp16 = const()[name = tensor("op_16809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1383_cast_fp16 = mul(x = var_16808_cast_fp16, y = var_16809_to_fp16)[name = tensor("aw_1383_cast_fp16")]; + tensor var_16812_equation_0 = const()[name = tensor("op_16812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16812_cast_fp16 = einsum(equation = var_16812_equation_0, values = (var_16654_cast_fp16, var_16571_cast_fp16))[name = tensor("op_16812_cast_fp16")]; + tensor var_16813_to_fp16 = const()[name = tensor("op_16813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1385_cast_fp16 = mul(x = var_16812_cast_fp16, y = var_16813_to_fp16)[name = tensor("aw_1385_cast_fp16")]; + tensor var_16816_equation_0 = const()[name = tensor("op_16816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16816_cast_fp16 = einsum(equation = var_16816_equation_0, values = (var_16658_cast_fp16, var_16575_cast_fp16))[name = tensor("op_16816_cast_fp16")]; + tensor var_16817_to_fp16 = const()[name = tensor("op_16817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1387_cast_fp16 = mul(x = var_16816_cast_fp16, y = var_16817_to_fp16)[name = tensor("aw_1387_cast_fp16")]; + tensor var_16820_equation_0 = const()[name = tensor("op_16820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16820_cast_fp16 = einsum(equation = var_16820_equation_0, values = (var_16662_cast_fp16, var_16579_cast_fp16))[name = tensor("op_16820_cast_fp16")]; + tensor var_16821_to_fp16 = const()[name = tensor("op_16821_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1389_cast_fp16 = mul(x = var_16820_cast_fp16, y = var_16821_to_fp16)[name = tensor("aw_1389_cast_fp16")]; + tensor var_16824_equation_0 = const()[name = tensor("op_16824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16824_cast_fp16 = einsum(equation = var_16824_equation_0, values = (var_16666_cast_fp16, var_16583_cast_fp16))[name = tensor("op_16824_cast_fp16")]; + tensor var_16825_to_fp16 = const()[name = tensor("op_16825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1391_cast_fp16 = mul(x = var_16824_cast_fp16, y = var_16825_to_fp16)[name = tensor("aw_1391_cast_fp16")]; + tensor var_16828_equation_0 = const()[name = tensor("op_16828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16828_cast_fp16 = einsum(equation = var_16828_equation_0, values = (var_16670_cast_fp16, var_16587_cast_fp16))[name = tensor("op_16828_cast_fp16")]; + tensor var_16829_to_fp16 = const()[name = tensor("op_16829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1393_cast_fp16 = mul(x = var_16828_cast_fp16, y = var_16829_to_fp16)[name = tensor("aw_1393_cast_fp16")]; + tensor var_16832_equation_0 = const()[name = tensor("op_16832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16832_cast_fp16 = einsum(equation = var_16832_equation_0, values = (var_16674_cast_fp16, var_16591_cast_fp16))[name = tensor("op_16832_cast_fp16")]; + tensor var_16833_to_fp16 = const()[name = tensor("op_16833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1395_cast_fp16 = mul(x = var_16832_cast_fp16, y = var_16833_to_fp16)[name = tensor("aw_1395_cast_fp16")]; + tensor var_16836_equation_0 = const()[name = tensor("op_16836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16836_cast_fp16 = einsum(equation = var_16836_equation_0, values = (var_16678_cast_fp16, var_16595_cast_fp16))[name = tensor("op_16836_cast_fp16")]; + tensor var_16837_to_fp16 = const()[name = tensor("op_16837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1397_cast_fp16 = mul(x = var_16836_cast_fp16, y = var_16837_to_fp16)[name = tensor("aw_1397_cast_fp16")]; + tensor var_16840_equation_0 = const()[name = tensor("op_16840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16840_cast_fp16 = einsum(equation = var_16840_equation_0, values = (var_16682_cast_fp16, var_16599_cast_fp16))[name = tensor("op_16840_cast_fp16")]; + tensor var_16841_to_fp16 = const()[name = tensor("op_16841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1399_cast_fp16 = mul(x = var_16840_cast_fp16, y = var_16841_to_fp16)[name = tensor("aw_1399_cast_fp16")]; + tensor var_16843_cast_fp16 = softmax(axis = var_2624, x = aw_1361_cast_fp16)[name = tensor("op_16843_cast_fp16")]; + tensor var_16844_cast_fp16 = softmax(axis = var_2624, x = aw_1363_cast_fp16)[name = tensor("op_16844_cast_fp16")]; + tensor var_16845_cast_fp16 = softmax(axis = var_2624, x = aw_1365_cast_fp16)[name = tensor("op_16845_cast_fp16")]; + tensor var_16846_cast_fp16 = softmax(axis = var_2624, x = aw_1367_cast_fp16)[name = tensor("op_16846_cast_fp16")]; + tensor var_16847_cast_fp16 = softmax(axis = var_2624, x = aw_1369_cast_fp16)[name = tensor("op_16847_cast_fp16")]; + tensor var_16848_cast_fp16 = softmax(axis = var_2624, x = aw_1371_cast_fp16)[name = tensor("op_16848_cast_fp16")]; + tensor var_16849_cast_fp16 = softmax(axis = var_2624, x = aw_1373_cast_fp16)[name = tensor("op_16849_cast_fp16")]; + tensor var_16850_cast_fp16 = softmax(axis = var_2624, x = aw_1375_cast_fp16)[name = tensor("op_16850_cast_fp16")]; + tensor var_16851_cast_fp16 = softmax(axis = var_2624, x = aw_1377_cast_fp16)[name = tensor("op_16851_cast_fp16")]; + tensor var_16852_cast_fp16 = softmax(axis = var_2624, x = aw_1379_cast_fp16)[name = tensor("op_16852_cast_fp16")]; + tensor var_16853_cast_fp16 = softmax(axis = var_2624, x = aw_1381_cast_fp16)[name = tensor("op_16853_cast_fp16")]; + tensor var_16854_cast_fp16 = softmax(axis = var_2624, x = aw_1383_cast_fp16)[name = tensor("op_16854_cast_fp16")]; + tensor var_16855_cast_fp16 = softmax(axis = var_2624, x = aw_1385_cast_fp16)[name = tensor("op_16855_cast_fp16")]; + tensor var_16856_cast_fp16 = softmax(axis = var_2624, x = aw_1387_cast_fp16)[name = tensor("op_16856_cast_fp16")]; + tensor var_16857_cast_fp16 = softmax(axis = var_2624, x = aw_1389_cast_fp16)[name = tensor("op_16857_cast_fp16")]; + tensor var_16858_cast_fp16 = softmax(axis = var_2624, x = aw_1391_cast_fp16)[name = tensor("op_16858_cast_fp16")]; + tensor var_16859_cast_fp16 = softmax(axis = var_2624, x = aw_1393_cast_fp16)[name = tensor("op_16859_cast_fp16")]; + tensor var_16860_cast_fp16 = softmax(axis = var_2624, x = aw_1395_cast_fp16)[name = tensor("op_16860_cast_fp16")]; + tensor var_16861_cast_fp16 = softmax(axis = var_2624, x = aw_1397_cast_fp16)[name = tensor("op_16861_cast_fp16")]; + tensor var_16862_cast_fp16 = softmax(axis = var_2624, x = aw_1399_cast_fp16)[name = tensor("op_16862_cast_fp16")]; + tensor var_16864_equation_0 = const()[name = tensor("op_16864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16864_cast_fp16 = einsum(equation = var_16864_equation_0, values = (var_16684_cast_fp16, var_16843_cast_fp16))[name = tensor("op_16864_cast_fp16")]; + tensor var_16866_equation_0 = const()[name = tensor("op_16866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16866_cast_fp16 = einsum(equation = var_16866_equation_0, values = (var_16688_cast_fp16, var_16844_cast_fp16))[name = tensor("op_16866_cast_fp16")]; + tensor var_16868_equation_0 = const()[name = tensor("op_16868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16868_cast_fp16 = einsum(equation = var_16868_equation_0, values = (var_16692_cast_fp16, var_16845_cast_fp16))[name = tensor("op_16868_cast_fp16")]; + tensor var_16870_equation_0 = const()[name = tensor("op_16870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16870_cast_fp16 = einsum(equation = var_16870_equation_0, values = (var_16696_cast_fp16, var_16846_cast_fp16))[name = tensor("op_16870_cast_fp16")]; + tensor var_16872_equation_0 = const()[name = tensor("op_16872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16872_cast_fp16 = einsum(equation = var_16872_equation_0, values = (var_16700_cast_fp16, var_16847_cast_fp16))[name = tensor("op_16872_cast_fp16")]; + tensor var_16874_equation_0 = const()[name = tensor("op_16874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16874_cast_fp16 = einsum(equation = var_16874_equation_0, values = (var_16704_cast_fp16, var_16848_cast_fp16))[name = tensor("op_16874_cast_fp16")]; + tensor var_16876_equation_0 = const()[name = tensor("op_16876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16876_cast_fp16 = einsum(equation = var_16876_equation_0, values = (var_16708_cast_fp16, var_16849_cast_fp16))[name = tensor("op_16876_cast_fp16")]; + tensor var_16878_equation_0 = const()[name = tensor("op_16878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16878_cast_fp16 = einsum(equation = var_16878_equation_0, values = (var_16712_cast_fp16, var_16850_cast_fp16))[name = tensor("op_16878_cast_fp16")]; + tensor var_16880_equation_0 = const()[name = tensor("op_16880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16880_cast_fp16 = einsum(equation = var_16880_equation_0, values = (var_16716_cast_fp16, var_16851_cast_fp16))[name = tensor("op_16880_cast_fp16")]; + tensor var_16882_equation_0 = const()[name = tensor("op_16882_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16882_cast_fp16 = einsum(equation = var_16882_equation_0, values = (var_16720_cast_fp16, var_16852_cast_fp16))[name = tensor("op_16882_cast_fp16")]; + tensor var_16884_equation_0 = const()[name = tensor("op_16884_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16884_cast_fp16 = einsum(equation = var_16884_equation_0, values = (var_16724_cast_fp16, var_16853_cast_fp16))[name = tensor("op_16884_cast_fp16")]; + tensor var_16886_equation_0 = const()[name = tensor("op_16886_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16886_cast_fp16 = einsum(equation = var_16886_equation_0, values = (var_16728_cast_fp16, var_16854_cast_fp16))[name = tensor("op_16886_cast_fp16")]; + tensor var_16888_equation_0 = const()[name = tensor("op_16888_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16888_cast_fp16 = einsum(equation = var_16888_equation_0, values = (var_16732_cast_fp16, var_16855_cast_fp16))[name = tensor("op_16888_cast_fp16")]; + tensor var_16890_equation_0 = const()[name = tensor("op_16890_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16890_cast_fp16 = einsum(equation = var_16890_equation_0, values = (var_16736_cast_fp16, var_16856_cast_fp16))[name = tensor("op_16890_cast_fp16")]; + tensor var_16892_equation_0 = const()[name = tensor("op_16892_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16892_cast_fp16 = einsum(equation = var_16892_equation_0, values = (var_16740_cast_fp16, var_16857_cast_fp16))[name = tensor("op_16892_cast_fp16")]; + tensor var_16894_equation_0 = const()[name = tensor("op_16894_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16894_cast_fp16 = einsum(equation = var_16894_equation_0, values = (var_16744_cast_fp16, var_16858_cast_fp16))[name = tensor("op_16894_cast_fp16")]; + tensor var_16896_equation_0 = const()[name = tensor("op_16896_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16896_cast_fp16 = einsum(equation = var_16896_equation_0, values = (var_16748_cast_fp16, var_16859_cast_fp16))[name = tensor("op_16896_cast_fp16")]; + tensor var_16898_equation_0 = const()[name = tensor("op_16898_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16898_cast_fp16 = einsum(equation = var_16898_equation_0, values = (var_16752_cast_fp16, var_16860_cast_fp16))[name = tensor("op_16898_cast_fp16")]; + tensor var_16900_equation_0 = const()[name = tensor("op_16900_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16900_cast_fp16 = einsum(equation = var_16900_equation_0, values = (var_16756_cast_fp16, var_16861_cast_fp16))[name = tensor("op_16900_cast_fp16")]; + tensor var_16902_equation_0 = const()[name = tensor("op_16902_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16902_cast_fp16 = einsum(equation = var_16902_equation_0, values = (var_16760_cast_fp16, var_16862_cast_fp16))[name = tensor("op_16902_cast_fp16")]; + tensor input_269_interleave_0 = const()[name = tensor("input_269_interleave_0"), val = tensor(false)]; + tensor input_269_cast_fp16 = concat(axis = var_2624, interleave = input_269_interleave_0, values = (var_16864_cast_fp16, var_16866_cast_fp16, var_16868_cast_fp16, var_16870_cast_fp16, var_16872_cast_fp16, var_16874_cast_fp16, var_16876_cast_fp16, var_16878_cast_fp16, var_16880_cast_fp16, var_16882_cast_fp16, var_16884_cast_fp16, var_16886_cast_fp16, var_16888_cast_fp16, var_16890_cast_fp16, var_16892_cast_fp16, var_16894_cast_fp16, var_16896_cast_fp16, var_16898_cast_fp16, var_16900_cast_fp16, var_16902_cast_fp16))[name = tensor("input_269_cast_fp16")]; + tensor var_16912_pad_type_0 = const()[name = tensor("op_16912_pad_type_0"), val = tensor("valid")]; + tensor var_16912_strides_0 = const()[name = tensor("op_16912_strides_0"), val = tensor([1, 1])]; + tensor var_16912_pad_0 = const()[name = tensor("op_16912_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16912_dilations_0 = const()[name = tensor("op_16912_dilations_0"), val = tensor([1, 1])]; + tensor var_16912_groups_0 = const()[name = tensor("op_16912_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(495426496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496655360))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496655552)))]; + tensor var_16912_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_16912_dilations_0, groups = var_16912_groups_0, pad = var_16912_pad_0, pad_type = var_16912_pad_type_0, strides = var_16912_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized, x = input_269_cast_fp16)[name = tensor("op_16912_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = var_16912_cast_fp16, y = inputs_115_cast_fp16)[name = tensor("inputs_117_cast_fp16")]; + tensor hidden_states_169_axes_0 = const()[name = tensor("hidden_states_169_axes_0"), val = tensor([1])]; + tensor hidden_states_169_gamma_0_to_fp16 = const()[name = tensor("hidden_states_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496658176)))]; + tensor hidden_states_169_beta_0_to_fp16 = const()[name = tensor("hidden_states_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496660800)))]; + tensor var_16922_to_fp16 = const()[name = tensor("op_16922_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_169_cast_fp16 = layer_norm(axes = hidden_states_169_axes_0, beta = hidden_states_169_beta_0_to_fp16, epsilon = var_16922_to_fp16, gamma = hidden_states_169_gamma_0_to_fp16, x = inputs_117_cast_fp16)[name = tensor("hidden_states_169_cast_fp16")]; + tensor q_79_pad_type_0 = const()[name = tensor("q_79_pad_type_0"), val = tensor("valid")]; + tensor q_79_strides_0 = const()[name = tensor("q_79_strides_0"), val = tensor([1, 1])]; + tensor q_79_pad_0 = const()[name = tensor("q_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_79_dilations_0 = const()[name = tensor("q_79_dilations_0"), val = tensor([1, 1])]; + tensor q_79_groups_0 = const()[name = tensor("q_79_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496663424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(497892288))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_79_cast_fp16 = conv(dilations = q_79_dilations_0, groups = q_79_groups_0, pad = q_79_pad_0, pad_type = q_79_pad_type_0, strides = q_79_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_169_cast_fp16)[name = tensor("q_79_cast_fp16")]; + tensor k_157_pad_type_0 = const()[name = tensor("k_157_pad_type_0"), val = tensor("valid")]; + tensor k_157_strides_0 = const()[name = tensor("k_157_strides_0"), val = tensor([1, 1])]; + tensor k_157_pad_0 = const()[name = tensor("k_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_157_dilations_0 = const()[name = tensor("k_157_dilations_0"), val = tensor([1, 1])]; + tensor k_157_groups_0 = const()[name = tensor("k_157_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(497892480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499858624))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_157_cast_fp16 = conv(dilations = k_157_dilations_0, groups = k_157_groups_0, pad = k_157_pad_0, pad_type = k_157_pad_type_0, strides = k_157_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_157_cast_fp16")]; + tensor v_79_pad_type_0 = const()[name = tensor("v_79_pad_type_0"), val = tensor("valid")]; + tensor v_79_strides_0 = const()[name = tensor("v_79_strides_0"), val = tensor([1, 1])]; + tensor v_79_pad_0 = const()[name = tensor("v_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_79_dilations_0 = const()[name = tensor("v_79_dilations_0"), val = tensor([1, 1])]; + tensor v_79_groups_0 = const()[name = tensor("v_79_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499858816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501824960))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_79_cast_fp16 = conv(dilations = v_79_dilations_0, groups = v_79_groups_0, pad = v_79_pad_0, pad_type = v_79_pad_type_0, strides = v_79_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_79_cast_fp16")]; + tensor var_16955_begin_0 = const()[name = tensor("op_16955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16955_end_0 = const()[name = tensor("op_16955_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16955_end_mask_0 = const()[name = tensor("op_16955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16955_cast_fp16 = slice_by_index(begin = var_16955_begin_0, end = var_16955_end_0, end_mask = var_16955_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16955_cast_fp16")]; + tensor var_16959_begin_0 = const()[name = tensor("op_16959_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16959_end_0 = const()[name = tensor("op_16959_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16959_end_mask_0 = const()[name = tensor("op_16959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16959_cast_fp16 = slice_by_index(begin = var_16959_begin_0, end = var_16959_end_0, end_mask = var_16959_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16959_cast_fp16")]; + tensor var_16963_begin_0 = const()[name = tensor("op_16963_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16963_end_0 = const()[name = tensor("op_16963_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16963_end_mask_0 = const()[name = tensor("op_16963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16963_cast_fp16 = slice_by_index(begin = var_16963_begin_0, end = var_16963_end_0, end_mask = var_16963_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16963_cast_fp16")]; + tensor var_16967_begin_0 = const()[name = tensor("op_16967_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16967_end_0 = const()[name = tensor("op_16967_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16967_end_mask_0 = const()[name = tensor("op_16967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16967_cast_fp16 = slice_by_index(begin = var_16967_begin_0, end = var_16967_end_0, end_mask = var_16967_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16967_cast_fp16")]; + tensor var_16971_begin_0 = const()[name = tensor("op_16971_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16971_end_0 = const()[name = tensor("op_16971_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16971_end_mask_0 = const()[name = tensor("op_16971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16971_cast_fp16 = slice_by_index(begin = var_16971_begin_0, end = var_16971_end_0, end_mask = var_16971_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16971_cast_fp16")]; + tensor var_16975_begin_0 = const()[name = tensor("op_16975_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16975_end_0 = const()[name = tensor("op_16975_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16975_end_mask_0 = const()[name = tensor("op_16975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16975_cast_fp16 = slice_by_index(begin = var_16975_begin_0, end = var_16975_end_0, end_mask = var_16975_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16975_cast_fp16")]; + tensor var_16979_begin_0 = const()[name = tensor("op_16979_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16979_end_0 = const()[name = tensor("op_16979_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16979_end_mask_0 = const()[name = tensor("op_16979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16979_cast_fp16 = slice_by_index(begin = var_16979_begin_0, end = var_16979_end_0, end_mask = var_16979_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16979_cast_fp16")]; + tensor var_16983_begin_0 = const()[name = tensor("op_16983_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16983_end_0 = const()[name = tensor("op_16983_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16983_end_mask_0 = const()[name = tensor("op_16983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16983_cast_fp16 = slice_by_index(begin = var_16983_begin_0, end = var_16983_end_0, end_mask = var_16983_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16983_cast_fp16")]; + tensor var_16987_begin_0 = const()[name = tensor("op_16987_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16987_end_0 = const()[name = tensor("op_16987_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16987_end_mask_0 = const()[name = tensor("op_16987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16987_cast_fp16 = slice_by_index(begin = var_16987_begin_0, end = var_16987_end_0, end_mask = var_16987_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16987_cast_fp16")]; + tensor var_16991_begin_0 = const()[name = tensor("op_16991_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16991_end_0 = const()[name = tensor("op_16991_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16991_end_mask_0 = const()[name = tensor("op_16991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16991_cast_fp16 = slice_by_index(begin = var_16991_begin_0, end = var_16991_end_0, end_mask = var_16991_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16991_cast_fp16")]; + tensor var_16995_begin_0 = const()[name = tensor("op_16995_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16995_end_0 = const()[name = tensor("op_16995_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16995_end_mask_0 = const()[name = tensor("op_16995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16995_cast_fp16 = slice_by_index(begin = var_16995_begin_0, end = var_16995_end_0, end_mask = var_16995_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16995_cast_fp16")]; + tensor var_16999_begin_0 = const()[name = tensor("op_16999_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16999_end_0 = const()[name = tensor("op_16999_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16999_end_mask_0 = const()[name = tensor("op_16999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16999_cast_fp16 = slice_by_index(begin = var_16999_begin_0, end = var_16999_end_0, end_mask = var_16999_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16999_cast_fp16")]; + tensor var_17003_begin_0 = const()[name = tensor("op_17003_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17003_end_0 = const()[name = tensor("op_17003_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17003_end_mask_0 = const()[name = tensor("op_17003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17003_cast_fp16 = slice_by_index(begin = var_17003_begin_0, end = var_17003_end_0, end_mask = var_17003_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17003_cast_fp16")]; + tensor var_17007_begin_0 = const()[name = tensor("op_17007_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17007_end_0 = const()[name = tensor("op_17007_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17007_end_mask_0 = const()[name = tensor("op_17007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17007_cast_fp16 = slice_by_index(begin = var_17007_begin_0, end = var_17007_end_0, end_mask = var_17007_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17007_cast_fp16")]; + tensor var_17011_begin_0 = const()[name = tensor("op_17011_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17011_end_0 = const()[name = tensor("op_17011_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17011_end_mask_0 = const()[name = tensor("op_17011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17011_cast_fp16 = slice_by_index(begin = var_17011_begin_0, end = var_17011_end_0, end_mask = var_17011_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17011_cast_fp16")]; + tensor var_17015_begin_0 = const()[name = tensor("op_17015_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17015_end_0 = const()[name = tensor("op_17015_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17015_end_mask_0 = const()[name = tensor("op_17015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17015_cast_fp16 = slice_by_index(begin = var_17015_begin_0, end = var_17015_end_0, end_mask = var_17015_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17015_cast_fp16")]; + tensor var_17019_begin_0 = const()[name = tensor("op_17019_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17019_end_0 = const()[name = tensor("op_17019_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17019_end_mask_0 = const()[name = tensor("op_17019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17019_cast_fp16 = slice_by_index(begin = var_17019_begin_0, end = var_17019_end_0, end_mask = var_17019_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17019_cast_fp16")]; + tensor var_17023_begin_0 = const()[name = tensor("op_17023_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17023_end_0 = const()[name = tensor("op_17023_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17023_end_mask_0 = const()[name = tensor("op_17023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17023_cast_fp16 = slice_by_index(begin = var_17023_begin_0, end = var_17023_end_0, end_mask = var_17023_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17023_cast_fp16")]; + tensor var_17027_begin_0 = const()[name = tensor("op_17027_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17027_end_0 = const()[name = tensor("op_17027_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17027_end_mask_0 = const()[name = tensor("op_17027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17027_cast_fp16 = slice_by_index(begin = var_17027_begin_0, end = var_17027_end_0, end_mask = var_17027_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17027_cast_fp16")]; + tensor var_17031_begin_0 = const()[name = tensor("op_17031_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17031_end_0 = const()[name = tensor("op_17031_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17031_end_mask_0 = const()[name = tensor("op_17031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17031_cast_fp16 = slice_by_index(begin = var_17031_begin_0, end = var_17031_end_0, end_mask = var_17031_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17031_cast_fp16")]; + tensor k_159_perm_0 = const()[name = tensor("k_159_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17038_begin_0 = const()[name = tensor("op_17038_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17038_end_0 = const()[name = tensor("op_17038_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_17038_end_mask_0 = const()[name = tensor("op_17038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_159_cast_fp16 = transpose(perm = k_159_perm_0, x = k_157_cast_fp16)[name = tensor("transpose_28")]; + tensor var_17038_cast_fp16 = slice_by_index(begin = var_17038_begin_0, end = var_17038_end_0, end_mask = var_17038_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17038_cast_fp16")]; + tensor var_17042_begin_0 = const()[name = tensor("op_17042_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17042_end_0 = const()[name = tensor("op_17042_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_17042_end_mask_0 = const()[name = tensor("op_17042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17042_cast_fp16 = slice_by_index(begin = var_17042_begin_0, end = var_17042_end_0, end_mask = var_17042_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17042_cast_fp16")]; + tensor var_17046_begin_0 = const()[name = tensor("op_17046_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17046_end_0 = const()[name = tensor("op_17046_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_17046_end_mask_0 = const()[name = tensor("op_17046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17046_cast_fp16 = slice_by_index(begin = var_17046_begin_0, end = var_17046_end_0, end_mask = var_17046_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17046_cast_fp16")]; + tensor var_17050_begin_0 = const()[name = tensor("op_17050_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17050_end_0 = const()[name = tensor("op_17050_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_17050_end_mask_0 = const()[name = tensor("op_17050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17050_cast_fp16 = slice_by_index(begin = var_17050_begin_0, end = var_17050_end_0, end_mask = var_17050_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17050_cast_fp16")]; + tensor var_17054_begin_0 = const()[name = tensor("op_17054_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17054_end_0 = const()[name = tensor("op_17054_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_17054_end_mask_0 = const()[name = tensor("op_17054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17054_cast_fp16 = slice_by_index(begin = var_17054_begin_0, end = var_17054_end_0, end_mask = var_17054_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17054_cast_fp16")]; + tensor var_17058_begin_0 = const()[name = tensor("op_17058_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17058_end_0 = const()[name = tensor("op_17058_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_17058_end_mask_0 = const()[name = tensor("op_17058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17058_cast_fp16 = slice_by_index(begin = var_17058_begin_0, end = var_17058_end_0, end_mask = var_17058_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17058_cast_fp16")]; + tensor var_17062_begin_0 = const()[name = tensor("op_17062_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17062_end_0 = const()[name = tensor("op_17062_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_17062_end_mask_0 = const()[name = tensor("op_17062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17062_cast_fp16 = slice_by_index(begin = var_17062_begin_0, end = var_17062_end_0, end_mask = var_17062_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17062_cast_fp16")]; + tensor var_17066_begin_0 = const()[name = tensor("op_17066_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17066_end_0 = const()[name = tensor("op_17066_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_17066_end_mask_0 = const()[name = tensor("op_17066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17066_cast_fp16 = slice_by_index(begin = var_17066_begin_0, end = var_17066_end_0, end_mask = var_17066_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17066_cast_fp16")]; + tensor var_17070_begin_0 = const()[name = tensor("op_17070_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17070_end_0 = const()[name = tensor("op_17070_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_17070_end_mask_0 = const()[name = tensor("op_17070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17070_cast_fp16 = slice_by_index(begin = var_17070_begin_0, end = var_17070_end_0, end_mask = var_17070_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17070_cast_fp16")]; + tensor var_17074_begin_0 = const()[name = tensor("op_17074_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17074_end_0 = const()[name = tensor("op_17074_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_17074_end_mask_0 = const()[name = tensor("op_17074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17074_cast_fp16 = slice_by_index(begin = var_17074_begin_0, end = var_17074_end_0, end_mask = var_17074_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17074_cast_fp16")]; + tensor var_17078_begin_0 = const()[name = tensor("op_17078_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17078_end_0 = const()[name = tensor("op_17078_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_17078_end_mask_0 = const()[name = tensor("op_17078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17078_cast_fp16 = slice_by_index(begin = var_17078_begin_0, end = var_17078_end_0, end_mask = var_17078_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17078_cast_fp16")]; + tensor var_17082_begin_0 = const()[name = tensor("op_17082_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17082_end_0 = const()[name = tensor("op_17082_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_17082_end_mask_0 = const()[name = tensor("op_17082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17082_cast_fp16 = slice_by_index(begin = var_17082_begin_0, end = var_17082_end_0, end_mask = var_17082_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17082_cast_fp16")]; + tensor var_17086_begin_0 = const()[name = tensor("op_17086_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17086_end_0 = const()[name = tensor("op_17086_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_17086_end_mask_0 = const()[name = tensor("op_17086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17086_cast_fp16 = slice_by_index(begin = var_17086_begin_0, end = var_17086_end_0, end_mask = var_17086_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17086_cast_fp16")]; + tensor var_17090_begin_0 = const()[name = tensor("op_17090_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_17090_end_0 = const()[name = tensor("op_17090_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_17090_end_mask_0 = const()[name = tensor("op_17090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17090_cast_fp16 = slice_by_index(begin = var_17090_begin_0, end = var_17090_end_0, end_mask = var_17090_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17090_cast_fp16")]; + tensor var_17094_begin_0 = const()[name = tensor("op_17094_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_17094_end_0 = const()[name = tensor("op_17094_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_17094_end_mask_0 = const()[name = tensor("op_17094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17094_cast_fp16 = slice_by_index(begin = var_17094_begin_0, end = var_17094_end_0, end_mask = var_17094_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17094_cast_fp16")]; + tensor var_17098_begin_0 = const()[name = tensor("op_17098_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_17098_end_0 = const()[name = tensor("op_17098_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_17098_end_mask_0 = const()[name = tensor("op_17098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17098_cast_fp16 = slice_by_index(begin = var_17098_begin_0, end = var_17098_end_0, end_mask = var_17098_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17098_cast_fp16")]; + tensor var_17102_begin_0 = const()[name = tensor("op_17102_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_17102_end_0 = const()[name = tensor("op_17102_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_17102_end_mask_0 = const()[name = tensor("op_17102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17102_cast_fp16 = slice_by_index(begin = var_17102_begin_0, end = var_17102_end_0, end_mask = var_17102_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17102_cast_fp16")]; + tensor var_17106_begin_0 = const()[name = tensor("op_17106_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_17106_end_0 = const()[name = tensor("op_17106_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_17106_end_mask_0 = const()[name = tensor("op_17106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17106_cast_fp16 = slice_by_index(begin = var_17106_begin_0, end = var_17106_end_0, end_mask = var_17106_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17106_cast_fp16")]; + tensor var_17110_begin_0 = const()[name = tensor("op_17110_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_17110_end_0 = const()[name = tensor("op_17110_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_17110_end_mask_0 = const()[name = tensor("op_17110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17110_cast_fp16 = slice_by_index(begin = var_17110_begin_0, end = var_17110_end_0, end_mask = var_17110_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17110_cast_fp16")]; + tensor var_17114_begin_0 = const()[name = tensor("op_17114_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_17114_end_0 = const()[name = tensor("op_17114_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_17114_end_mask_0 = const()[name = tensor("op_17114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17114_cast_fp16 = slice_by_index(begin = var_17114_begin_0, end = var_17114_end_0, end_mask = var_17114_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17114_cast_fp16")]; + tensor var_17116_begin_0 = const()[name = tensor("op_17116_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17116_end_0 = const()[name = tensor("op_17116_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_17116_end_mask_0 = const()[name = tensor("op_17116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17116_cast_fp16 = slice_by_index(begin = var_17116_begin_0, end = var_17116_end_0, end_mask = var_17116_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17116_cast_fp16")]; + tensor var_17120_begin_0 = const()[name = tensor("op_17120_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17120_end_0 = const()[name = tensor("op_17120_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_17120_end_mask_0 = const()[name = tensor("op_17120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17120_cast_fp16 = slice_by_index(begin = var_17120_begin_0, end = var_17120_end_0, end_mask = var_17120_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17120_cast_fp16")]; + tensor var_17124_begin_0 = const()[name = tensor("op_17124_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17124_end_0 = const()[name = tensor("op_17124_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_17124_end_mask_0 = const()[name = tensor("op_17124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17124_cast_fp16 = slice_by_index(begin = var_17124_begin_0, end = var_17124_end_0, end_mask = var_17124_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17124_cast_fp16")]; + tensor var_17128_begin_0 = const()[name = tensor("op_17128_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17128_end_0 = const()[name = tensor("op_17128_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_17128_end_mask_0 = const()[name = tensor("op_17128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17128_cast_fp16 = slice_by_index(begin = var_17128_begin_0, end = var_17128_end_0, end_mask = var_17128_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17128_cast_fp16")]; + tensor var_17132_begin_0 = const()[name = tensor("op_17132_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17132_end_0 = const()[name = tensor("op_17132_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_17132_end_mask_0 = const()[name = tensor("op_17132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17132_cast_fp16 = slice_by_index(begin = var_17132_begin_0, end = var_17132_end_0, end_mask = var_17132_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17132_cast_fp16")]; + tensor var_17136_begin_0 = const()[name = tensor("op_17136_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17136_end_0 = const()[name = tensor("op_17136_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_17136_end_mask_0 = const()[name = tensor("op_17136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17136_cast_fp16 = slice_by_index(begin = var_17136_begin_0, end = var_17136_end_0, end_mask = var_17136_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17136_cast_fp16")]; + tensor var_17140_begin_0 = const()[name = tensor("op_17140_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17140_end_0 = const()[name = tensor("op_17140_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_17140_end_mask_0 = const()[name = tensor("op_17140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17140_cast_fp16 = slice_by_index(begin = var_17140_begin_0, end = var_17140_end_0, end_mask = var_17140_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17140_cast_fp16")]; + tensor var_17144_begin_0 = const()[name = tensor("op_17144_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17144_end_0 = const()[name = tensor("op_17144_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_17144_end_mask_0 = const()[name = tensor("op_17144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17144_cast_fp16 = slice_by_index(begin = var_17144_begin_0, end = var_17144_end_0, end_mask = var_17144_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17144_cast_fp16")]; + tensor var_17148_begin_0 = const()[name = tensor("op_17148_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17148_end_0 = const()[name = tensor("op_17148_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_17148_end_mask_0 = const()[name = tensor("op_17148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17148_cast_fp16 = slice_by_index(begin = var_17148_begin_0, end = var_17148_end_0, end_mask = var_17148_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17148_cast_fp16")]; + tensor var_17152_begin_0 = const()[name = tensor("op_17152_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17152_end_0 = const()[name = tensor("op_17152_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_17152_end_mask_0 = const()[name = tensor("op_17152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17152_cast_fp16 = slice_by_index(begin = var_17152_begin_0, end = var_17152_end_0, end_mask = var_17152_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17152_cast_fp16")]; + tensor var_17156_begin_0 = const()[name = tensor("op_17156_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17156_end_0 = const()[name = tensor("op_17156_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_17156_end_mask_0 = const()[name = tensor("op_17156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17156_cast_fp16 = slice_by_index(begin = var_17156_begin_0, end = var_17156_end_0, end_mask = var_17156_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17156_cast_fp16")]; + tensor var_17160_begin_0 = const()[name = tensor("op_17160_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17160_end_0 = const()[name = tensor("op_17160_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_17160_end_mask_0 = const()[name = tensor("op_17160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17160_cast_fp16 = slice_by_index(begin = var_17160_begin_0, end = var_17160_end_0, end_mask = var_17160_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17160_cast_fp16")]; + tensor var_17164_begin_0 = const()[name = tensor("op_17164_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17164_end_0 = const()[name = tensor("op_17164_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_17164_end_mask_0 = const()[name = tensor("op_17164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17164_cast_fp16 = slice_by_index(begin = var_17164_begin_0, end = var_17164_end_0, end_mask = var_17164_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17164_cast_fp16")]; + tensor var_17168_begin_0 = const()[name = tensor("op_17168_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17168_end_0 = const()[name = tensor("op_17168_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_17168_end_mask_0 = const()[name = tensor("op_17168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17168_cast_fp16 = slice_by_index(begin = var_17168_begin_0, end = var_17168_end_0, end_mask = var_17168_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17168_cast_fp16")]; + tensor var_17172_begin_0 = const()[name = tensor("op_17172_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17172_end_0 = const()[name = tensor("op_17172_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_17172_end_mask_0 = const()[name = tensor("op_17172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17172_cast_fp16 = slice_by_index(begin = var_17172_begin_0, end = var_17172_end_0, end_mask = var_17172_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17172_cast_fp16")]; + tensor var_17176_begin_0 = const()[name = tensor("op_17176_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17176_end_0 = const()[name = tensor("op_17176_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_17176_end_mask_0 = const()[name = tensor("op_17176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17176_cast_fp16 = slice_by_index(begin = var_17176_begin_0, end = var_17176_end_0, end_mask = var_17176_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17176_cast_fp16")]; + tensor var_17180_begin_0 = const()[name = tensor("op_17180_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17180_end_0 = const()[name = tensor("op_17180_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_17180_end_mask_0 = const()[name = tensor("op_17180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17180_cast_fp16 = slice_by_index(begin = var_17180_begin_0, end = var_17180_end_0, end_mask = var_17180_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17180_cast_fp16")]; + tensor var_17184_begin_0 = const()[name = tensor("op_17184_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17184_end_0 = const()[name = tensor("op_17184_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_17184_end_mask_0 = const()[name = tensor("op_17184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17184_cast_fp16 = slice_by_index(begin = var_17184_begin_0, end = var_17184_end_0, end_mask = var_17184_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17184_cast_fp16")]; + tensor var_17188_begin_0 = const()[name = tensor("op_17188_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17188_end_0 = const()[name = tensor("op_17188_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_17188_end_mask_0 = const()[name = tensor("op_17188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17188_cast_fp16 = slice_by_index(begin = var_17188_begin_0, end = var_17188_end_0, end_mask = var_17188_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17188_cast_fp16")]; + tensor var_17192_begin_0 = const()[name = tensor("op_17192_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17192_end_0 = const()[name = tensor("op_17192_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_17192_end_mask_0 = const()[name = tensor("op_17192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17192_cast_fp16 = slice_by_index(begin = var_17192_begin_0, end = var_17192_end_0, end_mask = var_17192_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17192_cast_fp16")]; + tensor var_17196_equation_0 = const()[name = tensor("op_17196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17196_cast_fp16 = einsum(equation = var_17196_equation_0, values = (var_17038_cast_fp16, var_16955_cast_fp16))[name = tensor("op_17196_cast_fp16")]; + tensor var_17197_to_fp16 = const()[name = tensor("op_17197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1401_cast_fp16 = mul(x = var_17196_cast_fp16, y = var_17197_to_fp16)[name = tensor("aw_1401_cast_fp16")]; + tensor var_17200_equation_0 = const()[name = tensor("op_17200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17200_cast_fp16 = einsum(equation = var_17200_equation_0, values = (var_17042_cast_fp16, var_16959_cast_fp16))[name = tensor("op_17200_cast_fp16")]; + tensor var_17201_to_fp16 = const()[name = tensor("op_17201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1403_cast_fp16 = mul(x = var_17200_cast_fp16, y = var_17201_to_fp16)[name = tensor("aw_1403_cast_fp16")]; + tensor var_17204_equation_0 = const()[name = tensor("op_17204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17204_cast_fp16 = einsum(equation = var_17204_equation_0, values = (var_17046_cast_fp16, var_16963_cast_fp16))[name = tensor("op_17204_cast_fp16")]; + tensor var_17205_to_fp16 = const()[name = tensor("op_17205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1405_cast_fp16 = mul(x = var_17204_cast_fp16, y = var_17205_to_fp16)[name = tensor("aw_1405_cast_fp16")]; + tensor var_17208_equation_0 = const()[name = tensor("op_17208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17208_cast_fp16 = einsum(equation = var_17208_equation_0, values = (var_17050_cast_fp16, var_16967_cast_fp16))[name = tensor("op_17208_cast_fp16")]; + tensor var_17209_to_fp16 = const()[name = tensor("op_17209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1407_cast_fp16 = mul(x = var_17208_cast_fp16, y = var_17209_to_fp16)[name = tensor("aw_1407_cast_fp16")]; + tensor var_17212_equation_0 = const()[name = tensor("op_17212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17212_cast_fp16 = einsum(equation = var_17212_equation_0, values = (var_17054_cast_fp16, var_16971_cast_fp16))[name = tensor("op_17212_cast_fp16")]; + tensor var_17213_to_fp16 = const()[name = tensor("op_17213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1409_cast_fp16 = mul(x = var_17212_cast_fp16, y = var_17213_to_fp16)[name = tensor("aw_1409_cast_fp16")]; + tensor var_17216_equation_0 = const()[name = tensor("op_17216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17216_cast_fp16 = einsum(equation = var_17216_equation_0, values = (var_17058_cast_fp16, var_16975_cast_fp16))[name = tensor("op_17216_cast_fp16")]; + tensor var_17217_to_fp16 = const()[name = tensor("op_17217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1411_cast_fp16 = mul(x = var_17216_cast_fp16, y = var_17217_to_fp16)[name = tensor("aw_1411_cast_fp16")]; + tensor var_17220_equation_0 = const()[name = tensor("op_17220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17220_cast_fp16 = einsum(equation = var_17220_equation_0, values = (var_17062_cast_fp16, var_16979_cast_fp16))[name = tensor("op_17220_cast_fp16")]; + tensor var_17221_to_fp16 = const()[name = tensor("op_17221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1413_cast_fp16 = mul(x = var_17220_cast_fp16, y = var_17221_to_fp16)[name = tensor("aw_1413_cast_fp16")]; + tensor var_17224_equation_0 = const()[name = tensor("op_17224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17224_cast_fp16 = einsum(equation = var_17224_equation_0, values = (var_17066_cast_fp16, var_16983_cast_fp16))[name = tensor("op_17224_cast_fp16")]; + tensor var_17225_to_fp16 = const()[name = tensor("op_17225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1415_cast_fp16 = mul(x = var_17224_cast_fp16, y = var_17225_to_fp16)[name = tensor("aw_1415_cast_fp16")]; + tensor var_17228_equation_0 = const()[name = tensor("op_17228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17228_cast_fp16 = einsum(equation = var_17228_equation_0, values = (var_17070_cast_fp16, var_16987_cast_fp16))[name = tensor("op_17228_cast_fp16")]; + tensor var_17229_to_fp16 = const()[name = tensor("op_17229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1417_cast_fp16 = mul(x = var_17228_cast_fp16, y = var_17229_to_fp16)[name = tensor("aw_1417_cast_fp16")]; + tensor var_17232_equation_0 = const()[name = tensor("op_17232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17232_cast_fp16 = einsum(equation = var_17232_equation_0, values = (var_17074_cast_fp16, var_16991_cast_fp16))[name = tensor("op_17232_cast_fp16")]; + tensor var_17233_to_fp16 = const()[name = tensor("op_17233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1419_cast_fp16 = mul(x = var_17232_cast_fp16, y = var_17233_to_fp16)[name = tensor("aw_1419_cast_fp16")]; + tensor var_17236_equation_0 = const()[name = tensor("op_17236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17236_cast_fp16 = einsum(equation = var_17236_equation_0, values = (var_17078_cast_fp16, var_16995_cast_fp16))[name = tensor("op_17236_cast_fp16")]; + tensor var_17237_to_fp16 = const()[name = tensor("op_17237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1421_cast_fp16 = mul(x = var_17236_cast_fp16, y = var_17237_to_fp16)[name = tensor("aw_1421_cast_fp16")]; + tensor var_17240_equation_0 = const()[name = tensor("op_17240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17240_cast_fp16 = einsum(equation = var_17240_equation_0, values = (var_17082_cast_fp16, var_16999_cast_fp16))[name = tensor("op_17240_cast_fp16")]; + tensor var_17241_to_fp16 = const()[name = tensor("op_17241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1423_cast_fp16 = mul(x = var_17240_cast_fp16, y = var_17241_to_fp16)[name = tensor("aw_1423_cast_fp16")]; + tensor var_17244_equation_0 = const()[name = tensor("op_17244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17244_cast_fp16 = einsum(equation = var_17244_equation_0, values = (var_17086_cast_fp16, var_17003_cast_fp16))[name = tensor("op_17244_cast_fp16")]; + tensor var_17245_to_fp16 = const()[name = tensor("op_17245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1425_cast_fp16 = mul(x = var_17244_cast_fp16, y = var_17245_to_fp16)[name = tensor("aw_1425_cast_fp16")]; + tensor var_17248_equation_0 = const()[name = tensor("op_17248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17248_cast_fp16 = einsum(equation = var_17248_equation_0, values = (var_17090_cast_fp16, var_17007_cast_fp16))[name = tensor("op_17248_cast_fp16")]; + tensor var_17249_to_fp16 = const()[name = tensor("op_17249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1427_cast_fp16 = mul(x = var_17248_cast_fp16, y = var_17249_to_fp16)[name = tensor("aw_1427_cast_fp16")]; + tensor var_17252_equation_0 = const()[name = tensor("op_17252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17252_cast_fp16 = einsum(equation = var_17252_equation_0, values = (var_17094_cast_fp16, var_17011_cast_fp16))[name = tensor("op_17252_cast_fp16")]; + tensor var_17253_to_fp16 = const()[name = tensor("op_17253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1429_cast_fp16 = mul(x = var_17252_cast_fp16, y = var_17253_to_fp16)[name = tensor("aw_1429_cast_fp16")]; + tensor var_17256_equation_0 = const()[name = tensor("op_17256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17256_cast_fp16 = einsum(equation = var_17256_equation_0, values = (var_17098_cast_fp16, var_17015_cast_fp16))[name = tensor("op_17256_cast_fp16")]; + tensor var_17257_to_fp16 = const()[name = tensor("op_17257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1431_cast_fp16 = mul(x = var_17256_cast_fp16, y = var_17257_to_fp16)[name = tensor("aw_1431_cast_fp16")]; + tensor var_17260_equation_0 = const()[name = tensor("op_17260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17260_cast_fp16 = einsum(equation = var_17260_equation_0, values = (var_17102_cast_fp16, var_17019_cast_fp16))[name = tensor("op_17260_cast_fp16")]; + tensor var_17261_to_fp16 = const()[name = tensor("op_17261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1433_cast_fp16 = mul(x = var_17260_cast_fp16, y = var_17261_to_fp16)[name = tensor("aw_1433_cast_fp16")]; + tensor var_17264_equation_0 = const()[name = tensor("op_17264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17264_cast_fp16 = einsum(equation = var_17264_equation_0, values = (var_17106_cast_fp16, var_17023_cast_fp16))[name = tensor("op_17264_cast_fp16")]; + tensor var_17265_to_fp16 = const()[name = tensor("op_17265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1435_cast_fp16 = mul(x = var_17264_cast_fp16, y = var_17265_to_fp16)[name = tensor("aw_1435_cast_fp16")]; + tensor var_17268_equation_0 = const()[name = tensor("op_17268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17268_cast_fp16 = einsum(equation = var_17268_equation_0, values = (var_17110_cast_fp16, var_17027_cast_fp16))[name = tensor("op_17268_cast_fp16")]; + tensor var_17269_to_fp16 = const()[name = tensor("op_17269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1437_cast_fp16 = mul(x = var_17268_cast_fp16, y = var_17269_to_fp16)[name = tensor("aw_1437_cast_fp16")]; + tensor var_17272_equation_0 = const()[name = tensor("op_17272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17272_cast_fp16 = einsum(equation = var_17272_equation_0, values = (var_17114_cast_fp16, var_17031_cast_fp16))[name = tensor("op_17272_cast_fp16")]; + tensor var_17273_to_fp16 = const()[name = tensor("op_17273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1439_cast_fp16 = mul(x = var_17272_cast_fp16, y = var_17273_to_fp16)[name = tensor("aw_1439_cast_fp16")]; + tensor var_17275_cast_fp16 = softmax(axis = var_2624, x = aw_1401_cast_fp16)[name = tensor("op_17275_cast_fp16")]; + tensor var_17276_cast_fp16 = softmax(axis = var_2624, x = aw_1403_cast_fp16)[name = tensor("op_17276_cast_fp16")]; + tensor var_17277_cast_fp16 = softmax(axis = var_2624, x = aw_1405_cast_fp16)[name = tensor("op_17277_cast_fp16")]; + tensor var_17278_cast_fp16 = softmax(axis = var_2624, x = aw_1407_cast_fp16)[name = tensor("op_17278_cast_fp16")]; + tensor var_17279_cast_fp16 = softmax(axis = var_2624, x = aw_1409_cast_fp16)[name = tensor("op_17279_cast_fp16")]; + tensor var_17280_cast_fp16 = softmax(axis = var_2624, x = aw_1411_cast_fp16)[name = tensor("op_17280_cast_fp16")]; + tensor var_17281_cast_fp16 = softmax(axis = var_2624, x = aw_1413_cast_fp16)[name = tensor("op_17281_cast_fp16")]; + tensor var_17282_cast_fp16 = softmax(axis = var_2624, x = aw_1415_cast_fp16)[name = tensor("op_17282_cast_fp16")]; + tensor var_17283_cast_fp16 = softmax(axis = var_2624, x = aw_1417_cast_fp16)[name = tensor("op_17283_cast_fp16")]; + tensor var_17284_cast_fp16 = softmax(axis = var_2624, x = aw_1419_cast_fp16)[name = tensor("op_17284_cast_fp16")]; + tensor var_17285_cast_fp16 = softmax(axis = var_2624, x = aw_1421_cast_fp16)[name = tensor("op_17285_cast_fp16")]; + tensor var_17286_cast_fp16 = softmax(axis = var_2624, x = aw_1423_cast_fp16)[name = tensor("op_17286_cast_fp16")]; + tensor var_17287_cast_fp16 = softmax(axis = var_2624, x = aw_1425_cast_fp16)[name = tensor("op_17287_cast_fp16")]; + tensor var_17288_cast_fp16 = softmax(axis = var_2624, x = aw_1427_cast_fp16)[name = tensor("op_17288_cast_fp16")]; + tensor var_17289_cast_fp16 = softmax(axis = var_2624, x = aw_1429_cast_fp16)[name = tensor("op_17289_cast_fp16")]; + tensor var_17290_cast_fp16 = softmax(axis = var_2624, x = aw_1431_cast_fp16)[name = tensor("op_17290_cast_fp16")]; + tensor var_17291_cast_fp16 = softmax(axis = var_2624, x = aw_1433_cast_fp16)[name = tensor("op_17291_cast_fp16")]; + tensor var_17292_cast_fp16 = softmax(axis = var_2624, x = aw_1435_cast_fp16)[name = tensor("op_17292_cast_fp16")]; + tensor var_17293_cast_fp16 = softmax(axis = var_2624, x = aw_1437_cast_fp16)[name = tensor("op_17293_cast_fp16")]; + tensor var_17294_cast_fp16 = softmax(axis = var_2624, x = aw_1439_cast_fp16)[name = tensor("op_17294_cast_fp16")]; + tensor var_17296_equation_0 = const()[name = tensor("op_17296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17296_cast_fp16 = einsum(equation = var_17296_equation_0, values = (var_17116_cast_fp16, var_17275_cast_fp16))[name = tensor("op_17296_cast_fp16")]; + tensor var_17298_equation_0 = const()[name = tensor("op_17298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17298_cast_fp16 = einsum(equation = var_17298_equation_0, values = (var_17120_cast_fp16, var_17276_cast_fp16))[name = tensor("op_17298_cast_fp16")]; + tensor var_17300_equation_0 = const()[name = tensor("op_17300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17300_cast_fp16 = einsum(equation = var_17300_equation_0, values = (var_17124_cast_fp16, var_17277_cast_fp16))[name = tensor("op_17300_cast_fp16")]; + tensor var_17302_equation_0 = const()[name = tensor("op_17302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17302_cast_fp16 = einsum(equation = var_17302_equation_0, values = (var_17128_cast_fp16, var_17278_cast_fp16))[name = tensor("op_17302_cast_fp16")]; + tensor var_17304_equation_0 = const()[name = tensor("op_17304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17304_cast_fp16 = einsum(equation = var_17304_equation_0, values = (var_17132_cast_fp16, var_17279_cast_fp16))[name = tensor("op_17304_cast_fp16")]; + tensor var_17306_equation_0 = const()[name = tensor("op_17306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17306_cast_fp16 = einsum(equation = var_17306_equation_0, values = (var_17136_cast_fp16, var_17280_cast_fp16))[name = tensor("op_17306_cast_fp16")]; + tensor var_17308_equation_0 = const()[name = tensor("op_17308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17308_cast_fp16 = einsum(equation = var_17308_equation_0, values = (var_17140_cast_fp16, var_17281_cast_fp16))[name = tensor("op_17308_cast_fp16")]; + tensor var_17310_equation_0 = const()[name = tensor("op_17310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17310_cast_fp16 = einsum(equation = var_17310_equation_0, values = (var_17144_cast_fp16, var_17282_cast_fp16))[name = tensor("op_17310_cast_fp16")]; + tensor var_17312_equation_0 = const()[name = tensor("op_17312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17312_cast_fp16 = einsum(equation = var_17312_equation_0, values = (var_17148_cast_fp16, var_17283_cast_fp16))[name = tensor("op_17312_cast_fp16")]; + tensor var_17314_equation_0 = const()[name = tensor("op_17314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17314_cast_fp16 = einsum(equation = var_17314_equation_0, values = (var_17152_cast_fp16, var_17284_cast_fp16))[name = tensor("op_17314_cast_fp16")]; + tensor var_17316_equation_0 = const()[name = tensor("op_17316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17316_cast_fp16 = einsum(equation = var_17316_equation_0, values = (var_17156_cast_fp16, var_17285_cast_fp16))[name = tensor("op_17316_cast_fp16")]; + tensor var_17318_equation_0 = const()[name = tensor("op_17318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17318_cast_fp16 = einsum(equation = var_17318_equation_0, values = (var_17160_cast_fp16, var_17286_cast_fp16))[name = tensor("op_17318_cast_fp16")]; + tensor var_17320_equation_0 = const()[name = tensor("op_17320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17320_cast_fp16 = einsum(equation = var_17320_equation_0, values = (var_17164_cast_fp16, var_17287_cast_fp16))[name = tensor("op_17320_cast_fp16")]; + tensor var_17322_equation_0 = const()[name = tensor("op_17322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17322_cast_fp16 = einsum(equation = var_17322_equation_0, values = (var_17168_cast_fp16, var_17288_cast_fp16))[name = tensor("op_17322_cast_fp16")]; + tensor var_17324_equation_0 = const()[name = tensor("op_17324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17324_cast_fp16 = einsum(equation = var_17324_equation_0, values = (var_17172_cast_fp16, var_17289_cast_fp16))[name = tensor("op_17324_cast_fp16")]; + tensor var_17326_equation_0 = const()[name = tensor("op_17326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17326_cast_fp16 = einsum(equation = var_17326_equation_0, values = (var_17176_cast_fp16, var_17290_cast_fp16))[name = tensor("op_17326_cast_fp16")]; + tensor var_17328_equation_0 = const()[name = tensor("op_17328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17328_cast_fp16 = einsum(equation = var_17328_equation_0, values = (var_17180_cast_fp16, var_17291_cast_fp16))[name = tensor("op_17328_cast_fp16")]; + tensor var_17330_equation_0 = const()[name = tensor("op_17330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17330_cast_fp16 = einsum(equation = var_17330_equation_0, values = (var_17184_cast_fp16, var_17292_cast_fp16))[name = tensor("op_17330_cast_fp16")]; + tensor var_17332_equation_0 = const()[name = tensor("op_17332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17332_cast_fp16 = einsum(equation = var_17332_equation_0, values = (var_17188_cast_fp16, var_17293_cast_fp16))[name = tensor("op_17332_cast_fp16")]; + tensor var_17334_equation_0 = const()[name = tensor("op_17334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17334_cast_fp16 = einsum(equation = var_17334_equation_0, values = (var_17192_cast_fp16, var_17294_cast_fp16))[name = tensor("op_17334_cast_fp16")]; + tensor input_271_interleave_0 = const()[name = tensor("input_271_interleave_0"), val = tensor(false)]; + tensor input_271_cast_fp16 = concat(axis = var_2624, interleave = input_271_interleave_0, values = (var_17296_cast_fp16, var_17298_cast_fp16, var_17300_cast_fp16, var_17302_cast_fp16, var_17304_cast_fp16, var_17306_cast_fp16, var_17308_cast_fp16, var_17310_cast_fp16, var_17312_cast_fp16, var_17314_cast_fp16, var_17316_cast_fp16, var_17318_cast_fp16, var_17320_cast_fp16, var_17322_cast_fp16, var_17324_cast_fp16, var_17326_cast_fp16, var_17328_cast_fp16, var_17330_cast_fp16, var_17332_cast_fp16, var_17334_cast_fp16))[name = tensor("input_271_cast_fp16")]; + tensor var_17344_pad_type_0 = const()[name = tensor("op_17344_pad_type_0"), val = tensor("valid")]; + tensor var_17344_strides_0 = const()[name = tensor("op_17344_strides_0"), val = tensor([1, 1])]; + tensor var_17344_pad_0 = const()[name = tensor("op_17344_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17344_dilations_0 = const()[name = tensor("op_17344_dilations_0"), val = tensor([1, 1])]; + tensor var_17344_groups_0 = const()[name = tensor("op_17344_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501825152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503054016))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503054208)))]; + tensor var_17344_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_17344_dilations_0, groups = var_17344_groups_0, pad = var_17344_pad_0, pad_type = var_17344_pad_type_0, strides = var_17344_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = tensor("op_17344_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = var_17344_cast_fp16, y = inputs_117_cast_fp16)[name = tensor("inputs_119_cast_fp16")]; + tensor input_273_axes_0 = const()[name = tensor("input_273_axes_0"), val = tensor([1])]; + tensor input_273_gamma_0_to_fp16 = const()[name = tensor("input_273_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503056832)))]; + tensor input_273_beta_0_to_fp16 = const()[name = tensor("input_273_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503059456)))]; + tensor var_17354_to_fp16 = const()[name = tensor("op_17354_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_273_cast_fp16 = layer_norm(axes = input_273_axes_0, beta = input_273_beta_0_to_fp16, epsilon = var_17354_to_fp16, gamma = input_273_gamma_0_to_fp16, x = inputs_119_cast_fp16)[name = tensor("input_273_cast_fp16")]; + tensor var_17374_pad_type_0 = const()[name = tensor("op_17374_pad_type_0"), val = tensor("valid")]; + tensor var_17374_strides_0 = const()[name = tensor("op_17374_strides_0"), val = tensor([1, 1])]; + tensor var_17374_pad_0 = const()[name = tensor("op_17374_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17374_dilations_0 = const()[name = tensor("op_17374_dilations_0"), val = tensor([1, 1])]; + tensor var_17374_groups_0 = const()[name = tensor("op_17374_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503062080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512892544))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512892736)))]; + tensor var_17374_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_17374_dilations_0, groups = var_17374_groups_0, pad = var_17374_pad_0, pad_type = var_17374_pad_type_0, strides = var_17374_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized, x = input_273_cast_fp16)[name = tensor("op_17374_cast_fp16")]; + tensor var_17375_split_sizes_0 = const()[name = tensor("op_17375_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_17375_axis_0 = const()[name = tensor("op_17375_axis_0"), val = tensor(1)]; + tensor var_17375_cast_fp16_0, tensor var_17375_cast_fp16_1 = split(axis = var_17375_axis_0, split_sizes = var_17375_split_sizes_0, x = var_17374_cast_fp16)[name = tensor("op_17375_cast_fp16")]; + tensor var_17377_mode_0 = const()[name = tensor("op_17377_mode_0"), val = tensor("EXACT")]; + tensor var_17377_cast_fp16 = gelu(mode = var_17377_mode_0, x = var_17375_cast_fp16_1)[name = tensor("op_17377_cast_fp16")]; + tensor input_275_cast_fp16 = mul(x = var_17375_cast_fp16_0, y = var_17377_cast_fp16)[name = tensor("input_275_cast_fp16")]; + tensor var_17385_pad_type_0 = const()[name = tensor("op_17385_pad_type_0"), val = tensor("valid")]; + tensor var_17385_strides_0 = const()[name = tensor("op_17385_strides_0"), val = tensor([1, 1])]; + tensor var_17385_pad_0 = const()[name = tensor("op_17385_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17385_dilations_0 = const()[name = tensor("op_17385_dilations_0"), val = tensor([1, 1])]; + tensor var_17385_groups_0 = const()[name = tensor("op_17385_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512913280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517828544))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517828736)))]; + tensor var_17385_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_17385_dilations_0, groups = var_17385_groups_0, pad = var_17385_pad_0, pad_type = var_17385_pad_type_0, strides = var_17385_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized, x = input_275_cast_fp16)[name = tensor("op_17385_cast_fp16")]; + tensor inputs_121_cast_fp16 = add(x = var_17385_cast_fp16, y = inputs_119_cast_fp16)[name = tensor("inputs_121_cast_fp16")]; + tensor hidden_states_173_axes_0 = const()[name = tensor("hidden_states_173_axes_0"), val = tensor([1])]; + tensor hidden_states_173_gamma_0_to_fp16 = const()[name = tensor("hidden_states_173_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517831360)))]; + tensor hidden_states_173_beta_0_to_fp16 = const()[name = tensor("hidden_states_173_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517833984)))]; + tensor var_17401_to_fp16 = const()[name = tensor("op_17401_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_173_cast_fp16 = layer_norm(axes = hidden_states_173_axes_0, beta = hidden_states_173_beta_0_to_fp16, epsilon = var_17401_to_fp16, gamma = hidden_states_173_gamma_0_to_fp16, x = inputs_121_cast_fp16)[name = tensor("hidden_states_173_cast_fp16")]; + tensor q_81_pad_type_0 = const()[name = tensor("q_81_pad_type_0"), val = tensor("valid")]; + tensor q_81_strides_0 = const()[name = tensor("q_81_strides_0"), val = tensor([1, 1])]; + tensor q_81_pad_0 = const()[name = tensor("q_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_81_dilations_0 = const()[name = tensor("q_81_dilations_0"), val = tensor([1, 1])]; + tensor q_81_groups_0 = const()[name = tensor("q_81_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517836608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519065472))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_81_cast_fp16 = conv(dilations = q_81_dilations_0, groups = q_81_groups_0, pad = q_81_pad_0, pad_type = q_81_pad_type_0, strides = q_81_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_173_cast_fp16)[name = tensor("q_81_cast_fp16")]; + tensor k_161_pad_type_0 = const()[name = tensor("k_161_pad_type_0"), val = tensor("valid")]; + tensor k_161_strides_0 = const()[name = tensor("k_161_strides_0"), val = tensor([1, 1])]; + tensor k_161_pad_0 = const()[name = tensor("k_161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_161_dilations_0 = const()[name = tensor("k_161_dilations_0"), val = tensor([1, 1])]; + tensor k_161_groups_0 = const()[name = tensor("k_161_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519065664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520294528))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_161_cast_fp16 = conv(dilations = k_161_dilations_0, groups = k_161_groups_0, pad = k_161_pad_0, pad_type = k_161_pad_type_0, strides = k_161_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_173_cast_fp16)[name = tensor("k_161_cast_fp16")]; + tensor v_81_pad_type_0 = const()[name = tensor("v_81_pad_type_0"), val = tensor("valid")]; + tensor v_81_strides_0 = const()[name = tensor("v_81_strides_0"), val = tensor([1, 1])]; + tensor v_81_pad_0 = const()[name = tensor("v_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_81_dilations_0 = const()[name = tensor("v_81_dilations_0"), val = tensor([1, 1])]; + tensor v_81_groups_0 = const()[name = tensor("v_81_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520294720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521523584))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_81_cast_fp16 = conv(dilations = v_81_dilations_0, groups = v_81_groups_0, pad = v_81_pad_0, pad_type = v_81_pad_type_0, strides = v_81_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_173_cast_fp16)[name = tensor("v_81_cast_fp16")]; + tensor var_17434_begin_0 = const()[name = tensor("op_17434_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17434_end_0 = const()[name = tensor("op_17434_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_17434_end_mask_0 = const()[name = tensor("op_17434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17434_cast_fp16 = slice_by_index(begin = var_17434_begin_0, end = var_17434_end_0, end_mask = var_17434_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17434_cast_fp16")]; + tensor var_17438_begin_0 = const()[name = tensor("op_17438_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17438_end_0 = const()[name = tensor("op_17438_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_17438_end_mask_0 = const()[name = tensor("op_17438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17438_cast_fp16 = slice_by_index(begin = var_17438_begin_0, end = var_17438_end_0, end_mask = var_17438_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17438_cast_fp16")]; + tensor var_17442_begin_0 = const()[name = tensor("op_17442_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17442_end_0 = const()[name = tensor("op_17442_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_17442_end_mask_0 = const()[name = tensor("op_17442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17442_cast_fp16 = slice_by_index(begin = var_17442_begin_0, end = var_17442_end_0, end_mask = var_17442_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17442_cast_fp16")]; + tensor var_17446_begin_0 = const()[name = tensor("op_17446_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17446_end_0 = const()[name = tensor("op_17446_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_17446_end_mask_0 = const()[name = tensor("op_17446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17446_cast_fp16 = slice_by_index(begin = var_17446_begin_0, end = var_17446_end_0, end_mask = var_17446_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17446_cast_fp16")]; + tensor var_17450_begin_0 = const()[name = tensor("op_17450_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17450_end_0 = const()[name = tensor("op_17450_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_17450_end_mask_0 = const()[name = tensor("op_17450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17450_cast_fp16 = slice_by_index(begin = var_17450_begin_0, end = var_17450_end_0, end_mask = var_17450_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17450_cast_fp16")]; + tensor var_17454_begin_0 = const()[name = tensor("op_17454_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17454_end_0 = const()[name = tensor("op_17454_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_17454_end_mask_0 = const()[name = tensor("op_17454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17454_cast_fp16 = slice_by_index(begin = var_17454_begin_0, end = var_17454_end_0, end_mask = var_17454_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17454_cast_fp16")]; + tensor var_17458_begin_0 = const()[name = tensor("op_17458_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17458_end_0 = const()[name = tensor("op_17458_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_17458_end_mask_0 = const()[name = tensor("op_17458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17458_cast_fp16 = slice_by_index(begin = var_17458_begin_0, end = var_17458_end_0, end_mask = var_17458_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17458_cast_fp16")]; + tensor var_17462_begin_0 = const()[name = tensor("op_17462_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17462_end_0 = const()[name = tensor("op_17462_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_17462_end_mask_0 = const()[name = tensor("op_17462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17462_cast_fp16 = slice_by_index(begin = var_17462_begin_0, end = var_17462_end_0, end_mask = var_17462_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17462_cast_fp16")]; + tensor var_17466_begin_0 = const()[name = tensor("op_17466_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17466_end_0 = const()[name = tensor("op_17466_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_17466_end_mask_0 = const()[name = tensor("op_17466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17466_cast_fp16 = slice_by_index(begin = var_17466_begin_0, end = var_17466_end_0, end_mask = var_17466_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17466_cast_fp16")]; + tensor var_17470_begin_0 = const()[name = tensor("op_17470_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17470_end_0 = const()[name = tensor("op_17470_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_17470_end_mask_0 = const()[name = tensor("op_17470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17470_cast_fp16 = slice_by_index(begin = var_17470_begin_0, end = var_17470_end_0, end_mask = var_17470_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17470_cast_fp16")]; + tensor var_17474_begin_0 = const()[name = tensor("op_17474_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17474_end_0 = const()[name = tensor("op_17474_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_17474_end_mask_0 = const()[name = tensor("op_17474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17474_cast_fp16 = slice_by_index(begin = var_17474_begin_0, end = var_17474_end_0, end_mask = var_17474_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17474_cast_fp16")]; + tensor var_17478_begin_0 = const()[name = tensor("op_17478_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17478_end_0 = const()[name = tensor("op_17478_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_17478_end_mask_0 = const()[name = tensor("op_17478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17478_cast_fp16 = slice_by_index(begin = var_17478_begin_0, end = var_17478_end_0, end_mask = var_17478_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17478_cast_fp16")]; + tensor var_17482_begin_0 = const()[name = tensor("op_17482_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17482_end_0 = const()[name = tensor("op_17482_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17482_end_mask_0 = const()[name = tensor("op_17482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17482_cast_fp16 = slice_by_index(begin = var_17482_begin_0, end = var_17482_end_0, end_mask = var_17482_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17482_cast_fp16")]; + tensor var_17486_begin_0 = const()[name = tensor("op_17486_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17486_end_0 = const()[name = tensor("op_17486_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17486_end_mask_0 = const()[name = tensor("op_17486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17486_cast_fp16 = slice_by_index(begin = var_17486_begin_0, end = var_17486_end_0, end_mask = var_17486_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17486_cast_fp16")]; + tensor var_17490_begin_0 = const()[name = tensor("op_17490_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17490_end_0 = const()[name = tensor("op_17490_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17490_end_mask_0 = const()[name = tensor("op_17490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17490_cast_fp16 = slice_by_index(begin = var_17490_begin_0, end = var_17490_end_0, end_mask = var_17490_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17490_cast_fp16")]; + tensor var_17494_begin_0 = const()[name = tensor("op_17494_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17494_end_0 = const()[name = tensor("op_17494_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17494_end_mask_0 = const()[name = tensor("op_17494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17494_cast_fp16 = slice_by_index(begin = var_17494_begin_0, end = var_17494_end_0, end_mask = var_17494_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17494_cast_fp16")]; + tensor var_17498_begin_0 = const()[name = tensor("op_17498_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17498_end_0 = const()[name = tensor("op_17498_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17498_end_mask_0 = const()[name = tensor("op_17498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17498_cast_fp16 = slice_by_index(begin = var_17498_begin_0, end = var_17498_end_0, end_mask = var_17498_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17498_cast_fp16")]; + tensor var_17502_begin_0 = const()[name = tensor("op_17502_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17502_end_0 = const()[name = tensor("op_17502_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17502_end_mask_0 = const()[name = tensor("op_17502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17502_cast_fp16 = slice_by_index(begin = var_17502_begin_0, end = var_17502_end_0, end_mask = var_17502_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17502_cast_fp16")]; + tensor var_17506_begin_0 = const()[name = tensor("op_17506_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17506_end_0 = const()[name = tensor("op_17506_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17506_end_mask_0 = const()[name = tensor("op_17506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17506_cast_fp16 = slice_by_index(begin = var_17506_begin_0, end = var_17506_end_0, end_mask = var_17506_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17506_cast_fp16")]; + tensor var_17510_begin_0 = const()[name = tensor("op_17510_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17510_end_0 = const()[name = tensor("op_17510_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17510_end_mask_0 = const()[name = tensor("op_17510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17510_cast_fp16 = slice_by_index(begin = var_17510_begin_0, end = var_17510_end_0, end_mask = var_17510_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17510_cast_fp16")]; + tensor k_163_perm_0 = const()[name = tensor("k_163_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17517_begin_0 = const()[name = tensor("op_17517_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17517_end_0 = const()[name = tensor("op_17517_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_17517_end_mask_0 = const()[name = tensor("op_17517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_163_cast_fp16 = transpose(perm = k_163_perm_0, x = k_161_cast_fp16)[name = tensor("transpose_27")]; + tensor var_17517_cast_fp16 = slice_by_index(begin = var_17517_begin_0, end = var_17517_end_0, end_mask = var_17517_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17517_cast_fp16")]; + tensor var_17521_begin_0 = const()[name = tensor("op_17521_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17521_end_0 = const()[name = tensor("op_17521_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_17521_end_mask_0 = const()[name = tensor("op_17521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17521_cast_fp16 = slice_by_index(begin = var_17521_begin_0, end = var_17521_end_0, end_mask = var_17521_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17521_cast_fp16")]; + tensor var_17525_begin_0 = const()[name = tensor("op_17525_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17525_end_0 = const()[name = tensor("op_17525_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_17525_end_mask_0 = const()[name = tensor("op_17525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17525_cast_fp16 = slice_by_index(begin = var_17525_begin_0, end = var_17525_end_0, end_mask = var_17525_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17525_cast_fp16")]; + tensor var_17529_begin_0 = const()[name = tensor("op_17529_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17529_end_0 = const()[name = tensor("op_17529_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_17529_end_mask_0 = const()[name = tensor("op_17529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17529_cast_fp16 = slice_by_index(begin = var_17529_begin_0, end = var_17529_end_0, end_mask = var_17529_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17529_cast_fp16")]; + tensor var_17533_begin_0 = const()[name = tensor("op_17533_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17533_end_0 = const()[name = tensor("op_17533_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_17533_end_mask_0 = const()[name = tensor("op_17533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17533_cast_fp16 = slice_by_index(begin = var_17533_begin_0, end = var_17533_end_0, end_mask = var_17533_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17533_cast_fp16")]; + tensor var_17537_begin_0 = const()[name = tensor("op_17537_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17537_end_0 = const()[name = tensor("op_17537_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_17537_end_mask_0 = const()[name = tensor("op_17537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17537_cast_fp16 = slice_by_index(begin = var_17537_begin_0, end = var_17537_end_0, end_mask = var_17537_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17537_cast_fp16")]; + tensor var_17541_begin_0 = const()[name = tensor("op_17541_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17541_end_0 = const()[name = tensor("op_17541_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_17541_end_mask_0 = const()[name = tensor("op_17541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17541_cast_fp16 = slice_by_index(begin = var_17541_begin_0, end = var_17541_end_0, end_mask = var_17541_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17541_cast_fp16")]; + tensor var_17545_begin_0 = const()[name = tensor("op_17545_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17545_end_0 = const()[name = tensor("op_17545_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_17545_end_mask_0 = const()[name = tensor("op_17545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17545_cast_fp16 = slice_by_index(begin = var_17545_begin_0, end = var_17545_end_0, end_mask = var_17545_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17545_cast_fp16")]; + tensor var_17549_begin_0 = const()[name = tensor("op_17549_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17549_end_0 = const()[name = tensor("op_17549_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_17549_end_mask_0 = const()[name = tensor("op_17549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17549_cast_fp16 = slice_by_index(begin = var_17549_begin_0, end = var_17549_end_0, end_mask = var_17549_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17549_cast_fp16")]; + tensor var_17553_begin_0 = const()[name = tensor("op_17553_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17553_end_0 = const()[name = tensor("op_17553_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_17553_end_mask_0 = const()[name = tensor("op_17553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17553_cast_fp16 = slice_by_index(begin = var_17553_begin_0, end = var_17553_end_0, end_mask = var_17553_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17553_cast_fp16")]; + tensor var_17557_begin_0 = const()[name = tensor("op_17557_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17557_end_0 = const()[name = tensor("op_17557_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_17557_end_mask_0 = const()[name = tensor("op_17557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17557_cast_fp16 = slice_by_index(begin = var_17557_begin_0, end = var_17557_end_0, end_mask = var_17557_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17557_cast_fp16")]; + tensor var_17561_begin_0 = const()[name = tensor("op_17561_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17561_end_0 = const()[name = tensor("op_17561_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_17561_end_mask_0 = const()[name = tensor("op_17561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17561_cast_fp16 = slice_by_index(begin = var_17561_begin_0, end = var_17561_end_0, end_mask = var_17561_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17561_cast_fp16")]; + tensor var_17565_begin_0 = const()[name = tensor("op_17565_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17565_end_0 = const()[name = tensor("op_17565_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_17565_end_mask_0 = const()[name = tensor("op_17565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17565_cast_fp16 = slice_by_index(begin = var_17565_begin_0, end = var_17565_end_0, end_mask = var_17565_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17565_cast_fp16")]; + tensor var_17569_begin_0 = const()[name = tensor("op_17569_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_17569_end_0 = const()[name = tensor("op_17569_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_17569_end_mask_0 = const()[name = tensor("op_17569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17569_cast_fp16 = slice_by_index(begin = var_17569_begin_0, end = var_17569_end_0, end_mask = var_17569_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17569_cast_fp16")]; + tensor var_17573_begin_0 = const()[name = tensor("op_17573_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_17573_end_0 = const()[name = tensor("op_17573_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_17573_end_mask_0 = const()[name = tensor("op_17573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17573_cast_fp16 = slice_by_index(begin = var_17573_begin_0, end = var_17573_end_0, end_mask = var_17573_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17573_cast_fp16")]; + tensor var_17577_begin_0 = const()[name = tensor("op_17577_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_17577_end_0 = const()[name = tensor("op_17577_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17577_end_mask_0 = const()[name = tensor("op_17577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17577_cast_fp16 = slice_by_index(begin = var_17577_begin_0, end = var_17577_end_0, end_mask = var_17577_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17577_cast_fp16")]; + tensor var_17581_begin_0 = const()[name = tensor("op_17581_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_17581_end_0 = const()[name = tensor("op_17581_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_17581_end_mask_0 = const()[name = tensor("op_17581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17581_cast_fp16 = slice_by_index(begin = var_17581_begin_0, end = var_17581_end_0, end_mask = var_17581_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17581_cast_fp16")]; + tensor var_17585_begin_0 = const()[name = tensor("op_17585_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_17585_end_0 = const()[name = tensor("op_17585_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_17585_end_mask_0 = const()[name = tensor("op_17585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17585_cast_fp16 = slice_by_index(begin = var_17585_begin_0, end = var_17585_end_0, end_mask = var_17585_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17585_cast_fp16")]; + tensor var_17589_begin_0 = const()[name = tensor("op_17589_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_17589_end_0 = const()[name = tensor("op_17589_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_17589_end_mask_0 = const()[name = tensor("op_17589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17589_cast_fp16 = slice_by_index(begin = var_17589_begin_0, end = var_17589_end_0, end_mask = var_17589_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17589_cast_fp16")]; + tensor var_17593_begin_0 = const()[name = tensor("op_17593_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_17593_end_0 = const()[name = tensor("op_17593_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_17593_end_mask_0 = const()[name = tensor("op_17593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17593_cast_fp16 = slice_by_index(begin = var_17593_begin_0, end = var_17593_end_0, end_mask = var_17593_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17593_cast_fp16")]; + tensor var_17595_begin_0 = const()[name = tensor("op_17595_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17595_end_0 = const()[name = tensor("op_17595_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_17595_end_mask_0 = const()[name = tensor("op_17595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17595_cast_fp16 = slice_by_index(begin = var_17595_begin_0, end = var_17595_end_0, end_mask = var_17595_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17595_cast_fp16")]; + tensor var_17599_begin_0 = const()[name = tensor("op_17599_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17599_end_0 = const()[name = tensor("op_17599_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_17599_end_mask_0 = const()[name = tensor("op_17599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17599_cast_fp16 = slice_by_index(begin = var_17599_begin_0, end = var_17599_end_0, end_mask = var_17599_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17599_cast_fp16")]; + tensor var_17603_begin_0 = const()[name = tensor("op_17603_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17603_end_0 = const()[name = tensor("op_17603_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_17603_end_mask_0 = const()[name = tensor("op_17603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17603_cast_fp16 = slice_by_index(begin = var_17603_begin_0, end = var_17603_end_0, end_mask = var_17603_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17603_cast_fp16")]; + tensor var_17607_begin_0 = const()[name = tensor("op_17607_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17607_end_0 = const()[name = tensor("op_17607_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_17607_end_mask_0 = const()[name = tensor("op_17607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17607_cast_fp16 = slice_by_index(begin = var_17607_begin_0, end = var_17607_end_0, end_mask = var_17607_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17607_cast_fp16")]; + tensor var_17611_begin_0 = const()[name = tensor("op_17611_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17611_end_0 = const()[name = tensor("op_17611_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_17611_end_mask_0 = const()[name = tensor("op_17611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17611_cast_fp16 = slice_by_index(begin = var_17611_begin_0, end = var_17611_end_0, end_mask = var_17611_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17611_cast_fp16")]; + tensor var_17615_begin_0 = const()[name = tensor("op_17615_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17615_end_0 = const()[name = tensor("op_17615_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_17615_end_mask_0 = const()[name = tensor("op_17615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17615_cast_fp16 = slice_by_index(begin = var_17615_begin_0, end = var_17615_end_0, end_mask = var_17615_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17615_cast_fp16")]; + tensor var_17619_begin_0 = const()[name = tensor("op_17619_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17619_end_0 = const()[name = tensor("op_17619_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_17619_end_mask_0 = const()[name = tensor("op_17619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17619_cast_fp16 = slice_by_index(begin = var_17619_begin_0, end = var_17619_end_0, end_mask = var_17619_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17619_cast_fp16")]; + tensor var_17623_begin_0 = const()[name = tensor("op_17623_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17623_end_0 = const()[name = tensor("op_17623_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_17623_end_mask_0 = const()[name = tensor("op_17623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17623_cast_fp16 = slice_by_index(begin = var_17623_begin_0, end = var_17623_end_0, end_mask = var_17623_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17623_cast_fp16")]; + tensor var_17627_begin_0 = const()[name = tensor("op_17627_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17627_end_0 = const()[name = tensor("op_17627_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_17627_end_mask_0 = const()[name = tensor("op_17627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17627_cast_fp16 = slice_by_index(begin = var_17627_begin_0, end = var_17627_end_0, end_mask = var_17627_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17627_cast_fp16")]; + tensor var_17631_begin_0 = const()[name = tensor("op_17631_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17631_end_0 = const()[name = tensor("op_17631_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_17631_end_mask_0 = const()[name = tensor("op_17631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17631_cast_fp16 = slice_by_index(begin = var_17631_begin_0, end = var_17631_end_0, end_mask = var_17631_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17631_cast_fp16")]; + tensor var_17635_begin_0 = const()[name = tensor("op_17635_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17635_end_0 = const()[name = tensor("op_17635_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_17635_end_mask_0 = const()[name = tensor("op_17635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17635_cast_fp16 = slice_by_index(begin = var_17635_begin_0, end = var_17635_end_0, end_mask = var_17635_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17635_cast_fp16")]; + tensor var_17639_begin_0 = const()[name = tensor("op_17639_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17639_end_0 = const()[name = tensor("op_17639_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_17639_end_mask_0 = const()[name = tensor("op_17639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17639_cast_fp16 = slice_by_index(begin = var_17639_begin_0, end = var_17639_end_0, end_mask = var_17639_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17639_cast_fp16")]; + tensor var_17643_begin_0 = const()[name = tensor("op_17643_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17643_end_0 = const()[name = tensor("op_17643_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17643_end_mask_0 = const()[name = tensor("op_17643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17643_cast_fp16 = slice_by_index(begin = var_17643_begin_0, end = var_17643_end_0, end_mask = var_17643_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17643_cast_fp16")]; + tensor var_17647_begin_0 = const()[name = tensor("op_17647_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17647_end_0 = const()[name = tensor("op_17647_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17647_end_mask_0 = const()[name = tensor("op_17647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17647_cast_fp16 = slice_by_index(begin = var_17647_begin_0, end = var_17647_end_0, end_mask = var_17647_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17647_cast_fp16")]; + tensor var_17651_begin_0 = const()[name = tensor("op_17651_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17651_end_0 = const()[name = tensor("op_17651_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17651_end_mask_0 = const()[name = tensor("op_17651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17651_cast_fp16 = slice_by_index(begin = var_17651_begin_0, end = var_17651_end_0, end_mask = var_17651_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17651_cast_fp16")]; + tensor var_17655_begin_0 = const()[name = tensor("op_17655_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17655_end_0 = const()[name = tensor("op_17655_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17655_end_mask_0 = const()[name = tensor("op_17655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17655_cast_fp16 = slice_by_index(begin = var_17655_begin_0, end = var_17655_end_0, end_mask = var_17655_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17655_cast_fp16")]; + tensor var_17659_begin_0 = const()[name = tensor("op_17659_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17659_end_0 = const()[name = tensor("op_17659_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17659_end_mask_0 = const()[name = tensor("op_17659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17659_cast_fp16 = slice_by_index(begin = var_17659_begin_0, end = var_17659_end_0, end_mask = var_17659_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17659_cast_fp16")]; + tensor var_17663_begin_0 = const()[name = tensor("op_17663_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17663_end_0 = const()[name = tensor("op_17663_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17663_end_mask_0 = const()[name = tensor("op_17663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17663_cast_fp16 = slice_by_index(begin = var_17663_begin_0, end = var_17663_end_0, end_mask = var_17663_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17663_cast_fp16")]; + tensor var_17667_begin_0 = const()[name = tensor("op_17667_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17667_end_0 = const()[name = tensor("op_17667_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17667_end_mask_0 = const()[name = tensor("op_17667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17667_cast_fp16 = slice_by_index(begin = var_17667_begin_0, end = var_17667_end_0, end_mask = var_17667_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17667_cast_fp16")]; + tensor var_17671_begin_0 = const()[name = tensor("op_17671_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17671_end_0 = const()[name = tensor("op_17671_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17671_end_mask_0 = const()[name = tensor("op_17671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17671_cast_fp16 = slice_by_index(begin = var_17671_begin_0, end = var_17671_end_0, end_mask = var_17671_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17671_cast_fp16")]; + tensor var_17675_equation_0 = const()[name = tensor("op_17675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17675_cast_fp16 = einsum(equation = var_17675_equation_0, values = (var_17517_cast_fp16, var_17434_cast_fp16))[name = tensor("op_17675_cast_fp16")]; + tensor var_17676_to_fp16 = const()[name = tensor("op_17676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1441_cast_fp16 = mul(x = var_17675_cast_fp16, y = var_17676_to_fp16)[name = tensor("aw_1441_cast_fp16")]; + tensor var_17679_equation_0 = const()[name = tensor("op_17679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17679_cast_fp16 = einsum(equation = var_17679_equation_0, values = (var_17521_cast_fp16, var_17438_cast_fp16))[name = tensor("op_17679_cast_fp16")]; + tensor var_17680_to_fp16 = const()[name = tensor("op_17680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1443_cast_fp16 = mul(x = var_17679_cast_fp16, y = var_17680_to_fp16)[name = tensor("aw_1443_cast_fp16")]; + tensor var_17683_equation_0 = const()[name = tensor("op_17683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17683_cast_fp16 = einsum(equation = var_17683_equation_0, values = (var_17525_cast_fp16, var_17442_cast_fp16))[name = tensor("op_17683_cast_fp16")]; + tensor var_17684_to_fp16 = const()[name = tensor("op_17684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1445_cast_fp16 = mul(x = var_17683_cast_fp16, y = var_17684_to_fp16)[name = tensor("aw_1445_cast_fp16")]; + tensor var_17687_equation_0 = const()[name = tensor("op_17687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17687_cast_fp16 = einsum(equation = var_17687_equation_0, values = (var_17529_cast_fp16, var_17446_cast_fp16))[name = tensor("op_17687_cast_fp16")]; + tensor var_17688_to_fp16 = const()[name = tensor("op_17688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1447_cast_fp16 = mul(x = var_17687_cast_fp16, y = var_17688_to_fp16)[name = tensor("aw_1447_cast_fp16")]; + tensor var_17691_equation_0 = const()[name = tensor("op_17691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17691_cast_fp16 = einsum(equation = var_17691_equation_0, values = (var_17533_cast_fp16, var_17450_cast_fp16))[name = tensor("op_17691_cast_fp16")]; + tensor var_17692_to_fp16 = const()[name = tensor("op_17692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1449_cast_fp16 = mul(x = var_17691_cast_fp16, y = var_17692_to_fp16)[name = tensor("aw_1449_cast_fp16")]; + tensor var_17695_equation_0 = const()[name = tensor("op_17695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17695_cast_fp16 = einsum(equation = var_17695_equation_0, values = (var_17537_cast_fp16, var_17454_cast_fp16))[name = tensor("op_17695_cast_fp16")]; + tensor var_17696_to_fp16 = const()[name = tensor("op_17696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1451_cast_fp16 = mul(x = var_17695_cast_fp16, y = var_17696_to_fp16)[name = tensor("aw_1451_cast_fp16")]; + tensor var_17699_equation_0 = const()[name = tensor("op_17699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17699_cast_fp16 = einsum(equation = var_17699_equation_0, values = (var_17541_cast_fp16, var_17458_cast_fp16))[name = tensor("op_17699_cast_fp16")]; + tensor var_17700_to_fp16 = const()[name = tensor("op_17700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1453_cast_fp16 = mul(x = var_17699_cast_fp16, y = var_17700_to_fp16)[name = tensor("aw_1453_cast_fp16")]; + tensor var_17703_equation_0 = const()[name = tensor("op_17703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17703_cast_fp16 = einsum(equation = var_17703_equation_0, values = (var_17545_cast_fp16, var_17462_cast_fp16))[name = tensor("op_17703_cast_fp16")]; + tensor var_17704_to_fp16 = const()[name = tensor("op_17704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1455_cast_fp16 = mul(x = var_17703_cast_fp16, y = var_17704_to_fp16)[name = tensor("aw_1455_cast_fp16")]; + tensor var_17707_equation_0 = const()[name = tensor("op_17707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17707_cast_fp16 = einsum(equation = var_17707_equation_0, values = (var_17549_cast_fp16, var_17466_cast_fp16))[name = tensor("op_17707_cast_fp16")]; + tensor var_17708_to_fp16 = const()[name = tensor("op_17708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1457_cast_fp16 = mul(x = var_17707_cast_fp16, y = var_17708_to_fp16)[name = tensor("aw_1457_cast_fp16")]; + tensor var_17711_equation_0 = const()[name = tensor("op_17711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17711_cast_fp16 = einsum(equation = var_17711_equation_0, values = (var_17553_cast_fp16, var_17470_cast_fp16))[name = tensor("op_17711_cast_fp16")]; + tensor var_17712_to_fp16 = const()[name = tensor("op_17712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1459_cast_fp16 = mul(x = var_17711_cast_fp16, y = var_17712_to_fp16)[name = tensor("aw_1459_cast_fp16")]; + tensor var_17715_equation_0 = const()[name = tensor("op_17715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17715_cast_fp16 = einsum(equation = var_17715_equation_0, values = (var_17557_cast_fp16, var_17474_cast_fp16))[name = tensor("op_17715_cast_fp16")]; + tensor var_17716_to_fp16 = const()[name = tensor("op_17716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1461_cast_fp16 = mul(x = var_17715_cast_fp16, y = var_17716_to_fp16)[name = tensor("aw_1461_cast_fp16")]; + tensor var_17719_equation_0 = const()[name = tensor("op_17719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17719_cast_fp16 = einsum(equation = var_17719_equation_0, values = (var_17561_cast_fp16, var_17478_cast_fp16))[name = tensor("op_17719_cast_fp16")]; + tensor var_17720_to_fp16 = const()[name = tensor("op_17720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1463_cast_fp16 = mul(x = var_17719_cast_fp16, y = var_17720_to_fp16)[name = tensor("aw_1463_cast_fp16")]; + tensor var_17723_equation_0 = const()[name = tensor("op_17723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17723_cast_fp16 = einsum(equation = var_17723_equation_0, values = (var_17565_cast_fp16, var_17482_cast_fp16))[name = tensor("op_17723_cast_fp16")]; + tensor var_17724_to_fp16 = const()[name = tensor("op_17724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1465_cast_fp16 = mul(x = var_17723_cast_fp16, y = var_17724_to_fp16)[name = tensor("aw_1465_cast_fp16")]; + tensor var_17727_equation_0 = const()[name = tensor("op_17727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17727_cast_fp16 = einsum(equation = var_17727_equation_0, values = (var_17569_cast_fp16, var_17486_cast_fp16))[name = tensor("op_17727_cast_fp16")]; + tensor var_17728_to_fp16 = const()[name = tensor("op_17728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1467_cast_fp16 = mul(x = var_17727_cast_fp16, y = var_17728_to_fp16)[name = tensor("aw_1467_cast_fp16")]; + tensor var_17731_equation_0 = const()[name = tensor("op_17731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17731_cast_fp16 = einsum(equation = var_17731_equation_0, values = (var_17573_cast_fp16, var_17490_cast_fp16))[name = tensor("op_17731_cast_fp16")]; + tensor var_17732_to_fp16 = const()[name = tensor("op_17732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1469_cast_fp16 = mul(x = var_17731_cast_fp16, y = var_17732_to_fp16)[name = tensor("aw_1469_cast_fp16")]; + tensor var_17735_equation_0 = const()[name = tensor("op_17735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17735_cast_fp16 = einsum(equation = var_17735_equation_0, values = (var_17577_cast_fp16, var_17494_cast_fp16))[name = tensor("op_17735_cast_fp16")]; + tensor var_17736_to_fp16 = const()[name = tensor("op_17736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1471_cast_fp16 = mul(x = var_17735_cast_fp16, y = var_17736_to_fp16)[name = tensor("aw_1471_cast_fp16")]; + tensor var_17739_equation_0 = const()[name = tensor("op_17739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17739_cast_fp16 = einsum(equation = var_17739_equation_0, values = (var_17581_cast_fp16, var_17498_cast_fp16))[name = tensor("op_17739_cast_fp16")]; + tensor var_17740_to_fp16 = const()[name = tensor("op_17740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1473_cast_fp16 = mul(x = var_17739_cast_fp16, y = var_17740_to_fp16)[name = tensor("aw_1473_cast_fp16")]; + tensor var_17743_equation_0 = const()[name = tensor("op_17743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17743_cast_fp16 = einsum(equation = var_17743_equation_0, values = (var_17585_cast_fp16, var_17502_cast_fp16))[name = tensor("op_17743_cast_fp16")]; + tensor var_17744_to_fp16 = const()[name = tensor("op_17744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1475_cast_fp16 = mul(x = var_17743_cast_fp16, y = var_17744_to_fp16)[name = tensor("aw_1475_cast_fp16")]; + tensor var_17747_equation_0 = const()[name = tensor("op_17747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17747_cast_fp16 = einsum(equation = var_17747_equation_0, values = (var_17589_cast_fp16, var_17506_cast_fp16))[name = tensor("op_17747_cast_fp16")]; + tensor var_17748_to_fp16 = const()[name = tensor("op_17748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1477_cast_fp16 = mul(x = var_17747_cast_fp16, y = var_17748_to_fp16)[name = tensor("aw_1477_cast_fp16")]; + tensor var_17751_equation_0 = const()[name = tensor("op_17751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17751_cast_fp16 = einsum(equation = var_17751_equation_0, values = (var_17593_cast_fp16, var_17510_cast_fp16))[name = tensor("op_17751_cast_fp16")]; + tensor var_17752_to_fp16 = const()[name = tensor("op_17752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1479_cast_fp16 = mul(x = var_17751_cast_fp16, y = var_17752_to_fp16)[name = tensor("aw_1479_cast_fp16")]; + tensor var_17754_cast_fp16 = softmax(axis = var_2624, x = aw_1441_cast_fp16)[name = tensor("op_17754_cast_fp16")]; + tensor var_17755_cast_fp16 = softmax(axis = var_2624, x = aw_1443_cast_fp16)[name = tensor("op_17755_cast_fp16")]; + tensor var_17756_cast_fp16 = softmax(axis = var_2624, x = aw_1445_cast_fp16)[name = tensor("op_17756_cast_fp16")]; + tensor var_17757_cast_fp16 = softmax(axis = var_2624, x = aw_1447_cast_fp16)[name = tensor("op_17757_cast_fp16")]; + tensor var_17758_cast_fp16 = softmax(axis = var_2624, x = aw_1449_cast_fp16)[name = tensor("op_17758_cast_fp16")]; + tensor var_17759_cast_fp16 = softmax(axis = var_2624, x = aw_1451_cast_fp16)[name = tensor("op_17759_cast_fp16")]; + tensor var_17760_cast_fp16 = softmax(axis = var_2624, x = aw_1453_cast_fp16)[name = tensor("op_17760_cast_fp16")]; + tensor var_17761_cast_fp16 = softmax(axis = var_2624, x = aw_1455_cast_fp16)[name = tensor("op_17761_cast_fp16")]; + tensor var_17762_cast_fp16 = softmax(axis = var_2624, x = aw_1457_cast_fp16)[name = tensor("op_17762_cast_fp16")]; + tensor var_17763_cast_fp16 = softmax(axis = var_2624, x = aw_1459_cast_fp16)[name = tensor("op_17763_cast_fp16")]; + tensor var_17764_cast_fp16 = softmax(axis = var_2624, x = aw_1461_cast_fp16)[name = tensor("op_17764_cast_fp16")]; + tensor var_17765_cast_fp16 = softmax(axis = var_2624, x = aw_1463_cast_fp16)[name = tensor("op_17765_cast_fp16")]; + tensor var_17766_cast_fp16 = softmax(axis = var_2624, x = aw_1465_cast_fp16)[name = tensor("op_17766_cast_fp16")]; + tensor var_17767_cast_fp16 = softmax(axis = var_2624, x = aw_1467_cast_fp16)[name = tensor("op_17767_cast_fp16")]; + tensor var_17768_cast_fp16 = softmax(axis = var_2624, x = aw_1469_cast_fp16)[name = tensor("op_17768_cast_fp16")]; + tensor var_17769_cast_fp16 = softmax(axis = var_2624, x = aw_1471_cast_fp16)[name = tensor("op_17769_cast_fp16")]; + tensor var_17770_cast_fp16 = softmax(axis = var_2624, x = aw_1473_cast_fp16)[name = tensor("op_17770_cast_fp16")]; + tensor var_17771_cast_fp16 = softmax(axis = var_2624, x = aw_1475_cast_fp16)[name = tensor("op_17771_cast_fp16")]; + tensor var_17772_cast_fp16 = softmax(axis = var_2624, x = aw_1477_cast_fp16)[name = tensor("op_17772_cast_fp16")]; + tensor var_17773_cast_fp16 = softmax(axis = var_2624, x = aw_1479_cast_fp16)[name = tensor("op_17773_cast_fp16")]; + tensor var_17775_equation_0 = const()[name = tensor("op_17775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17775_cast_fp16 = einsum(equation = var_17775_equation_0, values = (var_17595_cast_fp16, var_17754_cast_fp16))[name = tensor("op_17775_cast_fp16")]; + tensor var_17777_equation_0 = const()[name = tensor("op_17777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17777_cast_fp16 = einsum(equation = var_17777_equation_0, values = (var_17599_cast_fp16, var_17755_cast_fp16))[name = tensor("op_17777_cast_fp16")]; + tensor var_17779_equation_0 = const()[name = tensor("op_17779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17779_cast_fp16 = einsum(equation = var_17779_equation_0, values = (var_17603_cast_fp16, var_17756_cast_fp16))[name = tensor("op_17779_cast_fp16")]; + tensor var_17781_equation_0 = const()[name = tensor("op_17781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17781_cast_fp16 = einsum(equation = var_17781_equation_0, values = (var_17607_cast_fp16, var_17757_cast_fp16))[name = tensor("op_17781_cast_fp16")]; + tensor var_17783_equation_0 = const()[name = tensor("op_17783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17783_cast_fp16 = einsum(equation = var_17783_equation_0, values = (var_17611_cast_fp16, var_17758_cast_fp16))[name = tensor("op_17783_cast_fp16")]; + tensor var_17785_equation_0 = const()[name = tensor("op_17785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17785_cast_fp16 = einsum(equation = var_17785_equation_0, values = (var_17615_cast_fp16, var_17759_cast_fp16))[name = tensor("op_17785_cast_fp16")]; + tensor var_17787_equation_0 = const()[name = tensor("op_17787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17787_cast_fp16 = einsum(equation = var_17787_equation_0, values = (var_17619_cast_fp16, var_17760_cast_fp16))[name = tensor("op_17787_cast_fp16")]; + tensor var_17789_equation_0 = const()[name = tensor("op_17789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17789_cast_fp16 = einsum(equation = var_17789_equation_0, values = (var_17623_cast_fp16, var_17761_cast_fp16))[name = tensor("op_17789_cast_fp16")]; + tensor var_17791_equation_0 = const()[name = tensor("op_17791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17791_cast_fp16 = einsum(equation = var_17791_equation_0, values = (var_17627_cast_fp16, var_17762_cast_fp16))[name = tensor("op_17791_cast_fp16")]; + tensor var_17793_equation_0 = const()[name = tensor("op_17793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17793_cast_fp16 = einsum(equation = var_17793_equation_0, values = (var_17631_cast_fp16, var_17763_cast_fp16))[name = tensor("op_17793_cast_fp16")]; + tensor var_17795_equation_0 = const()[name = tensor("op_17795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17795_cast_fp16 = einsum(equation = var_17795_equation_0, values = (var_17635_cast_fp16, var_17764_cast_fp16))[name = tensor("op_17795_cast_fp16")]; + tensor var_17797_equation_0 = const()[name = tensor("op_17797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17797_cast_fp16 = einsum(equation = var_17797_equation_0, values = (var_17639_cast_fp16, var_17765_cast_fp16))[name = tensor("op_17797_cast_fp16")]; + tensor var_17799_equation_0 = const()[name = tensor("op_17799_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17799_cast_fp16 = einsum(equation = var_17799_equation_0, values = (var_17643_cast_fp16, var_17766_cast_fp16))[name = tensor("op_17799_cast_fp16")]; + tensor var_17801_equation_0 = const()[name = tensor("op_17801_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17801_cast_fp16 = einsum(equation = var_17801_equation_0, values = (var_17647_cast_fp16, var_17767_cast_fp16))[name = tensor("op_17801_cast_fp16")]; + tensor var_17803_equation_0 = const()[name = tensor("op_17803_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17803_cast_fp16 = einsum(equation = var_17803_equation_0, values = (var_17651_cast_fp16, var_17768_cast_fp16))[name = tensor("op_17803_cast_fp16")]; + tensor var_17805_equation_0 = const()[name = tensor("op_17805_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17805_cast_fp16 = einsum(equation = var_17805_equation_0, values = (var_17655_cast_fp16, var_17769_cast_fp16))[name = tensor("op_17805_cast_fp16")]; + tensor var_17807_equation_0 = const()[name = tensor("op_17807_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17807_cast_fp16 = einsum(equation = var_17807_equation_0, values = (var_17659_cast_fp16, var_17770_cast_fp16))[name = tensor("op_17807_cast_fp16")]; + tensor var_17809_equation_0 = const()[name = tensor("op_17809_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17809_cast_fp16 = einsum(equation = var_17809_equation_0, values = (var_17663_cast_fp16, var_17771_cast_fp16))[name = tensor("op_17809_cast_fp16")]; + tensor var_17811_equation_0 = const()[name = tensor("op_17811_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17811_cast_fp16 = einsum(equation = var_17811_equation_0, values = (var_17667_cast_fp16, var_17772_cast_fp16))[name = tensor("op_17811_cast_fp16")]; + tensor var_17813_equation_0 = const()[name = tensor("op_17813_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17813_cast_fp16 = einsum(equation = var_17813_equation_0, values = (var_17671_cast_fp16, var_17773_cast_fp16))[name = tensor("op_17813_cast_fp16")]; + tensor input_277_interleave_0 = const()[name = tensor("input_277_interleave_0"), val = tensor(false)]; + tensor input_277_cast_fp16 = concat(axis = var_2624, interleave = input_277_interleave_0, values = (var_17775_cast_fp16, var_17777_cast_fp16, var_17779_cast_fp16, var_17781_cast_fp16, var_17783_cast_fp16, var_17785_cast_fp16, var_17787_cast_fp16, var_17789_cast_fp16, var_17791_cast_fp16, var_17793_cast_fp16, var_17795_cast_fp16, var_17797_cast_fp16, var_17799_cast_fp16, var_17801_cast_fp16, var_17803_cast_fp16, var_17805_cast_fp16, var_17807_cast_fp16, var_17809_cast_fp16, var_17811_cast_fp16, var_17813_cast_fp16))[name = tensor("input_277_cast_fp16")]; + tensor var_17823_pad_type_0 = const()[name = tensor("op_17823_pad_type_0"), val = tensor("valid")]; + tensor var_17823_strides_0 = const()[name = tensor("op_17823_strides_0"), val = tensor([1, 1])]; + tensor var_17823_pad_0 = const()[name = tensor("op_17823_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17823_dilations_0 = const()[name = tensor("op_17823_dilations_0"), val = tensor([1, 1])]; + tensor var_17823_groups_0 = const()[name = tensor("op_17823_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521523776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522752640))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522752832)))]; + tensor var_17823_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_17823_dilations_0, groups = var_17823_groups_0, pad = var_17823_pad_0, pad_type = var_17823_pad_type_0, strides = var_17823_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized, x = input_277_cast_fp16)[name = tensor("op_17823_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = var_17823_cast_fp16, y = inputs_121_cast_fp16)[name = tensor("inputs_123_cast_fp16")]; + tensor hidden_states_175_axes_0 = const()[name = tensor("hidden_states_175_axes_0"), val = tensor([1])]; + tensor hidden_states_175_gamma_0_to_fp16 = const()[name = tensor("hidden_states_175_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522755456)))]; + tensor hidden_states_175_beta_0_to_fp16 = const()[name = tensor("hidden_states_175_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522758080)))]; + tensor var_17833_to_fp16 = const()[name = tensor("op_17833_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_175_cast_fp16 = layer_norm(axes = hidden_states_175_axes_0, beta = hidden_states_175_beta_0_to_fp16, epsilon = var_17833_to_fp16, gamma = hidden_states_175_gamma_0_to_fp16, x = inputs_123_cast_fp16)[name = tensor("hidden_states_175_cast_fp16")]; + tensor q_83_pad_type_0 = const()[name = tensor("q_83_pad_type_0"), val = tensor("valid")]; + tensor q_83_strides_0 = const()[name = tensor("q_83_strides_0"), val = tensor([1, 1])]; + tensor q_83_pad_0 = const()[name = tensor("q_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_83_dilations_0 = const()[name = tensor("q_83_dilations_0"), val = tensor([1, 1])]; + tensor q_83_groups_0 = const()[name = tensor("q_83_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522760704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523989568))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_83_cast_fp16 = conv(dilations = q_83_dilations_0, groups = q_83_groups_0, pad = q_83_pad_0, pad_type = q_83_pad_type_0, strides = q_83_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_175_cast_fp16)[name = tensor("q_83_cast_fp16")]; + tensor k_165_pad_type_0 = const()[name = tensor("k_165_pad_type_0"), val = tensor("valid")]; + tensor k_165_strides_0 = const()[name = tensor("k_165_strides_0"), val = tensor([1, 1])]; + tensor k_165_pad_0 = const()[name = tensor("k_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_165_dilations_0 = const()[name = tensor("k_165_dilations_0"), val = tensor([1, 1])]; + tensor k_165_groups_0 = const()[name = tensor("k_165_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523989760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525955904))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_165_cast_fp16 = conv(dilations = k_165_dilations_0, groups = k_165_groups_0, pad = k_165_pad_0, pad_type = k_165_pad_type_0, strides = k_165_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_165_cast_fp16")]; + tensor v_83_pad_type_0 = const()[name = tensor("v_83_pad_type_0"), val = tensor("valid")]; + tensor v_83_strides_0 = const()[name = tensor("v_83_strides_0"), val = tensor([1, 1])]; + tensor v_83_pad_0 = const()[name = tensor("v_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_83_dilations_0 = const()[name = tensor("v_83_dilations_0"), val = tensor([1, 1])]; + tensor v_83_groups_0 = const()[name = tensor("v_83_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525956096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(527922240))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_83_cast_fp16 = conv(dilations = v_83_dilations_0, groups = v_83_groups_0, pad = v_83_pad_0, pad_type = v_83_pad_type_0, strides = v_83_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_83_cast_fp16")]; + tensor var_17866_begin_0 = const()[name = tensor("op_17866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17866_end_0 = const()[name = tensor("op_17866_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_17866_end_mask_0 = const()[name = tensor("op_17866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17866_cast_fp16 = slice_by_index(begin = var_17866_begin_0, end = var_17866_end_0, end_mask = var_17866_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17866_cast_fp16")]; + tensor var_17870_begin_0 = const()[name = tensor("op_17870_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17870_end_0 = const()[name = tensor("op_17870_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_17870_end_mask_0 = const()[name = tensor("op_17870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17870_cast_fp16 = slice_by_index(begin = var_17870_begin_0, end = var_17870_end_0, end_mask = var_17870_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17870_cast_fp16")]; + tensor var_17874_begin_0 = const()[name = tensor("op_17874_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17874_end_0 = const()[name = tensor("op_17874_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_17874_end_mask_0 = const()[name = tensor("op_17874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17874_cast_fp16 = slice_by_index(begin = var_17874_begin_0, end = var_17874_end_0, end_mask = var_17874_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17874_cast_fp16")]; + tensor var_17878_begin_0 = const()[name = tensor("op_17878_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17878_end_0 = const()[name = tensor("op_17878_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_17878_end_mask_0 = const()[name = tensor("op_17878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17878_cast_fp16 = slice_by_index(begin = var_17878_begin_0, end = var_17878_end_0, end_mask = var_17878_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17878_cast_fp16")]; + tensor var_17882_begin_0 = const()[name = tensor("op_17882_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17882_end_0 = const()[name = tensor("op_17882_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_17882_end_mask_0 = const()[name = tensor("op_17882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17882_cast_fp16 = slice_by_index(begin = var_17882_begin_0, end = var_17882_end_0, end_mask = var_17882_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17882_cast_fp16")]; + tensor var_17886_begin_0 = const()[name = tensor("op_17886_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17886_end_0 = const()[name = tensor("op_17886_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_17886_end_mask_0 = const()[name = tensor("op_17886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17886_cast_fp16 = slice_by_index(begin = var_17886_begin_0, end = var_17886_end_0, end_mask = var_17886_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17886_cast_fp16")]; + tensor var_17890_begin_0 = const()[name = tensor("op_17890_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17890_end_0 = const()[name = tensor("op_17890_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_17890_end_mask_0 = const()[name = tensor("op_17890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17890_cast_fp16 = slice_by_index(begin = var_17890_begin_0, end = var_17890_end_0, end_mask = var_17890_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17890_cast_fp16")]; + tensor var_17894_begin_0 = const()[name = tensor("op_17894_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17894_end_0 = const()[name = tensor("op_17894_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_17894_end_mask_0 = const()[name = tensor("op_17894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17894_cast_fp16 = slice_by_index(begin = var_17894_begin_0, end = var_17894_end_0, end_mask = var_17894_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17894_cast_fp16")]; + tensor var_17898_begin_0 = const()[name = tensor("op_17898_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17898_end_0 = const()[name = tensor("op_17898_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_17898_end_mask_0 = const()[name = tensor("op_17898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17898_cast_fp16 = slice_by_index(begin = var_17898_begin_0, end = var_17898_end_0, end_mask = var_17898_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17898_cast_fp16")]; + tensor var_17902_begin_0 = const()[name = tensor("op_17902_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17902_end_0 = const()[name = tensor("op_17902_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_17902_end_mask_0 = const()[name = tensor("op_17902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17902_cast_fp16 = slice_by_index(begin = var_17902_begin_0, end = var_17902_end_0, end_mask = var_17902_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17902_cast_fp16")]; + tensor var_17906_begin_0 = const()[name = tensor("op_17906_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17906_end_0 = const()[name = tensor("op_17906_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_17906_end_mask_0 = const()[name = tensor("op_17906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17906_cast_fp16 = slice_by_index(begin = var_17906_begin_0, end = var_17906_end_0, end_mask = var_17906_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17906_cast_fp16")]; + tensor var_17910_begin_0 = const()[name = tensor("op_17910_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17910_end_0 = const()[name = tensor("op_17910_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_17910_end_mask_0 = const()[name = tensor("op_17910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17910_cast_fp16 = slice_by_index(begin = var_17910_begin_0, end = var_17910_end_0, end_mask = var_17910_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17910_cast_fp16")]; + tensor var_17914_begin_0 = const()[name = tensor("op_17914_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17914_end_0 = const()[name = tensor("op_17914_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17914_end_mask_0 = const()[name = tensor("op_17914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17914_cast_fp16 = slice_by_index(begin = var_17914_begin_0, end = var_17914_end_0, end_mask = var_17914_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17914_cast_fp16")]; + tensor var_17918_begin_0 = const()[name = tensor("op_17918_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17918_end_0 = const()[name = tensor("op_17918_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17918_end_mask_0 = const()[name = tensor("op_17918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17918_cast_fp16 = slice_by_index(begin = var_17918_begin_0, end = var_17918_end_0, end_mask = var_17918_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17918_cast_fp16")]; + tensor var_17922_begin_0 = const()[name = tensor("op_17922_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17922_end_0 = const()[name = tensor("op_17922_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17922_end_mask_0 = const()[name = tensor("op_17922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17922_cast_fp16 = slice_by_index(begin = var_17922_begin_0, end = var_17922_end_0, end_mask = var_17922_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17922_cast_fp16")]; + tensor var_17926_begin_0 = const()[name = tensor("op_17926_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17926_end_0 = const()[name = tensor("op_17926_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17926_end_mask_0 = const()[name = tensor("op_17926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17926_cast_fp16 = slice_by_index(begin = var_17926_begin_0, end = var_17926_end_0, end_mask = var_17926_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17926_cast_fp16")]; + tensor var_17930_begin_0 = const()[name = tensor("op_17930_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17930_end_0 = const()[name = tensor("op_17930_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17930_end_mask_0 = const()[name = tensor("op_17930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17930_cast_fp16 = slice_by_index(begin = var_17930_begin_0, end = var_17930_end_0, end_mask = var_17930_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17930_cast_fp16")]; + tensor var_17934_begin_0 = const()[name = tensor("op_17934_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17934_end_0 = const()[name = tensor("op_17934_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17934_end_mask_0 = const()[name = tensor("op_17934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17934_cast_fp16 = slice_by_index(begin = var_17934_begin_0, end = var_17934_end_0, end_mask = var_17934_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17934_cast_fp16")]; + tensor var_17938_begin_0 = const()[name = tensor("op_17938_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17938_end_0 = const()[name = tensor("op_17938_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17938_end_mask_0 = const()[name = tensor("op_17938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17938_cast_fp16 = slice_by_index(begin = var_17938_begin_0, end = var_17938_end_0, end_mask = var_17938_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17938_cast_fp16")]; + tensor var_17942_begin_0 = const()[name = tensor("op_17942_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17942_end_0 = const()[name = tensor("op_17942_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17942_end_mask_0 = const()[name = tensor("op_17942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17942_cast_fp16 = slice_by_index(begin = var_17942_begin_0, end = var_17942_end_0, end_mask = var_17942_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17942_cast_fp16")]; + tensor k_167_perm_0 = const()[name = tensor("k_167_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17949_begin_0 = const()[name = tensor("op_17949_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17949_end_0 = const()[name = tensor("op_17949_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_17949_end_mask_0 = const()[name = tensor("op_17949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_167_cast_fp16 = transpose(perm = k_167_perm_0, x = k_165_cast_fp16)[name = tensor("transpose_26")]; + tensor var_17949_cast_fp16 = slice_by_index(begin = var_17949_begin_0, end = var_17949_end_0, end_mask = var_17949_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17949_cast_fp16")]; + tensor var_17953_begin_0 = const()[name = tensor("op_17953_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17953_end_0 = const()[name = tensor("op_17953_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_17953_end_mask_0 = const()[name = tensor("op_17953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17953_cast_fp16 = slice_by_index(begin = var_17953_begin_0, end = var_17953_end_0, end_mask = var_17953_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17953_cast_fp16")]; + tensor var_17957_begin_0 = const()[name = tensor("op_17957_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17957_end_0 = const()[name = tensor("op_17957_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_17957_end_mask_0 = const()[name = tensor("op_17957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17957_cast_fp16 = slice_by_index(begin = var_17957_begin_0, end = var_17957_end_0, end_mask = var_17957_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17957_cast_fp16")]; + tensor var_17961_begin_0 = const()[name = tensor("op_17961_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17961_end_0 = const()[name = tensor("op_17961_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_17961_end_mask_0 = const()[name = tensor("op_17961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17961_cast_fp16 = slice_by_index(begin = var_17961_begin_0, end = var_17961_end_0, end_mask = var_17961_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17961_cast_fp16")]; + tensor var_17965_begin_0 = const()[name = tensor("op_17965_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17965_end_0 = const()[name = tensor("op_17965_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_17965_end_mask_0 = const()[name = tensor("op_17965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17965_cast_fp16 = slice_by_index(begin = var_17965_begin_0, end = var_17965_end_0, end_mask = var_17965_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17965_cast_fp16")]; + tensor var_17969_begin_0 = const()[name = tensor("op_17969_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17969_end_0 = const()[name = tensor("op_17969_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_17969_end_mask_0 = const()[name = tensor("op_17969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17969_cast_fp16 = slice_by_index(begin = var_17969_begin_0, end = var_17969_end_0, end_mask = var_17969_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17969_cast_fp16")]; + tensor var_17973_begin_0 = const()[name = tensor("op_17973_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17973_end_0 = const()[name = tensor("op_17973_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_17973_end_mask_0 = const()[name = tensor("op_17973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17973_cast_fp16 = slice_by_index(begin = var_17973_begin_0, end = var_17973_end_0, end_mask = var_17973_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17973_cast_fp16")]; + tensor var_17977_begin_0 = const()[name = tensor("op_17977_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17977_end_0 = const()[name = tensor("op_17977_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_17977_end_mask_0 = const()[name = tensor("op_17977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17977_cast_fp16 = slice_by_index(begin = var_17977_begin_0, end = var_17977_end_0, end_mask = var_17977_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17977_cast_fp16")]; + tensor var_17981_begin_0 = const()[name = tensor("op_17981_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17981_end_0 = const()[name = tensor("op_17981_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_17981_end_mask_0 = const()[name = tensor("op_17981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17981_cast_fp16 = slice_by_index(begin = var_17981_begin_0, end = var_17981_end_0, end_mask = var_17981_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17981_cast_fp16")]; + tensor var_17985_begin_0 = const()[name = tensor("op_17985_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17985_end_0 = const()[name = tensor("op_17985_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_17985_end_mask_0 = const()[name = tensor("op_17985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17985_cast_fp16 = slice_by_index(begin = var_17985_begin_0, end = var_17985_end_0, end_mask = var_17985_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17985_cast_fp16")]; + tensor var_17989_begin_0 = const()[name = tensor("op_17989_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17989_end_0 = const()[name = tensor("op_17989_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_17989_end_mask_0 = const()[name = tensor("op_17989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17989_cast_fp16 = slice_by_index(begin = var_17989_begin_0, end = var_17989_end_0, end_mask = var_17989_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17989_cast_fp16")]; + tensor var_17993_begin_0 = const()[name = tensor("op_17993_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17993_end_0 = const()[name = tensor("op_17993_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_17993_end_mask_0 = const()[name = tensor("op_17993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17993_cast_fp16 = slice_by_index(begin = var_17993_begin_0, end = var_17993_end_0, end_mask = var_17993_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17993_cast_fp16")]; + tensor var_17997_begin_0 = const()[name = tensor("op_17997_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17997_end_0 = const()[name = tensor("op_17997_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_17997_end_mask_0 = const()[name = tensor("op_17997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17997_cast_fp16 = slice_by_index(begin = var_17997_begin_0, end = var_17997_end_0, end_mask = var_17997_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17997_cast_fp16")]; + tensor var_18001_begin_0 = const()[name = tensor("op_18001_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_18001_end_0 = const()[name = tensor("op_18001_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_18001_end_mask_0 = const()[name = tensor("op_18001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18001_cast_fp16 = slice_by_index(begin = var_18001_begin_0, end = var_18001_end_0, end_mask = var_18001_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18001_cast_fp16")]; + tensor var_18005_begin_0 = const()[name = tensor("op_18005_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_18005_end_0 = const()[name = tensor("op_18005_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_18005_end_mask_0 = const()[name = tensor("op_18005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18005_cast_fp16 = slice_by_index(begin = var_18005_begin_0, end = var_18005_end_0, end_mask = var_18005_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18005_cast_fp16")]; + tensor var_18009_begin_0 = const()[name = tensor("op_18009_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_18009_end_0 = const()[name = tensor("op_18009_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_18009_end_mask_0 = const()[name = tensor("op_18009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18009_cast_fp16 = slice_by_index(begin = var_18009_begin_0, end = var_18009_end_0, end_mask = var_18009_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18009_cast_fp16")]; + tensor var_18013_begin_0 = const()[name = tensor("op_18013_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_18013_end_0 = const()[name = tensor("op_18013_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_18013_end_mask_0 = const()[name = tensor("op_18013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18013_cast_fp16 = slice_by_index(begin = var_18013_begin_0, end = var_18013_end_0, end_mask = var_18013_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18013_cast_fp16")]; + tensor var_18017_begin_0 = const()[name = tensor("op_18017_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_18017_end_0 = const()[name = tensor("op_18017_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_18017_end_mask_0 = const()[name = tensor("op_18017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18017_cast_fp16 = slice_by_index(begin = var_18017_begin_0, end = var_18017_end_0, end_mask = var_18017_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18017_cast_fp16")]; + tensor var_18021_begin_0 = const()[name = tensor("op_18021_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_18021_end_0 = const()[name = tensor("op_18021_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_18021_end_mask_0 = const()[name = tensor("op_18021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18021_cast_fp16 = slice_by_index(begin = var_18021_begin_0, end = var_18021_end_0, end_mask = var_18021_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18021_cast_fp16")]; + tensor var_18025_begin_0 = const()[name = tensor("op_18025_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_18025_end_0 = const()[name = tensor("op_18025_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_18025_end_mask_0 = const()[name = tensor("op_18025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18025_cast_fp16 = slice_by_index(begin = var_18025_begin_0, end = var_18025_end_0, end_mask = var_18025_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18025_cast_fp16")]; + tensor var_18027_begin_0 = const()[name = tensor("op_18027_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18027_end_0 = const()[name = tensor("op_18027_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_18027_end_mask_0 = const()[name = tensor("op_18027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18027_cast_fp16 = slice_by_index(begin = var_18027_begin_0, end = var_18027_end_0, end_mask = var_18027_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18027_cast_fp16")]; + tensor var_18031_begin_0 = const()[name = tensor("op_18031_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18031_end_0 = const()[name = tensor("op_18031_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_18031_end_mask_0 = const()[name = tensor("op_18031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18031_cast_fp16 = slice_by_index(begin = var_18031_begin_0, end = var_18031_end_0, end_mask = var_18031_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18031_cast_fp16")]; + tensor var_18035_begin_0 = const()[name = tensor("op_18035_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18035_end_0 = const()[name = tensor("op_18035_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_18035_end_mask_0 = const()[name = tensor("op_18035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18035_cast_fp16 = slice_by_index(begin = var_18035_begin_0, end = var_18035_end_0, end_mask = var_18035_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18035_cast_fp16")]; + tensor var_18039_begin_0 = const()[name = tensor("op_18039_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18039_end_0 = const()[name = tensor("op_18039_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_18039_end_mask_0 = const()[name = tensor("op_18039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18039_cast_fp16 = slice_by_index(begin = var_18039_begin_0, end = var_18039_end_0, end_mask = var_18039_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18039_cast_fp16")]; + tensor var_18043_begin_0 = const()[name = tensor("op_18043_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18043_end_0 = const()[name = tensor("op_18043_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_18043_end_mask_0 = const()[name = tensor("op_18043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18043_cast_fp16 = slice_by_index(begin = var_18043_begin_0, end = var_18043_end_0, end_mask = var_18043_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18043_cast_fp16")]; + tensor var_18047_begin_0 = const()[name = tensor("op_18047_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18047_end_0 = const()[name = tensor("op_18047_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_18047_end_mask_0 = const()[name = tensor("op_18047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18047_cast_fp16 = slice_by_index(begin = var_18047_begin_0, end = var_18047_end_0, end_mask = var_18047_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18047_cast_fp16")]; + tensor var_18051_begin_0 = const()[name = tensor("op_18051_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18051_end_0 = const()[name = tensor("op_18051_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_18051_end_mask_0 = const()[name = tensor("op_18051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18051_cast_fp16 = slice_by_index(begin = var_18051_begin_0, end = var_18051_end_0, end_mask = var_18051_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18051_cast_fp16")]; + tensor var_18055_begin_0 = const()[name = tensor("op_18055_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18055_end_0 = const()[name = tensor("op_18055_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_18055_end_mask_0 = const()[name = tensor("op_18055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18055_cast_fp16 = slice_by_index(begin = var_18055_begin_0, end = var_18055_end_0, end_mask = var_18055_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18055_cast_fp16")]; + tensor var_18059_begin_0 = const()[name = tensor("op_18059_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18059_end_0 = const()[name = tensor("op_18059_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_18059_end_mask_0 = const()[name = tensor("op_18059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18059_cast_fp16 = slice_by_index(begin = var_18059_begin_0, end = var_18059_end_0, end_mask = var_18059_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18059_cast_fp16")]; + tensor var_18063_begin_0 = const()[name = tensor("op_18063_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18063_end_0 = const()[name = tensor("op_18063_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_18063_end_mask_0 = const()[name = tensor("op_18063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18063_cast_fp16 = slice_by_index(begin = var_18063_begin_0, end = var_18063_end_0, end_mask = var_18063_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18063_cast_fp16")]; + tensor var_18067_begin_0 = const()[name = tensor("op_18067_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18067_end_0 = const()[name = tensor("op_18067_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_18067_end_mask_0 = const()[name = tensor("op_18067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18067_cast_fp16 = slice_by_index(begin = var_18067_begin_0, end = var_18067_end_0, end_mask = var_18067_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18067_cast_fp16")]; + tensor var_18071_begin_0 = const()[name = tensor("op_18071_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18071_end_0 = const()[name = tensor("op_18071_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_18071_end_mask_0 = const()[name = tensor("op_18071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18071_cast_fp16 = slice_by_index(begin = var_18071_begin_0, end = var_18071_end_0, end_mask = var_18071_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18071_cast_fp16")]; + tensor var_18075_begin_0 = const()[name = tensor("op_18075_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18075_end_0 = const()[name = tensor("op_18075_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_18075_end_mask_0 = const()[name = tensor("op_18075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18075_cast_fp16 = slice_by_index(begin = var_18075_begin_0, end = var_18075_end_0, end_mask = var_18075_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18075_cast_fp16")]; + tensor var_18079_begin_0 = const()[name = tensor("op_18079_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18079_end_0 = const()[name = tensor("op_18079_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_18079_end_mask_0 = const()[name = tensor("op_18079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18079_cast_fp16 = slice_by_index(begin = var_18079_begin_0, end = var_18079_end_0, end_mask = var_18079_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18079_cast_fp16")]; + tensor var_18083_begin_0 = const()[name = tensor("op_18083_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18083_end_0 = const()[name = tensor("op_18083_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_18083_end_mask_0 = const()[name = tensor("op_18083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18083_cast_fp16 = slice_by_index(begin = var_18083_begin_0, end = var_18083_end_0, end_mask = var_18083_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18083_cast_fp16")]; + tensor var_18087_begin_0 = const()[name = tensor("op_18087_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18087_end_0 = const()[name = tensor("op_18087_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_18087_end_mask_0 = const()[name = tensor("op_18087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18087_cast_fp16 = slice_by_index(begin = var_18087_begin_0, end = var_18087_end_0, end_mask = var_18087_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18087_cast_fp16")]; + tensor var_18091_begin_0 = const()[name = tensor("op_18091_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18091_end_0 = const()[name = tensor("op_18091_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_18091_end_mask_0 = const()[name = tensor("op_18091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18091_cast_fp16 = slice_by_index(begin = var_18091_begin_0, end = var_18091_end_0, end_mask = var_18091_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18091_cast_fp16")]; + tensor var_18095_begin_0 = const()[name = tensor("op_18095_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18095_end_0 = const()[name = tensor("op_18095_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_18095_end_mask_0 = const()[name = tensor("op_18095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18095_cast_fp16 = slice_by_index(begin = var_18095_begin_0, end = var_18095_end_0, end_mask = var_18095_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18095_cast_fp16")]; + tensor var_18099_begin_0 = const()[name = tensor("op_18099_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18099_end_0 = const()[name = tensor("op_18099_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_18099_end_mask_0 = const()[name = tensor("op_18099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18099_cast_fp16 = slice_by_index(begin = var_18099_begin_0, end = var_18099_end_0, end_mask = var_18099_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18099_cast_fp16")]; + tensor var_18103_begin_0 = const()[name = tensor("op_18103_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18103_end_0 = const()[name = tensor("op_18103_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_18103_end_mask_0 = const()[name = tensor("op_18103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18103_cast_fp16 = slice_by_index(begin = var_18103_begin_0, end = var_18103_end_0, end_mask = var_18103_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18103_cast_fp16")]; + tensor var_18107_equation_0 = const()[name = tensor("op_18107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18107_cast_fp16 = einsum(equation = var_18107_equation_0, values = (var_17949_cast_fp16, var_17866_cast_fp16))[name = tensor("op_18107_cast_fp16")]; + tensor var_18108_to_fp16 = const()[name = tensor("op_18108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1481_cast_fp16 = mul(x = var_18107_cast_fp16, y = var_18108_to_fp16)[name = tensor("aw_1481_cast_fp16")]; + tensor var_18111_equation_0 = const()[name = tensor("op_18111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18111_cast_fp16 = einsum(equation = var_18111_equation_0, values = (var_17953_cast_fp16, var_17870_cast_fp16))[name = tensor("op_18111_cast_fp16")]; + tensor var_18112_to_fp16 = const()[name = tensor("op_18112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1483_cast_fp16 = mul(x = var_18111_cast_fp16, y = var_18112_to_fp16)[name = tensor("aw_1483_cast_fp16")]; + tensor var_18115_equation_0 = const()[name = tensor("op_18115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18115_cast_fp16 = einsum(equation = var_18115_equation_0, values = (var_17957_cast_fp16, var_17874_cast_fp16))[name = tensor("op_18115_cast_fp16")]; + tensor var_18116_to_fp16 = const()[name = tensor("op_18116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1485_cast_fp16 = mul(x = var_18115_cast_fp16, y = var_18116_to_fp16)[name = tensor("aw_1485_cast_fp16")]; + tensor var_18119_equation_0 = const()[name = tensor("op_18119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18119_cast_fp16 = einsum(equation = var_18119_equation_0, values = (var_17961_cast_fp16, var_17878_cast_fp16))[name = tensor("op_18119_cast_fp16")]; + tensor var_18120_to_fp16 = const()[name = tensor("op_18120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1487_cast_fp16 = mul(x = var_18119_cast_fp16, y = var_18120_to_fp16)[name = tensor("aw_1487_cast_fp16")]; + tensor var_18123_equation_0 = const()[name = tensor("op_18123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18123_cast_fp16 = einsum(equation = var_18123_equation_0, values = (var_17965_cast_fp16, var_17882_cast_fp16))[name = tensor("op_18123_cast_fp16")]; + tensor var_18124_to_fp16 = const()[name = tensor("op_18124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1489_cast_fp16 = mul(x = var_18123_cast_fp16, y = var_18124_to_fp16)[name = tensor("aw_1489_cast_fp16")]; + tensor var_18127_equation_0 = const()[name = tensor("op_18127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18127_cast_fp16 = einsum(equation = var_18127_equation_0, values = (var_17969_cast_fp16, var_17886_cast_fp16))[name = tensor("op_18127_cast_fp16")]; + tensor var_18128_to_fp16 = const()[name = tensor("op_18128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1491_cast_fp16 = mul(x = var_18127_cast_fp16, y = var_18128_to_fp16)[name = tensor("aw_1491_cast_fp16")]; + tensor var_18131_equation_0 = const()[name = tensor("op_18131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18131_cast_fp16 = einsum(equation = var_18131_equation_0, values = (var_17973_cast_fp16, var_17890_cast_fp16))[name = tensor("op_18131_cast_fp16")]; + tensor var_18132_to_fp16 = const()[name = tensor("op_18132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1493_cast_fp16 = mul(x = var_18131_cast_fp16, y = var_18132_to_fp16)[name = tensor("aw_1493_cast_fp16")]; + tensor var_18135_equation_0 = const()[name = tensor("op_18135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18135_cast_fp16 = einsum(equation = var_18135_equation_0, values = (var_17977_cast_fp16, var_17894_cast_fp16))[name = tensor("op_18135_cast_fp16")]; + tensor var_18136_to_fp16 = const()[name = tensor("op_18136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1495_cast_fp16 = mul(x = var_18135_cast_fp16, y = var_18136_to_fp16)[name = tensor("aw_1495_cast_fp16")]; + tensor var_18139_equation_0 = const()[name = tensor("op_18139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18139_cast_fp16 = einsum(equation = var_18139_equation_0, values = (var_17981_cast_fp16, var_17898_cast_fp16))[name = tensor("op_18139_cast_fp16")]; + tensor var_18140_to_fp16 = const()[name = tensor("op_18140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1497_cast_fp16 = mul(x = var_18139_cast_fp16, y = var_18140_to_fp16)[name = tensor("aw_1497_cast_fp16")]; + tensor var_18143_equation_0 = const()[name = tensor("op_18143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18143_cast_fp16 = einsum(equation = var_18143_equation_0, values = (var_17985_cast_fp16, var_17902_cast_fp16))[name = tensor("op_18143_cast_fp16")]; + tensor var_18144_to_fp16 = const()[name = tensor("op_18144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1499_cast_fp16 = mul(x = var_18143_cast_fp16, y = var_18144_to_fp16)[name = tensor("aw_1499_cast_fp16")]; + tensor var_18147_equation_0 = const()[name = tensor("op_18147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18147_cast_fp16 = einsum(equation = var_18147_equation_0, values = (var_17989_cast_fp16, var_17906_cast_fp16))[name = tensor("op_18147_cast_fp16")]; + tensor var_18148_to_fp16 = const()[name = tensor("op_18148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1501_cast_fp16 = mul(x = var_18147_cast_fp16, y = var_18148_to_fp16)[name = tensor("aw_1501_cast_fp16")]; + tensor var_18151_equation_0 = const()[name = tensor("op_18151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18151_cast_fp16 = einsum(equation = var_18151_equation_0, values = (var_17993_cast_fp16, var_17910_cast_fp16))[name = tensor("op_18151_cast_fp16")]; + tensor var_18152_to_fp16 = const()[name = tensor("op_18152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1503_cast_fp16 = mul(x = var_18151_cast_fp16, y = var_18152_to_fp16)[name = tensor("aw_1503_cast_fp16")]; + tensor var_18155_equation_0 = const()[name = tensor("op_18155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18155_cast_fp16 = einsum(equation = var_18155_equation_0, values = (var_17997_cast_fp16, var_17914_cast_fp16))[name = tensor("op_18155_cast_fp16")]; + tensor var_18156_to_fp16 = const()[name = tensor("op_18156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1505_cast_fp16 = mul(x = var_18155_cast_fp16, y = var_18156_to_fp16)[name = tensor("aw_1505_cast_fp16")]; + tensor var_18159_equation_0 = const()[name = tensor("op_18159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18159_cast_fp16 = einsum(equation = var_18159_equation_0, values = (var_18001_cast_fp16, var_17918_cast_fp16))[name = tensor("op_18159_cast_fp16")]; + tensor var_18160_to_fp16 = const()[name = tensor("op_18160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1507_cast_fp16 = mul(x = var_18159_cast_fp16, y = var_18160_to_fp16)[name = tensor("aw_1507_cast_fp16")]; + tensor var_18163_equation_0 = const()[name = tensor("op_18163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18163_cast_fp16 = einsum(equation = var_18163_equation_0, values = (var_18005_cast_fp16, var_17922_cast_fp16))[name = tensor("op_18163_cast_fp16")]; + tensor var_18164_to_fp16 = const()[name = tensor("op_18164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1509_cast_fp16 = mul(x = var_18163_cast_fp16, y = var_18164_to_fp16)[name = tensor("aw_1509_cast_fp16")]; + tensor var_18167_equation_0 = const()[name = tensor("op_18167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18167_cast_fp16 = einsum(equation = var_18167_equation_0, values = (var_18009_cast_fp16, var_17926_cast_fp16))[name = tensor("op_18167_cast_fp16")]; + tensor var_18168_to_fp16 = const()[name = tensor("op_18168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1511_cast_fp16 = mul(x = var_18167_cast_fp16, y = var_18168_to_fp16)[name = tensor("aw_1511_cast_fp16")]; + tensor var_18171_equation_0 = const()[name = tensor("op_18171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18171_cast_fp16 = einsum(equation = var_18171_equation_0, values = (var_18013_cast_fp16, var_17930_cast_fp16))[name = tensor("op_18171_cast_fp16")]; + tensor var_18172_to_fp16 = const()[name = tensor("op_18172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1513_cast_fp16 = mul(x = var_18171_cast_fp16, y = var_18172_to_fp16)[name = tensor("aw_1513_cast_fp16")]; + tensor var_18175_equation_0 = const()[name = tensor("op_18175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18175_cast_fp16 = einsum(equation = var_18175_equation_0, values = (var_18017_cast_fp16, var_17934_cast_fp16))[name = tensor("op_18175_cast_fp16")]; + tensor var_18176_to_fp16 = const()[name = tensor("op_18176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1515_cast_fp16 = mul(x = var_18175_cast_fp16, y = var_18176_to_fp16)[name = tensor("aw_1515_cast_fp16")]; + tensor var_18179_equation_0 = const()[name = tensor("op_18179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18179_cast_fp16 = einsum(equation = var_18179_equation_0, values = (var_18021_cast_fp16, var_17938_cast_fp16))[name = tensor("op_18179_cast_fp16")]; + tensor var_18180_to_fp16 = const()[name = tensor("op_18180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1517_cast_fp16 = mul(x = var_18179_cast_fp16, y = var_18180_to_fp16)[name = tensor("aw_1517_cast_fp16")]; + tensor var_18183_equation_0 = const()[name = tensor("op_18183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18183_cast_fp16 = einsum(equation = var_18183_equation_0, values = (var_18025_cast_fp16, var_17942_cast_fp16))[name = tensor("op_18183_cast_fp16")]; + tensor var_18184_to_fp16 = const()[name = tensor("op_18184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1519_cast_fp16 = mul(x = var_18183_cast_fp16, y = var_18184_to_fp16)[name = tensor("aw_1519_cast_fp16")]; + tensor var_18186_cast_fp16 = softmax(axis = var_2624, x = aw_1481_cast_fp16)[name = tensor("op_18186_cast_fp16")]; + tensor var_18187_cast_fp16 = softmax(axis = var_2624, x = aw_1483_cast_fp16)[name = tensor("op_18187_cast_fp16")]; + tensor var_18188_cast_fp16 = softmax(axis = var_2624, x = aw_1485_cast_fp16)[name = tensor("op_18188_cast_fp16")]; + tensor var_18189_cast_fp16 = softmax(axis = var_2624, x = aw_1487_cast_fp16)[name = tensor("op_18189_cast_fp16")]; + tensor var_18190_cast_fp16 = softmax(axis = var_2624, x = aw_1489_cast_fp16)[name = tensor("op_18190_cast_fp16")]; + tensor var_18191_cast_fp16 = softmax(axis = var_2624, x = aw_1491_cast_fp16)[name = tensor("op_18191_cast_fp16")]; + tensor var_18192_cast_fp16 = softmax(axis = var_2624, x = aw_1493_cast_fp16)[name = tensor("op_18192_cast_fp16")]; + tensor var_18193_cast_fp16 = softmax(axis = var_2624, x = aw_1495_cast_fp16)[name = tensor("op_18193_cast_fp16")]; + tensor var_18194_cast_fp16 = softmax(axis = var_2624, x = aw_1497_cast_fp16)[name = tensor("op_18194_cast_fp16")]; + tensor var_18195_cast_fp16 = softmax(axis = var_2624, x = aw_1499_cast_fp16)[name = tensor("op_18195_cast_fp16")]; + tensor var_18196_cast_fp16 = softmax(axis = var_2624, x = aw_1501_cast_fp16)[name = tensor("op_18196_cast_fp16")]; + tensor var_18197_cast_fp16 = softmax(axis = var_2624, x = aw_1503_cast_fp16)[name = tensor("op_18197_cast_fp16")]; + tensor var_18198_cast_fp16 = softmax(axis = var_2624, x = aw_1505_cast_fp16)[name = tensor("op_18198_cast_fp16")]; + tensor var_18199_cast_fp16 = softmax(axis = var_2624, x = aw_1507_cast_fp16)[name = tensor("op_18199_cast_fp16")]; + tensor var_18200_cast_fp16 = softmax(axis = var_2624, x = aw_1509_cast_fp16)[name = tensor("op_18200_cast_fp16")]; + tensor var_18201_cast_fp16 = softmax(axis = var_2624, x = aw_1511_cast_fp16)[name = tensor("op_18201_cast_fp16")]; + tensor var_18202_cast_fp16 = softmax(axis = var_2624, x = aw_1513_cast_fp16)[name = tensor("op_18202_cast_fp16")]; + tensor var_18203_cast_fp16 = softmax(axis = var_2624, x = aw_1515_cast_fp16)[name = tensor("op_18203_cast_fp16")]; + tensor var_18204_cast_fp16 = softmax(axis = var_2624, x = aw_1517_cast_fp16)[name = tensor("op_18204_cast_fp16")]; + tensor var_18205_cast_fp16 = softmax(axis = var_2624, x = aw_1519_cast_fp16)[name = tensor("op_18205_cast_fp16")]; + tensor var_18207_equation_0 = const()[name = tensor("op_18207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18207_cast_fp16 = einsum(equation = var_18207_equation_0, values = (var_18027_cast_fp16, var_18186_cast_fp16))[name = tensor("op_18207_cast_fp16")]; + tensor var_18209_equation_0 = const()[name = tensor("op_18209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18209_cast_fp16 = einsum(equation = var_18209_equation_0, values = (var_18031_cast_fp16, var_18187_cast_fp16))[name = tensor("op_18209_cast_fp16")]; + tensor var_18211_equation_0 = const()[name = tensor("op_18211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18211_cast_fp16 = einsum(equation = var_18211_equation_0, values = (var_18035_cast_fp16, var_18188_cast_fp16))[name = tensor("op_18211_cast_fp16")]; + tensor var_18213_equation_0 = const()[name = tensor("op_18213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18213_cast_fp16 = einsum(equation = var_18213_equation_0, values = (var_18039_cast_fp16, var_18189_cast_fp16))[name = tensor("op_18213_cast_fp16")]; + tensor var_18215_equation_0 = const()[name = tensor("op_18215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18215_cast_fp16 = einsum(equation = var_18215_equation_0, values = (var_18043_cast_fp16, var_18190_cast_fp16))[name = tensor("op_18215_cast_fp16")]; + tensor var_18217_equation_0 = const()[name = tensor("op_18217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18217_cast_fp16 = einsum(equation = var_18217_equation_0, values = (var_18047_cast_fp16, var_18191_cast_fp16))[name = tensor("op_18217_cast_fp16")]; + tensor var_18219_equation_0 = const()[name = tensor("op_18219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18219_cast_fp16 = einsum(equation = var_18219_equation_0, values = (var_18051_cast_fp16, var_18192_cast_fp16))[name = tensor("op_18219_cast_fp16")]; + tensor var_18221_equation_0 = const()[name = tensor("op_18221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18221_cast_fp16 = einsum(equation = var_18221_equation_0, values = (var_18055_cast_fp16, var_18193_cast_fp16))[name = tensor("op_18221_cast_fp16")]; + tensor var_18223_equation_0 = const()[name = tensor("op_18223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18223_cast_fp16 = einsum(equation = var_18223_equation_0, values = (var_18059_cast_fp16, var_18194_cast_fp16))[name = tensor("op_18223_cast_fp16")]; + tensor var_18225_equation_0 = const()[name = tensor("op_18225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18225_cast_fp16 = einsum(equation = var_18225_equation_0, values = (var_18063_cast_fp16, var_18195_cast_fp16))[name = tensor("op_18225_cast_fp16")]; + tensor var_18227_equation_0 = const()[name = tensor("op_18227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18227_cast_fp16 = einsum(equation = var_18227_equation_0, values = (var_18067_cast_fp16, var_18196_cast_fp16))[name = tensor("op_18227_cast_fp16")]; + tensor var_18229_equation_0 = const()[name = tensor("op_18229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18229_cast_fp16 = einsum(equation = var_18229_equation_0, values = (var_18071_cast_fp16, var_18197_cast_fp16))[name = tensor("op_18229_cast_fp16")]; + tensor var_18231_equation_0 = const()[name = tensor("op_18231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18231_cast_fp16 = einsum(equation = var_18231_equation_0, values = (var_18075_cast_fp16, var_18198_cast_fp16))[name = tensor("op_18231_cast_fp16")]; + tensor var_18233_equation_0 = const()[name = tensor("op_18233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18233_cast_fp16 = einsum(equation = var_18233_equation_0, values = (var_18079_cast_fp16, var_18199_cast_fp16))[name = tensor("op_18233_cast_fp16")]; + tensor var_18235_equation_0 = const()[name = tensor("op_18235_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18235_cast_fp16 = einsum(equation = var_18235_equation_0, values = (var_18083_cast_fp16, var_18200_cast_fp16))[name = tensor("op_18235_cast_fp16")]; + tensor var_18237_equation_0 = const()[name = tensor("op_18237_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18237_cast_fp16 = einsum(equation = var_18237_equation_0, values = (var_18087_cast_fp16, var_18201_cast_fp16))[name = tensor("op_18237_cast_fp16")]; + tensor var_18239_equation_0 = const()[name = tensor("op_18239_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18239_cast_fp16 = einsum(equation = var_18239_equation_0, values = (var_18091_cast_fp16, var_18202_cast_fp16))[name = tensor("op_18239_cast_fp16")]; + tensor var_18241_equation_0 = const()[name = tensor("op_18241_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18241_cast_fp16 = einsum(equation = var_18241_equation_0, values = (var_18095_cast_fp16, var_18203_cast_fp16))[name = tensor("op_18241_cast_fp16")]; + tensor var_18243_equation_0 = const()[name = tensor("op_18243_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18243_cast_fp16 = einsum(equation = var_18243_equation_0, values = (var_18099_cast_fp16, var_18204_cast_fp16))[name = tensor("op_18243_cast_fp16")]; + tensor var_18245_equation_0 = const()[name = tensor("op_18245_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18245_cast_fp16 = einsum(equation = var_18245_equation_0, values = (var_18103_cast_fp16, var_18205_cast_fp16))[name = tensor("op_18245_cast_fp16")]; + tensor input_279_interleave_0 = const()[name = tensor("input_279_interleave_0"), val = tensor(false)]; + tensor input_279_cast_fp16 = concat(axis = var_2624, interleave = input_279_interleave_0, values = (var_18207_cast_fp16, var_18209_cast_fp16, var_18211_cast_fp16, var_18213_cast_fp16, var_18215_cast_fp16, var_18217_cast_fp16, var_18219_cast_fp16, var_18221_cast_fp16, var_18223_cast_fp16, var_18225_cast_fp16, var_18227_cast_fp16, var_18229_cast_fp16, var_18231_cast_fp16, var_18233_cast_fp16, var_18235_cast_fp16, var_18237_cast_fp16, var_18239_cast_fp16, var_18241_cast_fp16, var_18243_cast_fp16, var_18245_cast_fp16))[name = tensor("input_279_cast_fp16")]; + tensor var_18255_pad_type_0 = const()[name = tensor("op_18255_pad_type_0"), val = tensor("valid")]; + tensor var_18255_strides_0 = const()[name = tensor("op_18255_strides_0"), val = tensor([1, 1])]; + tensor var_18255_pad_0 = const()[name = tensor("op_18255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18255_dilations_0 = const()[name = tensor("op_18255_dilations_0"), val = tensor([1, 1])]; + tensor var_18255_groups_0 = const()[name = tensor("op_18255_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(527922432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529151296))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529151488)))]; + tensor var_18255_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_18255_dilations_0, groups = var_18255_groups_0, pad = var_18255_pad_0, pad_type = var_18255_pad_type_0, strides = var_18255_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized, x = input_279_cast_fp16)[name = tensor("op_18255_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = var_18255_cast_fp16, y = inputs_123_cast_fp16)[name = tensor("inputs_125_cast_fp16")]; + tensor input_281_axes_0 = const()[name = tensor("input_281_axes_0"), val = tensor([1])]; + tensor input_281_gamma_0_to_fp16 = const()[name = tensor("input_281_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529154112)))]; + tensor input_281_beta_0_to_fp16 = const()[name = tensor("input_281_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529156736)))]; + tensor var_18265_to_fp16 = const()[name = tensor("op_18265_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = input_281_beta_0_to_fp16, epsilon = var_18265_to_fp16, gamma = input_281_gamma_0_to_fp16, x = inputs_125_cast_fp16)[name = tensor("input_281_cast_fp16")]; + tensor var_18285_pad_type_0 = const()[name = tensor("op_18285_pad_type_0"), val = tensor("valid")]; + tensor var_18285_strides_0 = const()[name = tensor("op_18285_strides_0"), val = tensor([1, 1])]; + tensor var_18285_pad_0 = const()[name = tensor("op_18285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18285_dilations_0 = const()[name = tensor("op_18285_dilations_0"), val = tensor([1, 1])]; + tensor var_18285_groups_0 = const()[name = tensor("op_18285_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529159360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538989824))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538990016)))]; + tensor var_18285_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_18285_dilations_0, groups = var_18285_groups_0, pad = var_18285_pad_0, pad_type = var_18285_pad_type_0, strides = var_18285_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized, x = input_281_cast_fp16)[name = tensor("op_18285_cast_fp16")]; + tensor var_18286_split_sizes_0 = const()[name = tensor("op_18286_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_18286_axis_0 = const()[name = tensor("op_18286_axis_0"), val = tensor(1)]; + tensor var_18286_cast_fp16_0, tensor var_18286_cast_fp16_1 = split(axis = var_18286_axis_0, split_sizes = var_18286_split_sizes_0, x = var_18285_cast_fp16)[name = tensor("op_18286_cast_fp16")]; + tensor var_18288_mode_0 = const()[name = tensor("op_18288_mode_0"), val = tensor("EXACT")]; + tensor var_18288_cast_fp16 = gelu(mode = var_18288_mode_0, x = var_18286_cast_fp16_1)[name = tensor("op_18288_cast_fp16")]; + tensor input_283_cast_fp16 = mul(x = var_18286_cast_fp16_0, y = var_18288_cast_fp16)[name = tensor("input_283_cast_fp16")]; + tensor var_18296_pad_type_0 = const()[name = tensor("op_18296_pad_type_0"), val = tensor("valid")]; + tensor var_18296_strides_0 = const()[name = tensor("op_18296_strides_0"), val = tensor([1, 1])]; + tensor var_18296_pad_0 = const()[name = tensor("op_18296_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18296_dilations_0 = const()[name = tensor("op_18296_dilations_0"), val = tensor([1, 1])]; + tensor var_18296_groups_0 = const()[name = tensor("op_18296_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539010560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543925824))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543926016)))]; + tensor var_18296_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_18296_dilations_0, groups = var_18296_groups_0, pad = var_18296_pad_0, pad_type = var_18296_pad_type_0, strides = var_18296_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized, x = input_283_cast_fp16)[name = tensor("op_18296_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = var_18296_cast_fp16, y = inputs_125_cast_fp16)[name = tensor("inputs_127_cast_fp16")]; + tensor hidden_states_179_axes_0 = const()[name = tensor("hidden_states_179_axes_0"), val = tensor([1])]; + tensor hidden_states_179_gamma_0_to_fp16 = const()[name = tensor("hidden_states_179_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543928640)))]; + tensor hidden_states_179_beta_0_to_fp16 = const()[name = tensor("hidden_states_179_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543931264)))]; + tensor var_18312_to_fp16 = const()[name = tensor("op_18312_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_179_cast_fp16 = layer_norm(axes = hidden_states_179_axes_0, beta = hidden_states_179_beta_0_to_fp16, epsilon = var_18312_to_fp16, gamma = hidden_states_179_gamma_0_to_fp16, x = inputs_127_cast_fp16)[name = tensor("hidden_states_179_cast_fp16")]; + tensor q_85_pad_type_0 = const()[name = tensor("q_85_pad_type_0"), val = tensor("valid")]; + tensor q_85_strides_0 = const()[name = tensor("q_85_strides_0"), val = tensor([1, 1])]; + tensor q_85_pad_0 = const()[name = tensor("q_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_85_dilations_0 = const()[name = tensor("q_85_dilations_0"), val = tensor([1, 1])]; + tensor q_85_groups_0 = const()[name = tensor("q_85_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543933888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545162752))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_85_cast_fp16 = conv(dilations = q_85_dilations_0, groups = q_85_groups_0, pad = q_85_pad_0, pad_type = q_85_pad_type_0, strides = q_85_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_179_cast_fp16)[name = tensor("q_85_cast_fp16")]; + tensor k_169_pad_type_0 = const()[name = tensor("k_169_pad_type_0"), val = tensor("valid")]; + tensor k_169_strides_0 = const()[name = tensor("k_169_strides_0"), val = tensor([1, 1])]; + tensor k_169_pad_0 = const()[name = tensor("k_169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_169_dilations_0 = const()[name = tensor("k_169_dilations_0"), val = tensor([1, 1])]; + tensor k_169_groups_0 = const()[name = tensor("k_169_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545162944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546391808))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_169_cast_fp16 = conv(dilations = k_169_dilations_0, groups = k_169_groups_0, pad = k_169_pad_0, pad_type = k_169_pad_type_0, strides = k_169_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_179_cast_fp16)[name = tensor("k_169_cast_fp16")]; + tensor v_85_pad_type_0 = const()[name = tensor("v_85_pad_type_0"), val = tensor("valid")]; + tensor v_85_strides_0 = const()[name = tensor("v_85_strides_0"), val = tensor([1, 1])]; + tensor v_85_pad_0 = const()[name = tensor("v_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_85_dilations_0 = const()[name = tensor("v_85_dilations_0"), val = tensor([1, 1])]; + tensor v_85_groups_0 = const()[name = tensor("v_85_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546392000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547620864))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_85_cast_fp16 = conv(dilations = v_85_dilations_0, groups = v_85_groups_0, pad = v_85_pad_0, pad_type = v_85_pad_type_0, strides = v_85_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_179_cast_fp16)[name = tensor("v_85_cast_fp16")]; + tensor var_18345_begin_0 = const()[name = tensor("op_18345_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18345_end_0 = const()[name = tensor("op_18345_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_18345_end_mask_0 = const()[name = tensor("op_18345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18345_cast_fp16 = slice_by_index(begin = var_18345_begin_0, end = var_18345_end_0, end_mask = var_18345_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18345_cast_fp16")]; + tensor var_18349_begin_0 = const()[name = tensor("op_18349_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18349_end_0 = const()[name = tensor("op_18349_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_18349_end_mask_0 = const()[name = tensor("op_18349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18349_cast_fp16 = slice_by_index(begin = var_18349_begin_0, end = var_18349_end_0, end_mask = var_18349_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18349_cast_fp16")]; + tensor var_18353_begin_0 = const()[name = tensor("op_18353_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18353_end_0 = const()[name = tensor("op_18353_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_18353_end_mask_0 = const()[name = tensor("op_18353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18353_cast_fp16 = slice_by_index(begin = var_18353_begin_0, end = var_18353_end_0, end_mask = var_18353_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18353_cast_fp16")]; + tensor var_18357_begin_0 = const()[name = tensor("op_18357_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18357_end_0 = const()[name = tensor("op_18357_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_18357_end_mask_0 = const()[name = tensor("op_18357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18357_cast_fp16 = slice_by_index(begin = var_18357_begin_0, end = var_18357_end_0, end_mask = var_18357_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18357_cast_fp16")]; + tensor var_18361_begin_0 = const()[name = tensor("op_18361_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18361_end_0 = const()[name = tensor("op_18361_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_18361_end_mask_0 = const()[name = tensor("op_18361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18361_cast_fp16 = slice_by_index(begin = var_18361_begin_0, end = var_18361_end_0, end_mask = var_18361_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18361_cast_fp16")]; + tensor var_18365_begin_0 = const()[name = tensor("op_18365_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18365_end_0 = const()[name = tensor("op_18365_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_18365_end_mask_0 = const()[name = tensor("op_18365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18365_cast_fp16 = slice_by_index(begin = var_18365_begin_0, end = var_18365_end_0, end_mask = var_18365_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18365_cast_fp16")]; + tensor var_18369_begin_0 = const()[name = tensor("op_18369_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18369_end_0 = const()[name = tensor("op_18369_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_18369_end_mask_0 = const()[name = tensor("op_18369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18369_cast_fp16 = slice_by_index(begin = var_18369_begin_0, end = var_18369_end_0, end_mask = var_18369_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18369_cast_fp16")]; + tensor var_18373_begin_0 = const()[name = tensor("op_18373_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18373_end_0 = const()[name = tensor("op_18373_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_18373_end_mask_0 = const()[name = tensor("op_18373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18373_cast_fp16 = slice_by_index(begin = var_18373_begin_0, end = var_18373_end_0, end_mask = var_18373_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18373_cast_fp16")]; + tensor var_18377_begin_0 = const()[name = tensor("op_18377_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18377_end_0 = const()[name = tensor("op_18377_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_18377_end_mask_0 = const()[name = tensor("op_18377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18377_cast_fp16 = slice_by_index(begin = var_18377_begin_0, end = var_18377_end_0, end_mask = var_18377_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18377_cast_fp16")]; + tensor var_18381_begin_0 = const()[name = tensor("op_18381_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18381_end_0 = const()[name = tensor("op_18381_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_18381_end_mask_0 = const()[name = tensor("op_18381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18381_cast_fp16 = slice_by_index(begin = var_18381_begin_0, end = var_18381_end_0, end_mask = var_18381_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18381_cast_fp16")]; + tensor var_18385_begin_0 = const()[name = tensor("op_18385_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18385_end_0 = const()[name = tensor("op_18385_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_18385_end_mask_0 = const()[name = tensor("op_18385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18385_cast_fp16 = slice_by_index(begin = var_18385_begin_0, end = var_18385_end_0, end_mask = var_18385_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18385_cast_fp16")]; + tensor var_18389_begin_0 = const()[name = tensor("op_18389_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18389_end_0 = const()[name = tensor("op_18389_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_18389_end_mask_0 = const()[name = tensor("op_18389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18389_cast_fp16 = slice_by_index(begin = var_18389_begin_0, end = var_18389_end_0, end_mask = var_18389_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18389_cast_fp16")]; + tensor var_18393_begin_0 = const()[name = tensor("op_18393_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18393_end_0 = const()[name = tensor("op_18393_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_18393_end_mask_0 = const()[name = tensor("op_18393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18393_cast_fp16 = slice_by_index(begin = var_18393_begin_0, end = var_18393_end_0, end_mask = var_18393_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18393_cast_fp16")]; + tensor var_18397_begin_0 = const()[name = tensor("op_18397_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18397_end_0 = const()[name = tensor("op_18397_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_18397_end_mask_0 = const()[name = tensor("op_18397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18397_cast_fp16 = slice_by_index(begin = var_18397_begin_0, end = var_18397_end_0, end_mask = var_18397_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18397_cast_fp16")]; + tensor var_18401_begin_0 = const()[name = tensor("op_18401_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18401_end_0 = const()[name = tensor("op_18401_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_18401_end_mask_0 = const()[name = tensor("op_18401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18401_cast_fp16 = slice_by_index(begin = var_18401_begin_0, end = var_18401_end_0, end_mask = var_18401_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18401_cast_fp16")]; + tensor var_18405_begin_0 = const()[name = tensor("op_18405_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18405_end_0 = const()[name = tensor("op_18405_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18405_end_mask_0 = const()[name = tensor("op_18405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18405_cast_fp16 = slice_by_index(begin = var_18405_begin_0, end = var_18405_end_0, end_mask = var_18405_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18405_cast_fp16")]; + tensor var_18409_begin_0 = const()[name = tensor("op_18409_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18409_end_0 = const()[name = tensor("op_18409_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_18409_end_mask_0 = const()[name = tensor("op_18409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18409_cast_fp16 = slice_by_index(begin = var_18409_begin_0, end = var_18409_end_0, end_mask = var_18409_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18409_cast_fp16")]; + tensor var_18413_begin_0 = const()[name = tensor("op_18413_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18413_end_0 = const()[name = tensor("op_18413_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_18413_end_mask_0 = const()[name = tensor("op_18413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18413_cast_fp16 = slice_by_index(begin = var_18413_begin_0, end = var_18413_end_0, end_mask = var_18413_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18413_cast_fp16")]; + tensor var_18417_begin_0 = const()[name = tensor("op_18417_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18417_end_0 = const()[name = tensor("op_18417_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_18417_end_mask_0 = const()[name = tensor("op_18417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18417_cast_fp16 = slice_by_index(begin = var_18417_begin_0, end = var_18417_end_0, end_mask = var_18417_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18417_cast_fp16")]; + tensor var_18421_begin_0 = const()[name = tensor("op_18421_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18421_end_0 = const()[name = tensor("op_18421_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_18421_end_mask_0 = const()[name = tensor("op_18421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18421_cast_fp16 = slice_by_index(begin = var_18421_begin_0, end = var_18421_end_0, end_mask = var_18421_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18421_cast_fp16")]; + tensor k_171_perm_0 = const()[name = tensor("k_171_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_18428_begin_0 = const()[name = tensor("op_18428_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18428_end_0 = const()[name = tensor("op_18428_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_18428_end_mask_0 = const()[name = tensor("op_18428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_171_cast_fp16 = transpose(perm = k_171_perm_0, x = k_169_cast_fp16)[name = tensor("transpose_25")]; + tensor var_18428_cast_fp16 = slice_by_index(begin = var_18428_begin_0, end = var_18428_end_0, end_mask = var_18428_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18428_cast_fp16")]; + tensor var_18432_begin_0 = const()[name = tensor("op_18432_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_18432_end_0 = const()[name = tensor("op_18432_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_18432_end_mask_0 = const()[name = tensor("op_18432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18432_cast_fp16 = slice_by_index(begin = var_18432_begin_0, end = var_18432_end_0, end_mask = var_18432_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18432_cast_fp16")]; + tensor var_18436_begin_0 = const()[name = tensor("op_18436_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_18436_end_0 = const()[name = tensor("op_18436_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_18436_end_mask_0 = const()[name = tensor("op_18436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18436_cast_fp16 = slice_by_index(begin = var_18436_begin_0, end = var_18436_end_0, end_mask = var_18436_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18436_cast_fp16")]; + tensor var_18440_begin_0 = const()[name = tensor("op_18440_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_18440_end_0 = const()[name = tensor("op_18440_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_18440_end_mask_0 = const()[name = tensor("op_18440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18440_cast_fp16 = slice_by_index(begin = var_18440_begin_0, end = var_18440_end_0, end_mask = var_18440_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18440_cast_fp16")]; + tensor var_18444_begin_0 = const()[name = tensor("op_18444_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_18444_end_0 = const()[name = tensor("op_18444_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_18444_end_mask_0 = const()[name = tensor("op_18444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18444_cast_fp16 = slice_by_index(begin = var_18444_begin_0, end = var_18444_end_0, end_mask = var_18444_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18444_cast_fp16")]; + tensor var_18448_begin_0 = const()[name = tensor("op_18448_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_18448_end_0 = const()[name = tensor("op_18448_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_18448_end_mask_0 = const()[name = tensor("op_18448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18448_cast_fp16 = slice_by_index(begin = var_18448_begin_0, end = var_18448_end_0, end_mask = var_18448_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18448_cast_fp16")]; + tensor var_18452_begin_0 = const()[name = tensor("op_18452_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_18452_end_0 = const()[name = tensor("op_18452_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_18452_end_mask_0 = const()[name = tensor("op_18452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18452_cast_fp16 = slice_by_index(begin = var_18452_begin_0, end = var_18452_end_0, end_mask = var_18452_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18452_cast_fp16")]; + tensor var_18456_begin_0 = const()[name = tensor("op_18456_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_18456_end_0 = const()[name = tensor("op_18456_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_18456_end_mask_0 = const()[name = tensor("op_18456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18456_cast_fp16 = slice_by_index(begin = var_18456_begin_0, end = var_18456_end_0, end_mask = var_18456_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18456_cast_fp16")]; + tensor var_18460_begin_0 = const()[name = tensor("op_18460_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_18460_end_0 = const()[name = tensor("op_18460_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_18460_end_mask_0 = const()[name = tensor("op_18460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18460_cast_fp16 = slice_by_index(begin = var_18460_begin_0, end = var_18460_end_0, end_mask = var_18460_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18460_cast_fp16")]; + tensor var_18464_begin_0 = const()[name = tensor("op_18464_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_18464_end_0 = const()[name = tensor("op_18464_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_18464_end_mask_0 = const()[name = tensor("op_18464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18464_cast_fp16 = slice_by_index(begin = var_18464_begin_0, end = var_18464_end_0, end_mask = var_18464_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18464_cast_fp16")]; + tensor var_18468_begin_0 = const()[name = tensor("op_18468_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_18468_end_0 = const()[name = tensor("op_18468_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_18468_end_mask_0 = const()[name = tensor("op_18468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18468_cast_fp16 = slice_by_index(begin = var_18468_begin_0, end = var_18468_end_0, end_mask = var_18468_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18468_cast_fp16")]; + tensor var_18472_begin_0 = const()[name = tensor("op_18472_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_18472_end_0 = const()[name = tensor("op_18472_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_18472_end_mask_0 = const()[name = tensor("op_18472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18472_cast_fp16 = slice_by_index(begin = var_18472_begin_0, end = var_18472_end_0, end_mask = var_18472_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18472_cast_fp16")]; + tensor var_18476_begin_0 = const()[name = tensor("op_18476_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_18476_end_0 = const()[name = tensor("op_18476_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_18476_end_mask_0 = const()[name = tensor("op_18476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18476_cast_fp16 = slice_by_index(begin = var_18476_begin_0, end = var_18476_end_0, end_mask = var_18476_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18476_cast_fp16")]; + tensor var_18480_begin_0 = const()[name = tensor("op_18480_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_18480_end_0 = const()[name = tensor("op_18480_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_18480_end_mask_0 = const()[name = tensor("op_18480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18480_cast_fp16 = slice_by_index(begin = var_18480_begin_0, end = var_18480_end_0, end_mask = var_18480_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18480_cast_fp16")]; + tensor var_18484_begin_0 = const()[name = tensor("op_18484_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_18484_end_0 = const()[name = tensor("op_18484_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_18484_end_mask_0 = const()[name = tensor("op_18484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18484_cast_fp16 = slice_by_index(begin = var_18484_begin_0, end = var_18484_end_0, end_mask = var_18484_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18484_cast_fp16")]; + tensor var_18488_begin_0 = const()[name = tensor("op_18488_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_18488_end_0 = const()[name = tensor("op_18488_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18488_end_mask_0 = const()[name = tensor("op_18488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18488_cast_fp16 = slice_by_index(begin = var_18488_begin_0, end = var_18488_end_0, end_mask = var_18488_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18488_cast_fp16")]; + tensor var_18492_begin_0 = const()[name = tensor("op_18492_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_18492_end_0 = const()[name = tensor("op_18492_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_18492_end_mask_0 = const()[name = tensor("op_18492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18492_cast_fp16 = slice_by_index(begin = var_18492_begin_0, end = var_18492_end_0, end_mask = var_18492_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18492_cast_fp16")]; + tensor var_18496_begin_0 = const()[name = tensor("op_18496_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_18496_end_0 = const()[name = tensor("op_18496_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_18496_end_mask_0 = const()[name = tensor("op_18496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18496_cast_fp16 = slice_by_index(begin = var_18496_begin_0, end = var_18496_end_0, end_mask = var_18496_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18496_cast_fp16")]; + tensor var_18500_begin_0 = const()[name = tensor("op_18500_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_18500_end_0 = const()[name = tensor("op_18500_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_18500_end_mask_0 = const()[name = tensor("op_18500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18500_cast_fp16 = slice_by_index(begin = var_18500_begin_0, end = var_18500_end_0, end_mask = var_18500_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18500_cast_fp16")]; + tensor var_18504_begin_0 = const()[name = tensor("op_18504_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_18504_end_0 = const()[name = tensor("op_18504_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_18504_end_mask_0 = const()[name = tensor("op_18504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18504_cast_fp16 = slice_by_index(begin = var_18504_begin_0, end = var_18504_end_0, end_mask = var_18504_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18504_cast_fp16")]; + tensor var_18506_begin_0 = const()[name = tensor("op_18506_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18506_end_0 = const()[name = tensor("op_18506_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_18506_end_mask_0 = const()[name = tensor("op_18506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18506_cast_fp16 = slice_by_index(begin = var_18506_begin_0, end = var_18506_end_0, end_mask = var_18506_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18506_cast_fp16")]; + tensor var_18510_begin_0 = const()[name = tensor("op_18510_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18510_end_0 = const()[name = tensor("op_18510_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_18510_end_mask_0 = const()[name = tensor("op_18510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18510_cast_fp16 = slice_by_index(begin = var_18510_begin_0, end = var_18510_end_0, end_mask = var_18510_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18510_cast_fp16")]; + tensor var_18514_begin_0 = const()[name = tensor("op_18514_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18514_end_0 = const()[name = tensor("op_18514_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_18514_end_mask_0 = const()[name = tensor("op_18514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18514_cast_fp16 = slice_by_index(begin = var_18514_begin_0, end = var_18514_end_0, end_mask = var_18514_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18514_cast_fp16")]; + tensor var_18518_begin_0 = const()[name = tensor("op_18518_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18518_end_0 = const()[name = tensor("op_18518_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_18518_end_mask_0 = const()[name = tensor("op_18518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18518_cast_fp16 = slice_by_index(begin = var_18518_begin_0, end = var_18518_end_0, end_mask = var_18518_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18518_cast_fp16")]; + tensor var_18522_begin_0 = const()[name = tensor("op_18522_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18522_end_0 = const()[name = tensor("op_18522_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_18522_end_mask_0 = const()[name = tensor("op_18522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18522_cast_fp16 = slice_by_index(begin = var_18522_begin_0, end = var_18522_end_0, end_mask = var_18522_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18522_cast_fp16")]; + tensor var_18526_begin_0 = const()[name = tensor("op_18526_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18526_end_0 = const()[name = tensor("op_18526_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_18526_end_mask_0 = const()[name = tensor("op_18526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18526_cast_fp16 = slice_by_index(begin = var_18526_begin_0, end = var_18526_end_0, end_mask = var_18526_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18526_cast_fp16")]; + tensor var_18530_begin_0 = const()[name = tensor("op_18530_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18530_end_0 = const()[name = tensor("op_18530_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_18530_end_mask_0 = const()[name = tensor("op_18530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18530_cast_fp16 = slice_by_index(begin = var_18530_begin_0, end = var_18530_end_0, end_mask = var_18530_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18530_cast_fp16")]; + tensor var_18534_begin_0 = const()[name = tensor("op_18534_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18534_end_0 = const()[name = tensor("op_18534_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_18534_end_mask_0 = const()[name = tensor("op_18534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18534_cast_fp16 = slice_by_index(begin = var_18534_begin_0, end = var_18534_end_0, end_mask = var_18534_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18534_cast_fp16")]; + tensor var_18538_begin_0 = const()[name = tensor("op_18538_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18538_end_0 = const()[name = tensor("op_18538_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_18538_end_mask_0 = const()[name = tensor("op_18538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18538_cast_fp16 = slice_by_index(begin = var_18538_begin_0, end = var_18538_end_0, end_mask = var_18538_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18538_cast_fp16")]; + tensor var_18542_begin_0 = const()[name = tensor("op_18542_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18542_end_0 = const()[name = tensor("op_18542_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_18542_end_mask_0 = const()[name = tensor("op_18542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18542_cast_fp16 = slice_by_index(begin = var_18542_begin_0, end = var_18542_end_0, end_mask = var_18542_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18542_cast_fp16")]; + tensor var_18546_begin_0 = const()[name = tensor("op_18546_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18546_end_0 = const()[name = tensor("op_18546_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_18546_end_mask_0 = const()[name = tensor("op_18546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18546_cast_fp16 = slice_by_index(begin = var_18546_begin_0, end = var_18546_end_0, end_mask = var_18546_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18546_cast_fp16")]; + tensor var_18550_begin_0 = const()[name = tensor("op_18550_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18550_end_0 = const()[name = tensor("op_18550_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_18550_end_mask_0 = const()[name = tensor("op_18550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18550_cast_fp16 = slice_by_index(begin = var_18550_begin_0, end = var_18550_end_0, end_mask = var_18550_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18550_cast_fp16")]; + tensor var_18554_begin_0 = const()[name = tensor("op_18554_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18554_end_0 = const()[name = tensor("op_18554_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_18554_end_mask_0 = const()[name = tensor("op_18554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18554_cast_fp16 = slice_by_index(begin = var_18554_begin_0, end = var_18554_end_0, end_mask = var_18554_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18554_cast_fp16")]; + tensor var_18558_begin_0 = const()[name = tensor("op_18558_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18558_end_0 = const()[name = tensor("op_18558_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_18558_end_mask_0 = const()[name = tensor("op_18558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18558_cast_fp16 = slice_by_index(begin = var_18558_begin_0, end = var_18558_end_0, end_mask = var_18558_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18558_cast_fp16")]; + tensor var_18562_begin_0 = const()[name = tensor("op_18562_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18562_end_0 = const()[name = tensor("op_18562_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_18562_end_mask_0 = const()[name = tensor("op_18562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18562_cast_fp16 = slice_by_index(begin = var_18562_begin_0, end = var_18562_end_0, end_mask = var_18562_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18562_cast_fp16")]; + tensor var_18566_begin_0 = const()[name = tensor("op_18566_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18566_end_0 = const()[name = tensor("op_18566_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18566_end_mask_0 = const()[name = tensor("op_18566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18566_cast_fp16 = slice_by_index(begin = var_18566_begin_0, end = var_18566_end_0, end_mask = var_18566_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18566_cast_fp16")]; + tensor var_18570_begin_0 = const()[name = tensor("op_18570_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18570_end_0 = const()[name = tensor("op_18570_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_18570_end_mask_0 = const()[name = tensor("op_18570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18570_cast_fp16 = slice_by_index(begin = var_18570_begin_0, end = var_18570_end_0, end_mask = var_18570_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18570_cast_fp16")]; + tensor var_18574_begin_0 = const()[name = tensor("op_18574_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18574_end_0 = const()[name = tensor("op_18574_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_18574_end_mask_0 = const()[name = tensor("op_18574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18574_cast_fp16 = slice_by_index(begin = var_18574_begin_0, end = var_18574_end_0, end_mask = var_18574_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18574_cast_fp16")]; + tensor var_18578_begin_0 = const()[name = tensor("op_18578_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18578_end_0 = const()[name = tensor("op_18578_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_18578_end_mask_0 = const()[name = tensor("op_18578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18578_cast_fp16 = slice_by_index(begin = var_18578_begin_0, end = var_18578_end_0, end_mask = var_18578_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18578_cast_fp16")]; + tensor var_18582_begin_0 = const()[name = tensor("op_18582_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18582_end_0 = const()[name = tensor("op_18582_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_18582_end_mask_0 = const()[name = tensor("op_18582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18582_cast_fp16 = slice_by_index(begin = var_18582_begin_0, end = var_18582_end_0, end_mask = var_18582_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18582_cast_fp16")]; + tensor var_18586_equation_0 = const()[name = tensor("op_18586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18586_cast_fp16 = einsum(equation = var_18586_equation_0, values = (var_18428_cast_fp16, var_18345_cast_fp16))[name = tensor("op_18586_cast_fp16")]; + tensor var_18587_to_fp16 = const()[name = tensor("op_18587_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1521_cast_fp16 = mul(x = var_18586_cast_fp16, y = var_18587_to_fp16)[name = tensor("aw_1521_cast_fp16")]; + tensor var_18590_equation_0 = const()[name = tensor("op_18590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18590_cast_fp16 = einsum(equation = var_18590_equation_0, values = (var_18432_cast_fp16, var_18349_cast_fp16))[name = tensor("op_18590_cast_fp16")]; + tensor var_18591_to_fp16 = const()[name = tensor("op_18591_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1523_cast_fp16 = mul(x = var_18590_cast_fp16, y = var_18591_to_fp16)[name = tensor("aw_1523_cast_fp16")]; + tensor var_18594_equation_0 = const()[name = tensor("op_18594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18594_cast_fp16 = einsum(equation = var_18594_equation_0, values = (var_18436_cast_fp16, var_18353_cast_fp16))[name = tensor("op_18594_cast_fp16")]; + tensor var_18595_to_fp16 = const()[name = tensor("op_18595_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1525_cast_fp16 = mul(x = var_18594_cast_fp16, y = var_18595_to_fp16)[name = tensor("aw_1525_cast_fp16")]; + tensor var_18598_equation_0 = const()[name = tensor("op_18598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18598_cast_fp16 = einsum(equation = var_18598_equation_0, values = (var_18440_cast_fp16, var_18357_cast_fp16))[name = tensor("op_18598_cast_fp16")]; + tensor var_18599_to_fp16 = const()[name = tensor("op_18599_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1527_cast_fp16 = mul(x = var_18598_cast_fp16, y = var_18599_to_fp16)[name = tensor("aw_1527_cast_fp16")]; + tensor var_18602_equation_0 = const()[name = tensor("op_18602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18602_cast_fp16 = einsum(equation = var_18602_equation_0, values = (var_18444_cast_fp16, var_18361_cast_fp16))[name = tensor("op_18602_cast_fp16")]; + tensor var_18603_to_fp16 = const()[name = tensor("op_18603_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1529_cast_fp16 = mul(x = var_18602_cast_fp16, y = var_18603_to_fp16)[name = tensor("aw_1529_cast_fp16")]; + tensor var_18606_equation_0 = const()[name = tensor("op_18606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18606_cast_fp16 = einsum(equation = var_18606_equation_0, values = (var_18448_cast_fp16, var_18365_cast_fp16))[name = tensor("op_18606_cast_fp16")]; + tensor var_18607_to_fp16 = const()[name = tensor("op_18607_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1531_cast_fp16 = mul(x = var_18606_cast_fp16, y = var_18607_to_fp16)[name = tensor("aw_1531_cast_fp16")]; + tensor var_18610_equation_0 = const()[name = tensor("op_18610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18610_cast_fp16 = einsum(equation = var_18610_equation_0, values = (var_18452_cast_fp16, var_18369_cast_fp16))[name = tensor("op_18610_cast_fp16")]; + tensor var_18611_to_fp16 = const()[name = tensor("op_18611_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1533_cast_fp16 = mul(x = var_18610_cast_fp16, y = var_18611_to_fp16)[name = tensor("aw_1533_cast_fp16")]; + tensor var_18614_equation_0 = const()[name = tensor("op_18614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18614_cast_fp16 = einsum(equation = var_18614_equation_0, values = (var_18456_cast_fp16, var_18373_cast_fp16))[name = tensor("op_18614_cast_fp16")]; + tensor var_18615_to_fp16 = const()[name = tensor("op_18615_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1535_cast_fp16 = mul(x = var_18614_cast_fp16, y = var_18615_to_fp16)[name = tensor("aw_1535_cast_fp16")]; + tensor var_18618_equation_0 = const()[name = tensor("op_18618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18618_cast_fp16 = einsum(equation = var_18618_equation_0, values = (var_18460_cast_fp16, var_18377_cast_fp16))[name = tensor("op_18618_cast_fp16")]; + tensor var_18619_to_fp16 = const()[name = tensor("op_18619_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1537_cast_fp16 = mul(x = var_18618_cast_fp16, y = var_18619_to_fp16)[name = tensor("aw_1537_cast_fp16")]; + tensor var_18622_equation_0 = const()[name = tensor("op_18622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18622_cast_fp16 = einsum(equation = var_18622_equation_0, values = (var_18464_cast_fp16, var_18381_cast_fp16))[name = tensor("op_18622_cast_fp16")]; + tensor var_18623_to_fp16 = const()[name = tensor("op_18623_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1539_cast_fp16 = mul(x = var_18622_cast_fp16, y = var_18623_to_fp16)[name = tensor("aw_1539_cast_fp16")]; + tensor var_18626_equation_0 = const()[name = tensor("op_18626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18626_cast_fp16 = einsum(equation = var_18626_equation_0, values = (var_18468_cast_fp16, var_18385_cast_fp16))[name = tensor("op_18626_cast_fp16")]; + tensor var_18627_to_fp16 = const()[name = tensor("op_18627_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1541_cast_fp16 = mul(x = var_18626_cast_fp16, y = var_18627_to_fp16)[name = tensor("aw_1541_cast_fp16")]; + tensor var_18630_equation_0 = const()[name = tensor("op_18630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18630_cast_fp16 = einsum(equation = var_18630_equation_0, values = (var_18472_cast_fp16, var_18389_cast_fp16))[name = tensor("op_18630_cast_fp16")]; + tensor var_18631_to_fp16 = const()[name = tensor("op_18631_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1543_cast_fp16 = mul(x = var_18630_cast_fp16, y = var_18631_to_fp16)[name = tensor("aw_1543_cast_fp16")]; + tensor var_18634_equation_0 = const()[name = tensor("op_18634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18634_cast_fp16 = einsum(equation = var_18634_equation_0, values = (var_18476_cast_fp16, var_18393_cast_fp16))[name = tensor("op_18634_cast_fp16")]; + tensor var_18635_to_fp16 = const()[name = tensor("op_18635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1545_cast_fp16 = mul(x = var_18634_cast_fp16, y = var_18635_to_fp16)[name = tensor("aw_1545_cast_fp16")]; + tensor var_18638_equation_0 = const()[name = tensor("op_18638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18638_cast_fp16 = einsum(equation = var_18638_equation_0, values = (var_18480_cast_fp16, var_18397_cast_fp16))[name = tensor("op_18638_cast_fp16")]; + tensor var_18639_to_fp16 = const()[name = tensor("op_18639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1547_cast_fp16 = mul(x = var_18638_cast_fp16, y = var_18639_to_fp16)[name = tensor("aw_1547_cast_fp16")]; + tensor var_18642_equation_0 = const()[name = tensor("op_18642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18642_cast_fp16 = einsum(equation = var_18642_equation_0, values = (var_18484_cast_fp16, var_18401_cast_fp16))[name = tensor("op_18642_cast_fp16")]; + tensor var_18643_to_fp16 = const()[name = tensor("op_18643_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1549_cast_fp16 = mul(x = var_18642_cast_fp16, y = var_18643_to_fp16)[name = tensor("aw_1549_cast_fp16")]; + tensor var_18646_equation_0 = const()[name = tensor("op_18646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18646_cast_fp16 = einsum(equation = var_18646_equation_0, values = (var_18488_cast_fp16, var_18405_cast_fp16))[name = tensor("op_18646_cast_fp16")]; + tensor var_18647_to_fp16 = const()[name = tensor("op_18647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1551_cast_fp16 = mul(x = var_18646_cast_fp16, y = var_18647_to_fp16)[name = tensor("aw_1551_cast_fp16")]; + tensor var_18650_equation_0 = const()[name = tensor("op_18650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18650_cast_fp16 = einsum(equation = var_18650_equation_0, values = (var_18492_cast_fp16, var_18409_cast_fp16))[name = tensor("op_18650_cast_fp16")]; + tensor var_18651_to_fp16 = const()[name = tensor("op_18651_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1553_cast_fp16 = mul(x = var_18650_cast_fp16, y = var_18651_to_fp16)[name = tensor("aw_1553_cast_fp16")]; + tensor var_18654_equation_0 = const()[name = tensor("op_18654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18654_cast_fp16 = einsum(equation = var_18654_equation_0, values = (var_18496_cast_fp16, var_18413_cast_fp16))[name = tensor("op_18654_cast_fp16")]; + tensor var_18655_to_fp16 = const()[name = tensor("op_18655_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1555_cast_fp16 = mul(x = var_18654_cast_fp16, y = var_18655_to_fp16)[name = tensor("aw_1555_cast_fp16")]; + tensor var_18658_equation_0 = const()[name = tensor("op_18658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18658_cast_fp16 = einsum(equation = var_18658_equation_0, values = (var_18500_cast_fp16, var_18417_cast_fp16))[name = tensor("op_18658_cast_fp16")]; + tensor var_18659_to_fp16 = const()[name = tensor("op_18659_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1557_cast_fp16 = mul(x = var_18658_cast_fp16, y = var_18659_to_fp16)[name = tensor("aw_1557_cast_fp16")]; + tensor var_18662_equation_0 = const()[name = tensor("op_18662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18662_cast_fp16 = einsum(equation = var_18662_equation_0, values = (var_18504_cast_fp16, var_18421_cast_fp16))[name = tensor("op_18662_cast_fp16")]; + tensor var_18663_to_fp16 = const()[name = tensor("op_18663_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1559_cast_fp16 = mul(x = var_18662_cast_fp16, y = var_18663_to_fp16)[name = tensor("aw_1559_cast_fp16")]; + tensor var_18665_cast_fp16 = softmax(axis = var_2624, x = aw_1521_cast_fp16)[name = tensor("op_18665_cast_fp16")]; + tensor var_18666_cast_fp16 = softmax(axis = var_2624, x = aw_1523_cast_fp16)[name = tensor("op_18666_cast_fp16")]; + tensor var_18667_cast_fp16 = softmax(axis = var_2624, x = aw_1525_cast_fp16)[name = tensor("op_18667_cast_fp16")]; + tensor var_18668_cast_fp16 = softmax(axis = var_2624, x = aw_1527_cast_fp16)[name = tensor("op_18668_cast_fp16")]; + tensor var_18669_cast_fp16 = softmax(axis = var_2624, x = aw_1529_cast_fp16)[name = tensor("op_18669_cast_fp16")]; + tensor var_18670_cast_fp16 = softmax(axis = var_2624, x = aw_1531_cast_fp16)[name = tensor("op_18670_cast_fp16")]; + tensor var_18671_cast_fp16 = softmax(axis = var_2624, x = aw_1533_cast_fp16)[name = tensor("op_18671_cast_fp16")]; + tensor var_18672_cast_fp16 = softmax(axis = var_2624, x = aw_1535_cast_fp16)[name = tensor("op_18672_cast_fp16")]; + tensor var_18673_cast_fp16 = softmax(axis = var_2624, x = aw_1537_cast_fp16)[name = tensor("op_18673_cast_fp16")]; + tensor var_18674_cast_fp16 = softmax(axis = var_2624, x = aw_1539_cast_fp16)[name = tensor("op_18674_cast_fp16")]; + tensor var_18675_cast_fp16 = softmax(axis = var_2624, x = aw_1541_cast_fp16)[name = tensor("op_18675_cast_fp16")]; + tensor var_18676_cast_fp16 = softmax(axis = var_2624, x = aw_1543_cast_fp16)[name = tensor("op_18676_cast_fp16")]; + tensor var_18677_cast_fp16 = softmax(axis = var_2624, x = aw_1545_cast_fp16)[name = tensor("op_18677_cast_fp16")]; + tensor var_18678_cast_fp16 = softmax(axis = var_2624, x = aw_1547_cast_fp16)[name = tensor("op_18678_cast_fp16")]; + tensor var_18679_cast_fp16 = softmax(axis = var_2624, x = aw_1549_cast_fp16)[name = tensor("op_18679_cast_fp16")]; + tensor var_18680_cast_fp16 = softmax(axis = var_2624, x = aw_1551_cast_fp16)[name = tensor("op_18680_cast_fp16")]; + tensor var_18681_cast_fp16 = softmax(axis = var_2624, x = aw_1553_cast_fp16)[name = tensor("op_18681_cast_fp16")]; + tensor var_18682_cast_fp16 = softmax(axis = var_2624, x = aw_1555_cast_fp16)[name = tensor("op_18682_cast_fp16")]; + tensor var_18683_cast_fp16 = softmax(axis = var_2624, x = aw_1557_cast_fp16)[name = tensor("op_18683_cast_fp16")]; + tensor var_18684_cast_fp16 = softmax(axis = var_2624, x = aw_1559_cast_fp16)[name = tensor("op_18684_cast_fp16")]; + tensor var_18686_equation_0 = const()[name = tensor("op_18686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18686_cast_fp16 = einsum(equation = var_18686_equation_0, values = (var_18506_cast_fp16, var_18665_cast_fp16))[name = tensor("op_18686_cast_fp16")]; + tensor var_18688_equation_0 = const()[name = tensor("op_18688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18688_cast_fp16 = einsum(equation = var_18688_equation_0, values = (var_18510_cast_fp16, var_18666_cast_fp16))[name = tensor("op_18688_cast_fp16")]; + tensor var_18690_equation_0 = const()[name = tensor("op_18690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18690_cast_fp16 = einsum(equation = var_18690_equation_0, values = (var_18514_cast_fp16, var_18667_cast_fp16))[name = tensor("op_18690_cast_fp16")]; + tensor var_18692_equation_0 = const()[name = tensor("op_18692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18692_cast_fp16 = einsum(equation = var_18692_equation_0, values = (var_18518_cast_fp16, var_18668_cast_fp16))[name = tensor("op_18692_cast_fp16")]; + tensor var_18694_equation_0 = const()[name = tensor("op_18694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18694_cast_fp16 = einsum(equation = var_18694_equation_0, values = (var_18522_cast_fp16, var_18669_cast_fp16))[name = tensor("op_18694_cast_fp16")]; + tensor var_18696_equation_0 = const()[name = tensor("op_18696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18696_cast_fp16 = einsum(equation = var_18696_equation_0, values = (var_18526_cast_fp16, var_18670_cast_fp16))[name = tensor("op_18696_cast_fp16")]; + tensor var_18698_equation_0 = const()[name = tensor("op_18698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18698_cast_fp16 = einsum(equation = var_18698_equation_0, values = (var_18530_cast_fp16, var_18671_cast_fp16))[name = tensor("op_18698_cast_fp16")]; + tensor var_18700_equation_0 = const()[name = tensor("op_18700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18700_cast_fp16 = einsum(equation = var_18700_equation_0, values = (var_18534_cast_fp16, var_18672_cast_fp16))[name = tensor("op_18700_cast_fp16")]; + tensor var_18702_equation_0 = const()[name = tensor("op_18702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18702_cast_fp16 = einsum(equation = var_18702_equation_0, values = (var_18538_cast_fp16, var_18673_cast_fp16))[name = tensor("op_18702_cast_fp16")]; + tensor var_18704_equation_0 = const()[name = tensor("op_18704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18704_cast_fp16 = einsum(equation = var_18704_equation_0, values = (var_18542_cast_fp16, var_18674_cast_fp16))[name = tensor("op_18704_cast_fp16")]; + tensor var_18706_equation_0 = const()[name = tensor("op_18706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18706_cast_fp16 = einsum(equation = var_18706_equation_0, values = (var_18546_cast_fp16, var_18675_cast_fp16))[name = tensor("op_18706_cast_fp16")]; + tensor var_18708_equation_0 = const()[name = tensor("op_18708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18708_cast_fp16 = einsum(equation = var_18708_equation_0, values = (var_18550_cast_fp16, var_18676_cast_fp16))[name = tensor("op_18708_cast_fp16")]; + tensor var_18710_equation_0 = const()[name = tensor("op_18710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18710_cast_fp16 = einsum(equation = var_18710_equation_0, values = (var_18554_cast_fp16, var_18677_cast_fp16))[name = tensor("op_18710_cast_fp16")]; + tensor var_18712_equation_0 = const()[name = tensor("op_18712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18712_cast_fp16 = einsum(equation = var_18712_equation_0, values = (var_18558_cast_fp16, var_18678_cast_fp16))[name = tensor("op_18712_cast_fp16")]; + tensor var_18714_equation_0 = const()[name = tensor("op_18714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18714_cast_fp16 = einsum(equation = var_18714_equation_0, values = (var_18562_cast_fp16, var_18679_cast_fp16))[name = tensor("op_18714_cast_fp16")]; + tensor var_18716_equation_0 = const()[name = tensor("op_18716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18716_cast_fp16 = einsum(equation = var_18716_equation_0, values = (var_18566_cast_fp16, var_18680_cast_fp16))[name = tensor("op_18716_cast_fp16")]; + tensor var_18718_equation_0 = const()[name = tensor("op_18718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18718_cast_fp16 = einsum(equation = var_18718_equation_0, values = (var_18570_cast_fp16, var_18681_cast_fp16))[name = tensor("op_18718_cast_fp16")]; + tensor var_18720_equation_0 = const()[name = tensor("op_18720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18720_cast_fp16 = einsum(equation = var_18720_equation_0, values = (var_18574_cast_fp16, var_18682_cast_fp16))[name = tensor("op_18720_cast_fp16")]; + tensor var_18722_equation_0 = const()[name = tensor("op_18722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18722_cast_fp16 = einsum(equation = var_18722_equation_0, values = (var_18578_cast_fp16, var_18683_cast_fp16))[name = tensor("op_18722_cast_fp16")]; + tensor var_18724_equation_0 = const()[name = tensor("op_18724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18724_cast_fp16 = einsum(equation = var_18724_equation_0, values = (var_18582_cast_fp16, var_18684_cast_fp16))[name = tensor("op_18724_cast_fp16")]; + tensor input_285_interleave_0 = const()[name = tensor("input_285_interleave_0"), val = tensor(false)]; + tensor input_285_cast_fp16 = concat(axis = var_2624, interleave = input_285_interleave_0, values = (var_18686_cast_fp16, var_18688_cast_fp16, var_18690_cast_fp16, var_18692_cast_fp16, var_18694_cast_fp16, var_18696_cast_fp16, var_18698_cast_fp16, var_18700_cast_fp16, var_18702_cast_fp16, var_18704_cast_fp16, var_18706_cast_fp16, var_18708_cast_fp16, var_18710_cast_fp16, var_18712_cast_fp16, var_18714_cast_fp16, var_18716_cast_fp16, var_18718_cast_fp16, var_18720_cast_fp16, var_18722_cast_fp16, var_18724_cast_fp16))[name = tensor("input_285_cast_fp16")]; + tensor var_18734_pad_type_0 = const()[name = tensor("op_18734_pad_type_0"), val = tensor("valid")]; + tensor var_18734_strides_0 = const()[name = tensor("op_18734_strides_0"), val = tensor([1, 1])]; + tensor var_18734_pad_0 = const()[name = tensor("op_18734_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18734_dilations_0 = const()[name = tensor("op_18734_dilations_0"), val = tensor([1, 1])]; + tensor var_18734_groups_0 = const()[name = tensor("op_18734_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547621056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548849920))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548850112)))]; + tensor var_18734_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_18734_dilations_0, groups = var_18734_groups_0, pad = var_18734_pad_0, pad_type = var_18734_pad_type_0, strides = var_18734_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = tensor("op_18734_cast_fp16")]; + tensor inputs_129_cast_fp16 = add(x = var_18734_cast_fp16, y = inputs_127_cast_fp16)[name = tensor("inputs_129_cast_fp16")]; + tensor hidden_states_181_axes_0 = const()[name = tensor("hidden_states_181_axes_0"), val = tensor([1])]; + tensor hidden_states_181_gamma_0_to_fp16 = const()[name = tensor("hidden_states_181_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548852736)))]; + tensor hidden_states_181_beta_0_to_fp16 = const()[name = tensor("hidden_states_181_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548855360)))]; + tensor var_18744_to_fp16 = const()[name = tensor("op_18744_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_181_cast_fp16 = layer_norm(axes = hidden_states_181_axes_0, beta = hidden_states_181_beta_0_to_fp16, epsilon = var_18744_to_fp16, gamma = hidden_states_181_gamma_0_to_fp16, x = inputs_129_cast_fp16)[name = tensor("hidden_states_181_cast_fp16")]; + tensor q_87_pad_type_0 = const()[name = tensor("q_87_pad_type_0"), val = tensor("valid")]; + tensor q_87_strides_0 = const()[name = tensor("q_87_strides_0"), val = tensor([1, 1])]; + tensor q_87_pad_0 = const()[name = tensor("q_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_87_dilations_0 = const()[name = tensor("q_87_dilations_0"), val = tensor([1, 1])]; + tensor q_87_groups_0 = const()[name = tensor("q_87_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548857984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(550086848))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_87_cast_fp16 = conv(dilations = q_87_dilations_0, groups = q_87_groups_0, pad = q_87_pad_0, pad_type = q_87_pad_type_0, strides = q_87_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_181_cast_fp16)[name = tensor("q_87_cast_fp16")]; + tensor k_173_pad_type_0 = const()[name = tensor("k_173_pad_type_0"), val = tensor("valid")]; + tensor k_173_strides_0 = const()[name = tensor("k_173_strides_0"), val = tensor([1, 1])]; + tensor k_173_pad_0 = const()[name = tensor("k_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_173_dilations_0 = const()[name = tensor("k_173_dilations_0"), val = tensor([1, 1])]; + tensor k_173_groups_0 = const()[name = tensor("k_173_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(550087040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552053184))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_173_cast_fp16 = conv(dilations = k_173_dilations_0, groups = k_173_groups_0, pad = k_173_pad_0, pad_type = k_173_pad_type_0, strides = k_173_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_173_cast_fp16")]; + tensor v_87_pad_type_0 = const()[name = tensor("v_87_pad_type_0"), val = tensor("valid")]; + tensor v_87_strides_0 = const()[name = tensor("v_87_strides_0"), val = tensor([1, 1])]; + tensor v_87_pad_0 = const()[name = tensor("v_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_87_dilations_0 = const()[name = tensor("v_87_dilations_0"), val = tensor([1, 1])]; + tensor v_87_groups_0 = const()[name = tensor("v_87_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552053376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554019520))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_87_cast_fp16 = conv(dilations = v_87_dilations_0, groups = v_87_groups_0, pad = v_87_pad_0, pad_type = v_87_pad_type_0, strides = v_87_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_87_cast_fp16")]; + tensor var_18777_begin_0 = const()[name = tensor("op_18777_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18777_end_0 = const()[name = tensor("op_18777_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_18777_end_mask_0 = const()[name = tensor("op_18777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18777_cast_fp16 = slice_by_index(begin = var_18777_begin_0, end = var_18777_end_0, end_mask = var_18777_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18777_cast_fp16")]; + tensor var_18781_begin_0 = const()[name = tensor("op_18781_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18781_end_0 = const()[name = tensor("op_18781_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_18781_end_mask_0 = const()[name = tensor("op_18781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18781_cast_fp16 = slice_by_index(begin = var_18781_begin_0, end = var_18781_end_0, end_mask = var_18781_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18781_cast_fp16")]; + tensor var_18785_begin_0 = const()[name = tensor("op_18785_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18785_end_0 = const()[name = tensor("op_18785_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_18785_end_mask_0 = const()[name = tensor("op_18785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18785_cast_fp16 = slice_by_index(begin = var_18785_begin_0, end = var_18785_end_0, end_mask = var_18785_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18785_cast_fp16")]; + tensor var_18789_begin_0 = const()[name = tensor("op_18789_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18789_end_0 = const()[name = tensor("op_18789_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_18789_end_mask_0 = const()[name = tensor("op_18789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18789_cast_fp16 = slice_by_index(begin = var_18789_begin_0, end = var_18789_end_0, end_mask = var_18789_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18789_cast_fp16")]; + tensor var_18793_begin_0 = const()[name = tensor("op_18793_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18793_end_0 = const()[name = tensor("op_18793_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_18793_end_mask_0 = const()[name = tensor("op_18793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18793_cast_fp16 = slice_by_index(begin = var_18793_begin_0, end = var_18793_end_0, end_mask = var_18793_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18793_cast_fp16")]; + tensor var_18797_begin_0 = const()[name = tensor("op_18797_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18797_end_0 = const()[name = tensor("op_18797_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_18797_end_mask_0 = const()[name = tensor("op_18797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18797_cast_fp16 = slice_by_index(begin = var_18797_begin_0, end = var_18797_end_0, end_mask = var_18797_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18797_cast_fp16")]; + tensor var_18801_begin_0 = const()[name = tensor("op_18801_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18801_end_0 = const()[name = tensor("op_18801_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_18801_end_mask_0 = const()[name = tensor("op_18801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18801_cast_fp16 = slice_by_index(begin = var_18801_begin_0, end = var_18801_end_0, end_mask = var_18801_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18801_cast_fp16")]; + tensor var_18805_begin_0 = const()[name = tensor("op_18805_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18805_end_0 = const()[name = tensor("op_18805_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_18805_end_mask_0 = const()[name = tensor("op_18805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18805_cast_fp16 = slice_by_index(begin = var_18805_begin_0, end = var_18805_end_0, end_mask = var_18805_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18805_cast_fp16")]; + tensor var_18809_begin_0 = const()[name = tensor("op_18809_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18809_end_0 = const()[name = tensor("op_18809_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_18809_end_mask_0 = const()[name = tensor("op_18809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18809_cast_fp16 = slice_by_index(begin = var_18809_begin_0, end = var_18809_end_0, end_mask = var_18809_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18809_cast_fp16")]; + tensor var_18813_begin_0 = const()[name = tensor("op_18813_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18813_end_0 = const()[name = tensor("op_18813_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_18813_end_mask_0 = const()[name = tensor("op_18813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18813_cast_fp16 = slice_by_index(begin = var_18813_begin_0, end = var_18813_end_0, end_mask = var_18813_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18813_cast_fp16")]; + tensor var_18817_begin_0 = const()[name = tensor("op_18817_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18817_end_0 = const()[name = tensor("op_18817_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_18817_end_mask_0 = const()[name = tensor("op_18817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18817_cast_fp16 = slice_by_index(begin = var_18817_begin_0, end = var_18817_end_0, end_mask = var_18817_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18817_cast_fp16")]; + tensor var_18821_begin_0 = const()[name = tensor("op_18821_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18821_end_0 = const()[name = tensor("op_18821_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_18821_end_mask_0 = const()[name = tensor("op_18821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18821_cast_fp16 = slice_by_index(begin = var_18821_begin_0, end = var_18821_end_0, end_mask = var_18821_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18821_cast_fp16")]; + tensor var_18825_begin_0 = const()[name = tensor("op_18825_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18825_end_0 = const()[name = tensor("op_18825_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_18825_end_mask_0 = const()[name = tensor("op_18825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18825_cast_fp16 = slice_by_index(begin = var_18825_begin_0, end = var_18825_end_0, end_mask = var_18825_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18825_cast_fp16")]; + tensor var_18829_begin_0 = const()[name = tensor("op_18829_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18829_end_0 = const()[name = tensor("op_18829_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_18829_end_mask_0 = const()[name = tensor("op_18829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18829_cast_fp16 = slice_by_index(begin = var_18829_begin_0, end = var_18829_end_0, end_mask = var_18829_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18829_cast_fp16")]; + tensor var_18833_begin_0 = const()[name = tensor("op_18833_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18833_end_0 = const()[name = tensor("op_18833_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_18833_end_mask_0 = const()[name = tensor("op_18833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18833_cast_fp16 = slice_by_index(begin = var_18833_begin_0, end = var_18833_end_0, end_mask = var_18833_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18833_cast_fp16")]; + tensor var_18837_begin_0 = const()[name = tensor("op_18837_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18837_end_0 = const()[name = tensor("op_18837_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18837_end_mask_0 = const()[name = tensor("op_18837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18837_cast_fp16 = slice_by_index(begin = var_18837_begin_0, end = var_18837_end_0, end_mask = var_18837_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18837_cast_fp16")]; + tensor var_18841_begin_0 = const()[name = tensor("op_18841_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18841_end_0 = const()[name = tensor("op_18841_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_18841_end_mask_0 = const()[name = tensor("op_18841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18841_cast_fp16 = slice_by_index(begin = var_18841_begin_0, end = var_18841_end_0, end_mask = var_18841_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18841_cast_fp16")]; + tensor var_18845_begin_0 = const()[name = tensor("op_18845_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18845_end_0 = const()[name = tensor("op_18845_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_18845_end_mask_0 = const()[name = tensor("op_18845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18845_cast_fp16 = slice_by_index(begin = var_18845_begin_0, end = var_18845_end_0, end_mask = var_18845_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18845_cast_fp16")]; + tensor var_18849_begin_0 = const()[name = tensor("op_18849_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18849_end_0 = const()[name = tensor("op_18849_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_18849_end_mask_0 = const()[name = tensor("op_18849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18849_cast_fp16 = slice_by_index(begin = var_18849_begin_0, end = var_18849_end_0, end_mask = var_18849_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18849_cast_fp16")]; + tensor var_18853_begin_0 = const()[name = tensor("op_18853_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18853_end_0 = const()[name = tensor("op_18853_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_18853_end_mask_0 = const()[name = tensor("op_18853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18853_cast_fp16 = slice_by_index(begin = var_18853_begin_0, end = var_18853_end_0, end_mask = var_18853_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18853_cast_fp16")]; + tensor k_175_perm_0 = const()[name = tensor("k_175_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_18860_begin_0 = const()[name = tensor("op_18860_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18860_end_0 = const()[name = tensor("op_18860_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_18860_end_mask_0 = const()[name = tensor("op_18860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_175_cast_fp16 = transpose(perm = k_175_perm_0, x = k_173_cast_fp16)[name = tensor("transpose_24")]; + tensor var_18860_cast_fp16 = slice_by_index(begin = var_18860_begin_0, end = var_18860_end_0, end_mask = var_18860_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18860_cast_fp16")]; + tensor var_18864_begin_0 = const()[name = tensor("op_18864_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_18864_end_0 = const()[name = tensor("op_18864_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_18864_end_mask_0 = const()[name = tensor("op_18864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18864_cast_fp16 = slice_by_index(begin = var_18864_begin_0, end = var_18864_end_0, end_mask = var_18864_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18864_cast_fp16")]; + tensor var_18868_begin_0 = const()[name = tensor("op_18868_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_18868_end_0 = const()[name = tensor("op_18868_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_18868_end_mask_0 = const()[name = tensor("op_18868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18868_cast_fp16 = slice_by_index(begin = var_18868_begin_0, end = var_18868_end_0, end_mask = var_18868_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18868_cast_fp16")]; + tensor var_18872_begin_0 = const()[name = tensor("op_18872_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_18872_end_0 = const()[name = tensor("op_18872_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_18872_end_mask_0 = const()[name = tensor("op_18872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18872_cast_fp16 = slice_by_index(begin = var_18872_begin_0, end = var_18872_end_0, end_mask = var_18872_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18872_cast_fp16")]; + tensor var_18876_begin_0 = const()[name = tensor("op_18876_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_18876_end_0 = const()[name = tensor("op_18876_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_18876_end_mask_0 = const()[name = tensor("op_18876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18876_cast_fp16 = slice_by_index(begin = var_18876_begin_0, end = var_18876_end_0, end_mask = var_18876_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18876_cast_fp16")]; + tensor var_18880_begin_0 = const()[name = tensor("op_18880_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_18880_end_0 = const()[name = tensor("op_18880_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_18880_end_mask_0 = const()[name = tensor("op_18880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18880_cast_fp16 = slice_by_index(begin = var_18880_begin_0, end = var_18880_end_0, end_mask = var_18880_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18880_cast_fp16")]; + tensor var_18884_begin_0 = const()[name = tensor("op_18884_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_18884_end_0 = const()[name = tensor("op_18884_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_18884_end_mask_0 = const()[name = tensor("op_18884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18884_cast_fp16 = slice_by_index(begin = var_18884_begin_0, end = var_18884_end_0, end_mask = var_18884_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18884_cast_fp16")]; + tensor var_18888_begin_0 = const()[name = tensor("op_18888_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_18888_end_0 = const()[name = tensor("op_18888_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_18888_end_mask_0 = const()[name = tensor("op_18888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18888_cast_fp16 = slice_by_index(begin = var_18888_begin_0, end = var_18888_end_0, end_mask = var_18888_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18888_cast_fp16")]; + tensor var_18892_begin_0 = const()[name = tensor("op_18892_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_18892_end_0 = const()[name = tensor("op_18892_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_18892_end_mask_0 = const()[name = tensor("op_18892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18892_cast_fp16 = slice_by_index(begin = var_18892_begin_0, end = var_18892_end_0, end_mask = var_18892_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18892_cast_fp16")]; + tensor var_18896_begin_0 = const()[name = tensor("op_18896_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_18896_end_0 = const()[name = tensor("op_18896_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_18896_end_mask_0 = const()[name = tensor("op_18896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18896_cast_fp16 = slice_by_index(begin = var_18896_begin_0, end = var_18896_end_0, end_mask = var_18896_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18896_cast_fp16")]; + tensor var_18900_begin_0 = const()[name = tensor("op_18900_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_18900_end_0 = const()[name = tensor("op_18900_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_18900_end_mask_0 = const()[name = tensor("op_18900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18900_cast_fp16 = slice_by_index(begin = var_18900_begin_0, end = var_18900_end_0, end_mask = var_18900_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18900_cast_fp16")]; + tensor var_18904_begin_0 = const()[name = tensor("op_18904_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_18904_end_0 = const()[name = tensor("op_18904_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_18904_end_mask_0 = const()[name = tensor("op_18904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18904_cast_fp16 = slice_by_index(begin = var_18904_begin_0, end = var_18904_end_0, end_mask = var_18904_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18904_cast_fp16")]; + tensor var_18908_begin_0 = const()[name = tensor("op_18908_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_18908_end_0 = const()[name = tensor("op_18908_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_18908_end_mask_0 = const()[name = tensor("op_18908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18908_cast_fp16 = slice_by_index(begin = var_18908_begin_0, end = var_18908_end_0, end_mask = var_18908_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18908_cast_fp16")]; + tensor var_18912_begin_0 = const()[name = tensor("op_18912_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_18912_end_0 = const()[name = tensor("op_18912_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_18912_end_mask_0 = const()[name = tensor("op_18912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18912_cast_fp16 = slice_by_index(begin = var_18912_begin_0, end = var_18912_end_0, end_mask = var_18912_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18912_cast_fp16")]; + tensor var_18916_begin_0 = const()[name = tensor("op_18916_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_18916_end_0 = const()[name = tensor("op_18916_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_18916_end_mask_0 = const()[name = tensor("op_18916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18916_cast_fp16 = slice_by_index(begin = var_18916_begin_0, end = var_18916_end_0, end_mask = var_18916_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18916_cast_fp16")]; + tensor var_18920_begin_0 = const()[name = tensor("op_18920_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_18920_end_0 = const()[name = tensor("op_18920_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_18920_end_mask_0 = const()[name = tensor("op_18920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18920_cast_fp16 = slice_by_index(begin = var_18920_begin_0, end = var_18920_end_0, end_mask = var_18920_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18920_cast_fp16")]; + tensor var_18924_begin_0 = const()[name = tensor("op_18924_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_18924_end_0 = const()[name = tensor("op_18924_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_18924_end_mask_0 = const()[name = tensor("op_18924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18924_cast_fp16 = slice_by_index(begin = var_18924_begin_0, end = var_18924_end_0, end_mask = var_18924_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18924_cast_fp16")]; + tensor var_18928_begin_0 = const()[name = tensor("op_18928_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_18928_end_0 = const()[name = tensor("op_18928_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_18928_end_mask_0 = const()[name = tensor("op_18928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18928_cast_fp16 = slice_by_index(begin = var_18928_begin_0, end = var_18928_end_0, end_mask = var_18928_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18928_cast_fp16")]; + tensor var_18932_begin_0 = const()[name = tensor("op_18932_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_18932_end_0 = const()[name = tensor("op_18932_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_18932_end_mask_0 = const()[name = tensor("op_18932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18932_cast_fp16 = slice_by_index(begin = var_18932_begin_0, end = var_18932_end_0, end_mask = var_18932_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18932_cast_fp16")]; + tensor var_18936_begin_0 = const()[name = tensor("op_18936_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_18936_end_0 = const()[name = tensor("op_18936_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_18936_end_mask_0 = const()[name = tensor("op_18936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18936_cast_fp16 = slice_by_index(begin = var_18936_begin_0, end = var_18936_end_0, end_mask = var_18936_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18936_cast_fp16")]; + tensor var_18938_begin_0 = const()[name = tensor("op_18938_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18938_end_0 = const()[name = tensor("op_18938_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_18938_end_mask_0 = const()[name = tensor("op_18938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18938_cast_fp16 = slice_by_index(begin = var_18938_begin_0, end = var_18938_end_0, end_mask = var_18938_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18938_cast_fp16")]; + tensor var_18942_begin_0 = const()[name = tensor("op_18942_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18942_end_0 = const()[name = tensor("op_18942_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_18942_end_mask_0 = const()[name = tensor("op_18942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18942_cast_fp16 = slice_by_index(begin = var_18942_begin_0, end = var_18942_end_0, end_mask = var_18942_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18942_cast_fp16")]; + tensor var_18946_begin_0 = const()[name = tensor("op_18946_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18946_end_0 = const()[name = tensor("op_18946_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_18946_end_mask_0 = const()[name = tensor("op_18946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18946_cast_fp16 = slice_by_index(begin = var_18946_begin_0, end = var_18946_end_0, end_mask = var_18946_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18946_cast_fp16")]; + tensor var_18950_begin_0 = const()[name = tensor("op_18950_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18950_end_0 = const()[name = tensor("op_18950_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_18950_end_mask_0 = const()[name = tensor("op_18950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18950_cast_fp16 = slice_by_index(begin = var_18950_begin_0, end = var_18950_end_0, end_mask = var_18950_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18950_cast_fp16")]; + tensor var_18954_begin_0 = const()[name = tensor("op_18954_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18954_end_0 = const()[name = tensor("op_18954_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_18954_end_mask_0 = const()[name = tensor("op_18954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18954_cast_fp16 = slice_by_index(begin = var_18954_begin_0, end = var_18954_end_0, end_mask = var_18954_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18954_cast_fp16")]; + tensor var_18958_begin_0 = const()[name = tensor("op_18958_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18958_end_0 = const()[name = tensor("op_18958_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_18958_end_mask_0 = const()[name = tensor("op_18958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18958_cast_fp16 = slice_by_index(begin = var_18958_begin_0, end = var_18958_end_0, end_mask = var_18958_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18958_cast_fp16")]; + tensor var_18962_begin_0 = const()[name = tensor("op_18962_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18962_end_0 = const()[name = tensor("op_18962_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_18962_end_mask_0 = const()[name = tensor("op_18962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18962_cast_fp16 = slice_by_index(begin = var_18962_begin_0, end = var_18962_end_0, end_mask = var_18962_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18962_cast_fp16")]; + tensor var_18966_begin_0 = const()[name = tensor("op_18966_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18966_end_0 = const()[name = tensor("op_18966_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_18966_end_mask_0 = const()[name = tensor("op_18966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18966_cast_fp16 = slice_by_index(begin = var_18966_begin_0, end = var_18966_end_0, end_mask = var_18966_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18966_cast_fp16")]; + tensor var_18970_begin_0 = const()[name = tensor("op_18970_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18970_end_0 = const()[name = tensor("op_18970_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_18970_end_mask_0 = const()[name = tensor("op_18970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18970_cast_fp16 = slice_by_index(begin = var_18970_begin_0, end = var_18970_end_0, end_mask = var_18970_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18970_cast_fp16")]; + tensor var_18974_begin_0 = const()[name = tensor("op_18974_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18974_end_0 = const()[name = tensor("op_18974_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_18974_end_mask_0 = const()[name = tensor("op_18974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18974_cast_fp16 = slice_by_index(begin = var_18974_begin_0, end = var_18974_end_0, end_mask = var_18974_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18974_cast_fp16")]; + tensor var_18978_begin_0 = const()[name = tensor("op_18978_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18978_end_0 = const()[name = tensor("op_18978_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_18978_end_mask_0 = const()[name = tensor("op_18978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18978_cast_fp16 = slice_by_index(begin = var_18978_begin_0, end = var_18978_end_0, end_mask = var_18978_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18978_cast_fp16")]; + tensor var_18982_begin_0 = const()[name = tensor("op_18982_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18982_end_0 = const()[name = tensor("op_18982_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_18982_end_mask_0 = const()[name = tensor("op_18982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18982_cast_fp16 = slice_by_index(begin = var_18982_begin_0, end = var_18982_end_0, end_mask = var_18982_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18982_cast_fp16")]; + tensor var_18986_begin_0 = const()[name = tensor("op_18986_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18986_end_0 = const()[name = tensor("op_18986_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_18986_end_mask_0 = const()[name = tensor("op_18986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18986_cast_fp16 = slice_by_index(begin = var_18986_begin_0, end = var_18986_end_0, end_mask = var_18986_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18986_cast_fp16")]; + tensor var_18990_begin_0 = const()[name = tensor("op_18990_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18990_end_0 = const()[name = tensor("op_18990_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_18990_end_mask_0 = const()[name = tensor("op_18990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18990_cast_fp16 = slice_by_index(begin = var_18990_begin_0, end = var_18990_end_0, end_mask = var_18990_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18990_cast_fp16")]; + tensor var_18994_begin_0 = const()[name = tensor("op_18994_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18994_end_0 = const()[name = tensor("op_18994_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_18994_end_mask_0 = const()[name = tensor("op_18994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18994_cast_fp16 = slice_by_index(begin = var_18994_begin_0, end = var_18994_end_0, end_mask = var_18994_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18994_cast_fp16")]; + tensor var_18998_begin_0 = const()[name = tensor("op_18998_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18998_end_0 = const()[name = tensor("op_18998_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_18998_end_mask_0 = const()[name = tensor("op_18998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18998_cast_fp16 = slice_by_index(begin = var_18998_begin_0, end = var_18998_end_0, end_mask = var_18998_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18998_cast_fp16")]; + tensor var_19002_begin_0 = const()[name = tensor("op_19002_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19002_end_0 = const()[name = tensor("op_19002_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_19002_end_mask_0 = const()[name = tensor("op_19002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19002_cast_fp16 = slice_by_index(begin = var_19002_begin_0, end = var_19002_end_0, end_mask = var_19002_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19002_cast_fp16")]; + tensor var_19006_begin_0 = const()[name = tensor("op_19006_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19006_end_0 = const()[name = tensor("op_19006_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_19006_end_mask_0 = const()[name = tensor("op_19006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19006_cast_fp16 = slice_by_index(begin = var_19006_begin_0, end = var_19006_end_0, end_mask = var_19006_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19006_cast_fp16")]; + tensor var_19010_begin_0 = const()[name = tensor("op_19010_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19010_end_0 = const()[name = tensor("op_19010_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_19010_end_mask_0 = const()[name = tensor("op_19010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19010_cast_fp16 = slice_by_index(begin = var_19010_begin_0, end = var_19010_end_0, end_mask = var_19010_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19010_cast_fp16")]; + tensor var_19014_begin_0 = const()[name = tensor("op_19014_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19014_end_0 = const()[name = tensor("op_19014_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_19014_end_mask_0 = const()[name = tensor("op_19014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19014_cast_fp16 = slice_by_index(begin = var_19014_begin_0, end = var_19014_end_0, end_mask = var_19014_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19014_cast_fp16")]; + tensor var_19018_equation_0 = const()[name = tensor("op_19018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19018_cast_fp16 = einsum(equation = var_19018_equation_0, values = (var_18860_cast_fp16, var_18777_cast_fp16))[name = tensor("op_19018_cast_fp16")]; + tensor var_19019_to_fp16 = const()[name = tensor("op_19019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1561_cast_fp16 = mul(x = var_19018_cast_fp16, y = var_19019_to_fp16)[name = tensor("aw_1561_cast_fp16")]; + tensor var_19022_equation_0 = const()[name = tensor("op_19022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19022_cast_fp16 = einsum(equation = var_19022_equation_0, values = (var_18864_cast_fp16, var_18781_cast_fp16))[name = tensor("op_19022_cast_fp16")]; + tensor var_19023_to_fp16 = const()[name = tensor("op_19023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1563_cast_fp16 = mul(x = var_19022_cast_fp16, y = var_19023_to_fp16)[name = tensor("aw_1563_cast_fp16")]; + tensor var_19026_equation_0 = const()[name = tensor("op_19026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19026_cast_fp16 = einsum(equation = var_19026_equation_0, values = (var_18868_cast_fp16, var_18785_cast_fp16))[name = tensor("op_19026_cast_fp16")]; + tensor var_19027_to_fp16 = const()[name = tensor("op_19027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1565_cast_fp16 = mul(x = var_19026_cast_fp16, y = var_19027_to_fp16)[name = tensor("aw_1565_cast_fp16")]; + tensor var_19030_equation_0 = const()[name = tensor("op_19030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19030_cast_fp16 = einsum(equation = var_19030_equation_0, values = (var_18872_cast_fp16, var_18789_cast_fp16))[name = tensor("op_19030_cast_fp16")]; + tensor var_19031_to_fp16 = const()[name = tensor("op_19031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1567_cast_fp16 = mul(x = var_19030_cast_fp16, y = var_19031_to_fp16)[name = tensor("aw_1567_cast_fp16")]; + tensor var_19034_equation_0 = const()[name = tensor("op_19034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19034_cast_fp16 = einsum(equation = var_19034_equation_0, values = (var_18876_cast_fp16, var_18793_cast_fp16))[name = tensor("op_19034_cast_fp16")]; + tensor var_19035_to_fp16 = const()[name = tensor("op_19035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1569_cast_fp16 = mul(x = var_19034_cast_fp16, y = var_19035_to_fp16)[name = tensor("aw_1569_cast_fp16")]; + tensor var_19038_equation_0 = const()[name = tensor("op_19038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19038_cast_fp16 = einsum(equation = var_19038_equation_0, values = (var_18880_cast_fp16, var_18797_cast_fp16))[name = tensor("op_19038_cast_fp16")]; + tensor var_19039_to_fp16 = const()[name = tensor("op_19039_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1571_cast_fp16 = mul(x = var_19038_cast_fp16, y = var_19039_to_fp16)[name = tensor("aw_1571_cast_fp16")]; + tensor var_19042_equation_0 = const()[name = tensor("op_19042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19042_cast_fp16 = einsum(equation = var_19042_equation_0, values = (var_18884_cast_fp16, var_18801_cast_fp16))[name = tensor("op_19042_cast_fp16")]; + tensor var_19043_to_fp16 = const()[name = tensor("op_19043_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1573_cast_fp16 = mul(x = var_19042_cast_fp16, y = var_19043_to_fp16)[name = tensor("aw_1573_cast_fp16")]; + tensor var_19046_equation_0 = const()[name = tensor("op_19046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19046_cast_fp16 = einsum(equation = var_19046_equation_0, values = (var_18888_cast_fp16, var_18805_cast_fp16))[name = tensor("op_19046_cast_fp16")]; + tensor var_19047_to_fp16 = const()[name = tensor("op_19047_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1575_cast_fp16 = mul(x = var_19046_cast_fp16, y = var_19047_to_fp16)[name = tensor("aw_1575_cast_fp16")]; + tensor var_19050_equation_0 = const()[name = tensor("op_19050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19050_cast_fp16 = einsum(equation = var_19050_equation_0, values = (var_18892_cast_fp16, var_18809_cast_fp16))[name = tensor("op_19050_cast_fp16")]; + tensor var_19051_to_fp16 = const()[name = tensor("op_19051_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1577_cast_fp16 = mul(x = var_19050_cast_fp16, y = var_19051_to_fp16)[name = tensor("aw_1577_cast_fp16")]; + tensor var_19054_equation_0 = const()[name = tensor("op_19054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19054_cast_fp16 = einsum(equation = var_19054_equation_0, values = (var_18896_cast_fp16, var_18813_cast_fp16))[name = tensor("op_19054_cast_fp16")]; + tensor var_19055_to_fp16 = const()[name = tensor("op_19055_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1579_cast_fp16 = mul(x = var_19054_cast_fp16, y = var_19055_to_fp16)[name = tensor("aw_1579_cast_fp16")]; + tensor var_19058_equation_0 = const()[name = tensor("op_19058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19058_cast_fp16 = einsum(equation = var_19058_equation_0, values = (var_18900_cast_fp16, var_18817_cast_fp16))[name = tensor("op_19058_cast_fp16")]; + tensor var_19059_to_fp16 = const()[name = tensor("op_19059_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1581_cast_fp16 = mul(x = var_19058_cast_fp16, y = var_19059_to_fp16)[name = tensor("aw_1581_cast_fp16")]; + tensor var_19062_equation_0 = const()[name = tensor("op_19062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19062_cast_fp16 = einsum(equation = var_19062_equation_0, values = (var_18904_cast_fp16, var_18821_cast_fp16))[name = tensor("op_19062_cast_fp16")]; + tensor var_19063_to_fp16 = const()[name = tensor("op_19063_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1583_cast_fp16 = mul(x = var_19062_cast_fp16, y = var_19063_to_fp16)[name = tensor("aw_1583_cast_fp16")]; + tensor var_19066_equation_0 = const()[name = tensor("op_19066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19066_cast_fp16 = einsum(equation = var_19066_equation_0, values = (var_18908_cast_fp16, var_18825_cast_fp16))[name = tensor("op_19066_cast_fp16")]; + tensor var_19067_to_fp16 = const()[name = tensor("op_19067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1585_cast_fp16 = mul(x = var_19066_cast_fp16, y = var_19067_to_fp16)[name = tensor("aw_1585_cast_fp16")]; + tensor var_19070_equation_0 = const()[name = tensor("op_19070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19070_cast_fp16 = einsum(equation = var_19070_equation_0, values = (var_18912_cast_fp16, var_18829_cast_fp16))[name = tensor("op_19070_cast_fp16")]; + tensor var_19071_to_fp16 = const()[name = tensor("op_19071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1587_cast_fp16 = mul(x = var_19070_cast_fp16, y = var_19071_to_fp16)[name = tensor("aw_1587_cast_fp16")]; + tensor var_19074_equation_0 = const()[name = tensor("op_19074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19074_cast_fp16 = einsum(equation = var_19074_equation_0, values = (var_18916_cast_fp16, var_18833_cast_fp16))[name = tensor("op_19074_cast_fp16")]; + tensor var_19075_to_fp16 = const()[name = tensor("op_19075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1589_cast_fp16 = mul(x = var_19074_cast_fp16, y = var_19075_to_fp16)[name = tensor("aw_1589_cast_fp16")]; + tensor var_19078_equation_0 = const()[name = tensor("op_19078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19078_cast_fp16 = einsum(equation = var_19078_equation_0, values = (var_18920_cast_fp16, var_18837_cast_fp16))[name = tensor("op_19078_cast_fp16")]; + tensor var_19079_to_fp16 = const()[name = tensor("op_19079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1591_cast_fp16 = mul(x = var_19078_cast_fp16, y = var_19079_to_fp16)[name = tensor("aw_1591_cast_fp16")]; + tensor var_19082_equation_0 = const()[name = tensor("op_19082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19082_cast_fp16 = einsum(equation = var_19082_equation_0, values = (var_18924_cast_fp16, var_18841_cast_fp16))[name = tensor("op_19082_cast_fp16")]; + tensor var_19083_to_fp16 = const()[name = tensor("op_19083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1593_cast_fp16 = mul(x = var_19082_cast_fp16, y = var_19083_to_fp16)[name = tensor("aw_1593_cast_fp16")]; + tensor var_19086_equation_0 = const()[name = tensor("op_19086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19086_cast_fp16 = einsum(equation = var_19086_equation_0, values = (var_18928_cast_fp16, var_18845_cast_fp16))[name = tensor("op_19086_cast_fp16")]; + tensor var_19087_to_fp16 = const()[name = tensor("op_19087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1595_cast_fp16 = mul(x = var_19086_cast_fp16, y = var_19087_to_fp16)[name = tensor("aw_1595_cast_fp16")]; + tensor var_19090_equation_0 = const()[name = tensor("op_19090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19090_cast_fp16 = einsum(equation = var_19090_equation_0, values = (var_18932_cast_fp16, var_18849_cast_fp16))[name = tensor("op_19090_cast_fp16")]; + tensor var_19091_to_fp16 = const()[name = tensor("op_19091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1597_cast_fp16 = mul(x = var_19090_cast_fp16, y = var_19091_to_fp16)[name = tensor("aw_1597_cast_fp16")]; + tensor var_19094_equation_0 = const()[name = tensor("op_19094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19094_cast_fp16 = einsum(equation = var_19094_equation_0, values = (var_18936_cast_fp16, var_18853_cast_fp16))[name = tensor("op_19094_cast_fp16")]; + tensor var_19095_to_fp16 = const()[name = tensor("op_19095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1599_cast_fp16 = mul(x = var_19094_cast_fp16, y = var_19095_to_fp16)[name = tensor("aw_1599_cast_fp16")]; + tensor var_19097_cast_fp16 = softmax(axis = var_2624, x = aw_1561_cast_fp16)[name = tensor("op_19097_cast_fp16")]; + tensor var_19098_cast_fp16 = softmax(axis = var_2624, x = aw_1563_cast_fp16)[name = tensor("op_19098_cast_fp16")]; + tensor var_19099_cast_fp16 = softmax(axis = var_2624, x = aw_1565_cast_fp16)[name = tensor("op_19099_cast_fp16")]; + tensor var_19100_cast_fp16 = softmax(axis = var_2624, x = aw_1567_cast_fp16)[name = tensor("op_19100_cast_fp16")]; + tensor var_19101_cast_fp16 = softmax(axis = var_2624, x = aw_1569_cast_fp16)[name = tensor("op_19101_cast_fp16")]; + tensor var_19102_cast_fp16 = softmax(axis = var_2624, x = aw_1571_cast_fp16)[name = tensor("op_19102_cast_fp16")]; + tensor var_19103_cast_fp16 = softmax(axis = var_2624, x = aw_1573_cast_fp16)[name = tensor("op_19103_cast_fp16")]; + tensor var_19104_cast_fp16 = softmax(axis = var_2624, x = aw_1575_cast_fp16)[name = tensor("op_19104_cast_fp16")]; + tensor var_19105_cast_fp16 = softmax(axis = var_2624, x = aw_1577_cast_fp16)[name = tensor("op_19105_cast_fp16")]; + tensor var_19106_cast_fp16 = softmax(axis = var_2624, x = aw_1579_cast_fp16)[name = tensor("op_19106_cast_fp16")]; + tensor var_19107_cast_fp16 = softmax(axis = var_2624, x = aw_1581_cast_fp16)[name = tensor("op_19107_cast_fp16")]; + tensor var_19108_cast_fp16 = softmax(axis = var_2624, x = aw_1583_cast_fp16)[name = tensor("op_19108_cast_fp16")]; + tensor var_19109_cast_fp16 = softmax(axis = var_2624, x = aw_1585_cast_fp16)[name = tensor("op_19109_cast_fp16")]; + tensor var_19110_cast_fp16 = softmax(axis = var_2624, x = aw_1587_cast_fp16)[name = tensor("op_19110_cast_fp16")]; + tensor var_19111_cast_fp16 = softmax(axis = var_2624, x = aw_1589_cast_fp16)[name = tensor("op_19111_cast_fp16")]; + tensor var_19112_cast_fp16 = softmax(axis = var_2624, x = aw_1591_cast_fp16)[name = tensor("op_19112_cast_fp16")]; + tensor var_19113_cast_fp16 = softmax(axis = var_2624, x = aw_1593_cast_fp16)[name = tensor("op_19113_cast_fp16")]; + tensor var_19114_cast_fp16 = softmax(axis = var_2624, x = aw_1595_cast_fp16)[name = tensor("op_19114_cast_fp16")]; + tensor var_19115_cast_fp16 = softmax(axis = var_2624, x = aw_1597_cast_fp16)[name = tensor("op_19115_cast_fp16")]; + tensor var_19116_cast_fp16 = softmax(axis = var_2624, x = aw_1599_cast_fp16)[name = tensor("op_19116_cast_fp16")]; + tensor var_19118_equation_0 = const()[name = tensor("op_19118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19118_cast_fp16 = einsum(equation = var_19118_equation_0, values = (var_18938_cast_fp16, var_19097_cast_fp16))[name = tensor("op_19118_cast_fp16")]; + tensor var_19120_equation_0 = const()[name = tensor("op_19120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19120_cast_fp16 = einsum(equation = var_19120_equation_0, values = (var_18942_cast_fp16, var_19098_cast_fp16))[name = tensor("op_19120_cast_fp16")]; + tensor var_19122_equation_0 = const()[name = tensor("op_19122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19122_cast_fp16 = einsum(equation = var_19122_equation_0, values = (var_18946_cast_fp16, var_19099_cast_fp16))[name = tensor("op_19122_cast_fp16")]; + tensor var_19124_equation_0 = const()[name = tensor("op_19124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19124_cast_fp16 = einsum(equation = var_19124_equation_0, values = (var_18950_cast_fp16, var_19100_cast_fp16))[name = tensor("op_19124_cast_fp16")]; + tensor var_19126_equation_0 = const()[name = tensor("op_19126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19126_cast_fp16 = einsum(equation = var_19126_equation_0, values = (var_18954_cast_fp16, var_19101_cast_fp16))[name = tensor("op_19126_cast_fp16")]; + tensor var_19128_equation_0 = const()[name = tensor("op_19128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19128_cast_fp16 = einsum(equation = var_19128_equation_0, values = (var_18958_cast_fp16, var_19102_cast_fp16))[name = tensor("op_19128_cast_fp16")]; + tensor var_19130_equation_0 = const()[name = tensor("op_19130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19130_cast_fp16 = einsum(equation = var_19130_equation_0, values = (var_18962_cast_fp16, var_19103_cast_fp16))[name = tensor("op_19130_cast_fp16")]; + tensor var_19132_equation_0 = const()[name = tensor("op_19132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19132_cast_fp16 = einsum(equation = var_19132_equation_0, values = (var_18966_cast_fp16, var_19104_cast_fp16))[name = tensor("op_19132_cast_fp16")]; + tensor var_19134_equation_0 = const()[name = tensor("op_19134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19134_cast_fp16 = einsum(equation = var_19134_equation_0, values = (var_18970_cast_fp16, var_19105_cast_fp16))[name = tensor("op_19134_cast_fp16")]; + tensor var_19136_equation_0 = const()[name = tensor("op_19136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19136_cast_fp16 = einsum(equation = var_19136_equation_0, values = (var_18974_cast_fp16, var_19106_cast_fp16))[name = tensor("op_19136_cast_fp16")]; + tensor var_19138_equation_0 = const()[name = tensor("op_19138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19138_cast_fp16 = einsum(equation = var_19138_equation_0, values = (var_18978_cast_fp16, var_19107_cast_fp16))[name = tensor("op_19138_cast_fp16")]; + tensor var_19140_equation_0 = const()[name = tensor("op_19140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19140_cast_fp16 = einsum(equation = var_19140_equation_0, values = (var_18982_cast_fp16, var_19108_cast_fp16))[name = tensor("op_19140_cast_fp16")]; + tensor var_19142_equation_0 = const()[name = tensor("op_19142_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19142_cast_fp16 = einsum(equation = var_19142_equation_0, values = (var_18986_cast_fp16, var_19109_cast_fp16))[name = tensor("op_19142_cast_fp16")]; + tensor var_19144_equation_0 = const()[name = tensor("op_19144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19144_cast_fp16 = einsum(equation = var_19144_equation_0, values = (var_18990_cast_fp16, var_19110_cast_fp16))[name = tensor("op_19144_cast_fp16")]; + tensor var_19146_equation_0 = const()[name = tensor("op_19146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19146_cast_fp16 = einsum(equation = var_19146_equation_0, values = (var_18994_cast_fp16, var_19111_cast_fp16))[name = tensor("op_19146_cast_fp16")]; + tensor var_19148_equation_0 = const()[name = tensor("op_19148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19148_cast_fp16 = einsum(equation = var_19148_equation_0, values = (var_18998_cast_fp16, var_19112_cast_fp16))[name = tensor("op_19148_cast_fp16")]; + tensor var_19150_equation_0 = const()[name = tensor("op_19150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19150_cast_fp16 = einsum(equation = var_19150_equation_0, values = (var_19002_cast_fp16, var_19113_cast_fp16))[name = tensor("op_19150_cast_fp16")]; + tensor var_19152_equation_0 = const()[name = tensor("op_19152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19152_cast_fp16 = einsum(equation = var_19152_equation_0, values = (var_19006_cast_fp16, var_19114_cast_fp16))[name = tensor("op_19152_cast_fp16")]; + tensor var_19154_equation_0 = const()[name = tensor("op_19154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19154_cast_fp16 = einsum(equation = var_19154_equation_0, values = (var_19010_cast_fp16, var_19115_cast_fp16))[name = tensor("op_19154_cast_fp16")]; + tensor var_19156_equation_0 = const()[name = tensor("op_19156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19156_cast_fp16 = einsum(equation = var_19156_equation_0, values = (var_19014_cast_fp16, var_19116_cast_fp16))[name = tensor("op_19156_cast_fp16")]; + tensor input_287_interleave_0 = const()[name = tensor("input_287_interleave_0"), val = tensor(false)]; + tensor input_287_cast_fp16 = concat(axis = var_2624, interleave = input_287_interleave_0, values = (var_19118_cast_fp16, var_19120_cast_fp16, var_19122_cast_fp16, var_19124_cast_fp16, var_19126_cast_fp16, var_19128_cast_fp16, var_19130_cast_fp16, var_19132_cast_fp16, var_19134_cast_fp16, var_19136_cast_fp16, var_19138_cast_fp16, var_19140_cast_fp16, var_19142_cast_fp16, var_19144_cast_fp16, var_19146_cast_fp16, var_19148_cast_fp16, var_19150_cast_fp16, var_19152_cast_fp16, var_19154_cast_fp16, var_19156_cast_fp16))[name = tensor("input_287_cast_fp16")]; + tensor var_19166_pad_type_0 = const()[name = tensor("op_19166_pad_type_0"), val = tensor("valid")]; + tensor var_19166_strides_0 = const()[name = tensor("op_19166_strides_0"), val = tensor([1, 1])]; + tensor var_19166_pad_0 = const()[name = tensor("op_19166_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19166_dilations_0 = const()[name = tensor("op_19166_dilations_0"), val = tensor([1, 1])]; + tensor var_19166_groups_0 = const()[name = tensor("op_19166_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554019712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555248576))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555248768)))]; + tensor var_19166_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_19166_dilations_0, groups = var_19166_groups_0, pad = var_19166_pad_0, pad_type = var_19166_pad_type_0, strides = var_19166_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized, x = input_287_cast_fp16)[name = tensor("op_19166_cast_fp16")]; + tensor inputs_131_cast_fp16 = add(x = var_19166_cast_fp16, y = inputs_129_cast_fp16)[name = tensor("inputs_131_cast_fp16")]; + tensor input_289_axes_0 = const()[name = tensor("input_289_axes_0"), val = tensor([1])]; + tensor input_289_gamma_0_to_fp16 = const()[name = tensor("input_289_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555251392)))]; + tensor input_289_beta_0_to_fp16 = const()[name = tensor("input_289_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555254016)))]; + tensor var_19176_to_fp16 = const()[name = tensor("op_19176_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_289_cast_fp16 = layer_norm(axes = input_289_axes_0, beta = input_289_beta_0_to_fp16, epsilon = var_19176_to_fp16, gamma = input_289_gamma_0_to_fp16, x = inputs_131_cast_fp16)[name = tensor("input_289_cast_fp16")]; + tensor var_19196_pad_type_0 = const()[name = tensor("op_19196_pad_type_0"), val = tensor("valid")]; + tensor var_19196_strides_0 = const()[name = tensor("op_19196_strides_0"), val = tensor([1, 1])]; + tensor var_19196_pad_0 = const()[name = tensor("op_19196_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19196_dilations_0 = const()[name = tensor("op_19196_dilations_0"), val = tensor([1, 1])]; + tensor var_19196_groups_0 = const()[name = tensor("op_19196_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555256640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565087104))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565087296)))]; + tensor var_19196_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_19196_dilations_0, groups = var_19196_groups_0, pad = var_19196_pad_0, pad_type = var_19196_pad_type_0, strides = var_19196_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized, x = input_289_cast_fp16)[name = tensor("op_19196_cast_fp16")]; + tensor var_19197_split_sizes_0 = const()[name = tensor("op_19197_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_19197_axis_0 = const()[name = tensor("op_19197_axis_0"), val = tensor(1)]; + tensor var_19197_cast_fp16_0, tensor var_19197_cast_fp16_1 = split(axis = var_19197_axis_0, split_sizes = var_19197_split_sizes_0, x = var_19196_cast_fp16)[name = tensor("op_19197_cast_fp16")]; + tensor var_19199_mode_0 = const()[name = tensor("op_19199_mode_0"), val = tensor("EXACT")]; + tensor var_19199_cast_fp16 = gelu(mode = var_19199_mode_0, x = var_19197_cast_fp16_1)[name = tensor("op_19199_cast_fp16")]; + tensor input_291_cast_fp16 = mul(x = var_19197_cast_fp16_0, y = var_19199_cast_fp16)[name = tensor("input_291_cast_fp16")]; + tensor var_19207_pad_type_0 = const()[name = tensor("op_19207_pad_type_0"), val = tensor("valid")]; + tensor var_19207_strides_0 = const()[name = tensor("op_19207_strides_0"), val = tensor([1, 1])]; + tensor var_19207_pad_0 = const()[name = tensor("op_19207_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19207_dilations_0 = const()[name = tensor("op_19207_dilations_0"), val = tensor([1, 1])]; + tensor var_19207_groups_0 = const()[name = tensor("op_19207_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565107840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570023104))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570023296)))]; + tensor var_19207_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_19207_dilations_0, groups = var_19207_groups_0, pad = var_19207_pad_0, pad_type = var_19207_pad_type_0, strides = var_19207_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized, x = input_291_cast_fp16)[name = tensor("op_19207_cast_fp16")]; + tensor inputs_133_cast_fp16 = add(x = var_19207_cast_fp16, y = inputs_131_cast_fp16)[name = tensor("inputs_133_cast_fp16")]; + tensor hidden_states_185_axes_0 = const()[name = tensor("hidden_states_185_axes_0"), val = tensor([1])]; + tensor hidden_states_185_gamma_0_to_fp16 = const()[name = tensor("hidden_states_185_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570025920)))]; + tensor hidden_states_185_beta_0_to_fp16 = const()[name = tensor("hidden_states_185_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570028544)))]; + tensor var_19223_to_fp16 = const()[name = tensor("op_19223_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_185_cast_fp16 = layer_norm(axes = hidden_states_185_axes_0, beta = hidden_states_185_beta_0_to_fp16, epsilon = var_19223_to_fp16, gamma = hidden_states_185_gamma_0_to_fp16, x = inputs_133_cast_fp16)[name = tensor("hidden_states_185_cast_fp16")]; + tensor q_89_pad_type_0 = const()[name = tensor("q_89_pad_type_0"), val = tensor("valid")]; + tensor q_89_strides_0 = const()[name = tensor("q_89_strides_0"), val = tensor([1, 1])]; + tensor q_89_pad_0 = const()[name = tensor("q_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_89_dilations_0 = const()[name = tensor("q_89_dilations_0"), val = tensor([1, 1])]; + tensor q_89_groups_0 = const()[name = tensor("q_89_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570031168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571260032))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_89_cast_fp16 = conv(dilations = q_89_dilations_0, groups = q_89_groups_0, pad = q_89_pad_0, pad_type = q_89_pad_type_0, strides = q_89_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_185_cast_fp16)[name = tensor("q_89_cast_fp16")]; + tensor k_177_pad_type_0 = const()[name = tensor("k_177_pad_type_0"), val = tensor("valid")]; + tensor k_177_strides_0 = const()[name = tensor("k_177_strides_0"), val = tensor([1, 1])]; + tensor k_177_pad_0 = const()[name = tensor("k_177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_177_dilations_0 = const()[name = tensor("k_177_dilations_0"), val = tensor([1, 1])]; + tensor k_177_groups_0 = const()[name = tensor("k_177_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571260224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572489088))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_177_cast_fp16 = conv(dilations = k_177_dilations_0, groups = k_177_groups_0, pad = k_177_pad_0, pad_type = k_177_pad_type_0, strides = k_177_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_185_cast_fp16)[name = tensor("k_177_cast_fp16")]; + tensor v_89_pad_type_0 = const()[name = tensor("v_89_pad_type_0"), val = tensor("valid")]; + tensor v_89_strides_0 = const()[name = tensor("v_89_strides_0"), val = tensor([1, 1])]; + tensor v_89_pad_0 = const()[name = tensor("v_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_89_dilations_0 = const()[name = tensor("v_89_dilations_0"), val = tensor([1, 1])]; + tensor v_89_groups_0 = const()[name = tensor("v_89_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572489280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(573718144))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_89_cast_fp16 = conv(dilations = v_89_dilations_0, groups = v_89_groups_0, pad = v_89_pad_0, pad_type = v_89_pad_type_0, strides = v_89_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_185_cast_fp16)[name = tensor("v_89_cast_fp16")]; + tensor var_19256_begin_0 = const()[name = tensor("op_19256_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19256_end_0 = const()[name = tensor("op_19256_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_19256_end_mask_0 = const()[name = tensor("op_19256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19256_cast_fp16 = slice_by_index(begin = var_19256_begin_0, end = var_19256_end_0, end_mask = var_19256_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19256_cast_fp16")]; + tensor var_19260_begin_0 = const()[name = tensor("op_19260_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19260_end_0 = const()[name = tensor("op_19260_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_19260_end_mask_0 = const()[name = tensor("op_19260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19260_cast_fp16 = slice_by_index(begin = var_19260_begin_0, end = var_19260_end_0, end_mask = var_19260_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19260_cast_fp16")]; + tensor var_19264_begin_0 = const()[name = tensor("op_19264_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19264_end_0 = const()[name = tensor("op_19264_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_19264_end_mask_0 = const()[name = tensor("op_19264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19264_cast_fp16 = slice_by_index(begin = var_19264_begin_0, end = var_19264_end_0, end_mask = var_19264_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19264_cast_fp16")]; + tensor var_19268_begin_0 = const()[name = tensor("op_19268_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19268_end_0 = const()[name = tensor("op_19268_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_19268_end_mask_0 = const()[name = tensor("op_19268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19268_cast_fp16 = slice_by_index(begin = var_19268_begin_0, end = var_19268_end_0, end_mask = var_19268_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19268_cast_fp16")]; + tensor var_19272_begin_0 = const()[name = tensor("op_19272_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19272_end_0 = const()[name = tensor("op_19272_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_19272_end_mask_0 = const()[name = tensor("op_19272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19272_cast_fp16 = slice_by_index(begin = var_19272_begin_0, end = var_19272_end_0, end_mask = var_19272_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19272_cast_fp16")]; + tensor var_19276_begin_0 = const()[name = tensor("op_19276_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19276_end_0 = const()[name = tensor("op_19276_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_19276_end_mask_0 = const()[name = tensor("op_19276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19276_cast_fp16 = slice_by_index(begin = var_19276_begin_0, end = var_19276_end_0, end_mask = var_19276_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19276_cast_fp16")]; + tensor var_19280_begin_0 = const()[name = tensor("op_19280_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19280_end_0 = const()[name = tensor("op_19280_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_19280_end_mask_0 = const()[name = tensor("op_19280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19280_cast_fp16 = slice_by_index(begin = var_19280_begin_0, end = var_19280_end_0, end_mask = var_19280_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19280_cast_fp16")]; + tensor var_19284_begin_0 = const()[name = tensor("op_19284_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19284_end_0 = const()[name = tensor("op_19284_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_19284_end_mask_0 = const()[name = tensor("op_19284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19284_cast_fp16 = slice_by_index(begin = var_19284_begin_0, end = var_19284_end_0, end_mask = var_19284_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19284_cast_fp16")]; + tensor var_19288_begin_0 = const()[name = tensor("op_19288_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19288_end_0 = const()[name = tensor("op_19288_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_19288_end_mask_0 = const()[name = tensor("op_19288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19288_cast_fp16 = slice_by_index(begin = var_19288_begin_0, end = var_19288_end_0, end_mask = var_19288_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19288_cast_fp16")]; + tensor var_19292_begin_0 = const()[name = tensor("op_19292_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19292_end_0 = const()[name = tensor("op_19292_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_19292_end_mask_0 = const()[name = tensor("op_19292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19292_cast_fp16 = slice_by_index(begin = var_19292_begin_0, end = var_19292_end_0, end_mask = var_19292_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19292_cast_fp16")]; + tensor var_19296_begin_0 = const()[name = tensor("op_19296_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19296_end_0 = const()[name = tensor("op_19296_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_19296_end_mask_0 = const()[name = tensor("op_19296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19296_cast_fp16 = slice_by_index(begin = var_19296_begin_0, end = var_19296_end_0, end_mask = var_19296_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19296_cast_fp16")]; + tensor var_19300_begin_0 = const()[name = tensor("op_19300_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19300_end_0 = const()[name = tensor("op_19300_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_19300_end_mask_0 = const()[name = tensor("op_19300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19300_cast_fp16 = slice_by_index(begin = var_19300_begin_0, end = var_19300_end_0, end_mask = var_19300_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19300_cast_fp16")]; + tensor var_19304_begin_0 = const()[name = tensor("op_19304_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19304_end_0 = const()[name = tensor("op_19304_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_19304_end_mask_0 = const()[name = tensor("op_19304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19304_cast_fp16 = slice_by_index(begin = var_19304_begin_0, end = var_19304_end_0, end_mask = var_19304_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19304_cast_fp16")]; + tensor var_19308_begin_0 = const()[name = tensor("op_19308_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19308_end_0 = const()[name = tensor("op_19308_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_19308_end_mask_0 = const()[name = tensor("op_19308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19308_cast_fp16 = slice_by_index(begin = var_19308_begin_0, end = var_19308_end_0, end_mask = var_19308_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19308_cast_fp16")]; + tensor var_19312_begin_0 = const()[name = tensor("op_19312_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19312_end_0 = const()[name = tensor("op_19312_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_19312_end_mask_0 = const()[name = tensor("op_19312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19312_cast_fp16 = slice_by_index(begin = var_19312_begin_0, end = var_19312_end_0, end_mask = var_19312_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19312_cast_fp16")]; + tensor var_19316_begin_0 = const()[name = tensor("op_19316_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19316_end_0 = const()[name = tensor("op_19316_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19316_end_mask_0 = const()[name = tensor("op_19316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19316_cast_fp16 = slice_by_index(begin = var_19316_begin_0, end = var_19316_end_0, end_mask = var_19316_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19316_cast_fp16")]; + tensor var_19320_begin_0 = const()[name = tensor("op_19320_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19320_end_0 = const()[name = tensor("op_19320_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_19320_end_mask_0 = const()[name = tensor("op_19320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19320_cast_fp16 = slice_by_index(begin = var_19320_begin_0, end = var_19320_end_0, end_mask = var_19320_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19320_cast_fp16")]; + tensor var_19324_begin_0 = const()[name = tensor("op_19324_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19324_end_0 = const()[name = tensor("op_19324_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_19324_end_mask_0 = const()[name = tensor("op_19324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19324_cast_fp16 = slice_by_index(begin = var_19324_begin_0, end = var_19324_end_0, end_mask = var_19324_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19324_cast_fp16")]; + tensor var_19328_begin_0 = const()[name = tensor("op_19328_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19328_end_0 = const()[name = tensor("op_19328_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_19328_end_mask_0 = const()[name = tensor("op_19328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19328_cast_fp16 = slice_by_index(begin = var_19328_begin_0, end = var_19328_end_0, end_mask = var_19328_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19328_cast_fp16")]; + tensor var_19332_begin_0 = const()[name = tensor("op_19332_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19332_end_0 = const()[name = tensor("op_19332_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_19332_end_mask_0 = const()[name = tensor("op_19332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19332_cast_fp16 = slice_by_index(begin = var_19332_begin_0, end = var_19332_end_0, end_mask = var_19332_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19332_cast_fp16")]; + tensor k_179_perm_0 = const()[name = tensor("k_179_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_19339_begin_0 = const()[name = tensor("op_19339_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19339_end_0 = const()[name = tensor("op_19339_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_19339_end_mask_0 = const()[name = tensor("op_19339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_179_cast_fp16 = transpose(perm = k_179_perm_0, x = k_177_cast_fp16)[name = tensor("transpose_23")]; + tensor var_19339_cast_fp16 = slice_by_index(begin = var_19339_begin_0, end = var_19339_end_0, end_mask = var_19339_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19339_cast_fp16")]; + tensor var_19343_begin_0 = const()[name = tensor("op_19343_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_19343_end_0 = const()[name = tensor("op_19343_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_19343_end_mask_0 = const()[name = tensor("op_19343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19343_cast_fp16 = slice_by_index(begin = var_19343_begin_0, end = var_19343_end_0, end_mask = var_19343_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19343_cast_fp16")]; + tensor var_19347_begin_0 = const()[name = tensor("op_19347_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_19347_end_0 = const()[name = tensor("op_19347_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_19347_end_mask_0 = const()[name = tensor("op_19347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19347_cast_fp16 = slice_by_index(begin = var_19347_begin_0, end = var_19347_end_0, end_mask = var_19347_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19347_cast_fp16")]; + tensor var_19351_begin_0 = const()[name = tensor("op_19351_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_19351_end_0 = const()[name = tensor("op_19351_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_19351_end_mask_0 = const()[name = tensor("op_19351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19351_cast_fp16 = slice_by_index(begin = var_19351_begin_0, end = var_19351_end_0, end_mask = var_19351_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19351_cast_fp16")]; + tensor var_19355_begin_0 = const()[name = tensor("op_19355_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_19355_end_0 = const()[name = tensor("op_19355_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_19355_end_mask_0 = const()[name = tensor("op_19355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19355_cast_fp16 = slice_by_index(begin = var_19355_begin_0, end = var_19355_end_0, end_mask = var_19355_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19355_cast_fp16")]; + tensor var_19359_begin_0 = const()[name = tensor("op_19359_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_19359_end_0 = const()[name = tensor("op_19359_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_19359_end_mask_0 = const()[name = tensor("op_19359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19359_cast_fp16 = slice_by_index(begin = var_19359_begin_0, end = var_19359_end_0, end_mask = var_19359_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19359_cast_fp16")]; + tensor var_19363_begin_0 = const()[name = tensor("op_19363_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_19363_end_0 = const()[name = tensor("op_19363_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_19363_end_mask_0 = const()[name = tensor("op_19363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19363_cast_fp16 = slice_by_index(begin = var_19363_begin_0, end = var_19363_end_0, end_mask = var_19363_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19363_cast_fp16")]; + tensor var_19367_begin_0 = const()[name = tensor("op_19367_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_19367_end_0 = const()[name = tensor("op_19367_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_19367_end_mask_0 = const()[name = tensor("op_19367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19367_cast_fp16 = slice_by_index(begin = var_19367_begin_0, end = var_19367_end_0, end_mask = var_19367_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19367_cast_fp16")]; + tensor var_19371_begin_0 = const()[name = tensor("op_19371_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_19371_end_0 = const()[name = tensor("op_19371_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_19371_end_mask_0 = const()[name = tensor("op_19371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19371_cast_fp16 = slice_by_index(begin = var_19371_begin_0, end = var_19371_end_0, end_mask = var_19371_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19371_cast_fp16")]; + tensor var_19375_begin_0 = const()[name = tensor("op_19375_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_19375_end_0 = const()[name = tensor("op_19375_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_19375_end_mask_0 = const()[name = tensor("op_19375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19375_cast_fp16 = slice_by_index(begin = var_19375_begin_0, end = var_19375_end_0, end_mask = var_19375_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19375_cast_fp16")]; + tensor var_19379_begin_0 = const()[name = tensor("op_19379_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_19379_end_0 = const()[name = tensor("op_19379_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_19379_end_mask_0 = const()[name = tensor("op_19379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19379_cast_fp16 = slice_by_index(begin = var_19379_begin_0, end = var_19379_end_0, end_mask = var_19379_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19379_cast_fp16")]; + tensor var_19383_begin_0 = const()[name = tensor("op_19383_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_19383_end_0 = const()[name = tensor("op_19383_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_19383_end_mask_0 = const()[name = tensor("op_19383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19383_cast_fp16 = slice_by_index(begin = var_19383_begin_0, end = var_19383_end_0, end_mask = var_19383_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19383_cast_fp16")]; + tensor var_19387_begin_0 = const()[name = tensor("op_19387_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_19387_end_0 = const()[name = tensor("op_19387_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_19387_end_mask_0 = const()[name = tensor("op_19387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19387_cast_fp16 = slice_by_index(begin = var_19387_begin_0, end = var_19387_end_0, end_mask = var_19387_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19387_cast_fp16")]; + tensor var_19391_begin_0 = const()[name = tensor("op_19391_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_19391_end_0 = const()[name = tensor("op_19391_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_19391_end_mask_0 = const()[name = tensor("op_19391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19391_cast_fp16 = slice_by_index(begin = var_19391_begin_0, end = var_19391_end_0, end_mask = var_19391_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19391_cast_fp16")]; + tensor var_19395_begin_0 = const()[name = tensor("op_19395_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_19395_end_0 = const()[name = tensor("op_19395_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_19395_end_mask_0 = const()[name = tensor("op_19395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19395_cast_fp16 = slice_by_index(begin = var_19395_begin_0, end = var_19395_end_0, end_mask = var_19395_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19395_cast_fp16")]; + tensor var_19399_begin_0 = const()[name = tensor("op_19399_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_19399_end_0 = const()[name = tensor("op_19399_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19399_end_mask_0 = const()[name = tensor("op_19399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19399_cast_fp16 = slice_by_index(begin = var_19399_begin_0, end = var_19399_end_0, end_mask = var_19399_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19399_cast_fp16")]; + tensor var_19403_begin_0 = const()[name = tensor("op_19403_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_19403_end_0 = const()[name = tensor("op_19403_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_19403_end_mask_0 = const()[name = tensor("op_19403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19403_cast_fp16 = slice_by_index(begin = var_19403_begin_0, end = var_19403_end_0, end_mask = var_19403_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19403_cast_fp16")]; + tensor var_19407_begin_0 = const()[name = tensor("op_19407_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_19407_end_0 = const()[name = tensor("op_19407_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_19407_end_mask_0 = const()[name = tensor("op_19407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19407_cast_fp16 = slice_by_index(begin = var_19407_begin_0, end = var_19407_end_0, end_mask = var_19407_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19407_cast_fp16")]; + tensor var_19411_begin_0 = const()[name = tensor("op_19411_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_19411_end_0 = const()[name = tensor("op_19411_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_19411_end_mask_0 = const()[name = tensor("op_19411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19411_cast_fp16 = slice_by_index(begin = var_19411_begin_0, end = var_19411_end_0, end_mask = var_19411_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19411_cast_fp16")]; + tensor var_19415_begin_0 = const()[name = tensor("op_19415_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_19415_end_0 = const()[name = tensor("op_19415_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_19415_end_mask_0 = const()[name = tensor("op_19415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19415_cast_fp16 = slice_by_index(begin = var_19415_begin_0, end = var_19415_end_0, end_mask = var_19415_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19415_cast_fp16")]; + tensor var_19417_begin_0 = const()[name = tensor("op_19417_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19417_end_0 = const()[name = tensor("op_19417_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_19417_end_mask_0 = const()[name = tensor("op_19417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19417_cast_fp16 = slice_by_index(begin = var_19417_begin_0, end = var_19417_end_0, end_mask = var_19417_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19417_cast_fp16")]; + tensor var_19421_begin_0 = const()[name = tensor("op_19421_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19421_end_0 = const()[name = tensor("op_19421_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_19421_end_mask_0 = const()[name = tensor("op_19421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19421_cast_fp16 = slice_by_index(begin = var_19421_begin_0, end = var_19421_end_0, end_mask = var_19421_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19421_cast_fp16")]; + tensor var_19425_begin_0 = const()[name = tensor("op_19425_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19425_end_0 = const()[name = tensor("op_19425_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_19425_end_mask_0 = const()[name = tensor("op_19425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19425_cast_fp16 = slice_by_index(begin = var_19425_begin_0, end = var_19425_end_0, end_mask = var_19425_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19425_cast_fp16")]; + tensor var_19429_begin_0 = const()[name = tensor("op_19429_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19429_end_0 = const()[name = tensor("op_19429_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_19429_end_mask_0 = const()[name = tensor("op_19429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19429_cast_fp16 = slice_by_index(begin = var_19429_begin_0, end = var_19429_end_0, end_mask = var_19429_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19429_cast_fp16")]; + tensor var_19433_begin_0 = const()[name = tensor("op_19433_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19433_end_0 = const()[name = tensor("op_19433_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_19433_end_mask_0 = const()[name = tensor("op_19433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19433_cast_fp16 = slice_by_index(begin = var_19433_begin_0, end = var_19433_end_0, end_mask = var_19433_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19433_cast_fp16")]; + tensor var_19437_begin_0 = const()[name = tensor("op_19437_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19437_end_0 = const()[name = tensor("op_19437_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_19437_end_mask_0 = const()[name = tensor("op_19437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19437_cast_fp16 = slice_by_index(begin = var_19437_begin_0, end = var_19437_end_0, end_mask = var_19437_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19437_cast_fp16")]; + tensor var_19441_begin_0 = const()[name = tensor("op_19441_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19441_end_0 = const()[name = tensor("op_19441_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_19441_end_mask_0 = const()[name = tensor("op_19441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19441_cast_fp16 = slice_by_index(begin = var_19441_begin_0, end = var_19441_end_0, end_mask = var_19441_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19441_cast_fp16")]; + tensor var_19445_begin_0 = const()[name = tensor("op_19445_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19445_end_0 = const()[name = tensor("op_19445_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_19445_end_mask_0 = const()[name = tensor("op_19445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19445_cast_fp16 = slice_by_index(begin = var_19445_begin_0, end = var_19445_end_0, end_mask = var_19445_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19445_cast_fp16")]; + tensor var_19449_begin_0 = const()[name = tensor("op_19449_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19449_end_0 = const()[name = tensor("op_19449_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_19449_end_mask_0 = const()[name = tensor("op_19449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19449_cast_fp16 = slice_by_index(begin = var_19449_begin_0, end = var_19449_end_0, end_mask = var_19449_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19449_cast_fp16")]; + tensor var_19453_begin_0 = const()[name = tensor("op_19453_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19453_end_0 = const()[name = tensor("op_19453_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_19453_end_mask_0 = const()[name = tensor("op_19453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19453_cast_fp16 = slice_by_index(begin = var_19453_begin_0, end = var_19453_end_0, end_mask = var_19453_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19453_cast_fp16")]; + tensor var_19457_begin_0 = const()[name = tensor("op_19457_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19457_end_0 = const()[name = tensor("op_19457_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_19457_end_mask_0 = const()[name = tensor("op_19457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19457_cast_fp16 = slice_by_index(begin = var_19457_begin_0, end = var_19457_end_0, end_mask = var_19457_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19457_cast_fp16")]; + tensor var_19461_begin_0 = const()[name = tensor("op_19461_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19461_end_0 = const()[name = tensor("op_19461_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_19461_end_mask_0 = const()[name = tensor("op_19461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19461_cast_fp16 = slice_by_index(begin = var_19461_begin_0, end = var_19461_end_0, end_mask = var_19461_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19461_cast_fp16")]; + tensor var_19465_begin_0 = const()[name = tensor("op_19465_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19465_end_0 = const()[name = tensor("op_19465_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_19465_end_mask_0 = const()[name = tensor("op_19465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19465_cast_fp16 = slice_by_index(begin = var_19465_begin_0, end = var_19465_end_0, end_mask = var_19465_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19465_cast_fp16")]; + tensor var_19469_begin_0 = const()[name = tensor("op_19469_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19469_end_0 = const()[name = tensor("op_19469_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_19469_end_mask_0 = const()[name = tensor("op_19469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19469_cast_fp16 = slice_by_index(begin = var_19469_begin_0, end = var_19469_end_0, end_mask = var_19469_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19469_cast_fp16")]; + tensor var_19473_begin_0 = const()[name = tensor("op_19473_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19473_end_0 = const()[name = tensor("op_19473_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_19473_end_mask_0 = const()[name = tensor("op_19473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19473_cast_fp16 = slice_by_index(begin = var_19473_begin_0, end = var_19473_end_0, end_mask = var_19473_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19473_cast_fp16")]; + tensor var_19477_begin_0 = const()[name = tensor("op_19477_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19477_end_0 = const()[name = tensor("op_19477_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19477_end_mask_0 = const()[name = tensor("op_19477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19477_cast_fp16 = slice_by_index(begin = var_19477_begin_0, end = var_19477_end_0, end_mask = var_19477_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19477_cast_fp16")]; + tensor var_19481_begin_0 = const()[name = tensor("op_19481_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19481_end_0 = const()[name = tensor("op_19481_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_19481_end_mask_0 = const()[name = tensor("op_19481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19481_cast_fp16 = slice_by_index(begin = var_19481_begin_0, end = var_19481_end_0, end_mask = var_19481_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19481_cast_fp16")]; + tensor var_19485_begin_0 = const()[name = tensor("op_19485_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19485_end_0 = const()[name = tensor("op_19485_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_19485_end_mask_0 = const()[name = tensor("op_19485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19485_cast_fp16 = slice_by_index(begin = var_19485_begin_0, end = var_19485_end_0, end_mask = var_19485_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19485_cast_fp16")]; + tensor var_19489_begin_0 = const()[name = tensor("op_19489_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19489_end_0 = const()[name = tensor("op_19489_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_19489_end_mask_0 = const()[name = tensor("op_19489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19489_cast_fp16 = slice_by_index(begin = var_19489_begin_0, end = var_19489_end_0, end_mask = var_19489_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19489_cast_fp16")]; + tensor var_19493_begin_0 = const()[name = tensor("op_19493_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19493_end_0 = const()[name = tensor("op_19493_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_19493_end_mask_0 = const()[name = tensor("op_19493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19493_cast_fp16 = slice_by_index(begin = var_19493_begin_0, end = var_19493_end_0, end_mask = var_19493_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19493_cast_fp16")]; + tensor var_19497_equation_0 = const()[name = tensor("op_19497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19497_cast_fp16 = einsum(equation = var_19497_equation_0, values = (var_19339_cast_fp16, var_19256_cast_fp16))[name = tensor("op_19497_cast_fp16")]; + tensor var_19498_to_fp16 = const()[name = tensor("op_19498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1601_cast_fp16 = mul(x = var_19497_cast_fp16, y = var_19498_to_fp16)[name = tensor("aw_1601_cast_fp16")]; + tensor var_19501_equation_0 = const()[name = tensor("op_19501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19501_cast_fp16 = einsum(equation = var_19501_equation_0, values = (var_19343_cast_fp16, var_19260_cast_fp16))[name = tensor("op_19501_cast_fp16")]; + tensor var_19502_to_fp16 = const()[name = tensor("op_19502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1603_cast_fp16 = mul(x = var_19501_cast_fp16, y = var_19502_to_fp16)[name = tensor("aw_1603_cast_fp16")]; + tensor var_19505_equation_0 = const()[name = tensor("op_19505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19505_cast_fp16 = einsum(equation = var_19505_equation_0, values = (var_19347_cast_fp16, var_19264_cast_fp16))[name = tensor("op_19505_cast_fp16")]; + tensor var_19506_to_fp16 = const()[name = tensor("op_19506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1605_cast_fp16 = mul(x = var_19505_cast_fp16, y = var_19506_to_fp16)[name = tensor("aw_1605_cast_fp16")]; + tensor var_19509_equation_0 = const()[name = tensor("op_19509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19509_cast_fp16 = einsum(equation = var_19509_equation_0, values = (var_19351_cast_fp16, var_19268_cast_fp16))[name = tensor("op_19509_cast_fp16")]; + tensor var_19510_to_fp16 = const()[name = tensor("op_19510_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1607_cast_fp16 = mul(x = var_19509_cast_fp16, y = var_19510_to_fp16)[name = tensor("aw_1607_cast_fp16")]; + tensor var_19513_equation_0 = const()[name = tensor("op_19513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19513_cast_fp16 = einsum(equation = var_19513_equation_0, values = (var_19355_cast_fp16, var_19272_cast_fp16))[name = tensor("op_19513_cast_fp16")]; + tensor var_19514_to_fp16 = const()[name = tensor("op_19514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1609_cast_fp16 = mul(x = var_19513_cast_fp16, y = var_19514_to_fp16)[name = tensor("aw_1609_cast_fp16")]; + tensor var_19517_equation_0 = const()[name = tensor("op_19517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19517_cast_fp16 = einsum(equation = var_19517_equation_0, values = (var_19359_cast_fp16, var_19276_cast_fp16))[name = tensor("op_19517_cast_fp16")]; + tensor var_19518_to_fp16 = const()[name = tensor("op_19518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1611_cast_fp16 = mul(x = var_19517_cast_fp16, y = var_19518_to_fp16)[name = tensor("aw_1611_cast_fp16")]; + tensor var_19521_equation_0 = const()[name = tensor("op_19521_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19521_cast_fp16 = einsum(equation = var_19521_equation_0, values = (var_19363_cast_fp16, var_19280_cast_fp16))[name = tensor("op_19521_cast_fp16")]; + tensor var_19522_to_fp16 = const()[name = tensor("op_19522_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1613_cast_fp16 = mul(x = var_19521_cast_fp16, y = var_19522_to_fp16)[name = tensor("aw_1613_cast_fp16")]; + tensor var_19525_equation_0 = const()[name = tensor("op_19525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19525_cast_fp16 = einsum(equation = var_19525_equation_0, values = (var_19367_cast_fp16, var_19284_cast_fp16))[name = tensor("op_19525_cast_fp16")]; + tensor var_19526_to_fp16 = const()[name = tensor("op_19526_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1615_cast_fp16 = mul(x = var_19525_cast_fp16, y = var_19526_to_fp16)[name = tensor("aw_1615_cast_fp16")]; + tensor var_19529_equation_0 = const()[name = tensor("op_19529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19529_cast_fp16 = einsum(equation = var_19529_equation_0, values = (var_19371_cast_fp16, var_19288_cast_fp16))[name = tensor("op_19529_cast_fp16")]; + tensor var_19530_to_fp16 = const()[name = tensor("op_19530_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1617_cast_fp16 = mul(x = var_19529_cast_fp16, y = var_19530_to_fp16)[name = tensor("aw_1617_cast_fp16")]; + tensor var_19533_equation_0 = const()[name = tensor("op_19533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19533_cast_fp16 = einsum(equation = var_19533_equation_0, values = (var_19375_cast_fp16, var_19292_cast_fp16))[name = tensor("op_19533_cast_fp16")]; + tensor var_19534_to_fp16 = const()[name = tensor("op_19534_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1619_cast_fp16 = mul(x = var_19533_cast_fp16, y = var_19534_to_fp16)[name = tensor("aw_1619_cast_fp16")]; + tensor var_19537_equation_0 = const()[name = tensor("op_19537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19537_cast_fp16 = einsum(equation = var_19537_equation_0, values = (var_19379_cast_fp16, var_19296_cast_fp16))[name = tensor("op_19537_cast_fp16")]; + tensor var_19538_to_fp16 = const()[name = tensor("op_19538_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1621_cast_fp16 = mul(x = var_19537_cast_fp16, y = var_19538_to_fp16)[name = tensor("aw_1621_cast_fp16")]; + tensor var_19541_equation_0 = const()[name = tensor("op_19541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19541_cast_fp16 = einsum(equation = var_19541_equation_0, values = (var_19383_cast_fp16, var_19300_cast_fp16))[name = tensor("op_19541_cast_fp16")]; + tensor var_19542_to_fp16 = const()[name = tensor("op_19542_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1623_cast_fp16 = mul(x = var_19541_cast_fp16, y = var_19542_to_fp16)[name = tensor("aw_1623_cast_fp16")]; + tensor var_19545_equation_0 = const()[name = tensor("op_19545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19545_cast_fp16 = einsum(equation = var_19545_equation_0, values = (var_19387_cast_fp16, var_19304_cast_fp16))[name = tensor("op_19545_cast_fp16")]; + tensor var_19546_to_fp16 = const()[name = tensor("op_19546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1625_cast_fp16 = mul(x = var_19545_cast_fp16, y = var_19546_to_fp16)[name = tensor("aw_1625_cast_fp16")]; + tensor var_19549_equation_0 = const()[name = tensor("op_19549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19549_cast_fp16 = einsum(equation = var_19549_equation_0, values = (var_19391_cast_fp16, var_19308_cast_fp16))[name = tensor("op_19549_cast_fp16")]; + tensor var_19550_to_fp16 = const()[name = tensor("op_19550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1627_cast_fp16 = mul(x = var_19549_cast_fp16, y = var_19550_to_fp16)[name = tensor("aw_1627_cast_fp16")]; + tensor var_19553_equation_0 = const()[name = tensor("op_19553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19553_cast_fp16 = einsum(equation = var_19553_equation_0, values = (var_19395_cast_fp16, var_19312_cast_fp16))[name = tensor("op_19553_cast_fp16")]; + tensor var_19554_to_fp16 = const()[name = tensor("op_19554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1629_cast_fp16 = mul(x = var_19553_cast_fp16, y = var_19554_to_fp16)[name = tensor("aw_1629_cast_fp16")]; + tensor var_19557_equation_0 = const()[name = tensor("op_19557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19557_cast_fp16 = einsum(equation = var_19557_equation_0, values = (var_19399_cast_fp16, var_19316_cast_fp16))[name = tensor("op_19557_cast_fp16")]; + tensor var_19558_to_fp16 = const()[name = tensor("op_19558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1631_cast_fp16 = mul(x = var_19557_cast_fp16, y = var_19558_to_fp16)[name = tensor("aw_1631_cast_fp16")]; + tensor var_19561_equation_0 = const()[name = tensor("op_19561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19561_cast_fp16 = einsum(equation = var_19561_equation_0, values = (var_19403_cast_fp16, var_19320_cast_fp16))[name = tensor("op_19561_cast_fp16")]; + tensor var_19562_to_fp16 = const()[name = tensor("op_19562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1633_cast_fp16 = mul(x = var_19561_cast_fp16, y = var_19562_to_fp16)[name = tensor("aw_1633_cast_fp16")]; + tensor var_19565_equation_0 = const()[name = tensor("op_19565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19565_cast_fp16 = einsum(equation = var_19565_equation_0, values = (var_19407_cast_fp16, var_19324_cast_fp16))[name = tensor("op_19565_cast_fp16")]; + tensor var_19566_to_fp16 = const()[name = tensor("op_19566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1635_cast_fp16 = mul(x = var_19565_cast_fp16, y = var_19566_to_fp16)[name = tensor("aw_1635_cast_fp16")]; + tensor var_19569_equation_0 = const()[name = tensor("op_19569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19569_cast_fp16 = einsum(equation = var_19569_equation_0, values = (var_19411_cast_fp16, var_19328_cast_fp16))[name = tensor("op_19569_cast_fp16")]; + tensor var_19570_to_fp16 = const()[name = tensor("op_19570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1637_cast_fp16 = mul(x = var_19569_cast_fp16, y = var_19570_to_fp16)[name = tensor("aw_1637_cast_fp16")]; + tensor var_19573_equation_0 = const()[name = tensor("op_19573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19573_cast_fp16 = einsum(equation = var_19573_equation_0, values = (var_19415_cast_fp16, var_19332_cast_fp16))[name = tensor("op_19573_cast_fp16")]; + tensor var_19574_to_fp16 = const()[name = tensor("op_19574_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1639_cast_fp16 = mul(x = var_19573_cast_fp16, y = var_19574_to_fp16)[name = tensor("aw_1639_cast_fp16")]; + tensor var_19576_cast_fp16 = softmax(axis = var_2624, x = aw_1601_cast_fp16)[name = tensor("op_19576_cast_fp16")]; + tensor var_19577_cast_fp16 = softmax(axis = var_2624, x = aw_1603_cast_fp16)[name = tensor("op_19577_cast_fp16")]; + tensor var_19578_cast_fp16 = softmax(axis = var_2624, x = aw_1605_cast_fp16)[name = tensor("op_19578_cast_fp16")]; + tensor var_19579_cast_fp16 = softmax(axis = var_2624, x = aw_1607_cast_fp16)[name = tensor("op_19579_cast_fp16")]; + tensor var_19580_cast_fp16 = softmax(axis = var_2624, x = aw_1609_cast_fp16)[name = tensor("op_19580_cast_fp16")]; + tensor var_19581_cast_fp16 = softmax(axis = var_2624, x = aw_1611_cast_fp16)[name = tensor("op_19581_cast_fp16")]; + tensor var_19582_cast_fp16 = softmax(axis = var_2624, x = aw_1613_cast_fp16)[name = tensor("op_19582_cast_fp16")]; + tensor var_19583_cast_fp16 = softmax(axis = var_2624, x = aw_1615_cast_fp16)[name = tensor("op_19583_cast_fp16")]; + tensor var_19584_cast_fp16 = softmax(axis = var_2624, x = aw_1617_cast_fp16)[name = tensor("op_19584_cast_fp16")]; + tensor var_19585_cast_fp16 = softmax(axis = var_2624, x = aw_1619_cast_fp16)[name = tensor("op_19585_cast_fp16")]; + tensor var_19586_cast_fp16 = softmax(axis = var_2624, x = aw_1621_cast_fp16)[name = tensor("op_19586_cast_fp16")]; + tensor var_19587_cast_fp16 = softmax(axis = var_2624, x = aw_1623_cast_fp16)[name = tensor("op_19587_cast_fp16")]; + tensor var_19588_cast_fp16 = softmax(axis = var_2624, x = aw_1625_cast_fp16)[name = tensor("op_19588_cast_fp16")]; + tensor var_19589_cast_fp16 = softmax(axis = var_2624, x = aw_1627_cast_fp16)[name = tensor("op_19589_cast_fp16")]; + tensor var_19590_cast_fp16 = softmax(axis = var_2624, x = aw_1629_cast_fp16)[name = tensor("op_19590_cast_fp16")]; + tensor var_19591_cast_fp16 = softmax(axis = var_2624, x = aw_1631_cast_fp16)[name = tensor("op_19591_cast_fp16")]; + tensor var_19592_cast_fp16 = softmax(axis = var_2624, x = aw_1633_cast_fp16)[name = tensor("op_19592_cast_fp16")]; + tensor var_19593_cast_fp16 = softmax(axis = var_2624, x = aw_1635_cast_fp16)[name = tensor("op_19593_cast_fp16")]; + tensor var_19594_cast_fp16 = softmax(axis = var_2624, x = aw_1637_cast_fp16)[name = tensor("op_19594_cast_fp16")]; + tensor var_19595_cast_fp16 = softmax(axis = var_2624, x = aw_1639_cast_fp16)[name = tensor("op_19595_cast_fp16")]; + tensor var_19597_equation_0 = const()[name = tensor("op_19597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19597_cast_fp16 = einsum(equation = var_19597_equation_0, values = (var_19417_cast_fp16, var_19576_cast_fp16))[name = tensor("op_19597_cast_fp16")]; + tensor var_19599_equation_0 = const()[name = tensor("op_19599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19599_cast_fp16 = einsum(equation = var_19599_equation_0, values = (var_19421_cast_fp16, var_19577_cast_fp16))[name = tensor("op_19599_cast_fp16")]; + tensor var_19601_equation_0 = const()[name = tensor("op_19601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19601_cast_fp16 = einsum(equation = var_19601_equation_0, values = (var_19425_cast_fp16, var_19578_cast_fp16))[name = tensor("op_19601_cast_fp16")]; + tensor var_19603_equation_0 = const()[name = tensor("op_19603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19603_cast_fp16 = einsum(equation = var_19603_equation_0, values = (var_19429_cast_fp16, var_19579_cast_fp16))[name = tensor("op_19603_cast_fp16")]; + tensor var_19605_equation_0 = const()[name = tensor("op_19605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19605_cast_fp16 = einsum(equation = var_19605_equation_0, values = (var_19433_cast_fp16, var_19580_cast_fp16))[name = tensor("op_19605_cast_fp16")]; + tensor var_19607_equation_0 = const()[name = tensor("op_19607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19607_cast_fp16 = einsum(equation = var_19607_equation_0, values = (var_19437_cast_fp16, var_19581_cast_fp16))[name = tensor("op_19607_cast_fp16")]; + tensor var_19609_equation_0 = const()[name = tensor("op_19609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19609_cast_fp16 = einsum(equation = var_19609_equation_0, values = (var_19441_cast_fp16, var_19582_cast_fp16))[name = tensor("op_19609_cast_fp16")]; + tensor var_19611_equation_0 = const()[name = tensor("op_19611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19611_cast_fp16 = einsum(equation = var_19611_equation_0, values = (var_19445_cast_fp16, var_19583_cast_fp16))[name = tensor("op_19611_cast_fp16")]; + tensor var_19613_equation_0 = const()[name = tensor("op_19613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19613_cast_fp16 = einsum(equation = var_19613_equation_0, values = (var_19449_cast_fp16, var_19584_cast_fp16))[name = tensor("op_19613_cast_fp16")]; + tensor var_19615_equation_0 = const()[name = tensor("op_19615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19615_cast_fp16 = einsum(equation = var_19615_equation_0, values = (var_19453_cast_fp16, var_19585_cast_fp16))[name = tensor("op_19615_cast_fp16")]; + tensor var_19617_equation_0 = const()[name = tensor("op_19617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19617_cast_fp16 = einsum(equation = var_19617_equation_0, values = (var_19457_cast_fp16, var_19586_cast_fp16))[name = tensor("op_19617_cast_fp16")]; + tensor var_19619_equation_0 = const()[name = tensor("op_19619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19619_cast_fp16 = einsum(equation = var_19619_equation_0, values = (var_19461_cast_fp16, var_19587_cast_fp16))[name = tensor("op_19619_cast_fp16")]; + tensor var_19621_equation_0 = const()[name = tensor("op_19621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19621_cast_fp16 = einsum(equation = var_19621_equation_0, values = (var_19465_cast_fp16, var_19588_cast_fp16))[name = tensor("op_19621_cast_fp16")]; + tensor var_19623_equation_0 = const()[name = tensor("op_19623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19623_cast_fp16 = einsum(equation = var_19623_equation_0, values = (var_19469_cast_fp16, var_19589_cast_fp16))[name = tensor("op_19623_cast_fp16")]; + tensor var_19625_equation_0 = const()[name = tensor("op_19625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19625_cast_fp16 = einsum(equation = var_19625_equation_0, values = (var_19473_cast_fp16, var_19590_cast_fp16))[name = tensor("op_19625_cast_fp16")]; + tensor var_19627_equation_0 = const()[name = tensor("op_19627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19627_cast_fp16 = einsum(equation = var_19627_equation_0, values = (var_19477_cast_fp16, var_19591_cast_fp16))[name = tensor("op_19627_cast_fp16")]; + tensor var_19629_equation_0 = const()[name = tensor("op_19629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19629_cast_fp16 = einsum(equation = var_19629_equation_0, values = (var_19481_cast_fp16, var_19592_cast_fp16))[name = tensor("op_19629_cast_fp16")]; + tensor var_19631_equation_0 = const()[name = tensor("op_19631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19631_cast_fp16 = einsum(equation = var_19631_equation_0, values = (var_19485_cast_fp16, var_19593_cast_fp16))[name = tensor("op_19631_cast_fp16")]; + tensor var_19633_equation_0 = const()[name = tensor("op_19633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19633_cast_fp16 = einsum(equation = var_19633_equation_0, values = (var_19489_cast_fp16, var_19594_cast_fp16))[name = tensor("op_19633_cast_fp16")]; + tensor var_19635_equation_0 = const()[name = tensor("op_19635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19635_cast_fp16 = einsum(equation = var_19635_equation_0, values = (var_19493_cast_fp16, var_19595_cast_fp16))[name = tensor("op_19635_cast_fp16")]; + tensor input_293_interleave_0 = const()[name = tensor("input_293_interleave_0"), val = tensor(false)]; + tensor input_293_cast_fp16 = concat(axis = var_2624, interleave = input_293_interleave_0, values = (var_19597_cast_fp16, var_19599_cast_fp16, var_19601_cast_fp16, var_19603_cast_fp16, var_19605_cast_fp16, var_19607_cast_fp16, var_19609_cast_fp16, var_19611_cast_fp16, var_19613_cast_fp16, var_19615_cast_fp16, var_19617_cast_fp16, var_19619_cast_fp16, var_19621_cast_fp16, var_19623_cast_fp16, var_19625_cast_fp16, var_19627_cast_fp16, var_19629_cast_fp16, var_19631_cast_fp16, var_19633_cast_fp16, var_19635_cast_fp16))[name = tensor("input_293_cast_fp16")]; + tensor var_19645_pad_type_0 = const()[name = tensor("op_19645_pad_type_0"), val = tensor("valid")]; + tensor var_19645_strides_0 = const()[name = tensor("op_19645_strides_0"), val = tensor([1, 1])]; + tensor var_19645_pad_0 = const()[name = tensor("op_19645_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19645_dilations_0 = const()[name = tensor("op_19645_dilations_0"), val = tensor([1, 1])]; + tensor var_19645_groups_0 = const()[name = tensor("op_19645_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(573718336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(574947200))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(574947392)))]; + tensor var_19645_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_19645_dilations_0, groups = var_19645_groups_0, pad = var_19645_pad_0, pad_type = var_19645_pad_type_0, strides = var_19645_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized, x = input_293_cast_fp16)[name = tensor("op_19645_cast_fp16")]; + tensor inputs_135_cast_fp16 = add(x = var_19645_cast_fp16, y = inputs_133_cast_fp16)[name = tensor("inputs_135_cast_fp16")]; + tensor hidden_states_187_axes_0 = const()[name = tensor("hidden_states_187_axes_0"), val = tensor([1])]; + tensor hidden_states_187_gamma_0_to_fp16 = const()[name = tensor("hidden_states_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(574950016)))]; + tensor hidden_states_187_beta_0_to_fp16 = const()[name = tensor("hidden_states_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(574952640)))]; + tensor var_19655_to_fp16 = const()[name = tensor("op_19655_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_187_cast_fp16 = layer_norm(axes = hidden_states_187_axes_0, beta = hidden_states_187_beta_0_to_fp16, epsilon = var_19655_to_fp16, gamma = hidden_states_187_gamma_0_to_fp16, x = inputs_135_cast_fp16)[name = tensor("hidden_states_187_cast_fp16")]; + tensor q_91_pad_type_0 = const()[name = tensor("q_91_pad_type_0"), val = tensor("valid")]; + tensor q_91_strides_0 = const()[name = tensor("q_91_strides_0"), val = tensor([1, 1])]; + tensor q_91_pad_0 = const()[name = tensor("q_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_91_dilations_0 = const()[name = tensor("q_91_dilations_0"), val = tensor([1, 1])]; + tensor q_91_groups_0 = const()[name = tensor("q_91_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(574955264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(576184128))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_91_cast_fp16 = conv(dilations = q_91_dilations_0, groups = q_91_groups_0, pad = q_91_pad_0, pad_type = q_91_pad_type_0, strides = q_91_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_187_cast_fp16)[name = tensor("q_91_cast_fp16")]; + tensor k_181_pad_type_0 = const()[name = tensor("k_181_pad_type_0"), val = tensor("valid")]; + tensor k_181_strides_0 = const()[name = tensor("k_181_strides_0"), val = tensor([1, 1])]; + tensor k_181_pad_0 = const()[name = tensor("k_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_181_dilations_0 = const()[name = tensor("k_181_dilations_0"), val = tensor([1, 1])]; + tensor k_181_groups_0 = const()[name = tensor("k_181_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(576184320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578150464))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_181_cast_fp16 = conv(dilations = k_181_dilations_0, groups = k_181_groups_0, pad = k_181_pad_0, pad_type = k_181_pad_type_0, strides = k_181_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_181_cast_fp16")]; + tensor v_91_pad_type_0 = const()[name = tensor("v_91_pad_type_0"), val = tensor("valid")]; + tensor v_91_strides_0 = const()[name = tensor("v_91_strides_0"), val = tensor([1, 1])]; + tensor v_91_pad_0 = const()[name = tensor("v_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_91_dilations_0 = const()[name = tensor("v_91_dilations_0"), val = tensor([1, 1])]; + tensor v_91_groups_0 = const()[name = tensor("v_91_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578150656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580116800))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_91_cast_fp16 = conv(dilations = v_91_dilations_0, groups = v_91_groups_0, pad = v_91_pad_0, pad_type = v_91_pad_type_0, strides = v_91_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_91_cast_fp16")]; + tensor var_19688_begin_0 = const()[name = tensor("op_19688_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19688_end_0 = const()[name = tensor("op_19688_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_19688_end_mask_0 = const()[name = tensor("op_19688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19688_cast_fp16 = slice_by_index(begin = var_19688_begin_0, end = var_19688_end_0, end_mask = var_19688_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19688_cast_fp16")]; + tensor var_19692_begin_0 = const()[name = tensor("op_19692_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19692_end_0 = const()[name = tensor("op_19692_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_19692_end_mask_0 = const()[name = tensor("op_19692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19692_cast_fp16 = slice_by_index(begin = var_19692_begin_0, end = var_19692_end_0, end_mask = var_19692_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19692_cast_fp16")]; + tensor var_19696_begin_0 = const()[name = tensor("op_19696_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19696_end_0 = const()[name = tensor("op_19696_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_19696_end_mask_0 = const()[name = tensor("op_19696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19696_cast_fp16 = slice_by_index(begin = var_19696_begin_0, end = var_19696_end_0, end_mask = var_19696_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19696_cast_fp16")]; + tensor var_19700_begin_0 = const()[name = tensor("op_19700_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19700_end_0 = const()[name = tensor("op_19700_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_19700_end_mask_0 = const()[name = tensor("op_19700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19700_cast_fp16 = slice_by_index(begin = var_19700_begin_0, end = var_19700_end_0, end_mask = var_19700_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19700_cast_fp16")]; + tensor var_19704_begin_0 = const()[name = tensor("op_19704_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19704_end_0 = const()[name = tensor("op_19704_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_19704_end_mask_0 = const()[name = tensor("op_19704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19704_cast_fp16 = slice_by_index(begin = var_19704_begin_0, end = var_19704_end_0, end_mask = var_19704_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19704_cast_fp16")]; + tensor var_19708_begin_0 = const()[name = tensor("op_19708_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19708_end_0 = const()[name = tensor("op_19708_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_19708_end_mask_0 = const()[name = tensor("op_19708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19708_cast_fp16 = slice_by_index(begin = var_19708_begin_0, end = var_19708_end_0, end_mask = var_19708_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19708_cast_fp16")]; + tensor var_19712_begin_0 = const()[name = tensor("op_19712_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19712_end_0 = const()[name = tensor("op_19712_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_19712_end_mask_0 = const()[name = tensor("op_19712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19712_cast_fp16 = slice_by_index(begin = var_19712_begin_0, end = var_19712_end_0, end_mask = var_19712_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19712_cast_fp16")]; + tensor var_19716_begin_0 = const()[name = tensor("op_19716_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19716_end_0 = const()[name = tensor("op_19716_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_19716_end_mask_0 = const()[name = tensor("op_19716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19716_cast_fp16 = slice_by_index(begin = var_19716_begin_0, end = var_19716_end_0, end_mask = var_19716_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19716_cast_fp16")]; + tensor var_19720_begin_0 = const()[name = tensor("op_19720_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19720_end_0 = const()[name = tensor("op_19720_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_19720_end_mask_0 = const()[name = tensor("op_19720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19720_cast_fp16 = slice_by_index(begin = var_19720_begin_0, end = var_19720_end_0, end_mask = var_19720_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19720_cast_fp16")]; + tensor var_19724_begin_0 = const()[name = tensor("op_19724_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19724_end_0 = const()[name = tensor("op_19724_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_19724_end_mask_0 = const()[name = tensor("op_19724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19724_cast_fp16 = slice_by_index(begin = var_19724_begin_0, end = var_19724_end_0, end_mask = var_19724_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19724_cast_fp16")]; + tensor var_19728_begin_0 = const()[name = tensor("op_19728_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19728_end_0 = const()[name = tensor("op_19728_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_19728_end_mask_0 = const()[name = tensor("op_19728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19728_cast_fp16 = slice_by_index(begin = var_19728_begin_0, end = var_19728_end_0, end_mask = var_19728_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19728_cast_fp16")]; + tensor var_19732_begin_0 = const()[name = tensor("op_19732_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19732_end_0 = const()[name = tensor("op_19732_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_19732_end_mask_0 = const()[name = tensor("op_19732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19732_cast_fp16 = slice_by_index(begin = var_19732_begin_0, end = var_19732_end_0, end_mask = var_19732_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19732_cast_fp16")]; + tensor var_19736_begin_0 = const()[name = tensor("op_19736_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19736_end_0 = const()[name = tensor("op_19736_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_19736_end_mask_0 = const()[name = tensor("op_19736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19736_cast_fp16 = slice_by_index(begin = var_19736_begin_0, end = var_19736_end_0, end_mask = var_19736_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19736_cast_fp16")]; + tensor var_19740_begin_0 = const()[name = tensor("op_19740_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19740_end_0 = const()[name = tensor("op_19740_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_19740_end_mask_0 = const()[name = tensor("op_19740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19740_cast_fp16 = slice_by_index(begin = var_19740_begin_0, end = var_19740_end_0, end_mask = var_19740_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19740_cast_fp16")]; + tensor var_19744_begin_0 = const()[name = tensor("op_19744_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19744_end_0 = const()[name = tensor("op_19744_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_19744_end_mask_0 = const()[name = tensor("op_19744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19744_cast_fp16 = slice_by_index(begin = var_19744_begin_0, end = var_19744_end_0, end_mask = var_19744_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19744_cast_fp16")]; + tensor var_19748_begin_0 = const()[name = tensor("op_19748_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19748_end_0 = const()[name = tensor("op_19748_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19748_end_mask_0 = const()[name = tensor("op_19748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19748_cast_fp16 = slice_by_index(begin = var_19748_begin_0, end = var_19748_end_0, end_mask = var_19748_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19748_cast_fp16")]; + tensor var_19752_begin_0 = const()[name = tensor("op_19752_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19752_end_0 = const()[name = tensor("op_19752_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_19752_end_mask_0 = const()[name = tensor("op_19752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19752_cast_fp16 = slice_by_index(begin = var_19752_begin_0, end = var_19752_end_0, end_mask = var_19752_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19752_cast_fp16")]; + tensor var_19756_begin_0 = const()[name = tensor("op_19756_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19756_end_0 = const()[name = tensor("op_19756_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_19756_end_mask_0 = const()[name = tensor("op_19756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19756_cast_fp16 = slice_by_index(begin = var_19756_begin_0, end = var_19756_end_0, end_mask = var_19756_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19756_cast_fp16")]; + tensor var_19760_begin_0 = const()[name = tensor("op_19760_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19760_end_0 = const()[name = tensor("op_19760_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_19760_end_mask_0 = const()[name = tensor("op_19760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19760_cast_fp16 = slice_by_index(begin = var_19760_begin_0, end = var_19760_end_0, end_mask = var_19760_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19760_cast_fp16")]; + tensor var_19764_begin_0 = const()[name = tensor("op_19764_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19764_end_0 = const()[name = tensor("op_19764_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_19764_end_mask_0 = const()[name = tensor("op_19764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19764_cast_fp16 = slice_by_index(begin = var_19764_begin_0, end = var_19764_end_0, end_mask = var_19764_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19764_cast_fp16")]; + tensor k_183_perm_0 = const()[name = tensor("k_183_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_19771_begin_0 = const()[name = tensor("op_19771_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19771_end_0 = const()[name = tensor("op_19771_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_19771_end_mask_0 = const()[name = tensor("op_19771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_183_cast_fp16 = transpose(perm = k_183_perm_0, x = k_181_cast_fp16)[name = tensor("transpose_22")]; + tensor var_19771_cast_fp16 = slice_by_index(begin = var_19771_begin_0, end = var_19771_end_0, end_mask = var_19771_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19771_cast_fp16")]; + tensor var_19775_begin_0 = const()[name = tensor("op_19775_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_19775_end_0 = const()[name = tensor("op_19775_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_19775_end_mask_0 = const()[name = tensor("op_19775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19775_cast_fp16 = slice_by_index(begin = var_19775_begin_0, end = var_19775_end_0, end_mask = var_19775_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19775_cast_fp16")]; + tensor var_19779_begin_0 = const()[name = tensor("op_19779_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_19779_end_0 = const()[name = tensor("op_19779_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_19779_end_mask_0 = const()[name = tensor("op_19779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19779_cast_fp16 = slice_by_index(begin = var_19779_begin_0, end = var_19779_end_0, end_mask = var_19779_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19779_cast_fp16")]; + tensor var_19783_begin_0 = const()[name = tensor("op_19783_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_19783_end_0 = const()[name = tensor("op_19783_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_19783_end_mask_0 = const()[name = tensor("op_19783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19783_cast_fp16 = slice_by_index(begin = var_19783_begin_0, end = var_19783_end_0, end_mask = var_19783_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19783_cast_fp16")]; + tensor var_19787_begin_0 = const()[name = tensor("op_19787_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_19787_end_0 = const()[name = tensor("op_19787_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_19787_end_mask_0 = const()[name = tensor("op_19787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19787_cast_fp16 = slice_by_index(begin = var_19787_begin_0, end = var_19787_end_0, end_mask = var_19787_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19787_cast_fp16")]; + tensor var_19791_begin_0 = const()[name = tensor("op_19791_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_19791_end_0 = const()[name = tensor("op_19791_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_19791_end_mask_0 = const()[name = tensor("op_19791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19791_cast_fp16 = slice_by_index(begin = var_19791_begin_0, end = var_19791_end_0, end_mask = var_19791_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19791_cast_fp16")]; + tensor var_19795_begin_0 = const()[name = tensor("op_19795_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_19795_end_0 = const()[name = tensor("op_19795_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_19795_end_mask_0 = const()[name = tensor("op_19795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19795_cast_fp16 = slice_by_index(begin = var_19795_begin_0, end = var_19795_end_0, end_mask = var_19795_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19795_cast_fp16")]; + tensor var_19799_begin_0 = const()[name = tensor("op_19799_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_19799_end_0 = const()[name = tensor("op_19799_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_19799_end_mask_0 = const()[name = tensor("op_19799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19799_cast_fp16 = slice_by_index(begin = var_19799_begin_0, end = var_19799_end_0, end_mask = var_19799_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19799_cast_fp16")]; + tensor var_19803_begin_0 = const()[name = tensor("op_19803_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_19803_end_0 = const()[name = tensor("op_19803_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_19803_end_mask_0 = const()[name = tensor("op_19803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19803_cast_fp16 = slice_by_index(begin = var_19803_begin_0, end = var_19803_end_0, end_mask = var_19803_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19803_cast_fp16")]; + tensor var_19807_begin_0 = const()[name = tensor("op_19807_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_19807_end_0 = const()[name = tensor("op_19807_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_19807_end_mask_0 = const()[name = tensor("op_19807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19807_cast_fp16 = slice_by_index(begin = var_19807_begin_0, end = var_19807_end_0, end_mask = var_19807_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19807_cast_fp16")]; + tensor var_19811_begin_0 = const()[name = tensor("op_19811_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_19811_end_0 = const()[name = tensor("op_19811_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_19811_end_mask_0 = const()[name = tensor("op_19811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19811_cast_fp16 = slice_by_index(begin = var_19811_begin_0, end = var_19811_end_0, end_mask = var_19811_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19811_cast_fp16")]; + tensor var_19815_begin_0 = const()[name = tensor("op_19815_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_19815_end_0 = const()[name = tensor("op_19815_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_19815_end_mask_0 = const()[name = tensor("op_19815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19815_cast_fp16 = slice_by_index(begin = var_19815_begin_0, end = var_19815_end_0, end_mask = var_19815_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19815_cast_fp16")]; + tensor var_19819_begin_0 = const()[name = tensor("op_19819_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_19819_end_0 = const()[name = tensor("op_19819_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_19819_end_mask_0 = const()[name = tensor("op_19819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19819_cast_fp16 = slice_by_index(begin = var_19819_begin_0, end = var_19819_end_0, end_mask = var_19819_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19819_cast_fp16")]; + tensor var_19823_begin_0 = const()[name = tensor("op_19823_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_19823_end_0 = const()[name = tensor("op_19823_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_19823_end_mask_0 = const()[name = tensor("op_19823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19823_cast_fp16 = slice_by_index(begin = var_19823_begin_0, end = var_19823_end_0, end_mask = var_19823_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19823_cast_fp16")]; + tensor var_19827_begin_0 = const()[name = tensor("op_19827_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_19827_end_0 = const()[name = tensor("op_19827_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_19827_end_mask_0 = const()[name = tensor("op_19827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19827_cast_fp16 = slice_by_index(begin = var_19827_begin_0, end = var_19827_end_0, end_mask = var_19827_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19827_cast_fp16")]; + tensor var_19831_begin_0 = const()[name = tensor("op_19831_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_19831_end_0 = const()[name = tensor("op_19831_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_19831_end_mask_0 = const()[name = tensor("op_19831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19831_cast_fp16 = slice_by_index(begin = var_19831_begin_0, end = var_19831_end_0, end_mask = var_19831_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19831_cast_fp16")]; + tensor var_19835_begin_0 = const()[name = tensor("op_19835_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_19835_end_0 = const()[name = tensor("op_19835_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_19835_end_mask_0 = const()[name = tensor("op_19835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19835_cast_fp16 = slice_by_index(begin = var_19835_begin_0, end = var_19835_end_0, end_mask = var_19835_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19835_cast_fp16")]; + tensor var_19839_begin_0 = const()[name = tensor("op_19839_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_19839_end_0 = const()[name = tensor("op_19839_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_19839_end_mask_0 = const()[name = tensor("op_19839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19839_cast_fp16 = slice_by_index(begin = var_19839_begin_0, end = var_19839_end_0, end_mask = var_19839_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19839_cast_fp16")]; + tensor var_19843_begin_0 = const()[name = tensor("op_19843_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_19843_end_0 = const()[name = tensor("op_19843_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_19843_end_mask_0 = const()[name = tensor("op_19843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19843_cast_fp16 = slice_by_index(begin = var_19843_begin_0, end = var_19843_end_0, end_mask = var_19843_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19843_cast_fp16")]; + tensor var_19847_begin_0 = const()[name = tensor("op_19847_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_19847_end_0 = const()[name = tensor("op_19847_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_19847_end_mask_0 = const()[name = tensor("op_19847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19847_cast_fp16 = slice_by_index(begin = var_19847_begin_0, end = var_19847_end_0, end_mask = var_19847_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19847_cast_fp16")]; + tensor var_19849_begin_0 = const()[name = tensor("op_19849_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19849_end_0 = const()[name = tensor("op_19849_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_19849_end_mask_0 = const()[name = tensor("op_19849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19849_cast_fp16 = slice_by_index(begin = var_19849_begin_0, end = var_19849_end_0, end_mask = var_19849_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19849_cast_fp16")]; + tensor var_19853_begin_0 = const()[name = tensor("op_19853_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19853_end_0 = const()[name = tensor("op_19853_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_19853_end_mask_0 = const()[name = tensor("op_19853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19853_cast_fp16 = slice_by_index(begin = var_19853_begin_0, end = var_19853_end_0, end_mask = var_19853_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19853_cast_fp16")]; + tensor var_19857_begin_0 = const()[name = tensor("op_19857_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19857_end_0 = const()[name = tensor("op_19857_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_19857_end_mask_0 = const()[name = tensor("op_19857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19857_cast_fp16 = slice_by_index(begin = var_19857_begin_0, end = var_19857_end_0, end_mask = var_19857_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19857_cast_fp16")]; + tensor var_19861_begin_0 = const()[name = tensor("op_19861_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19861_end_0 = const()[name = tensor("op_19861_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_19861_end_mask_0 = const()[name = tensor("op_19861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19861_cast_fp16 = slice_by_index(begin = var_19861_begin_0, end = var_19861_end_0, end_mask = var_19861_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19861_cast_fp16")]; + tensor var_19865_begin_0 = const()[name = tensor("op_19865_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19865_end_0 = const()[name = tensor("op_19865_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_19865_end_mask_0 = const()[name = tensor("op_19865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19865_cast_fp16 = slice_by_index(begin = var_19865_begin_0, end = var_19865_end_0, end_mask = var_19865_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19865_cast_fp16")]; + tensor var_19869_begin_0 = const()[name = tensor("op_19869_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19869_end_0 = const()[name = tensor("op_19869_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_19869_end_mask_0 = const()[name = tensor("op_19869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19869_cast_fp16 = slice_by_index(begin = var_19869_begin_0, end = var_19869_end_0, end_mask = var_19869_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19869_cast_fp16")]; + tensor var_19873_begin_0 = const()[name = tensor("op_19873_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19873_end_0 = const()[name = tensor("op_19873_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_19873_end_mask_0 = const()[name = tensor("op_19873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19873_cast_fp16 = slice_by_index(begin = var_19873_begin_0, end = var_19873_end_0, end_mask = var_19873_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19873_cast_fp16")]; + tensor var_19877_begin_0 = const()[name = tensor("op_19877_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19877_end_0 = const()[name = tensor("op_19877_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_19877_end_mask_0 = const()[name = tensor("op_19877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19877_cast_fp16 = slice_by_index(begin = var_19877_begin_0, end = var_19877_end_0, end_mask = var_19877_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19877_cast_fp16")]; + tensor var_19881_begin_0 = const()[name = tensor("op_19881_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19881_end_0 = const()[name = tensor("op_19881_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_19881_end_mask_0 = const()[name = tensor("op_19881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19881_cast_fp16 = slice_by_index(begin = var_19881_begin_0, end = var_19881_end_0, end_mask = var_19881_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19881_cast_fp16")]; + tensor var_19885_begin_0 = const()[name = tensor("op_19885_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19885_end_0 = const()[name = tensor("op_19885_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_19885_end_mask_0 = const()[name = tensor("op_19885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19885_cast_fp16 = slice_by_index(begin = var_19885_begin_0, end = var_19885_end_0, end_mask = var_19885_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19885_cast_fp16")]; + tensor var_19889_begin_0 = const()[name = tensor("op_19889_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19889_end_0 = const()[name = tensor("op_19889_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_19889_end_mask_0 = const()[name = tensor("op_19889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19889_cast_fp16 = slice_by_index(begin = var_19889_begin_0, end = var_19889_end_0, end_mask = var_19889_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19889_cast_fp16")]; + tensor var_19893_begin_0 = const()[name = tensor("op_19893_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19893_end_0 = const()[name = tensor("op_19893_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_19893_end_mask_0 = const()[name = tensor("op_19893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19893_cast_fp16 = slice_by_index(begin = var_19893_begin_0, end = var_19893_end_0, end_mask = var_19893_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19893_cast_fp16")]; + tensor var_19897_begin_0 = const()[name = tensor("op_19897_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19897_end_0 = const()[name = tensor("op_19897_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_19897_end_mask_0 = const()[name = tensor("op_19897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19897_cast_fp16 = slice_by_index(begin = var_19897_begin_0, end = var_19897_end_0, end_mask = var_19897_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19897_cast_fp16")]; + tensor var_19901_begin_0 = const()[name = tensor("op_19901_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19901_end_0 = const()[name = tensor("op_19901_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_19901_end_mask_0 = const()[name = tensor("op_19901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19901_cast_fp16 = slice_by_index(begin = var_19901_begin_0, end = var_19901_end_0, end_mask = var_19901_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19901_cast_fp16")]; + tensor var_19905_begin_0 = const()[name = tensor("op_19905_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19905_end_0 = const()[name = tensor("op_19905_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_19905_end_mask_0 = const()[name = tensor("op_19905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19905_cast_fp16 = slice_by_index(begin = var_19905_begin_0, end = var_19905_end_0, end_mask = var_19905_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19905_cast_fp16")]; + tensor var_19909_begin_0 = const()[name = tensor("op_19909_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19909_end_0 = const()[name = tensor("op_19909_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_19909_end_mask_0 = const()[name = tensor("op_19909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19909_cast_fp16 = slice_by_index(begin = var_19909_begin_0, end = var_19909_end_0, end_mask = var_19909_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19909_cast_fp16")]; + tensor var_19913_begin_0 = const()[name = tensor("op_19913_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19913_end_0 = const()[name = tensor("op_19913_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_19913_end_mask_0 = const()[name = tensor("op_19913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19913_cast_fp16 = slice_by_index(begin = var_19913_begin_0, end = var_19913_end_0, end_mask = var_19913_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19913_cast_fp16")]; + tensor var_19917_begin_0 = const()[name = tensor("op_19917_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19917_end_0 = const()[name = tensor("op_19917_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_19917_end_mask_0 = const()[name = tensor("op_19917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19917_cast_fp16 = slice_by_index(begin = var_19917_begin_0, end = var_19917_end_0, end_mask = var_19917_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19917_cast_fp16")]; + tensor var_19921_begin_0 = const()[name = tensor("op_19921_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19921_end_0 = const()[name = tensor("op_19921_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_19921_end_mask_0 = const()[name = tensor("op_19921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19921_cast_fp16 = slice_by_index(begin = var_19921_begin_0, end = var_19921_end_0, end_mask = var_19921_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19921_cast_fp16")]; + tensor var_19925_begin_0 = const()[name = tensor("op_19925_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19925_end_0 = const()[name = tensor("op_19925_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_19925_end_mask_0 = const()[name = tensor("op_19925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19925_cast_fp16 = slice_by_index(begin = var_19925_begin_0, end = var_19925_end_0, end_mask = var_19925_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19925_cast_fp16")]; + tensor var_19929_equation_0 = const()[name = tensor("op_19929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19929_cast_fp16 = einsum(equation = var_19929_equation_0, values = (var_19771_cast_fp16, var_19688_cast_fp16))[name = tensor("op_19929_cast_fp16")]; + tensor var_19930_to_fp16 = const()[name = tensor("op_19930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1641_cast_fp16 = mul(x = var_19929_cast_fp16, y = var_19930_to_fp16)[name = tensor("aw_1641_cast_fp16")]; + tensor var_19933_equation_0 = const()[name = tensor("op_19933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19933_cast_fp16 = einsum(equation = var_19933_equation_0, values = (var_19775_cast_fp16, var_19692_cast_fp16))[name = tensor("op_19933_cast_fp16")]; + tensor var_19934_to_fp16 = const()[name = tensor("op_19934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1643_cast_fp16 = mul(x = var_19933_cast_fp16, y = var_19934_to_fp16)[name = tensor("aw_1643_cast_fp16")]; + tensor var_19937_equation_0 = const()[name = tensor("op_19937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19937_cast_fp16 = einsum(equation = var_19937_equation_0, values = (var_19779_cast_fp16, var_19696_cast_fp16))[name = tensor("op_19937_cast_fp16")]; + tensor var_19938_to_fp16 = const()[name = tensor("op_19938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1645_cast_fp16 = mul(x = var_19937_cast_fp16, y = var_19938_to_fp16)[name = tensor("aw_1645_cast_fp16")]; + tensor var_19941_equation_0 = const()[name = tensor("op_19941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19941_cast_fp16 = einsum(equation = var_19941_equation_0, values = (var_19783_cast_fp16, var_19700_cast_fp16))[name = tensor("op_19941_cast_fp16")]; + tensor var_19942_to_fp16 = const()[name = tensor("op_19942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1647_cast_fp16 = mul(x = var_19941_cast_fp16, y = var_19942_to_fp16)[name = tensor("aw_1647_cast_fp16")]; + tensor var_19945_equation_0 = const()[name = tensor("op_19945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19945_cast_fp16 = einsum(equation = var_19945_equation_0, values = (var_19787_cast_fp16, var_19704_cast_fp16))[name = tensor("op_19945_cast_fp16")]; + tensor var_19946_to_fp16 = const()[name = tensor("op_19946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1649_cast_fp16 = mul(x = var_19945_cast_fp16, y = var_19946_to_fp16)[name = tensor("aw_1649_cast_fp16")]; + tensor var_19949_equation_0 = const()[name = tensor("op_19949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19949_cast_fp16 = einsum(equation = var_19949_equation_0, values = (var_19791_cast_fp16, var_19708_cast_fp16))[name = tensor("op_19949_cast_fp16")]; + tensor var_19950_to_fp16 = const()[name = tensor("op_19950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1651_cast_fp16 = mul(x = var_19949_cast_fp16, y = var_19950_to_fp16)[name = tensor("aw_1651_cast_fp16")]; + tensor var_19953_equation_0 = const()[name = tensor("op_19953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19953_cast_fp16 = einsum(equation = var_19953_equation_0, values = (var_19795_cast_fp16, var_19712_cast_fp16))[name = tensor("op_19953_cast_fp16")]; + tensor var_19954_to_fp16 = const()[name = tensor("op_19954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1653_cast_fp16 = mul(x = var_19953_cast_fp16, y = var_19954_to_fp16)[name = tensor("aw_1653_cast_fp16")]; + tensor var_19957_equation_0 = const()[name = tensor("op_19957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19957_cast_fp16 = einsum(equation = var_19957_equation_0, values = (var_19799_cast_fp16, var_19716_cast_fp16))[name = tensor("op_19957_cast_fp16")]; + tensor var_19958_to_fp16 = const()[name = tensor("op_19958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1655_cast_fp16 = mul(x = var_19957_cast_fp16, y = var_19958_to_fp16)[name = tensor("aw_1655_cast_fp16")]; + tensor var_19961_equation_0 = const()[name = tensor("op_19961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19961_cast_fp16 = einsum(equation = var_19961_equation_0, values = (var_19803_cast_fp16, var_19720_cast_fp16))[name = tensor("op_19961_cast_fp16")]; + tensor var_19962_to_fp16 = const()[name = tensor("op_19962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1657_cast_fp16 = mul(x = var_19961_cast_fp16, y = var_19962_to_fp16)[name = tensor("aw_1657_cast_fp16")]; + tensor var_19965_equation_0 = const()[name = tensor("op_19965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19965_cast_fp16 = einsum(equation = var_19965_equation_0, values = (var_19807_cast_fp16, var_19724_cast_fp16))[name = tensor("op_19965_cast_fp16")]; + tensor var_19966_to_fp16 = const()[name = tensor("op_19966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1659_cast_fp16 = mul(x = var_19965_cast_fp16, y = var_19966_to_fp16)[name = tensor("aw_1659_cast_fp16")]; + tensor var_19969_equation_0 = const()[name = tensor("op_19969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19969_cast_fp16 = einsum(equation = var_19969_equation_0, values = (var_19811_cast_fp16, var_19728_cast_fp16))[name = tensor("op_19969_cast_fp16")]; + tensor var_19970_to_fp16 = const()[name = tensor("op_19970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1661_cast_fp16 = mul(x = var_19969_cast_fp16, y = var_19970_to_fp16)[name = tensor("aw_1661_cast_fp16")]; + tensor var_19973_equation_0 = const()[name = tensor("op_19973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19973_cast_fp16 = einsum(equation = var_19973_equation_0, values = (var_19815_cast_fp16, var_19732_cast_fp16))[name = tensor("op_19973_cast_fp16")]; + tensor var_19974_to_fp16 = const()[name = tensor("op_19974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1663_cast_fp16 = mul(x = var_19973_cast_fp16, y = var_19974_to_fp16)[name = tensor("aw_1663_cast_fp16")]; + tensor var_19977_equation_0 = const()[name = tensor("op_19977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19977_cast_fp16 = einsum(equation = var_19977_equation_0, values = (var_19819_cast_fp16, var_19736_cast_fp16))[name = tensor("op_19977_cast_fp16")]; + tensor var_19978_to_fp16 = const()[name = tensor("op_19978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1665_cast_fp16 = mul(x = var_19977_cast_fp16, y = var_19978_to_fp16)[name = tensor("aw_1665_cast_fp16")]; + tensor var_19981_equation_0 = const()[name = tensor("op_19981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19981_cast_fp16 = einsum(equation = var_19981_equation_0, values = (var_19823_cast_fp16, var_19740_cast_fp16))[name = tensor("op_19981_cast_fp16")]; + tensor var_19982_to_fp16 = const()[name = tensor("op_19982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1667_cast_fp16 = mul(x = var_19981_cast_fp16, y = var_19982_to_fp16)[name = tensor("aw_1667_cast_fp16")]; + tensor var_19985_equation_0 = const()[name = tensor("op_19985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19985_cast_fp16 = einsum(equation = var_19985_equation_0, values = (var_19827_cast_fp16, var_19744_cast_fp16))[name = tensor("op_19985_cast_fp16")]; + tensor var_19986_to_fp16 = const()[name = tensor("op_19986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1669_cast_fp16 = mul(x = var_19985_cast_fp16, y = var_19986_to_fp16)[name = tensor("aw_1669_cast_fp16")]; + tensor var_19989_equation_0 = const()[name = tensor("op_19989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19989_cast_fp16 = einsum(equation = var_19989_equation_0, values = (var_19831_cast_fp16, var_19748_cast_fp16))[name = tensor("op_19989_cast_fp16")]; + tensor var_19990_to_fp16 = const()[name = tensor("op_19990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1671_cast_fp16 = mul(x = var_19989_cast_fp16, y = var_19990_to_fp16)[name = tensor("aw_1671_cast_fp16")]; + tensor var_19993_equation_0 = const()[name = tensor("op_19993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19993_cast_fp16 = einsum(equation = var_19993_equation_0, values = (var_19835_cast_fp16, var_19752_cast_fp16))[name = tensor("op_19993_cast_fp16")]; + tensor var_19994_to_fp16 = const()[name = tensor("op_19994_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1673_cast_fp16 = mul(x = var_19993_cast_fp16, y = var_19994_to_fp16)[name = tensor("aw_1673_cast_fp16")]; + tensor var_19997_equation_0 = const()[name = tensor("op_19997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19997_cast_fp16 = einsum(equation = var_19997_equation_0, values = (var_19839_cast_fp16, var_19756_cast_fp16))[name = tensor("op_19997_cast_fp16")]; + tensor var_19998_to_fp16 = const()[name = tensor("op_19998_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1675_cast_fp16 = mul(x = var_19997_cast_fp16, y = var_19998_to_fp16)[name = tensor("aw_1675_cast_fp16")]; + tensor var_20001_equation_0 = const()[name = tensor("op_20001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20001_cast_fp16 = einsum(equation = var_20001_equation_0, values = (var_19843_cast_fp16, var_19760_cast_fp16))[name = tensor("op_20001_cast_fp16")]; + tensor var_20002_to_fp16 = const()[name = tensor("op_20002_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1677_cast_fp16 = mul(x = var_20001_cast_fp16, y = var_20002_to_fp16)[name = tensor("aw_1677_cast_fp16")]; + tensor var_20005_equation_0 = const()[name = tensor("op_20005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20005_cast_fp16 = einsum(equation = var_20005_equation_0, values = (var_19847_cast_fp16, var_19764_cast_fp16))[name = tensor("op_20005_cast_fp16")]; + tensor var_20006_to_fp16 = const()[name = tensor("op_20006_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1679_cast_fp16 = mul(x = var_20005_cast_fp16, y = var_20006_to_fp16)[name = tensor("aw_1679_cast_fp16")]; + tensor var_20008_cast_fp16 = softmax(axis = var_2624, x = aw_1641_cast_fp16)[name = tensor("op_20008_cast_fp16")]; + tensor var_20009_cast_fp16 = softmax(axis = var_2624, x = aw_1643_cast_fp16)[name = tensor("op_20009_cast_fp16")]; + tensor var_20010_cast_fp16 = softmax(axis = var_2624, x = aw_1645_cast_fp16)[name = tensor("op_20010_cast_fp16")]; + tensor var_20011_cast_fp16 = softmax(axis = var_2624, x = aw_1647_cast_fp16)[name = tensor("op_20011_cast_fp16")]; + tensor var_20012_cast_fp16 = softmax(axis = var_2624, x = aw_1649_cast_fp16)[name = tensor("op_20012_cast_fp16")]; + tensor var_20013_cast_fp16 = softmax(axis = var_2624, x = aw_1651_cast_fp16)[name = tensor("op_20013_cast_fp16")]; + tensor var_20014_cast_fp16 = softmax(axis = var_2624, x = aw_1653_cast_fp16)[name = tensor("op_20014_cast_fp16")]; + tensor var_20015_cast_fp16 = softmax(axis = var_2624, x = aw_1655_cast_fp16)[name = tensor("op_20015_cast_fp16")]; + tensor var_20016_cast_fp16 = softmax(axis = var_2624, x = aw_1657_cast_fp16)[name = tensor("op_20016_cast_fp16")]; + tensor var_20017_cast_fp16 = softmax(axis = var_2624, x = aw_1659_cast_fp16)[name = tensor("op_20017_cast_fp16")]; + tensor var_20018_cast_fp16 = softmax(axis = var_2624, x = aw_1661_cast_fp16)[name = tensor("op_20018_cast_fp16")]; + tensor var_20019_cast_fp16 = softmax(axis = var_2624, x = aw_1663_cast_fp16)[name = tensor("op_20019_cast_fp16")]; + tensor var_20020_cast_fp16 = softmax(axis = var_2624, x = aw_1665_cast_fp16)[name = tensor("op_20020_cast_fp16")]; + tensor var_20021_cast_fp16 = softmax(axis = var_2624, x = aw_1667_cast_fp16)[name = tensor("op_20021_cast_fp16")]; + tensor var_20022_cast_fp16 = softmax(axis = var_2624, x = aw_1669_cast_fp16)[name = tensor("op_20022_cast_fp16")]; + tensor var_20023_cast_fp16 = softmax(axis = var_2624, x = aw_1671_cast_fp16)[name = tensor("op_20023_cast_fp16")]; + tensor var_20024_cast_fp16 = softmax(axis = var_2624, x = aw_1673_cast_fp16)[name = tensor("op_20024_cast_fp16")]; + tensor var_20025_cast_fp16 = softmax(axis = var_2624, x = aw_1675_cast_fp16)[name = tensor("op_20025_cast_fp16")]; + tensor var_20026_cast_fp16 = softmax(axis = var_2624, x = aw_1677_cast_fp16)[name = tensor("op_20026_cast_fp16")]; + tensor var_20027_cast_fp16 = softmax(axis = var_2624, x = aw_1679_cast_fp16)[name = tensor("op_20027_cast_fp16")]; + tensor var_20029_equation_0 = const()[name = tensor("op_20029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20029_cast_fp16 = einsum(equation = var_20029_equation_0, values = (var_19849_cast_fp16, var_20008_cast_fp16))[name = tensor("op_20029_cast_fp16")]; + tensor var_20031_equation_0 = const()[name = tensor("op_20031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20031_cast_fp16 = einsum(equation = var_20031_equation_0, values = (var_19853_cast_fp16, var_20009_cast_fp16))[name = tensor("op_20031_cast_fp16")]; + tensor var_20033_equation_0 = const()[name = tensor("op_20033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20033_cast_fp16 = einsum(equation = var_20033_equation_0, values = (var_19857_cast_fp16, var_20010_cast_fp16))[name = tensor("op_20033_cast_fp16")]; + tensor var_20035_equation_0 = const()[name = tensor("op_20035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20035_cast_fp16 = einsum(equation = var_20035_equation_0, values = (var_19861_cast_fp16, var_20011_cast_fp16))[name = tensor("op_20035_cast_fp16")]; + tensor var_20037_equation_0 = const()[name = tensor("op_20037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20037_cast_fp16 = einsum(equation = var_20037_equation_0, values = (var_19865_cast_fp16, var_20012_cast_fp16))[name = tensor("op_20037_cast_fp16")]; + tensor var_20039_equation_0 = const()[name = tensor("op_20039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20039_cast_fp16 = einsum(equation = var_20039_equation_0, values = (var_19869_cast_fp16, var_20013_cast_fp16))[name = tensor("op_20039_cast_fp16")]; + tensor var_20041_equation_0 = const()[name = tensor("op_20041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20041_cast_fp16 = einsum(equation = var_20041_equation_0, values = (var_19873_cast_fp16, var_20014_cast_fp16))[name = tensor("op_20041_cast_fp16")]; + tensor var_20043_equation_0 = const()[name = tensor("op_20043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20043_cast_fp16 = einsum(equation = var_20043_equation_0, values = (var_19877_cast_fp16, var_20015_cast_fp16))[name = tensor("op_20043_cast_fp16")]; + tensor var_20045_equation_0 = const()[name = tensor("op_20045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20045_cast_fp16 = einsum(equation = var_20045_equation_0, values = (var_19881_cast_fp16, var_20016_cast_fp16))[name = tensor("op_20045_cast_fp16")]; + tensor var_20047_equation_0 = const()[name = tensor("op_20047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20047_cast_fp16 = einsum(equation = var_20047_equation_0, values = (var_19885_cast_fp16, var_20017_cast_fp16))[name = tensor("op_20047_cast_fp16")]; + tensor var_20049_equation_0 = const()[name = tensor("op_20049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20049_cast_fp16 = einsum(equation = var_20049_equation_0, values = (var_19889_cast_fp16, var_20018_cast_fp16))[name = tensor("op_20049_cast_fp16")]; + tensor var_20051_equation_0 = const()[name = tensor("op_20051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20051_cast_fp16 = einsum(equation = var_20051_equation_0, values = (var_19893_cast_fp16, var_20019_cast_fp16))[name = tensor("op_20051_cast_fp16")]; + tensor var_20053_equation_0 = const()[name = tensor("op_20053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20053_cast_fp16 = einsum(equation = var_20053_equation_0, values = (var_19897_cast_fp16, var_20020_cast_fp16))[name = tensor("op_20053_cast_fp16")]; + tensor var_20055_equation_0 = const()[name = tensor("op_20055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20055_cast_fp16 = einsum(equation = var_20055_equation_0, values = (var_19901_cast_fp16, var_20021_cast_fp16))[name = tensor("op_20055_cast_fp16")]; + tensor var_20057_equation_0 = const()[name = tensor("op_20057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20057_cast_fp16 = einsum(equation = var_20057_equation_0, values = (var_19905_cast_fp16, var_20022_cast_fp16))[name = tensor("op_20057_cast_fp16")]; + tensor var_20059_equation_0 = const()[name = tensor("op_20059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20059_cast_fp16 = einsum(equation = var_20059_equation_0, values = (var_19909_cast_fp16, var_20023_cast_fp16))[name = tensor("op_20059_cast_fp16")]; + tensor var_20061_equation_0 = const()[name = tensor("op_20061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20061_cast_fp16 = einsum(equation = var_20061_equation_0, values = (var_19913_cast_fp16, var_20024_cast_fp16))[name = tensor("op_20061_cast_fp16")]; + tensor var_20063_equation_0 = const()[name = tensor("op_20063_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20063_cast_fp16 = einsum(equation = var_20063_equation_0, values = (var_19917_cast_fp16, var_20025_cast_fp16))[name = tensor("op_20063_cast_fp16")]; + tensor var_20065_equation_0 = const()[name = tensor("op_20065_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20065_cast_fp16 = einsum(equation = var_20065_equation_0, values = (var_19921_cast_fp16, var_20026_cast_fp16))[name = tensor("op_20065_cast_fp16")]; + tensor var_20067_equation_0 = const()[name = tensor("op_20067_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20067_cast_fp16 = einsum(equation = var_20067_equation_0, values = (var_19925_cast_fp16, var_20027_cast_fp16))[name = tensor("op_20067_cast_fp16")]; + tensor input_295_interleave_0 = const()[name = tensor("input_295_interleave_0"), val = tensor(false)]; + tensor input_295_cast_fp16 = concat(axis = var_2624, interleave = input_295_interleave_0, values = (var_20029_cast_fp16, var_20031_cast_fp16, var_20033_cast_fp16, var_20035_cast_fp16, var_20037_cast_fp16, var_20039_cast_fp16, var_20041_cast_fp16, var_20043_cast_fp16, var_20045_cast_fp16, var_20047_cast_fp16, var_20049_cast_fp16, var_20051_cast_fp16, var_20053_cast_fp16, var_20055_cast_fp16, var_20057_cast_fp16, var_20059_cast_fp16, var_20061_cast_fp16, var_20063_cast_fp16, var_20065_cast_fp16, var_20067_cast_fp16))[name = tensor("input_295_cast_fp16")]; + tensor var_20077_pad_type_0 = const()[name = tensor("op_20077_pad_type_0"), val = tensor("valid")]; + tensor var_20077_strides_0 = const()[name = tensor("op_20077_strides_0"), val = tensor([1, 1])]; + tensor var_20077_pad_0 = const()[name = tensor("op_20077_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20077_dilations_0 = const()[name = tensor("op_20077_dilations_0"), val = tensor([1, 1])]; + tensor var_20077_groups_0 = const()[name = tensor("op_20077_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580116992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581345856))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581346048)))]; + tensor var_20077_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_20077_dilations_0, groups = var_20077_groups_0, pad = var_20077_pad_0, pad_type = var_20077_pad_type_0, strides = var_20077_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized, x = input_295_cast_fp16)[name = tensor("op_20077_cast_fp16")]; + tensor inputs_137_cast_fp16 = add(x = var_20077_cast_fp16, y = inputs_135_cast_fp16)[name = tensor("inputs_137_cast_fp16")]; + tensor input_297_axes_0 = const()[name = tensor("input_297_axes_0"), val = tensor([1])]; + tensor input_297_gamma_0_to_fp16 = const()[name = tensor("input_297_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581348672)))]; + tensor input_297_beta_0_to_fp16 = const()[name = tensor("input_297_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581351296)))]; + tensor var_20087_to_fp16 = const()[name = tensor("op_20087_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_297_cast_fp16 = layer_norm(axes = input_297_axes_0, beta = input_297_beta_0_to_fp16, epsilon = var_20087_to_fp16, gamma = input_297_gamma_0_to_fp16, x = inputs_137_cast_fp16)[name = tensor("input_297_cast_fp16")]; + tensor var_20107_pad_type_0 = const()[name = tensor("op_20107_pad_type_0"), val = tensor("valid")]; + tensor var_20107_strides_0 = const()[name = tensor("op_20107_strides_0"), val = tensor([1, 1])]; + tensor var_20107_pad_0 = const()[name = tensor("op_20107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20107_dilations_0 = const()[name = tensor("op_20107_dilations_0"), val = tensor([1, 1])]; + tensor var_20107_groups_0 = const()[name = tensor("op_20107_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581353920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591184384))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591184576)))]; + tensor var_20107_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_20107_dilations_0, groups = var_20107_groups_0, pad = var_20107_pad_0, pad_type = var_20107_pad_type_0, strides = var_20107_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized, x = input_297_cast_fp16)[name = tensor("op_20107_cast_fp16")]; + tensor var_20108_split_sizes_0 = const()[name = tensor("op_20108_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_20108_axis_0 = const()[name = tensor("op_20108_axis_0"), val = tensor(1)]; + tensor var_20108_cast_fp16_0, tensor var_20108_cast_fp16_1 = split(axis = var_20108_axis_0, split_sizes = var_20108_split_sizes_0, x = var_20107_cast_fp16)[name = tensor("op_20108_cast_fp16")]; + tensor var_20110_mode_0 = const()[name = tensor("op_20110_mode_0"), val = tensor("EXACT")]; + tensor var_20110_cast_fp16 = gelu(mode = var_20110_mode_0, x = var_20108_cast_fp16_1)[name = tensor("op_20110_cast_fp16")]; + tensor input_299_cast_fp16 = mul(x = var_20108_cast_fp16_0, y = var_20110_cast_fp16)[name = tensor("input_299_cast_fp16")]; + tensor var_20118_pad_type_0 = const()[name = tensor("op_20118_pad_type_0"), val = tensor("valid")]; + tensor var_20118_strides_0 = const()[name = tensor("op_20118_strides_0"), val = tensor([1, 1])]; + tensor var_20118_pad_0 = const()[name = tensor("op_20118_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20118_dilations_0 = const()[name = tensor("op_20118_dilations_0"), val = tensor([1, 1])]; + tensor var_20118_groups_0 = const()[name = tensor("op_20118_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591205120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596120384))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596120576)))]; + tensor var_20118_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_20118_dilations_0, groups = var_20118_groups_0, pad = var_20118_pad_0, pad_type = var_20118_pad_type_0, strides = var_20118_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized, x = input_299_cast_fp16)[name = tensor("op_20118_cast_fp16")]; + tensor inputs_139_cast_fp16 = add(x = var_20118_cast_fp16, y = inputs_137_cast_fp16)[name = tensor("inputs_139_cast_fp16")]; + tensor hidden_states_191_axes_0 = const()[name = tensor("hidden_states_191_axes_0"), val = tensor([1])]; + tensor hidden_states_191_gamma_0_to_fp16 = const()[name = tensor("hidden_states_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596123200)))]; + tensor hidden_states_191_beta_0_to_fp16 = const()[name = tensor("hidden_states_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596125824)))]; + tensor var_20134_to_fp16 = const()[name = tensor("op_20134_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_191_cast_fp16 = layer_norm(axes = hidden_states_191_axes_0, beta = hidden_states_191_beta_0_to_fp16, epsilon = var_20134_to_fp16, gamma = hidden_states_191_gamma_0_to_fp16, x = inputs_139_cast_fp16)[name = tensor("hidden_states_191_cast_fp16")]; + tensor q_93_pad_type_0 = const()[name = tensor("q_93_pad_type_0"), val = tensor("valid")]; + tensor q_93_strides_0 = const()[name = tensor("q_93_strides_0"), val = tensor([1, 1])]; + tensor q_93_pad_0 = const()[name = tensor("q_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_93_dilations_0 = const()[name = tensor("q_93_dilations_0"), val = tensor([1, 1])]; + tensor q_93_groups_0 = const()[name = tensor("q_93_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596128448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597357312))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_93_cast_fp16 = conv(dilations = q_93_dilations_0, groups = q_93_groups_0, pad = q_93_pad_0, pad_type = q_93_pad_type_0, strides = q_93_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_191_cast_fp16)[name = tensor("q_93_cast_fp16")]; + tensor k_185_pad_type_0 = const()[name = tensor("k_185_pad_type_0"), val = tensor("valid")]; + tensor k_185_strides_0 = const()[name = tensor("k_185_strides_0"), val = tensor([1, 1])]; + tensor k_185_pad_0 = const()[name = tensor("k_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_185_dilations_0 = const()[name = tensor("k_185_dilations_0"), val = tensor([1, 1])]; + tensor k_185_groups_0 = const()[name = tensor("k_185_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597357504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(598586368))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_185_cast_fp16 = conv(dilations = k_185_dilations_0, groups = k_185_groups_0, pad = k_185_pad_0, pad_type = k_185_pad_type_0, strides = k_185_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_191_cast_fp16)[name = tensor("k_185_cast_fp16")]; + tensor v_93_pad_type_0 = const()[name = tensor("v_93_pad_type_0"), val = tensor("valid")]; + tensor v_93_strides_0 = const()[name = tensor("v_93_strides_0"), val = tensor([1, 1])]; + tensor v_93_pad_0 = const()[name = tensor("v_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_93_dilations_0 = const()[name = tensor("v_93_dilations_0"), val = tensor([1, 1])]; + tensor v_93_groups_0 = const()[name = tensor("v_93_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(598586560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(599815424))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_93_cast_fp16 = conv(dilations = v_93_dilations_0, groups = v_93_groups_0, pad = v_93_pad_0, pad_type = v_93_pad_type_0, strides = v_93_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_191_cast_fp16)[name = tensor("v_93_cast_fp16")]; + tensor var_20167_begin_0 = const()[name = tensor("op_20167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20167_end_0 = const()[name = tensor("op_20167_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_20167_end_mask_0 = const()[name = tensor("op_20167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20167_cast_fp16 = slice_by_index(begin = var_20167_begin_0, end = var_20167_end_0, end_mask = var_20167_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20167_cast_fp16")]; + tensor var_20171_begin_0 = const()[name = tensor("op_20171_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20171_end_0 = const()[name = tensor("op_20171_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_20171_end_mask_0 = const()[name = tensor("op_20171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20171_cast_fp16 = slice_by_index(begin = var_20171_begin_0, end = var_20171_end_0, end_mask = var_20171_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20171_cast_fp16")]; + tensor var_20175_begin_0 = const()[name = tensor("op_20175_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20175_end_0 = const()[name = tensor("op_20175_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_20175_end_mask_0 = const()[name = tensor("op_20175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20175_cast_fp16 = slice_by_index(begin = var_20175_begin_0, end = var_20175_end_0, end_mask = var_20175_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20175_cast_fp16")]; + tensor var_20179_begin_0 = const()[name = tensor("op_20179_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20179_end_0 = const()[name = tensor("op_20179_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_20179_end_mask_0 = const()[name = tensor("op_20179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20179_cast_fp16 = slice_by_index(begin = var_20179_begin_0, end = var_20179_end_0, end_mask = var_20179_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20179_cast_fp16")]; + tensor var_20183_begin_0 = const()[name = tensor("op_20183_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20183_end_0 = const()[name = tensor("op_20183_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_20183_end_mask_0 = const()[name = tensor("op_20183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20183_cast_fp16 = slice_by_index(begin = var_20183_begin_0, end = var_20183_end_0, end_mask = var_20183_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20183_cast_fp16")]; + tensor var_20187_begin_0 = const()[name = tensor("op_20187_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20187_end_0 = const()[name = tensor("op_20187_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_20187_end_mask_0 = const()[name = tensor("op_20187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20187_cast_fp16 = slice_by_index(begin = var_20187_begin_0, end = var_20187_end_0, end_mask = var_20187_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20187_cast_fp16")]; + tensor var_20191_begin_0 = const()[name = tensor("op_20191_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20191_end_0 = const()[name = tensor("op_20191_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_20191_end_mask_0 = const()[name = tensor("op_20191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20191_cast_fp16 = slice_by_index(begin = var_20191_begin_0, end = var_20191_end_0, end_mask = var_20191_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20191_cast_fp16")]; + tensor var_20195_begin_0 = const()[name = tensor("op_20195_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20195_end_0 = const()[name = tensor("op_20195_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_20195_end_mask_0 = const()[name = tensor("op_20195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20195_cast_fp16 = slice_by_index(begin = var_20195_begin_0, end = var_20195_end_0, end_mask = var_20195_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20195_cast_fp16")]; + tensor var_20199_begin_0 = const()[name = tensor("op_20199_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20199_end_0 = const()[name = tensor("op_20199_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_20199_end_mask_0 = const()[name = tensor("op_20199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20199_cast_fp16 = slice_by_index(begin = var_20199_begin_0, end = var_20199_end_0, end_mask = var_20199_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20199_cast_fp16")]; + tensor var_20203_begin_0 = const()[name = tensor("op_20203_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20203_end_0 = const()[name = tensor("op_20203_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_20203_end_mask_0 = const()[name = tensor("op_20203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20203_cast_fp16 = slice_by_index(begin = var_20203_begin_0, end = var_20203_end_0, end_mask = var_20203_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20203_cast_fp16")]; + tensor var_20207_begin_0 = const()[name = tensor("op_20207_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20207_end_0 = const()[name = tensor("op_20207_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_20207_end_mask_0 = const()[name = tensor("op_20207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20207_cast_fp16 = slice_by_index(begin = var_20207_begin_0, end = var_20207_end_0, end_mask = var_20207_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20207_cast_fp16")]; + tensor var_20211_begin_0 = const()[name = tensor("op_20211_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20211_end_0 = const()[name = tensor("op_20211_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_20211_end_mask_0 = const()[name = tensor("op_20211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20211_cast_fp16 = slice_by_index(begin = var_20211_begin_0, end = var_20211_end_0, end_mask = var_20211_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20211_cast_fp16")]; + tensor var_20215_begin_0 = const()[name = tensor("op_20215_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20215_end_0 = const()[name = tensor("op_20215_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_20215_end_mask_0 = const()[name = tensor("op_20215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20215_cast_fp16 = slice_by_index(begin = var_20215_begin_0, end = var_20215_end_0, end_mask = var_20215_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20215_cast_fp16")]; + tensor var_20219_begin_0 = const()[name = tensor("op_20219_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20219_end_0 = const()[name = tensor("op_20219_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_20219_end_mask_0 = const()[name = tensor("op_20219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20219_cast_fp16 = slice_by_index(begin = var_20219_begin_0, end = var_20219_end_0, end_mask = var_20219_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20219_cast_fp16")]; + tensor var_20223_begin_0 = const()[name = tensor("op_20223_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20223_end_0 = const()[name = tensor("op_20223_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_20223_end_mask_0 = const()[name = tensor("op_20223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20223_cast_fp16 = slice_by_index(begin = var_20223_begin_0, end = var_20223_end_0, end_mask = var_20223_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20223_cast_fp16")]; + tensor var_20227_begin_0 = const()[name = tensor("op_20227_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20227_end_0 = const()[name = tensor("op_20227_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20227_end_mask_0 = const()[name = tensor("op_20227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20227_cast_fp16 = slice_by_index(begin = var_20227_begin_0, end = var_20227_end_0, end_mask = var_20227_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20227_cast_fp16")]; + tensor var_20231_begin_0 = const()[name = tensor("op_20231_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20231_end_0 = const()[name = tensor("op_20231_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_20231_end_mask_0 = const()[name = tensor("op_20231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20231_cast_fp16 = slice_by_index(begin = var_20231_begin_0, end = var_20231_end_0, end_mask = var_20231_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20231_cast_fp16")]; + tensor var_20235_begin_0 = const()[name = tensor("op_20235_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20235_end_0 = const()[name = tensor("op_20235_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_20235_end_mask_0 = const()[name = tensor("op_20235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20235_cast_fp16 = slice_by_index(begin = var_20235_begin_0, end = var_20235_end_0, end_mask = var_20235_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20235_cast_fp16")]; + tensor var_20239_begin_0 = const()[name = tensor("op_20239_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20239_end_0 = const()[name = tensor("op_20239_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_20239_end_mask_0 = const()[name = tensor("op_20239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20239_cast_fp16 = slice_by_index(begin = var_20239_begin_0, end = var_20239_end_0, end_mask = var_20239_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20239_cast_fp16")]; + tensor var_20243_begin_0 = const()[name = tensor("op_20243_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20243_end_0 = const()[name = tensor("op_20243_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_20243_end_mask_0 = const()[name = tensor("op_20243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20243_cast_fp16 = slice_by_index(begin = var_20243_begin_0, end = var_20243_end_0, end_mask = var_20243_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20243_cast_fp16")]; + tensor k_187_perm_0 = const()[name = tensor("k_187_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_20250_begin_0 = const()[name = tensor("op_20250_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20250_end_0 = const()[name = tensor("op_20250_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_20250_end_mask_0 = const()[name = tensor("op_20250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_187_cast_fp16 = transpose(perm = k_187_perm_0, x = k_185_cast_fp16)[name = tensor("transpose_21")]; + tensor var_20250_cast_fp16 = slice_by_index(begin = var_20250_begin_0, end = var_20250_end_0, end_mask = var_20250_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20250_cast_fp16")]; + tensor var_20254_begin_0 = const()[name = tensor("op_20254_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_20254_end_0 = const()[name = tensor("op_20254_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_20254_end_mask_0 = const()[name = tensor("op_20254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20254_cast_fp16 = slice_by_index(begin = var_20254_begin_0, end = var_20254_end_0, end_mask = var_20254_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20254_cast_fp16")]; + tensor var_20258_begin_0 = const()[name = tensor("op_20258_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_20258_end_0 = const()[name = tensor("op_20258_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_20258_end_mask_0 = const()[name = tensor("op_20258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20258_cast_fp16 = slice_by_index(begin = var_20258_begin_0, end = var_20258_end_0, end_mask = var_20258_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20258_cast_fp16")]; + tensor var_20262_begin_0 = const()[name = tensor("op_20262_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_20262_end_0 = const()[name = tensor("op_20262_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_20262_end_mask_0 = const()[name = tensor("op_20262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20262_cast_fp16 = slice_by_index(begin = var_20262_begin_0, end = var_20262_end_0, end_mask = var_20262_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20262_cast_fp16")]; + tensor var_20266_begin_0 = const()[name = tensor("op_20266_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_20266_end_0 = const()[name = tensor("op_20266_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_20266_end_mask_0 = const()[name = tensor("op_20266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20266_cast_fp16 = slice_by_index(begin = var_20266_begin_0, end = var_20266_end_0, end_mask = var_20266_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20266_cast_fp16")]; + tensor var_20270_begin_0 = const()[name = tensor("op_20270_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_20270_end_0 = const()[name = tensor("op_20270_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_20270_end_mask_0 = const()[name = tensor("op_20270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20270_cast_fp16 = slice_by_index(begin = var_20270_begin_0, end = var_20270_end_0, end_mask = var_20270_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20270_cast_fp16")]; + tensor var_20274_begin_0 = const()[name = tensor("op_20274_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_20274_end_0 = const()[name = tensor("op_20274_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_20274_end_mask_0 = const()[name = tensor("op_20274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20274_cast_fp16 = slice_by_index(begin = var_20274_begin_0, end = var_20274_end_0, end_mask = var_20274_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20274_cast_fp16")]; + tensor var_20278_begin_0 = const()[name = tensor("op_20278_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_20278_end_0 = const()[name = tensor("op_20278_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_20278_end_mask_0 = const()[name = tensor("op_20278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20278_cast_fp16 = slice_by_index(begin = var_20278_begin_0, end = var_20278_end_0, end_mask = var_20278_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20278_cast_fp16")]; + tensor var_20282_begin_0 = const()[name = tensor("op_20282_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_20282_end_0 = const()[name = tensor("op_20282_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_20282_end_mask_0 = const()[name = tensor("op_20282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20282_cast_fp16 = slice_by_index(begin = var_20282_begin_0, end = var_20282_end_0, end_mask = var_20282_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20282_cast_fp16")]; + tensor var_20286_begin_0 = const()[name = tensor("op_20286_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_20286_end_0 = const()[name = tensor("op_20286_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_20286_end_mask_0 = const()[name = tensor("op_20286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20286_cast_fp16 = slice_by_index(begin = var_20286_begin_0, end = var_20286_end_0, end_mask = var_20286_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20286_cast_fp16")]; + tensor var_20290_begin_0 = const()[name = tensor("op_20290_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_20290_end_0 = const()[name = tensor("op_20290_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_20290_end_mask_0 = const()[name = tensor("op_20290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20290_cast_fp16 = slice_by_index(begin = var_20290_begin_0, end = var_20290_end_0, end_mask = var_20290_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20290_cast_fp16")]; + tensor var_20294_begin_0 = const()[name = tensor("op_20294_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_20294_end_0 = const()[name = tensor("op_20294_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_20294_end_mask_0 = const()[name = tensor("op_20294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20294_cast_fp16 = slice_by_index(begin = var_20294_begin_0, end = var_20294_end_0, end_mask = var_20294_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20294_cast_fp16")]; + tensor var_20298_begin_0 = const()[name = tensor("op_20298_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_20298_end_0 = const()[name = tensor("op_20298_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_20298_end_mask_0 = const()[name = tensor("op_20298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20298_cast_fp16 = slice_by_index(begin = var_20298_begin_0, end = var_20298_end_0, end_mask = var_20298_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20298_cast_fp16")]; + tensor var_20302_begin_0 = const()[name = tensor("op_20302_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_20302_end_0 = const()[name = tensor("op_20302_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_20302_end_mask_0 = const()[name = tensor("op_20302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20302_cast_fp16 = slice_by_index(begin = var_20302_begin_0, end = var_20302_end_0, end_mask = var_20302_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20302_cast_fp16")]; + tensor var_20306_begin_0 = const()[name = tensor("op_20306_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_20306_end_0 = const()[name = tensor("op_20306_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_20306_end_mask_0 = const()[name = tensor("op_20306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20306_cast_fp16 = slice_by_index(begin = var_20306_begin_0, end = var_20306_end_0, end_mask = var_20306_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20306_cast_fp16")]; + tensor var_20310_begin_0 = const()[name = tensor("op_20310_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_20310_end_0 = const()[name = tensor("op_20310_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20310_end_mask_0 = const()[name = tensor("op_20310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20310_cast_fp16 = slice_by_index(begin = var_20310_begin_0, end = var_20310_end_0, end_mask = var_20310_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20310_cast_fp16")]; + tensor var_20314_begin_0 = const()[name = tensor("op_20314_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_20314_end_0 = const()[name = tensor("op_20314_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_20314_end_mask_0 = const()[name = tensor("op_20314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20314_cast_fp16 = slice_by_index(begin = var_20314_begin_0, end = var_20314_end_0, end_mask = var_20314_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20314_cast_fp16")]; + tensor var_20318_begin_0 = const()[name = tensor("op_20318_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_20318_end_0 = const()[name = tensor("op_20318_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_20318_end_mask_0 = const()[name = tensor("op_20318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20318_cast_fp16 = slice_by_index(begin = var_20318_begin_0, end = var_20318_end_0, end_mask = var_20318_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20318_cast_fp16")]; + tensor var_20322_begin_0 = const()[name = tensor("op_20322_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_20322_end_0 = const()[name = tensor("op_20322_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_20322_end_mask_0 = const()[name = tensor("op_20322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20322_cast_fp16 = slice_by_index(begin = var_20322_begin_0, end = var_20322_end_0, end_mask = var_20322_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20322_cast_fp16")]; + tensor var_20326_begin_0 = const()[name = tensor("op_20326_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_20326_end_0 = const()[name = tensor("op_20326_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_20326_end_mask_0 = const()[name = tensor("op_20326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20326_cast_fp16 = slice_by_index(begin = var_20326_begin_0, end = var_20326_end_0, end_mask = var_20326_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20326_cast_fp16")]; + tensor var_20328_begin_0 = const()[name = tensor("op_20328_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20328_end_0 = const()[name = tensor("op_20328_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_20328_end_mask_0 = const()[name = tensor("op_20328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20328_cast_fp16 = slice_by_index(begin = var_20328_begin_0, end = var_20328_end_0, end_mask = var_20328_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20328_cast_fp16")]; + tensor var_20332_begin_0 = const()[name = tensor("op_20332_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20332_end_0 = const()[name = tensor("op_20332_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_20332_end_mask_0 = const()[name = tensor("op_20332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20332_cast_fp16 = slice_by_index(begin = var_20332_begin_0, end = var_20332_end_0, end_mask = var_20332_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20332_cast_fp16")]; + tensor var_20336_begin_0 = const()[name = tensor("op_20336_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20336_end_0 = const()[name = tensor("op_20336_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_20336_end_mask_0 = const()[name = tensor("op_20336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20336_cast_fp16 = slice_by_index(begin = var_20336_begin_0, end = var_20336_end_0, end_mask = var_20336_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20336_cast_fp16")]; + tensor var_20340_begin_0 = const()[name = tensor("op_20340_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20340_end_0 = const()[name = tensor("op_20340_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_20340_end_mask_0 = const()[name = tensor("op_20340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20340_cast_fp16 = slice_by_index(begin = var_20340_begin_0, end = var_20340_end_0, end_mask = var_20340_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20340_cast_fp16")]; + tensor var_20344_begin_0 = const()[name = tensor("op_20344_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20344_end_0 = const()[name = tensor("op_20344_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_20344_end_mask_0 = const()[name = tensor("op_20344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20344_cast_fp16 = slice_by_index(begin = var_20344_begin_0, end = var_20344_end_0, end_mask = var_20344_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20344_cast_fp16")]; + tensor var_20348_begin_0 = const()[name = tensor("op_20348_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20348_end_0 = const()[name = tensor("op_20348_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_20348_end_mask_0 = const()[name = tensor("op_20348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20348_cast_fp16 = slice_by_index(begin = var_20348_begin_0, end = var_20348_end_0, end_mask = var_20348_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20348_cast_fp16")]; + tensor var_20352_begin_0 = const()[name = tensor("op_20352_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20352_end_0 = const()[name = tensor("op_20352_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_20352_end_mask_0 = const()[name = tensor("op_20352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20352_cast_fp16 = slice_by_index(begin = var_20352_begin_0, end = var_20352_end_0, end_mask = var_20352_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20352_cast_fp16")]; + tensor var_20356_begin_0 = const()[name = tensor("op_20356_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20356_end_0 = const()[name = tensor("op_20356_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_20356_end_mask_0 = const()[name = tensor("op_20356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20356_cast_fp16 = slice_by_index(begin = var_20356_begin_0, end = var_20356_end_0, end_mask = var_20356_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20356_cast_fp16")]; + tensor var_20360_begin_0 = const()[name = tensor("op_20360_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20360_end_0 = const()[name = tensor("op_20360_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_20360_end_mask_0 = const()[name = tensor("op_20360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20360_cast_fp16 = slice_by_index(begin = var_20360_begin_0, end = var_20360_end_0, end_mask = var_20360_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20360_cast_fp16")]; + tensor var_20364_begin_0 = const()[name = tensor("op_20364_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20364_end_0 = const()[name = tensor("op_20364_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_20364_end_mask_0 = const()[name = tensor("op_20364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20364_cast_fp16 = slice_by_index(begin = var_20364_begin_0, end = var_20364_end_0, end_mask = var_20364_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20364_cast_fp16")]; + tensor var_20368_begin_0 = const()[name = tensor("op_20368_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20368_end_0 = const()[name = tensor("op_20368_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_20368_end_mask_0 = const()[name = tensor("op_20368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20368_cast_fp16 = slice_by_index(begin = var_20368_begin_0, end = var_20368_end_0, end_mask = var_20368_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20368_cast_fp16")]; + tensor var_20372_begin_0 = const()[name = tensor("op_20372_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20372_end_0 = const()[name = tensor("op_20372_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_20372_end_mask_0 = const()[name = tensor("op_20372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20372_cast_fp16 = slice_by_index(begin = var_20372_begin_0, end = var_20372_end_0, end_mask = var_20372_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20372_cast_fp16")]; + tensor var_20376_begin_0 = const()[name = tensor("op_20376_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20376_end_0 = const()[name = tensor("op_20376_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_20376_end_mask_0 = const()[name = tensor("op_20376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20376_cast_fp16 = slice_by_index(begin = var_20376_begin_0, end = var_20376_end_0, end_mask = var_20376_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20376_cast_fp16")]; + tensor var_20380_begin_0 = const()[name = tensor("op_20380_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20380_end_0 = const()[name = tensor("op_20380_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_20380_end_mask_0 = const()[name = tensor("op_20380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20380_cast_fp16 = slice_by_index(begin = var_20380_begin_0, end = var_20380_end_0, end_mask = var_20380_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20380_cast_fp16")]; + tensor var_20384_begin_0 = const()[name = tensor("op_20384_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20384_end_0 = const()[name = tensor("op_20384_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_20384_end_mask_0 = const()[name = tensor("op_20384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20384_cast_fp16 = slice_by_index(begin = var_20384_begin_0, end = var_20384_end_0, end_mask = var_20384_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20384_cast_fp16")]; + tensor var_20388_begin_0 = const()[name = tensor("op_20388_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20388_end_0 = const()[name = tensor("op_20388_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20388_end_mask_0 = const()[name = tensor("op_20388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20388_cast_fp16 = slice_by_index(begin = var_20388_begin_0, end = var_20388_end_0, end_mask = var_20388_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20388_cast_fp16")]; + tensor var_20392_begin_0 = const()[name = tensor("op_20392_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20392_end_0 = const()[name = tensor("op_20392_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_20392_end_mask_0 = const()[name = tensor("op_20392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20392_cast_fp16 = slice_by_index(begin = var_20392_begin_0, end = var_20392_end_0, end_mask = var_20392_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20392_cast_fp16")]; + tensor var_20396_begin_0 = const()[name = tensor("op_20396_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20396_end_0 = const()[name = tensor("op_20396_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_20396_end_mask_0 = const()[name = tensor("op_20396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20396_cast_fp16 = slice_by_index(begin = var_20396_begin_0, end = var_20396_end_0, end_mask = var_20396_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20396_cast_fp16")]; + tensor var_20400_begin_0 = const()[name = tensor("op_20400_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20400_end_0 = const()[name = tensor("op_20400_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_20400_end_mask_0 = const()[name = tensor("op_20400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20400_cast_fp16 = slice_by_index(begin = var_20400_begin_0, end = var_20400_end_0, end_mask = var_20400_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20400_cast_fp16")]; + tensor var_20404_begin_0 = const()[name = tensor("op_20404_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20404_end_0 = const()[name = tensor("op_20404_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_20404_end_mask_0 = const()[name = tensor("op_20404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20404_cast_fp16 = slice_by_index(begin = var_20404_begin_0, end = var_20404_end_0, end_mask = var_20404_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20404_cast_fp16")]; + tensor var_20408_equation_0 = const()[name = tensor("op_20408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20408_cast_fp16 = einsum(equation = var_20408_equation_0, values = (var_20250_cast_fp16, var_20167_cast_fp16))[name = tensor("op_20408_cast_fp16")]; + tensor var_20409_to_fp16 = const()[name = tensor("op_20409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1681_cast_fp16 = mul(x = var_20408_cast_fp16, y = var_20409_to_fp16)[name = tensor("aw_1681_cast_fp16")]; + tensor var_20412_equation_0 = const()[name = tensor("op_20412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20412_cast_fp16 = einsum(equation = var_20412_equation_0, values = (var_20254_cast_fp16, var_20171_cast_fp16))[name = tensor("op_20412_cast_fp16")]; + tensor var_20413_to_fp16 = const()[name = tensor("op_20413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1683_cast_fp16 = mul(x = var_20412_cast_fp16, y = var_20413_to_fp16)[name = tensor("aw_1683_cast_fp16")]; + tensor var_20416_equation_0 = const()[name = tensor("op_20416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20416_cast_fp16 = einsum(equation = var_20416_equation_0, values = (var_20258_cast_fp16, var_20175_cast_fp16))[name = tensor("op_20416_cast_fp16")]; + tensor var_20417_to_fp16 = const()[name = tensor("op_20417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1685_cast_fp16 = mul(x = var_20416_cast_fp16, y = var_20417_to_fp16)[name = tensor("aw_1685_cast_fp16")]; + tensor var_20420_equation_0 = const()[name = tensor("op_20420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20420_cast_fp16 = einsum(equation = var_20420_equation_0, values = (var_20262_cast_fp16, var_20179_cast_fp16))[name = tensor("op_20420_cast_fp16")]; + tensor var_20421_to_fp16 = const()[name = tensor("op_20421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1687_cast_fp16 = mul(x = var_20420_cast_fp16, y = var_20421_to_fp16)[name = tensor("aw_1687_cast_fp16")]; + tensor var_20424_equation_0 = const()[name = tensor("op_20424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20424_cast_fp16 = einsum(equation = var_20424_equation_0, values = (var_20266_cast_fp16, var_20183_cast_fp16))[name = tensor("op_20424_cast_fp16")]; + tensor var_20425_to_fp16 = const()[name = tensor("op_20425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1689_cast_fp16 = mul(x = var_20424_cast_fp16, y = var_20425_to_fp16)[name = tensor("aw_1689_cast_fp16")]; + tensor var_20428_equation_0 = const()[name = tensor("op_20428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20428_cast_fp16 = einsum(equation = var_20428_equation_0, values = (var_20270_cast_fp16, var_20187_cast_fp16))[name = tensor("op_20428_cast_fp16")]; + tensor var_20429_to_fp16 = const()[name = tensor("op_20429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1691_cast_fp16 = mul(x = var_20428_cast_fp16, y = var_20429_to_fp16)[name = tensor("aw_1691_cast_fp16")]; + tensor var_20432_equation_0 = const()[name = tensor("op_20432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20432_cast_fp16 = einsum(equation = var_20432_equation_0, values = (var_20274_cast_fp16, var_20191_cast_fp16))[name = tensor("op_20432_cast_fp16")]; + tensor var_20433_to_fp16 = const()[name = tensor("op_20433_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1693_cast_fp16 = mul(x = var_20432_cast_fp16, y = var_20433_to_fp16)[name = tensor("aw_1693_cast_fp16")]; + tensor var_20436_equation_0 = const()[name = tensor("op_20436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20436_cast_fp16 = einsum(equation = var_20436_equation_0, values = (var_20278_cast_fp16, var_20195_cast_fp16))[name = tensor("op_20436_cast_fp16")]; + tensor var_20437_to_fp16 = const()[name = tensor("op_20437_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1695_cast_fp16 = mul(x = var_20436_cast_fp16, y = var_20437_to_fp16)[name = tensor("aw_1695_cast_fp16")]; + tensor var_20440_equation_0 = const()[name = tensor("op_20440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20440_cast_fp16 = einsum(equation = var_20440_equation_0, values = (var_20282_cast_fp16, var_20199_cast_fp16))[name = tensor("op_20440_cast_fp16")]; + tensor var_20441_to_fp16 = const()[name = tensor("op_20441_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1697_cast_fp16 = mul(x = var_20440_cast_fp16, y = var_20441_to_fp16)[name = tensor("aw_1697_cast_fp16")]; + tensor var_20444_equation_0 = const()[name = tensor("op_20444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20444_cast_fp16 = einsum(equation = var_20444_equation_0, values = (var_20286_cast_fp16, var_20203_cast_fp16))[name = tensor("op_20444_cast_fp16")]; + tensor var_20445_to_fp16 = const()[name = tensor("op_20445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1699_cast_fp16 = mul(x = var_20444_cast_fp16, y = var_20445_to_fp16)[name = tensor("aw_1699_cast_fp16")]; + tensor var_20448_equation_0 = const()[name = tensor("op_20448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20448_cast_fp16 = einsum(equation = var_20448_equation_0, values = (var_20290_cast_fp16, var_20207_cast_fp16))[name = tensor("op_20448_cast_fp16")]; + tensor var_20449_to_fp16 = const()[name = tensor("op_20449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1701_cast_fp16 = mul(x = var_20448_cast_fp16, y = var_20449_to_fp16)[name = tensor("aw_1701_cast_fp16")]; + tensor var_20452_equation_0 = const()[name = tensor("op_20452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20452_cast_fp16 = einsum(equation = var_20452_equation_0, values = (var_20294_cast_fp16, var_20211_cast_fp16))[name = tensor("op_20452_cast_fp16")]; + tensor var_20453_to_fp16 = const()[name = tensor("op_20453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1703_cast_fp16 = mul(x = var_20452_cast_fp16, y = var_20453_to_fp16)[name = tensor("aw_1703_cast_fp16")]; + tensor var_20456_equation_0 = const()[name = tensor("op_20456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20456_cast_fp16 = einsum(equation = var_20456_equation_0, values = (var_20298_cast_fp16, var_20215_cast_fp16))[name = tensor("op_20456_cast_fp16")]; + tensor var_20457_to_fp16 = const()[name = tensor("op_20457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1705_cast_fp16 = mul(x = var_20456_cast_fp16, y = var_20457_to_fp16)[name = tensor("aw_1705_cast_fp16")]; + tensor var_20460_equation_0 = const()[name = tensor("op_20460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20460_cast_fp16 = einsum(equation = var_20460_equation_0, values = (var_20302_cast_fp16, var_20219_cast_fp16))[name = tensor("op_20460_cast_fp16")]; + tensor var_20461_to_fp16 = const()[name = tensor("op_20461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1707_cast_fp16 = mul(x = var_20460_cast_fp16, y = var_20461_to_fp16)[name = tensor("aw_1707_cast_fp16")]; + tensor var_20464_equation_0 = const()[name = tensor("op_20464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20464_cast_fp16 = einsum(equation = var_20464_equation_0, values = (var_20306_cast_fp16, var_20223_cast_fp16))[name = tensor("op_20464_cast_fp16")]; + tensor var_20465_to_fp16 = const()[name = tensor("op_20465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1709_cast_fp16 = mul(x = var_20464_cast_fp16, y = var_20465_to_fp16)[name = tensor("aw_1709_cast_fp16")]; + tensor var_20468_equation_0 = const()[name = tensor("op_20468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20468_cast_fp16 = einsum(equation = var_20468_equation_0, values = (var_20310_cast_fp16, var_20227_cast_fp16))[name = tensor("op_20468_cast_fp16")]; + tensor var_20469_to_fp16 = const()[name = tensor("op_20469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1711_cast_fp16 = mul(x = var_20468_cast_fp16, y = var_20469_to_fp16)[name = tensor("aw_1711_cast_fp16")]; + tensor var_20472_equation_0 = const()[name = tensor("op_20472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20472_cast_fp16 = einsum(equation = var_20472_equation_0, values = (var_20314_cast_fp16, var_20231_cast_fp16))[name = tensor("op_20472_cast_fp16")]; + tensor var_20473_to_fp16 = const()[name = tensor("op_20473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1713_cast_fp16 = mul(x = var_20472_cast_fp16, y = var_20473_to_fp16)[name = tensor("aw_1713_cast_fp16")]; + tensor var_20476_equation_0 = const()[name = tensor("op_20476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20476_cast_fp16 = einsum(equation = var_20476_equation_0, values = (var_20318_cast_fp16, var_20235_cast_fp16))[name = tensor("op_20476_cast_fp16")]; + tensor var_20477_to_fp16 = const()[name = tensor("op_20477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1715_cast_fp16 = mul(x = var_20476_cast_fp16, y = var_20477_to_fp16)[name = tensor("aw_1715_cast_fp16")]; + tensor var_20480_equation_0 = const()[name = tensor("op_20480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20480_cast_fp16 = einsum(equation = var_20480_equation_0, values = (var_20322_cast_fp16, var_20239_cast_fp16))[name = tensor("op_20480_cast_fp16")]; + tensor var_20481_to_fp16 = const()[name = tensor("op_20481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1717_cast_fp16 = mul(x = var_20480_cast_fp16, y = var_20481_to_fp16)[name = tensor("aw_1717_cast_fp16")]; + tensor var_20484_equation_0 = const()[name = tensor("op_20484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20484_cast_fp16 = einsum(equation = var_20484_equation_0, values = (var_20326_cast_fp16, var_20243_cast_fp16))[name = tensor("op_20484_cast_fp16")]; + tensor var_20485_to_fp16 = const()[name = tensor("op_20485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1719_cast_fp16 = mul(x = var_20484_cast_fp16, y = var_20485_to_fp16)[name = tensor("aw_1719_cast_fp16")]; + tensor var_20487_cast_fp16 = softmax(axis = var_2624, x = aw_1681_cast_fp16)[name = tensor("op_20487_cast_fp16")]; + tensor var_20488_cast_fp16 = softmax(axis = var_2624, x = aw_1683_cast_fp16)[name = tensor("op_20488_cast_fp16")]; + tensor var_20489_cast_fp16 = softmax(axis = var_2624, x = aw_1685_cast_fp16)[name = tensor("op_20489_cast_fp16")]; + tensor var_20490_cast_fp16 = softmax(axis = var_2624, x = aw_1687_cast_fp16)[name = tensor("op_20490_cast_fp16")]; + tensor var_20491_cast_fp16 = softmax(axis = var_2624, x = aw_1689_cast_fp16)[name = tensor("op_20491_cast_fp16")]; + tensor var_20492_cast_fp16 = softmax(axis = var_2624, x = aw_1691_cast_fp16)[name = tensor("op_20492_cast_fp16")]; + tensor var_20493_cast_fp16 = softmax(axis = var_2624, x = aw_1693_cast_fp16)[name = tensor("op_20493_cast_fp16")]; + tensor var_20494_cast_fp16 = softmax(axis = var_2624, x = aw_1695_cast_fp16)[name = tensor("op_20494_cast_fp16")]; + tensor var_20495_cast_fp16 = softmax(axis = var_2624, x = aw_1697_cast_fp16)[name = tensor("op_20495_cast_fp16")]; + tensor var_20496_cast_fp16 = softmax(axis = var_2624, x = aw_1699_cast_fp16)[name = tensor("op_20496_cast_fp16")]; + tensor var_20497_cast_fp16 = softmax(axis = var_2624, x = aw_1701_cast_fp16)[name = tensor("op_20497_cast_fp16")]; + tensor var_20498_cast_fp16 = softmax(axis = var_2624, x = aw_1703_cast_fp16)[name = tensor("op_20498_cast_fp16")]; + tensor var_20499_cast_fp16 = softmax(axis = var_2624, x = aw_1705_cast_fp16)[name = tensor("op_20499_cast_fp16")]; + tensor var_20500_cast_fp16 = softmax(axis = var_2624, x = aw_1707_cast_fp16)[name = tensor("op_20500_cast_fp16")]; + tensor var_20501_cast_fp16 = softmax(axis = var_2624, x = aw_1709_cast_fp16)[name = tensor("op_20501_cast_fp16")]; + tensor var_20502_cast_fp16 = softmax(axis = var_2624, x = aw_1711_cast_fp16)[name = tensor("op_20502_cast_fp16")]; + tensor var_20503_cast_fp16 = softmax(axis = var_2624, x = aw_1713_cast_fp16)[name = tensor("op_20503_cast_fp16")]; + tensor var_20504_cast_fp16 = softmax(axis = var_2624, x = aw_1715_cast_fp16)[name = tensor("op_20504_cast_fp16")]; + tensor var_20505_cast_fp16 = softmax(axis = var_2624, x = aw_1717_cast_fp16)[name = tensor("op_20505_cast_fp16")]; + tensor var_20506_cast_fp16 = softmax(axis = var_2624, x = aw_1719_cast_fp16)[name = tensor("op_20506_cast_fp16")]; + tensor var_20508_equation_0 = const()[name = tensor("op_20508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20508_cast_fp16 = einsum(equation = var_20508_equation_0, values = (var_20328_cast_fp16, var_20487_cast_fp16))[name = tensor("op_20508_cast_fp16")]; + tensor var_20510_equation_0 = const()[name = tensor("op_20510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20510_cast_fp16 = einsum(equation = var_20510_equation_0, values = (var_20332_cast_fp16, var_20488_cast_fp16))[name = tensor("op_20510_cast_fp16")]; + tensor var_20512_equation_0 = const()[name = tensor("op_20512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20512_cast_fp16 = einsum(equation = var_20512_equation_0, values = (var_20336_cast_fp16, var_20489_cast_fp16))[name = tensor("op_20512_cast_fp16")]; + tensor var_20514_equation_0 = const()[name = tensor("op_20514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20514_cast_fp16 = einsum(equation = var_20514_equation_0, values = (var_20340_cast_fp16, var_20490_cast_fp16))[name = tensor("op_20514_cast_fp16")]; + tensor var_20516_equation_0 = const()[name = tensor("op_20516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20516_cast_fp16 = einsum(equation = var_20516_equation_0, values = (var_20344_cast_fp16, var_20491_cast_fp16))[name = tensor("op_20516_cast_fp16")]; + tensor var_20518_equation_0 = const()[name = tensor("op_20518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20518_cast_fp16 = einsum(equation = var_20518_equation_0, values = (var_20348_cast_fp16, var_20492_cast_fp16))[name = tensor("op_20518_cast_fp16")]; + tensor var_20520_equation_0 = const()[name = tensor("op_20520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20520_cast_fp16 = einsum(equation = var_20520_equation_0, values = (var_20352_cast_fp16, var_20493_cast_fp16))[name = tensor("op_20520_cast_fp16")]; + tensor var_20522_equation_0 = const()[name = tensor("op_20522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20522_cast_fp16 = einsum(equation = var_20522_equation_0, values = (var_20356_cast_fp16, var_20494_cast_fp16))[name = tensor("op_20522_cast_fp16")]; + tensor var_20524_equation_0 = const()[name = tensor("op_20524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20524_cast_fp16 = einsum(equation = var_20524_equation_0, values = (var_20360_cast_fp16, var_20495_cast_fp16))[name = tensor("op_20524_cast_fp16")]; + tensor var_20526_equation_0 = const()[name = tensor("op_20526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20526_cast_fp16 = einsum(equation = var_20526_equation_0, values = (var_20364_cast_fp16, var_20496_cast_fp16))[name = tensor("op_20526_cast_fp16")]; + tensor var_20528_equation_0 = const()[name = tensor("op_20528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20528_cast_fp16 = einsum(equation = var_20528_equation_0, values = (var_20368_cast_fp16, var_20497_cast_fp16))[name = tensor("op_20528_cast_fp16")]; + tensor var_20530_equation_0 = const()[name = tensor("op_20530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20530_cast_fp16 = einsum(equation = var_20530_equation_0, values = (var_20372_cast_fp16, var_20498_cast_fp16))[name = tensor("op_20530_cast_fp16")]; + tensor var_20532_equation_0 = const()[name = tensor("op_20532_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20532_cast_fp16 = einsum(equation = var_20532_equation_0, values = (var_20376_cast_fp16, var_20499_cast_fp16))[name = tensor("op_20532_cast_fp16")]; + tensor var_20534_equation_0 = const()[name = tensor("op_20534_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20534_cast_fp16 = einsum(equation = var_20534_equation_0, values = (var_20380_cast_fp16, var_20500_cast_fp16))[name = tensor("op_20534_cast_fp16")]; + tensor var_20536_equation_0 = const()[name = tensor("op_20536_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20536_cast_fp16 = einsum(equation = var_20536_equation_0, values = (var_20384_cast_fp16, var_20501_cast_fp16))[name = tensor("op_20536_cast_fp16")]; + tensor var_20538_equation_0 = const()[name = tensor("op_20538_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20538_cast_fp16 = einsum(equation = var_20538_equation_0, values = (var_20388_cast_fp16, var_20502_cast_fp16))[name = tensor("op_20538_cast_fp16")]; + tensor var_20540_equation_0 = const()[name = tensor("op_20540_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20540_cast_fp16 = einsum(equation = var_20540_equation_0, values = (var_20392_cast_fp16, var_20503_cast_fp16))[name = tensor("op_20540_cast_fp16")]; + tensor var_20542_equation_0 = const()[name = tensor("op_20542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20542_cast_fp16 = einsum(equation = var_20542_equation_0, values = (var_20396_cast_fp16, var_20504_cast_fp16))[name = tensor("op_20542_cast_fp16")]; + tensor var_20544_equation_0 = const()[name = tensor("op_20544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20544_cast_fp16 = einsum(equation = var_20544_equation_0, values = (var_20400_cast_fp16, var_20505_cast_fp16))[name = tensor("op_20544_cast_fp16")]; + tensor var_20546_equation_0 = const()[name = tensor("op_20546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20546_cast_fp16 = einsum(equation = var_20546_equation_0, values = (var_20404_cast_fp16, var_20506_cast_fp16))[name = tensor("op_20546_cast_fp16")]; + tensor input_301_interleave_0 = const()[name = tensor("input_301_interleave_0"), val = tensor(false)]; + tensor input_301_cast_fp16 = concat(axis = var_2624, interleave = input_301_interleave_0, values = (var_20508_cast_fp16, var_20510_cast_fp16, var_20512_cast_fp16, var_20514_cast_fp16, var_20516_cast_fp16, var_20518_cast_fp16, var_20520_cast_fp16, var_20522_cast_fp16, var_20524_cast_fp16, var_20526_cast_fp16, var_20528_cast_fp16, var_20530_cast_fp16, var_20532_cast_fp16, var_20534_cast_fp16, var_20536_cast_fp16, var_20538_cast_fp16, var_20540_cast_fp16, var_20542_cast_fp16, var_20544_cast_fp16, var_20546_cast_fp16))[name = tensor("input_301_cast_fp16")]; + tensor var_20556_pad_type_0 = const()[name = tensor("op_20556_pad_type_0"), val = tensor("valid")]; + tensor var_20556_strides_0 = const()[name = tensor("op_20556_strides_0"), val = tensor([1, 1])]; + tensor var_20556_pad_0 = const()[name = tensor("op_20556_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20556_dilations_0 = const()[name = tensor("op_20556_dilations_0"), val = tensor([1, 1])]; + tensor var_20556_groups_0 = const()[name = tensor("op_20556_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(599815616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(601044480))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(601044672)))]; + tensor var_20556_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_20556_dilations_0, groups = var_20556_groups_0, pad = var_20556_pad_0, pad_type = var_20556_pad_type_0, strides = var_20556_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized, x = input_301_cast_fp16)[name = tensor("op_20556_cast_fp16")]; + tensor inputs_141_cast_fp16 = add(x = var_20556_cast_fp16, y = inputs_139_cast_fp16)[name = tensor("inputs_141_cast_fp16")]; + tensor hidden_states_193_axes_0 = const()[name = tensor("hidden_states_193_axes_0"), val = tensor([1])]; + tensor hidden_states_193_gamma_0_to_fp16 = const()[name = tensor("hidden_states_193_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(601047296)))]; + tensor hidden_states_193_beta_0_to_fp16 = const()[name = tensor("hidden_states_193_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(601049920)))]; + tensor var_20566_to_fp16 = const()[name = tensor("op_20566_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_193_cast_fp16 = layer_norm(axes = hidden_states_193_axes_0, beta = hidden_states_193_beta_0_to_fp16, epsilon = var_20566_to_fp16, gamma = hidden_states_193_gamma_0_to_fp16, x = inputs_141_cast_fp16)[name = tensor("hidden_states_193_cast_fp16")]; + tensor q_95_pad_type_0 = const()[name = tensor("q_95_pad_type_0"), val = tensor("valid")]; + tensor q_95_strides_0 = const()[name = tensor("q_95_strides_0"), val = tensor([1, 1])]; + tensor q_95_pad_0 = const()[name = tensor("q_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_95_dilations_0 = const()[name = tensor("q_95_dilations_0"), val = tensor([1, 1])]; + tensor q_95_groups_0 = const()[name = tensor("q_95_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(601052544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(602281408))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_95_cast_fp16 = conv(dilations = q_95_dilations_0, groups = q_95_groups_0, pad = q_95_pad_0, pad_type = q_95_pad_type_0, strides = q_95_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_193_cast_fp16)[name = tensor("q_95_cast_fp16")]; + tensor k_189_pad_type_0 = const()[name = tensor("k_189_pad_type_0"), val = tensor("valid")]; + tensor k_189_strides_0 = const()[name = tensor("k_189_strides_0"), val = tensor([1, 1])]; + tensor k_189_pad_0 = const()[name = tensor("k_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_189_dilations_0 = const()[name = tensor("k_189_dilations_0"), val = tensor([1, 1])]; + tensor k_189_groups_0 = const()[name = tensor("k_189_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(602281600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604247744))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_189_cast_fp16 = conv(dilations = k_189_dilations_0, groups = k_189_groups_0, pad = k_189_pad_0, pad_type = k_189_pad_type_0, strides = k_189_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_189_cast_fp16")]; + tensor v_95_pad_type_0 = const()[name = tensor("v_95_pad_type_0"), val = tensor("valid")]; + tensor v_95_strides_0 = const()[name = tensor("v_95_strides_0"), val = tensor([1, 1])]; + tensor v_95_pad_0 = const()[name = tensor("v_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_95_dilations_0 = const()[name = tensor("v_95_dilations_0"), val = tensor([1, 1])]; + tensor v_95_groups_0 = const()[name = tensor("v_95_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604247936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606214080))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_95_cast_fp16 = conv(dilations = v_95_dilations_0, groups = v_95_groups_0, pad = v_95_pad_0, pad_type = v_95_pad_type_0, strides = v_95_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_95_cast_fp16")]; + tensor var_20599_begin_0 = const()[name = tensor("op_20599_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20599_end_0 = const()[name = tensor("op_20599_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_20599_end_mask_0 = const()[name = tensor("op_20599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20599_cast_fp16 = slice_by_index(begin = var_20599_begin_0, end = var_20599_end_0, end_mask = var_20599_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20599_cast_fp16")]; + tensor var_20603_begin_0 = const()[name = tensor("op_20603_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20603_end_0 = const()[name = tensor("op_20603_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_20603_end_mask_0 = const()[name = tensor("op_20603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20603_cast_fp16 = slice_by_index(begin = var_20603_begin_0, end = var_20603_end_0, end_mask = var_20603_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20603_cast_fp16")]; + tensor var_20607_begin_0 = const()[name = tensor("op_20607_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20607_end_0 = const()[name = tensor("op_20607_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_20607_end_mask_0 = const()[name = tensor("op_20607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20607_cast_fp16 = slice_by_index(begin = var_20607_begin_0, end = var_20607_end_0, end_mask = var_20607_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20607_cast_fp16")]; + tensor var_20611_begin_0 = const()[name = tensor("op_20611_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20611_end_0 = const()[name = tensor("op_20611_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_20611_end_mask_0 = const()[name = tensor("op_20611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20611_cast_fp16 = slice_by_index(begin = var_20611_begin_0, end = var_20611_end_0, end_mask = var_20611_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20611_cast_fp16")]; + tensor var_20615_begin_0 = const()[name = tensor("op_20615_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20615_end_0 = const()[name = tensor("op_20615_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_20615_end_mask_0 = const()[name = tensor("op_20615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20615_cast_fp16 = slice_by_index(begin = var_20615_begin_0, end = var_20615_end_0, end_mask = var_20615_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20615_cast_fp16")]; + tensor var_20619_begin_0 = const()[name = tensor("op_20619_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20619_end_0 = const()[name = tensor("op_20619_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_20619_end_mask_0 = const()[name = tensor("op_20619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20619_cast_fp16 = slice_by_index(begin = var_20619_begin_0, end = var_20619_end_0, end_mask = var_20619_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20619_cast_fp16")]; + tensor var_20623_begin_0 = const()[name = tensor("op_20623_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20623_end_0 = const()[name = tensor("op_20623_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_20623_end_mask_0 = const()[name = tensor("op_20623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20623_cast_fp16 = slice_by_index(begin = var_20623_begin_0, end = var_20623_end_0, end_mask = var_20623_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20623_cast_fp16")]; + tensor var_20627_begin_0 = const()[name = tensor("op_20627_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20627_end_0 = const()[name = tensor("op_20627_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_20627_end_mask_0 = const()[name = tensor("op_20627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20627_cast_fp16 = slice_by_index(begin = var_20627_begin_0, end = var_20627_end_0, end_mask = var_20627_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20627_cast_fp16")]; + tensor var_20631_begin_0 = const()[name = tensor("op_20631_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20631_end_0 = const()[name = tensor("op_20631_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_20631_end_mask_0 = const()[name = tensor("op_20631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20631_cast_fp16 = slice_by_index(begin = var_20631_begin_0, end = var_20631_end_0, end_mask = var_20631_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20631_cast_fp16")]; + tensor var_20635_begin_0 = const()[name = tensor("op_20635_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20635_end_0 = const()[name = tensor("op_20635_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_20635_end_mask_0 = const()[name = tensor("op_20635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20635_cast_fp16 = slice_by_index(begin = var_20635_begin_0, end = var_20635_end_0, end_mask = var_20635_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20635_cast_fp16")]; + tensor var_20639_begin_0 = const()[name = tensor("op_20639_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20639_end_0 = const()[name = tensor("op_20639_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_20639_end_mask_0 = const()[name = tensor("op_20639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20639_cast_fp16 = slice_by_index(begin = var_20639_begin_0, end = var_20639_end_0, end_mask = var_20639_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20639_cast_fp16")]; + tensor var_20643_begin_0 = const()[name = tensor("op_20643_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20643_end_0 = const()[name = tensor("op_20643_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_20643_end_mask_0 = const()[name = tensor("op_20643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20643_cast_fp16 = slice_by_index(begin = var_20643_begin_0, end = var_20643_end_0, end_mask = var_20643_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20643_cast_fp16")]; + tensor var_20647_begin_0 = const()[name = tensor("op_20647_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20647_end_0 = const()[name = tensor("op_20647_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_20647_end_mask_0 = const()[name = tensor("op_20647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20647_cast_fp16 = slice_by_index(begin = var_20647_begin_0, end = var_20647_end_0, end_mask = var_20647_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20647_cast_fp16")]; + tensor var_20651_begin_0 = const()[name = tensor("op_20651_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20651_end_0 = const()[name = tensor("op_20651_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_20651_end_mask_0 = const()[name = tensor("op_20651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20651_cast_fp16 = slice_by_index(begin = var_20651_begin_0, end = var_20651_end_0, end_mask = var_20651_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20651_cast_fp16")]; + tensor var_20655_begin_0 = const()[name = tensor("op_20655_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20655_end_0 = const()[name = tensor("op_20655_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_20655_end_mask_0 = const()[name = tensor("op_20655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20655_cast_fp16 = slice_by_index(begin = var_20655_begin_0, end = var_20655_end_0, end_mask = var_20655_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20655_cast_fp16")]; + tensor var_20659_begin_0 = const()[name = tensor("op_20659_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20659_end_0 = const()[name = tensor("op_20659_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20659_end_mask_0 = const()[name = tensor("op_20659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20659_cast_fp16 = slice_by_index(begin = var_20659_begin_0, end = var_20659_end_0, end_mask = var_20659_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20659_cast_fp16")]; + tensor var_20663_begin_0 = const()[name = tensor("op_20663_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20663_end_0 = const()[name = tensor("op_20663_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_20663_end_mask_0 = const()[name = tensor("op_20663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20663_cast_fp16 = slice_by_index(begin = var_20663_begin_0, end = var_20663_end_0, end_mask = var_20663_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20663_cast_fp16")]; + tensor var_20667_begin_0 = const()[name = tensor("op_20667_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20667_end_0 = const()[name = tensor("op_20667_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_20667_end_mask_0 = const()[name = tensor("op_20667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20667_cast_fp16 = slice_by_index(begin = var_20667_begin_0, end = var_20667_end_0, end_mask = var_20667_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20667_cast_fp16")]; + tensor var_20671_begin_0 = const()[name = tensor("op_20671_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20671_end_0 = const()[name = tensor("op_20671_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_20671_end_mask_0 = const()[name = tensor("op_20671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20671_cast_fp16 = slice_by_index(begin = var_20671_begin_0, end = var_20671_end_0, end_mask = var_20671_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20671_cast_fp16")]; + tensor var_20675_begin_0 = const()[name = tensor("op_20675_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20675_end_0 = const()[name = tensor("op_20675_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_20675_end_mask_0 = const()[name = tensor("op_20675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20675_cast_fp16 = slice_by_index(begin = var_20675_begin_0, end = var_20675_end_0, end_mask = var_20675_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20675_cast_fp16")]; + tensor k_191_perm_0 = const()[name = tensor("k_191_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_20682_begin_0 = const()[name = tensor("op_20682_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20682_end_0 = const()[name = tensor("op_20682_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_20682_end_mask_0 = const()[name = tensor("op_20682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_191_cast_fp16 = transpose(perm = k_191_perm_0, x = k_189_cast_fp16)[name = tensor("transpose_20")]; + tensor var_20682_cast_fp16 = slice_by_index(begin = var_20682_begin_0, end = var_20682_end_0, end_mask = var_20682_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20682_cast_fp16")]; + tensor var_20686_begin_0 = const()[name = tensor("op_20686_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_20686_end_0 = const()[name = tensor("op_20686_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_20686_end_mask_0 = const()[name = tensor("op_20686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20686_cast_fp16 = slice_by_index(begin = var_20686_begin_0, end = var_20686_end_0, end_mask = var_20686_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20686_cast_fp16")]; + tensor var_20690_begin_0 = const()[name = tensor("op_20690_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_20690_end_0 = const()[name = tensor("op_20690_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_20690_end_mask_0 = const()[name = tensor("op_20690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20690_cast_fp16 = slice_by_index(begin = var_20690_begin_0, end = var_20690_end_0, end_mask = var_20690_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20690_cast_fp16")]; + tensor var_20694_begin_0 = const()[name = tensor("op_20694_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_20694_end_0 = const()[name = tensor("op_20694_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_20694_end_mask_0 = const()[name = tensor("op_20694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20694_cast_fp16 = slice_by_index(begin = var_20694_begin_0, end = var_20694_end_0, end_mask = var_20694_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20694_cast_fp16")]; + tensor var_20698_begin_0 = const()[name = tensor("op_20698_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_20698_end_0 = const()[name = tensor("op_20698_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_20698_end_mask_0 = const()[name = tensor("op_20698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20698_cast_fp16 = slice_by_index(begin = var_20698_begin_0, end = var_20698_end_0, end_mask = var_20698_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20698_cast_fp16")]; + tensor var_20702_begin_0 = const()[name = tensor("op_20702_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_20702_end_0 = const()[name = tensor("op_20702_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_20702_end_mask_0 = const()[name = tensor("op_20702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20702_cast_fp16 = slice_by_index(begin = var_20702_begin_0, end = var_20702_end_0, end_mask = var_20702_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20702_cast_fp16")]; + tensor var_20706_begin_0 = const()[name = tensor("op_20706_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_20706_end_0 = const()[name = tensor("op_20706_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_20706_end_mask_0 = const()[name = tensor("op_20706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20706_cast_fp16 = slice_by_index(begin = var_20706_begin_0, end = var_20706_end_0, end_mask = var_20706_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20706_cast_fp16")]; + tensor var_20710_begin_0 = const()[name = tensor("op_20710_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_20710_end_0 = const()[name = tensor("op_20710_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_20710_end_mask_0 = const()[name = tensor("op_20710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20710_cast_fp16 = slice_by_index(begin = var_20710_begin_0, end = var_20710_end_0, end_mask = var_20710_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20710_cast_fp16")]; + tensor var_20714_begin_0 = const()[name = tensor("op_20714_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_20714_end_0 = const()[name = tensor("op_20714_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_20714_end_mask_0 = const()[name = tensor("op_20714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20714_cast_fp16 = slice_by_index(begin = var_20714_begin_0, end = var_20714_end_0, end_mask = var_20714_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20714_cast_fp16")]; + tensor var_20718_begin_0 = const()[name = tensor("op_20718_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_20718_end_0 = const()[name = tensor("op_20718_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_20718_end_mask_0 = const()[name = tensor("op_20718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20718_cast_fp16 = slice_by_index(begin = var_20718_begin_0, end = var_20718_end_0, end_mask = var_20718_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20718_cast_fp16")]; + tensor var_20722_begin_0 = const()[name = tensor("op_20722_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_20722_end_0 = const()[name = tensor("op_20722_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_20722_end_mask_0 = const()[name = tensor("op_20722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20722_cast_fp16 = slice_by_index(begin = var_20722_begin_0, end = var_20722_end_0, end_mask = var_20722_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20722_cast_fp16")]; + tensor var_20726_begin_0 = const()[name = tensor("op_20726_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_20726_end_0 = const()[name = tensor("op_20726_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_20726_end_mask_0 = const()[name = tensor("op_20726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20726_cast_fp16 = slice_by_index(begin = var_20726_begin_0, end = var_20726_end_0, end_mask = var_20726_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20726_cast_fp16")]; + tensor var_20730_begin_0 = const()[name = tensor("op_20730_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_20730_end_0 = const()[name = tensor("op_20730_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_20730_end_mask_0 = const()[name = tensor("op_20730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20730_cast_fp16 = slice_by_index(begin = var_20730_begin_0, end = var_20730_end_0, end_mask = var_20730_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20730_cast_fp16")]; + tensor var_20734_begin_0 = const()[name = tensor("op_20734_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_20734_end_0 = const()[name = tensor("op_20734_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_20734_end_mask_0 = const()[name = tensor("op_20734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20734_cast_fp16 = slice_by_index(begin = var_20734_begin_0, end = var_20734_end_0, end_mask = var_20734_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20734_cast_fp16")]; + tensor var_20738_begin_0 = const()[name = tensor("op_20738_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_20738_end_0 = const()[name = tensor("op_20738_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_20738_end_mask_0 = const()[name = tensor("op_20738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20738_cast_fp16 = slice_by_index(begin = var_20738_begin_0, end = var_20738_end_0, end_mask = var_20738_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20738_cast_fp16")]; + tensor var_20742_begin_0 = const()[name = tensor("op_20742_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_20742_end_0 = const()[name = tensor("op_20742_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_20742_end_mask_0 = const()[name = tensor("op_20742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20742_cast_fp16 = slice_by_index(begin = var_20742_begin_0, end = var_20742_end_0, end_mask = var_20742_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20742_cast_fp16")]; + tensor var_20746_begin_0 = const()[name = tensor("op_20746_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_20746_end_0 = const()[name = tensor("op_20746_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_20746_end_mask_0 = const()[name = tensor("op_20746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20746_cast_fp16 = slice_by_index(begin = var_20746_begin_0, end = var_20746_end_0, end_mask = var_20746_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20746_cast_fp16")]; + tensor var_20750_begin_0 = const()[name = tensor("op_20750_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_20750_end_0 = const()[name = tensor("op_20750_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_20750_end_mask_0 = const()[name = tensor("op_20750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20750_cast_fp16 = slice_by_index(begin = var_20750_begin_0, end = var_20750_end_0, end_mask = var_20750_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20750_cast_fp16")]; + tensor var_20754_begin_0 = const()[name = tensor("op_20754_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_20754_end_0 = const()[name = tensor("op_20754_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_20754_end_mask_0 = const()[name = tensor("op_20754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20754_cast_fp16 = slice_by_index(begin = var_20754_begin_0, end = var_20754_end_0, end_mask = var_20754_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20754_cast_fp16")]; + tensor var_20758_begin_0 = const()[name = tensor("op_20758_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_20758_end_0 = const()[name = tensor("op_20758_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_20758_end_mask_0 = const()[name = tensor("op_20758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20758_cast_fp16 = slice_by_index(begin = var_20758_begin_0, end = var_20758_end_0, end_mask = var_20758_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20758_cast_fp16")]; + tensor var_20760_begin_0 = const()[name = tensor("op_20760_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20760_end_0 = const()[name = tensor("op_20760_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_20760_end_mask_0 = const()[name = tensor("op_20760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20760_cast_fp16 = slice_by_index(begin = var_20760_begin_0, end = var_20760_end_0, end_mask = var_20760_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20760_cast_fp16")]; + tensor var_20764_begin_0 = const()[name = tensor("op_20764_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20764_end_0 = const()[name = tensor("op_20764_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_20764_end_mask_0 = const()[name = tensor("op_20764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20764_cast_fp16 = slice_by_index(begin = var_20764_begin_0, end = var_20764_end_0, end_mask = var_20764_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20764_cast_fp16")]; + tensor var_20768_begin_0 = const()[name = tensor("op_20768_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20768_end_0 = const()[name = tensor("op_20768_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_20768_end_mask_0 = const()[name = tensor("op_20768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20768_cast_fp16 = slice_by_index(begin = var_20768_begin_0, end = var_20768_end_0, end_mask = var_20768_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20768_cast_fp16")]; + tensor var_20772_begin_0 = const()[name = tensor("op_20772_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20772_end_0 = const()[name = tensor("op_20772_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_20772_end_mask_0 = const()[name = tensor("op_20772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20772_cast_fp16 = slice_by_index(begin = var_20772_begin_0, end = var_20772_end_0, end_mask = var_20772_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20772_cast_fp16")]; + tensor var_20776_begin_0 = const()[name = tensor("op_20776_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20776_end_0 = const()[name = tensor("op_20776_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_20776_end_mask_0 = const()[name = tensor("op_20776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20776_cast_fp16 = slice_by_index(begin = var_20776_begin_0, end = var_20776_end_0, end_mask = var_20776_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20776_cast_fp16")]; + tensor var_20780_begin_0 = const()[name = tensor("op_20780_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20780_end_0 = const()[name = tensor("op_20780_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_20780_end_mask_0 = const()[name = tensor("op_20780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20780_cast_fp16 = slice_by_index(begin = var_20780_begin_0, end = var_20780_end_0, end_mask = var_20780_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20780_cast_fp16")]; + tensor var_20784_begin_0 = const()[name = tensor("op_20784_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20784_end_0 = const()[name = tensor("op_20784_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_20784_end_mask_0 = const()[name = tensor("op_20784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20784_cast_fp16 = slice_by_index(begin = var_20784_begin_0, end = var_20784_end_0, end_mask = var_20784_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20784_cast_fp16")]; + tensor var_20788_begin_0 = const()[name = tensor("op_20788_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20788_end_0 = const()[name = tensor("op_20788_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_20788_end_mask_0 = const()[name = tensor("op_20788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20788_cast_fp16 = slice_by_index(begin = var_20788_begin_0, end = var_20788_end_0, end_mask = var_20788_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20788_cast_fp16")]; + tensor var_20792_begin_0 = const()[name = tensor("op_20792_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20792_end_0 = const()[name = tensor("op_20792_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_20792_end_mask_0 = const()[name = tensor("op_20792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20792_cast_fp16 = slice_by_index(begin = var_20792_begin_0, end = var_20792_end_0, end_mask = var_20792_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20792_cast_fp16")]; + tensor var_20796_begin_0 = const()[name = tensor("op_20796_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20796_end_0 = const()[name = tensor("op_20796_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_20796_end_mask_0 = const()[name = tensor("op_20796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20796_cast_fp16 = slice_by_index(begin = var_20796_begin_0, end = var_20796_end_0, end_mask = var_20796_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20796_cast_fp16")]; + tensor var_20800_begin_0 = const()[name = tensor("op_20800_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20800_end_0 = const()[name = tensor("op_20800_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_20800_end_mask_0 = const()[name = tensor("op_20800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20800_cast_fp16 = slice_by_index(begin = var_20800_begin_0, end = var_20800_end_0, end_mask = var_20800_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20800_cast_fp16")]; + tensor var_20804_begin_0 = const()[name = tensor("op_20804_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20804_end_0 = const()[name = tensor("op_20804_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_20804_end_mask_0 = const()[name = tensor("op_20804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20804_cast_fp16 = slice_by_index(begin = var_20804_begin_0, end = var_20804_end_0, end_mask = var_20804_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20804_cast_fp16")]; + tensor var_20808_begin_0 = const()[name = tensor("op_20808_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20808_end_0 = const()[name = tensor("op_20808_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_20808_end_mask_0 = const()[name = tensor("op_20808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20808_cast_fp16 = slice_by_index(begin = var_20808_begin_0, end = var_20808_end_0, end_mask = var_20808_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20808_cast_fp16")]; + tensor var_20812_begin_0 = const()[name = tensor("op_20812_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20812_end_0 = const()[name = tensor("op_20812_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_20812_end_mask_0 = const()[name = tensor("op_20812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20812_cast_fp16 = slice_by_index(begin = var_20812_begin_0, end = var_20812_end_0, end_mask = var_20812_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20812_cast_fp16")]; + tensor var_20816_begin_0 = const()[name = tensor("op_20816_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20816_end_0 = const()[name = tensor("op_20816_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_20816_end_mask_0 = const()[name = tensor("op_20816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20816_cast_fp16 = slice_by_index(begin = var_20816_begin_0, end = var_20816_end_0, end_mask = var_20816_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20816_cast_fp16")]; + tensor var_20820_begin_0 = const()[name = tensor("op_20820_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20820_end_0 = const()[name = tensor("op_20820_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_20820_end_mask_0 = const()[name = tensor("op_20820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20820_cast_fp16 = slice_by_index(begin = var_20820_begin_0, end = var_20820_end_0, end_mask = var_20820_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20820_cast_fp16")]; + tensor var_20824_begin_0 = const()[name = tensor("op_20824_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20824_end_0 = const()[name = tensor("op_20824_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_20824_end_mask_0 = const()[name = tensor("op_20824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20824_cast_fp16 = slice_by_index(begin = var_20824_begin_0, end = var_20824_end_0, end_mask = var_20824_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20824_cast_fp16")]; + tensor var_20828_begin_0 = const()[name = tensor("op_20828_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20828_end_0 = const()[name = tensor("op_20828_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_20828_end_mask_0 = const()[name = tensor("op_20828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20828_cast_fp16 = slice_by_index(begin = var_20828_begin_0, end = var_20828_end_0, end_mask = var_20828_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20828_cast_fp16")]; + tensor var_20832_begin_0 = const()[name = tensor("op_20832_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20832_end_0 = const()[name = tensor("op_20832_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_20832_end_mask_0 = const()[name = tensor("op_20832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20832_cast_fp16 = slice_by_index(begin = var_20832_begin_0, end = var_20832_end_0, end_mask = var_20832_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20832_cast_fp16")]; + tensor var_20836_begin_0 = const()[name = tensor("op_20836_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20836_end_0 = const()[name = tensor("op_20836_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_20836_end_mask_0 = const()[name = tensor("op_20836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20836_cast_fp16 = slice_by_index(begin = var_20836_begin_0, end = var_20836_end_0, end_mask = var_20836_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20836_cast_fp16")]; + tensor var_20840_equation_0 = const()[name = tensor("op_20840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20840_cast_fp16 = einsum(equation = var_20840_equation_0, values = (var_20682_cast_fp16, var_20599_cast_fp16))[name = tensor("op_20840_cast_fp16")]; + tensor var_20841_to_fp16 = const()[name = tensor("op_20841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1721_cast_fp16 = mul(x = var_20840_cast_fp16, y = var_20841_to_fp16)[name = tensor("aw_1721_cast_fp16")]; + tensor var_20844_equation_0 = const()[name = tensor("op_20844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20844_cast_fp16 = einsum(equation = var_20844_equation_0, values = (var_20686_cast_fp16, var_20603_cast_fp16))[name = tensor("op_20844_cast_fp16")]; + tensor var_20845_to_fp16 = const()[name = tensor("op_20845_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1723_cast_fp16 = mul(x = var_20844_cast_fp16, y = var_20845_to_fp16)[name = tensor("aw_1723_cast_fp16")]; + tensor var_20848_equation_0 = const()[name = tensor("op_20848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20848_cast_fp16 = einsum(equation = var_20848_equation_0, values = (var_20690_cast_fp16, var_20607_cast_fp16))[name = tensor("op_20848_cast_fp16")]; + tensor var_20849_to_fp16 = const()[name = tensor("op_20849_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1725_cast_fp16 = mul(x = var_20848_cast_fp16, y = var_20849_to_fp16)[name = tensor("aw_1725_cast_fp16")]; + tensor var_20852_equation_0 = const()[name = tensor("op_20852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20852_cast_fp16 = einsum(equation = var_20852_equation_0, values = (var_20694_cast_fp16, var_20611_cast_fp16))[name = tensor("op_20852_cast_fp16")]; + tensor var_20853_to_fp16 = const()[name = tensor("op_20853_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1727_cast_fp16 = mul(x = var_20852_cast_fp16, y = var_20853_to_fp16)[name = tensor("aw_1727_cast_fp16")]; + tensor var_20856_equation_0 = const()[name = tensor("op_20856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20856_cast_fp16 = einsum(equation = var_20856_equation_0, values = (var_20698_cast_fp16, var_20615_cast_fp16))[name = tensor("op_20856_cast_fp16")]; + tensor var_20857_to_fp16 = const()[name = tensor("op_20857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1729_cast_fp16 = mul(x = var_20856_cast_fp16, y = var_20857_to_fp16)[name = tensor("aw_1729_cast_fp16")]; + tensor var_20860_equation_0 = const()[name = tensor("op_20860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20860_cast_fp16 = einsum(equation = var_20860_equation_0, values = (var_20702_cast_fp16, var_20619_cast_fp16))[name = tensor("op_20860_cast_fp16")]; + tensor var_20861_to_fp16 = const()[name = tensor("op_20861_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1731_cast_fp16 = mul(x = var_20860_cast_fp16, y = var_20861_to_fp16)[name = tensor("aw_1731_cast_fp16")]; + tensor var_20864_equation_0 = const()[name = tensor("op_20864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20864_cast_fp16 = einsum(equation = var_20864_equation_0, values = (var_20706_cast_fp16, var_20623_cast_fp16))[name = tensor("op_20864_cast_fp16")]; + tensor var_20865_to_fp16 = const()[name = tensor("op_20865_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1733_cast_fp16 = mul(x = var_20864_cast_fp16, y = var_20865_to_fp16)[name = tensor("aw_1733_cast_fp16")]; + tensor var_20868_equation_0 = const()[name = tensor("op_20868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20868_cast_fp16 = einsum(equation = var_20868_equation_0, values = (var_20710_cast_fp16, var_20627_cast_fp16))[name = tensor("op_20868_cast_fp16")]; + tensor var_20869_to_fp16 = const()[name = tensor("op_20869_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1735_cast_fp16 = mul(x = var_20868_cast_fp16, y = var_20869_to_fp16)[name = tensor("aw_1735_cast_fp16")]; + tensor var_20872_equation_0 = const()[name = tensor("op_20872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20872_cast_fp16 = einsum(equation = var_20872_equation_0, values = (var_20714_cast_fp16, var_20631_cast_fp16))[name = tensor("op_20872_cast_fp16")]; + tensor var_20873_to_fp16 = const()[name = tensor("op_20873_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1737_cast_fp16 = mul(x = var_20872_cast_fp16, y = var_20873_to_fp16)[name = tensor("aw_1737_cast_fp16")]; + tensor var_20876_equation_0 = const()[name = tensor("op_20876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20876_cast_fp16 = einsum(equation = var_20876_equation_0, values = (var_20718_cast_fp16, var_20635_cast_fp16))[name = tensor("op_20876_cast_fp16")]; + tensor var_20877_to_fp16 = const()[name = tensor("op_20877_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1739_cast_fp16 = mul(x = var_20876_cast_fp16, y = var_20877_to_fp16)[name = tensor("aw_1739_cast_fp16")]; + tensor var_20880_equation_0 = const()[name = tensor("op_20880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20880_cast_fp16 = einsum(equation = var_20880_equation_0, values = (var_20722_cast_fp16, var_20639_cast_fp16))[name = tensor("op_20880_cast_fp16")]; + tensor var_20881_to_fp16 = const()[name = tensor("op_20881_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1741_cast_fp16 = mul(x = var_20880_cast_fp16, y = var_20881_to_fp16)[name = tensor("aw_1741_cast_fp16")]; + tensor var_20884_equation_0 = const()[name = tensor("op_20884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20884_cast_fp16 = einsum(equation = var_20884_equation_0, values = (var_20726_cast_fp16, var_20643_cast_fp16))[name = tensor("op_20884_cast_fp16")]; + tensor var_20885_to_fp16 = const()[name = tensor("op_20885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1743_cast_fp16 = mul(x = var_20884_cast_fp16, y = var_20885_to_fp16)[name = tensor("aw_1743_cast_fp16")]; + tensor var_20888_equation_0 = const()[name = tensor("op_20888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20888_cast_fp16 = einsum(equation = var_20888_equation_0, values = (var_20730_cast_fp16, var_20647_cast_fp16))[name = tensor("op_20888_cast_fp16")]; + tensor var_20889_to_fp16 = const()[name = tensor("op_20889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1745_cast_fp16 = mul(x = var_20888_cast_fp16, y = var_20889_to_fp16)[name = tensor("aw_1745_cast_fp16")]; + tensor var_20892_equation_0 = const()[name = tensor("op_20892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20892_cast_fp16 = einsum(equation = var_20892_equation_0, values = (var_20734_cast_fp16, var_20651_cast_fp16))[name = tensor("op_20892_cast_fp16")]; + tensor var_20893_to_fp16 = const()[name = tensor("op_20893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1747_cast_fp16 = mul(x = var_20892_cast_fp16, y = var_20893_to_fp16)[name = tensor("aw_1747_cast_fp16")]; + tensor var_20896_equation_0 = const()[name = tensor("op_20896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20896_cast_fp16 = einsum(equation = var_20896_equation_0, values = (var_20738_cast_fp16, var_20655_cast_fp16))[name = tensor("op_20896_cast_fp16")]; + tensor var_20897_to_fp16 = const()[name = tensor("op_20897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1749_cast_fp16 = mul(x = var_20896_cast_fp16, y = var_20897_to_fp16)[name = tensor("aw_1749_cast_fp16")]; + tensor var_20900_equation_0 = const()[name = tensor("op_20900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20900_cast_fp16 = einsum(equation = var_20900_equation_0, values = (var_20742_cast_fp16, var_20659_cast_fp16))[name = tensor("op_20900_cast_fp16")]; + tensor var_20901_to_fp16 = const()[name = tensor("op_20901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1751_cast_fp16 = mul(x = var_20900_cast_fp16, y = var_20901_to_fp16)[name = tensor("aw_1751_cast_fp16")]; + tensor var_20904_equation_0 = const()[name = tensor("op_20904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20904_cast_fp16 = einsum(equation = var_20904_equation_0, values = (var_20746_cast_fp16, var_20663_cast_fp16))[name = tensor("op_20904_cast_fp16")]; + tensor var_20905_to_fp16 = const()[name = tensor("op_20905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1753_cast_fp16 = mul(x = var_20904_cast_fp16, y = var_20905_to_fp16)[name = tensor("aw_1753_cast_fp16")]; + tensor var_20908_equation_0 = const()[name = tensor("op_20908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20908_cast_fp16 = einsum(equation = var_20908_equation_0, values = (var_20750_cast_fp16, var_20667_cast_fp16))[name = tensor("op_20908_cast_fp16")]; + tensor var_20909_to_fp16 = const()[name = tensor("op_20909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1755_cast_fp16 = mul(x = var_20908_cast_fp16, y = var_20909_to_fp16)[name = tensor("aw_1755_cast_fp16")]; + tensor var_20912_equation_0 = const()[name = tensor("op_20912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20912_cast_fp16 = einsum(equation = var_20912_equation_0, values = (var_20754_cast_fp16, var_20671_cast_fp16))[name = tensor("op_20912_cast_fp16")]; + tensor var_20913_to_fp16 = const()[name = tensor("op_20913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1757_cast_fp16 = mul(x = var_20912_cast_fp16, y = var_20913_to_fp16)[name = tensor("aw_1757_cast_fp16")]; + tensor var_20916_equation_0 = const()[name = tensor("op_20916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20916_cast_fp16 = einsum(equation = var_20916_equation_0, values = (var_20758_cast_fp16, var_20675_cast_fp16))[name = tensor("op_20916_cast_fp16")]; + tensor var_20917_to_fp16 = const()[name = tensor("op_20917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1759_cast_fp16 = mul(x = var_20916_cast_fp16, y = var_20917_to_fp16)[name = tensor("aw_1759_cast_fp16")]; + tensor var_20919_cast_fp16 = softmax(axis = var_2624, x = aw_1721_cast_fp16)[name = tensor("op_20919_cast_fp16")]; + tensor var_20920_cast_fp16 = softmax(axis = var_2624, x = aw_1723_cast_fp16)[name = tensor("op_20920_cast_fp16")]; + tensor var_20921_cast_fp16 = softmax(axis = var_2624, x = aw_1725_cast_fp16)[name = tensor("op_20921_cast_fp16")]; + tensor var_20922_cast_fp16 = softmax(axis = var_2624, x = aw_1727_cast_fp16)[name = tensor("op_20922_cast_fp16")]; + tensor var_20923_cast_fp16 = softmax(axis = var_2624, x = aw_1729_cast_fp16)[name = tensor("op_20923_cast_fp16")]; + tensor var_20924_cast_fp16 = softmax(axis = var_2624, x = aw_1731_cast_fp16)[name = tensor("op_20924_cast_fp16")]; + tensor var_20925_cast_fp16 = softmax(axis = var_2624, x = aw_1733_cast_fp16)[name = tensor("op_20925_cast_fp16")]; + tensor var_20926_cast_fp16 = softmax(axis = var_2624, x = aw_1735_cast_fp16)[name = tensor("op_20926_cast_fp16")]; + tensor var_20927_cast_fp16 = softmax(axis = var_2624, x = aw_1737_cast_fp16)[name = tensor("op_20927_cast_fp16")]; + tensor var_20928_cast_fp16 = softmax(axis = var_2624, x = aw_1739_cast_fp16)[name = tensor("op_20928_cast_fp16")]; + tensor var_20929_cast_fp16 = softmax(axis = var_2624, x = aw_1741_cast_fp16)[name = tensor("op_20929_cast_fp16")]; + tensor var_20930_cast_fp16 = softmax(axis = var_2624, x = aw_1743_cast_fp16)[name = tensor("op_20930_cast_fp16")]; + tensor var_20931_cast_fp16 = softmax(axis = var_2624, x = aw_1745_cast_fp16)[name = tensor("op_20931_cast_fp16")]; + tensor var_20932_cast_fp16 = softmax(axis = var_2624, x = aw_1747_cast_fp16)[name = tensor("op_20932_cast_fp16")]; + tensor var_20933_cast_fp16 = softmax(axis = var_2624, x = aw_1749_cast_fp16)[name = tensor("op_20933_cast_fp16")]; + tensor var_20934_cast_fp16 = softmax(axis = var_2624, x = aw_1751_cast_fp16)[name = tensor("op_20934_cast_fp16")]; + tensor var_20935_cast_fp16 = softmax(axis = var_2624, x = aw_1753_cast_fp16)[name = tensor("op_20935_cast_fp16")]; + tensor var_20936_cast_fp16 = softmax(axis = var_2624, x = aw_1755_cast_fp16)[name = tensor("op_20936_cast_fp16")]; + tensor var_20937_cast_fp16 = softmax(axis = var_2624, x = aw_1757_cast_fp16)[name = tensor("op_20937_cast_fp16")]; + tensor var_20938_cast_fp16 = softmax(axis = var_2624, x = aw_1759_cast_fp16)[name = tensor("op_20938_cast_fp16")]; + tensor var_20940_equation_0 = const()[name = tensor("op_20940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20940_cast_fp16 = einsum(equation = var_20940_equation_0, values = (var_20760_cast_fp16, var_20919_cast_fp16))[name = tensor("op_20940_cast_fp16")]; + tensor var_20942_equation_0 = const()[name = tensor("op_20942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20942_cast_fp16 = einsum(equation = var_20942_equation_0, values = (var_20764_cast_fp16, var_20920_cast_fp16))[name = tensor("op_20942_cast_fp16")]; + tensor var_20944_equation_0 = const()[name = tensor("op_20944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20944_cast_fp16 = einsum(equation = var_20944_equation_0, values = (var_20768_cast_fp16, var_20921_cast_fp16))[name = tensor("op_20944_cast_fp16")]; + tensor var_20946_equation_0 = const()[name = tensor("op_20946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20946_cast_fp16 = einsum(equation = var_20946_equation_0, values = (var_20772_cast_fp16, var_20922_cast_fp16))[name = tensor("op_20946_cast_fp16")]; + tensor var_20948_equation_0 = const()[name = tensor("op_20948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20948_cast_fp16 = einsum(equation = var_20948_equation_0, values = (var_20776_cast_fp16, var_20923_cast_fp16))[name = tensor("op_20948_cast_fp16")]; + tensor var_20950_equation_0 = const()[name = tensor("op_20950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20950_cast_fp16 = einsum(equation = var_20950_equation_0, values = (var_20780_cast_fp16, var_20924_cast_fp16))[name = tensor("op_20950_cast_fp16")]; + tensor var_20952_equation_0 = const()[name = tensor("op_20952_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20952_cast_fp16 = einsum(equation = var_20952_equation_0, values = (var_20784_cast_fp16, var_20925_cast_fp16))[name = tensor("op_20952_cast_fp16")]; + tensor var_20954_equation_0 = const()[name = tensor("op_20954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20954_cast_fp16 = einsum(equation = var_20954_equation_0, values = (var_20788_cast_fp16, var_20926_cast_fp16))[name = tensor("op_20954_cast_fp16")]; + tensor var_20956_equation_0 = const()[name = tensor("op_20956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20956_cast_fp16 = einsum(equation = var_20956_equation_0, values = (var_20792_cast_fp16, var_20927_cast_fp16))[name = tensor("op_20956_cast_fp16")]; + tensor var_20958_equation_0 = const()[name = tensor("op_20958_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20958_cast_fp16 = einsum(equation = var_20958_equation_0, values = (var_20796_cast_fp16, var_20928_cast_fp16))[name = tensor("op_20958_cast_fp16")]; + tensor var_20960_equation_0 = const()[name = tensor("op_20960_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20960_cast_fp16 = einsum(equation = var_20960_equation_0, values = (var_20800_cast_fp16, var_20929_cast_fp16))[name = tensor("op_20960_cast_fp16")]; + tensor var_20962_equation_0 = const()[name = tensor("op_20962_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20962_cast_fp16 = einsum(equation = var_20962_equation_0, values = (var_20804_cast_fp16, var_20930_cast_fp16))[name = tensor("op_20962_cast_fp16")]; + tensor var_20964_equation_0 = const()[name = tensor("op_20964_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20964_cast_fp16 = einsum(equation = var_20964_equation_0, values = (var_20808_cast_fp16, var_20931_cast_fp16))[name = tensor("op_20964_cast_fp16")]; + tensor var_20966_equation_0 = const()[name = tensor("op_20966_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20966_cast_fp16 = einsum(equation = var_20966_equation_0, values = (var_20812_cast_fp16, var_20932_cast_fp16))[name = tensor("op_20966_cast_fp16")]; + tensor var_20968_equation_0 = const()[name = tensor("op_20968_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20968_cast_fp16 = einsum(equation = var_20968_equation_0, values = (var_20816_cast_fp16, var_20933_cast_fp16))[name = tensor("op_20968_cast_fp16")]; + tensor var_20970_equation_0 = const()[name = tensor("op_20970_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20970_cast_fp16 = einsum(equation = var_20970_equation_0, values = (var_20820_cast_fp16, var_20934_cast_fp16))[name = tensor("op_20970_cast_fp16")]; + tensor var_20972_equation_0 = const()[name = tensor("op_20972_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20972_cast_fp16 = einsum(equation = var_20972_equation_0, values = (var_20824_cast_fp16, var_20935_cast_fp16))[name = tensor("op_20972_cast_fp16")]; + tensor var_20974_equation_0 = const()[name = tensor("op_20974_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20974_cast_fp16 = einsum(equation = var_20974_equation_0, values = (var_20828_cast_fp16, var_20936_cast_fp16))[name = tensor("op_20974_cast_fp16")]; + tensor var_20976_equation_0 = const()[name = tensor("op_20976_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20976_cast_fp16 = einsum(equation = var_20976_equation_0, values = (var_20832_cast_fp16, var_20937_cast_fp16))[name = tensor("op_20976_cast_fp16")]; + tensor var_20978_equation_0 = const()[name = tensor("op_20978_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20978_cast_fp16 = einsum(equation = var_20978_equation_0, values = (var_20836_cast_fp16, var_20938_cast_fp16))[name = tensor("op_20978_cast_fp16")]; + tensor input_303_interleave_0 = const()[name = tensor("input_303_interleave_0"), val = tensor(false)]; + tensor input_303_cast_fp16 = concat(axis = var_2624, interleave = input_303_interleave_0, values = (var_20940_cast_fp16, var_20942_cast_fp16, var_20944_cast_fp16, var_20946_cast_fp16, var_20948_cast_fp16, var_20950_cast_fp16, var_20952_cast_fp16, var_20954_cast_fp16, var_20956_cast_fp16, var_20958_cast_fp16, var_20960_cast_fp16, var_20962_cast_fp16, var_20964_cast_fp16, var_20966_cast_fp16, var_20968_cast_fp16, var_20970_cast_fp16, var_20972_cast_fp16, var_20974_cast_fp16, var_20976_cast_fp16, var_20978_cast_fp16))[name = tensor("input_303_cast_fp16")]; + tensor var_20988_pad_type_0 = const()[name = tensor("op_20988_pad_type_0"), val = tensor("valid")]; + tensor var_20988_strides_0 = const()[name = tensor("op_20988_strides_0"), val = tensor([1, 1])]; + tensor var_20988_pad_0 = const()[name = tensor("op_20988_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20988_dilations_0 = const()[name = tensor("op_20988_dilations_0"), val = tensor([1, 1])]; + tensor var_20988_groups_0 = const()[name = tensor("op_20988_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606214272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607443136))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607443328)))]; + tensor var_20988_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_20988_dilations_0, groups = var_20988_groups_0, pad = var_20988_pad_0, pad_type = var_20988_pad_type_0, strides = var_20988_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized, x = input_303_cast_fp16)[name = tensor("op_20988_cast_fp16")]; + tensor inputs_143_cast_fp16 = add(x = var_20988_cast_fp16, y = inputs_141_cast_fp16)[name = tensor("inputs_143_cast_fp16")]; + tensor input_305_axes_0 = const()[name = tensor("input_305_axes_0"), val = tensor([1])]; + tensor input_305_gamma_0_to_fp16 = const()[name = tensor("input_305_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607445952)))]; + tensor input_305_beta_0_to_fp16 = const()[name = tensor("input_305_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607448576)))]; + tensor var_20998_to_fp16 = const()[name = tensor("op_20998_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_305_cast_fp16 = layer_norm(axes = input_305_axes_0, beta = input_305_beta_0_to_fp16, epsilon = var_20998_to_fp16, gamma = input_305_gamma_0_to_fp16, x = inputs_143_cast_fp16)[name = tensor("input_305_cast_fp16")]; + tensor var_21018_pad_type_0 = const()[name = tensor("op_21018_pad_type_0"), val = tensor("valid")]; + tensor var_21018_strides_0 = const()[name = tensor("op_21018_strides_0"), val = tensor([1, 1])]; + tensor var_21018_pad_0 = const()[name = tensor("op_21018_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21018_dilations_0 = const()[name = tensor("op_21018_dilations_0"), val = tensor([1, 1])]; + tensor var_21018_groups_0 = const()[name = tensor("op_21018_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607451200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617281664))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617281856)))]; + tensor var_21018_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_21018_dilations_0, groups = var_21018_groups_0, pad = var_21018_pad_0, pad_type = var_21018_pad_type_0, strides = var_21018_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized, x = input_305_cast_fp16)[name = tensor("op_21018_cast_fp16")]; + tensor var_21019_split_sizes_0 = const()[name = tensor("op_21019_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_21019_axis_0 = const()[name = tensor("op_21019_axis_0"), val = tensor(1)]; + tensor var_21019_cast_fp16_0, tensor var_21019_cast_fp16_1 = split(axis = var_21019_axis_0, split_sizes = var_21019_split_sizes_0, x = var_21018_cast_fp16)[name = tensor("op_21019_cast_fp16")]; + tensor var_21021_mode_0 = const()[name = tensor("op_21021_mode_0"), val = tensor("EXACT")]; + tensor var_21021_cast_fp16 = gelu(mode = var_21021_mode_0, x = var_21019_cast_fp16_1)[name = tensor("op_21021_cast_fp16")]; + tensor input_307_cast_fp16 = mul(x = var_21019_cast_fp16_0, y = var_21021_cast_fp16)[name = tensor("input_307_cast_fp16")]; + tensor var_21029_pad_type_0 = const()[name = tensor("op_21029_pad_type_0"), val = tensor("valid")]; + tensor var_21029_strides_0 = const()[name = tensor("op_21029_strides_0"), val = tensor([1, 1])]; + tensor var_21029_pad_0 = const()[name = tensor("op_21029_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21029_dilations_0 = const()[name = tensor("op_21029_dilations_0"), val = tensor([1, 1])]; + tensor var_21029_groups_0 = const()[name = tensor("op_21029_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617302400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622217664))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622217856)))]; + tensor var_21029_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_21029_dilations_0, groups = var_21029_groups_0, pad = var_21029_pad_0, pad_type = var_21029_pad_type_0, strides = var_21029_strides_0, weight = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized, x = input_307_cast_fp16)[name = tensor("op_21029_cast_fp16")]; + tensor hidden_states_197_cast_fp16 = add(x = var_21029_cast_fp16, y = inputs_143_cast_fp16)[name = tensor("hidden_states_197_cast_fp16")]; + tensor var_21031 = const()[name = tensor("op_21031"), val = tensor([2, 1280, 32, 32])]; + tensor input_309_cast_fp16 = reshape(shape = var_21031, x = hidden_states_197_cast_fp16)[name = tensor("input_309_cast_fp16")]; + tensor hidden_states_199_pad_type_0 = const()[name = tensor("hidden_states_199_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_199_strides_0 = const()[name = tensor("hidden_states_199_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_199_pad_0 = const()[name = tensor("hidden_states_199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_199_dilations_0 = const()[name = tensor("hidden_states_199_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_199_groups_0 = const()[name = tensor("hidden_states_199_groups_0"), val = tensor(1)]; + tensor down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622220480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(623449344))), name = tensor("down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(623449536)))]; + tensor hidden_states_199_cast_fp16 = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_199_dilations_0, groups = hidden_states_199_groups_0, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = hidden_states_199_strides_0, weight = down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized, x = input_309_cast_fp16)[name = tensor("hidden_states_199_cast_fp16")]; + tensor input_311_cast_fp16 = add(x = hidden_states_199_cast_fp16, y = hidden_states_133_cast_fp16)[name = tensor("input_311_cast_fp16")]; + tensor var_21077 = const()[name = tensor("op_21077"), val = tensor(1)]; + tensor reshape_64_shape_0 = const()[name = tensor("reshape_64_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_64_cast_fp16 = reshape(shape = reshape_64_shape_0, x = input_311_cast_fp16)[name = tensor("reshape_64_cast_fp16")]; + tensor reduce_mean_48_axes_0 = const()[name = tensor("reduce_mean_48_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_48_keep_dims_0 = const()[name = tensor("reduce_mean_48_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_48_cast_fp16 = reduce_mean(axes = reduce_mean_48_axes_0, keep_dims = reduce_mean_48_keep_dims_0, x = reshape_64_cast_fp16)[name = tensor("reduce_mean_48_cast_fp16")]; + tensor sub_32_cast_fp16 = sub(x = reshape_64_cast_fp16, y = reduce_mean_48_cast_fp16)[name = tensor("sub_32_cast_fp16")]; + tensor square_16_cast_fp16 = square(x = sub_32_cast_fp16)[name = tensor("square_16_cast_fp16")]; + tensor reduce_mean_50_axes_0 = const()[name = tensor("reduce_mean_50_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_50_keep_dims_0 = const()[name = tensor("reduce_mean_50_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_50_cast_fp16 = reduce_mean(axes = reduce_mean_50_axes_0, keep_dims = reduce_mean_50_keep_dims_0, x = square_16_cast_fp16)[name = tensor("reduce_mean_50_cast_fp16")]; + tensor add_32_y_0_to_fp16 = const()[name = tensor("add_32_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_32_cast_fp16 = add(x = reduce_mean_50_cast_fp16, y = add_32_y_0_to_fp16)[name = tensor("add_32_cast_fp16")]; + tensor sqrt_16_cast_fp16 = sqrt(x = add_32_cast_fp16)[name = tensor("sqrt_16_cast_fp16")]; + tensor real_div_16_cast_fp16 = real_div(x = sub_32_cast_fp16, y = sqrt_16_cast_fp16)[name = tensor("real_div_16_cast_fp16")]; + tensor reshape_65_shape_0 = const()[name = tensor("reshape_65_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_65_cast_fp16 = reshape(shape = reshape_65_shape_0, x = real_div_16_cast_fp16)[name = tensor("reshape_65_cast_fp16")]; + tensor add_33_gamma_0_to_fp16 = const()[name = tensor("add_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(623452160)))]; + tensor add_33_beta_0_to_fp16 = const()[name = tensor("add_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(623454784)))]; + tensor add_33_epsilon_0_to_fp16 = const()[name = tensor("add_33_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_33_cast_fp16 = batch_norm(beta = add_33_beta_0_to_fp16, epsilon = add_33_epsilon_0_to_fp16, gamma = add_33_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_65_cast_fp16)[name = tensor("add_33_cast_fp16")]; + tensor input_315_cast_fp16 = silu(x = add_33_cast_fp16)[name = tensor("input_315_cast_fp16")]; + tensor hidden_states_201_pad_type_0 = const()[name = tensor("hidden_states_201_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_201_pad_0 = const()[name = tensor("hidden_states_201_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_201_strides_0 = const()[name = tensor("hidden_states_201_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_201_dilations_0 = const()[name = tensor("hidden_states_201_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_201_groups_0 = const()[name = tensor("hidden_states_201_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(623457408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(634516672))), name = tensor("mid_block_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(634516864)))]; + tensor hidden_states_201_cast_fp16 = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_201_dilations_0, groups = hidden_states_201_groups_0, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = hidden_states_201_strides_0, weight = mid_block_resnets_0_conv1_weight_to_fp16_palettized, x = input_315_cast_fp16)[name = tensor("hidden_states_201_cast_fp16")]; + tensor temb_13_pad_type_0 = const()[name = tensor("temb_13_pad_type_0"), val = tensor("valid")]; + tensor temb_13_strides_0 = const()[name = tensor("temb_13_strides_0"), val = tensor([1, 1])]; + tensor temb_13_pad_0 = const()[name = tensor("temb_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_13_dilations_0 = const()[name = tensor("temb_13_dilations_0"), val = tensor([1, 1])]; + tensor temb_13_groups_0 = const()[name = tensor("temb_13_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(634519488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635748352))), name = tensor("mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635748544)))]; + tensor temb_13_cast_fp16 = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_13_dilations_0, groups = temb_13_groups_0, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = temb_13_strides_0, weight = mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_13_cast_fp16")]; + tensor input_319_cast_fp16 = add(x = hidden_states_201_cast_fp16, y = temb_13_cast_fp16)[name = tensor("input_319_cast_fp16")]; + tensor reshape_68_shape_0 = const()[name = tensor("reshape_68_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_68_cast_fp16 = reshape(shape = reshape_68_shape_0, x = input_319_cast_fp16)[name = tensor("reshape_68_cast_fp16")]; + tensor reduce_mean_51_axes_0 = const()[name = tensor("reduce_mean_51_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_51_keep_dims_0 = const()[name = tensor("reduce_mean_51_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_51_cast_fp16 = reduce_mean(axes = reduce_mean_51_axes_0, keep_dims = reduce_mean_51_keep_dims_0, x = reshape_68_cast_fp16)[name = tensor("reduce_mean_51_cast_fp16")]; + tensor sub_34_cast_fp16 = sub(x = reshape_68_cast_fp16, y = reduce_mean_51_cast_fp16)[name = tensor("sub_34_cast_fp16")]; + tensor square_17_cast_fp16 = square(x = sub_34_cast_fp16)[name = tensor("square_17_cast_fp16")]; + tensor reduce_mean_53_axes_0 = const()[name = tensor("reduce_mean_53_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_53_keep_dims_0 = const()[name = tensor("reduce_mean_53_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_53_cast_fp16 = reduce_mean(axes = reduce_mean_53_axes_0, keep_dims = reduce_mean_53_keep_dims_0, x = square_17_cast_fp16)[name = tensor("reduce_mean_53_cast_fp16")]; + tensor add_34_y_0_to_fp16 = const()[name = tensor("add_34_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_34_cast_fp16 = add(x = reduce_mean_53_cast_fp16, y = add_34_y_0_to_fp16)[name = tensor("add_34_cast_fp16")]; + tensor sqrt_17_cast_fp16 = sqrt(x = add_34_cast_fp16)[name = tensor("sqrt_17_cast_fp16")]; + tensor real_div_17_cast_fp16 = real_div(x = sub_34_cast_fp16, y = sqrt_17_cast_fp16)[name = tensor("real_div_17_cast_fp16")]; + tensor reshape_69_shape_0 = const()[name = tensor("reshape_69_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_69_cast_fp16 = reshape(shape = reshape_69_shape_0, x = real_div_17_cast_fp16)[name = tensor("reshape_69_cast_fp16")]; + tensor add_35_gamma_0_to_fp16 = const()[name = tensor("add_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635751168)))]; + tensor add_35_beta_0_to_fp16 = const()[name = tensor("add_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635753792)))]; + tensor add_35_epsilon_0_to_fp16 = const()[name = tensor("add_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_35_cast_fp16 = batch_norm(beta = add_35_beta_0_to_fp16, epsilon = add_35_epsilon_0_to_fp16, gamma = add_35_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_69_cast_fp16)[name = tensor("add_35_cast_fp16")]; + tensor input_323_cast_fp16 = silu(x = add_35_cast_fp16)[name = tensor("input_323_cast_fp16")]; + tensor hidden_states_203_pad_type_0 = const()[name = tensor("hidden_states_203_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_203_pad_0 = const()[name = tensor("hidden_states_203_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_203_strides_0 = const()[name = tensor("hidden_states_203_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_203_dilations_0 = const()[name = tensor("hidden_states_203_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_203_groups_0 = const()[name = tensor("hidden_states_203_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635756416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646815680))), name = tensor("mid_block_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646815872)))]; + tensor hidden_states_203_cast_fp16 = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_203_dilations_0, groups = hidden_states_203_groups_0, pad = hidden_states_203_pad_0, pad_type = hidden_states_203_pad_type_0, strides = hidden_states_203_strides_0, weight = mid_block_resnets_0_conv2_weight_to_fp16_palettized, x = input_323_cast_fp16)[name = tensor("hidden_states_203_cast_fp16")]; + tensor hidden_states_205_cast_fp16 = add(x = input_311_cast_fp16, y = hidden_states_203_cast_fp16)[name = tensor("hidden_states_205_cast_fp16")]; + tensor reshape_72_shape_0 = const()[name = tensor("reshape_72_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_72_cast_fp16 = reshape(shape = reshape_72_shape_0, x = hidden_states_205_cast_fp16)[name = tensor("reshape_72_cast_fp16")]; + tensor reduce_mean_54_axes_0 = const()[name = tensor("reduce_mean_54_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_54_keep_dims_0 = const()[name = tensor("reduce_mean_54_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_54_cast_fp16 = reduce_mean(axes = reduce_mean_54_axes_0, keep_dims = reduce_mean_54_keep_dims_0, x = reshape_72_cast_fp16)[name = tensor("reduce_mean_54_cast_fp16")]; + tensor sub_36_cast_fp16 = sub(x = reshape_72_cast_fp16, y = reduce_mean_54_cast_fp16)[name = tensor("sub_36_cast_fp16")]; + tensor square_18_cast_fp16 = square(x = sub_36_cast_fp16)[name = tensor("square_18_cast_fp16")]; + tensor reduce_mean_56_axes_0 = const()[name = tensor("reduce_mean_56_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_56_keep_dims_0 = const()[name = tensor("reduce_mean_56_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_56_cast_fp16 = reduce_mean(axes = reduce_mean_56_axes_0, keep_dims = reduce_mean_56_keep_dims_0, x = square_18_cast_fp16)[name = tensor("reduce_mean_56_cast_fp16")]; + tensor add_36_y_0_to_fp16 = const()[name = tensor("add_36_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_36_cast_fp16 = add(x = reduce_mean_56_cast_fp16, y = add_36_y_0_to_fp16)[name = tensor("add_36_cast_fp16")]; + tensor sqrt_18_cast_fp16 = sqrt(x = add_36_cast_fp16)[name = tensor("sqrt_18_cast_fp16")]; + tensor real_div_18_cast_fp16 = real_div(x = sub_36_cast_fp16, y = sqrt_18_cast_fp16)[name = tensor("real_div_18_cast_fp16")]; + tensor reshape_73_shape_0 = const()[name = tensor("reshape_73_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_73_cast_fp16 = reshape(shape = reshape_73_shape_0, x = real_div_18_cast_fp16)[name = tensor("reshape_73_cast_fp16")]; + tensor add_37_gamma_0_to_fp16 = const()[name = tensor("add_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646818496)))]; + tensor add_37_beta_0_to_fp16 = const()[name = tensor("add_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646821120)))]; + tensor add_37_epsilon_0_to_fp16 = const()[name = tensor("add_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_37_cast_fp16 = batch_norm(beta = add_37_beta_0_to_fp16, epsilon = add_37_epsilon_0_to_fp16, gamma = add_37_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_73_cast_fp16)[name = tensor("add_37_cast_fp16")]; + tensor hidden_states_207_pad_type_0 = const()[name = tensor("hidden_states_207_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_207_strides_0 = const()[name = tensor("hidden_states_207_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_207_pad_0 = const()[name = tensor("hidden_states_207_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_207_dilations_0 = const()[name = tensor("hidden_states_207_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_207_groups_0 = const()[name = tensor("hidden_states_207_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646823744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648052608))), name = tensor("mid_block_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648052800)))]; + tensor hidden_states_207_cast_fp16 = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_207_dilations_0, groups = hidden_states_207_groups_0, pad = hidden_states_207_pad_0, pad_type = hidden_states_207_pad_type_0, strides = hidden_states_207_strides_0, weight = mid_block_attentions_0_proj_in_weight_to_fp16_palettized, x = add_37_cast_fp16)[name = tensor("hidden_states_207_cast_fp16")]; + tensor var_21162 = const()[name = tensor("op_21162"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_145_cast_fp16 = reshape(shape = var_21162, x = hidden_states_207_cast_fp16)[name = tensor("inputs_145_cast_fp16")]; + tensor hidden_states_209_axes_0 = const()[name = tensor("hidden_states_209_axes_0"), val = tensor([1])]; + tensor hidden_states_209_gamma_0_to_fp16 = const()[name = tensor("hidden_states_209_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648055424)))]; + tensor hidden_states_209_beta_0_to_fp16 = const()[name = tensor("hidden_states_209_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648058048)))]; + tensor var_21178_to_fp16 = const()[name = tensor("op_21178_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_209_cast_fp16 = layer_norm(axes = hidden_states_209_axes_0, beta = hidden_states_209_beta_0_to_fp16, epsilon = var_21178_to_fp16, gamma = hidden_states_209_gamma_0_to_fp16, x = inputs_145_cast_fp16)[name = tensor("hidden_states_209_cast_fp16")]; + tensor q_97_pad_type_0 = const()[name = tensor("q_97_pad_type_0"), val = tensor("valid")]; + tensor q_97_strides_0 = const()[name = tensor("q_97_strides_0"), val = tensor([1, 1])]; + tensor q_97_pad_0 = const()[name = tensor("q_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_97_dilations_0 = const()[name = tensor("q_97_dilations_0"), val = tensor([1, 1])]; + tensor q_97_groups_0 = const()[name = tensor("q_97_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648060672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649289536))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_97_cast_fp16 = conv(dilations = q_97_dilations_0, groups = q_97_groups_0, pad = q_97_pad_0, pad_type = q_97_pad_type_0, strides = q_97_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_209_cast_fp16)[name = tensor("q_97_cast_fp16")]; + tensor k_193_pad_type_0 = const()[name = tensor("k_193_pad_type_0"), val = tensor("valid")]; + tensor k_193_strides_0 = const()[name = tensor("k_193_strides_0"), val = tensor([1, 1])]; + tensor k_193_pad_0 = const()[name = tensor("k_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_193_dilations_0 = const()[name = tensor("k_193_dilations_0"), val = tensor([1, 1])]; + tensor k_193_groups_0 = const()[name = tensor("k_193_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649289728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650518592))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_193_cast_fp16 = conv(dilations = k_193_dilations_0, groups = k_193_groups_0, pad = k_193_pad_0, pad_type = k_193_pad_type_0, strides = k_193_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_209_cast_fp16)[name = tensor("k_193_cast_fp16")]; + tensor v_97_pad_type_0 = const()[name = tensor("v_97_pad_type_0"), val = tensor("valid")]; + tensor v_97_strides_0 = const()[name = tensor("v_97_strides_0"), val = tensor([1, 1])]; + tensor v_97_pad_0 = const()[name = tensor("v_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_97_dilations_0 = const()[name = tensor("v_97_dilations_0"), val = tensor([1, 1])]; + tensor v_97_groups_0 = const()[name = tensor("v_97_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650518784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(651747648))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_97_cast_fp16 = conv(dilations = v_97_dilations_0, groups = v_97_groups_0, pad = v_97_pad_0, pad_type = v_97_pad_type_0, strides = v_97_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_209_cast_fp16)[name = tensor("v_97_cast_fp16")]; + tensor var_21211_begin_0 = const()[name = tensor("op_21211_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21211_end_0 = const()[name = tensor("op_21211_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_21211_end_mask_0 = const()[name = tensor("op_21211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21211_cast_fp16 = slice_by_index(begin = var_21211_begin_0, end = var_21211_end_0, end_mask = var_21211_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21211_cast_fp16")]; + tensor var_21215_begin_0 = const()[name = tensor("op_21215_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21215_end_0 = const()[name = tensor("op_21215_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_21215_end_mask_0 = const()[name = tensor("op_21215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21215_cast_fp16 = slice_by_index(begin = var_21215_begin_0, end = var_21215_end_0, end_mask = var_21215_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21215_cast_fp16")]; + tensor var_21219_begin_0 = const()[name = tensor("op_21219_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21219_end_0 = const()[name = tensor("op_21219_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_21219_end_mask_0 = const()[name = tensor("op_21219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21219_cast_fp16 = slice_by_index(begin = var_21219_begin_0, end = var_21219_end_0, end_mask = var_21219_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21219_cast_fp16")]; + tensor var_21223_begin_0 = const()[name = tensor("op_21223_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21223_end_0 = const()[name = tensor("op_21223_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_21223_end_mask_0 = const()[name = tensor("op_21223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21223_cast_fp16 = slice_by_index(begin = var_21223_begin_0, end = var_21223_end_0, end_mask = var_21223_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21223_cast_fp16")]; + tensor var_21227_begin_0 = const()[name = tensor("op_21227_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21227_end_0 = const()[name = tensor("op_21227_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_21227_end_mask_0 = const()[name = tensor("op_21227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21227_cast_fp16 = slice_by_index(begin = var_21227_begin_0, end = var_21227_end_0, end_mask = var_21227_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21227_cast_fp16")]; + tensor var_21231_begin_0 = const()[name = tensor("op_21231_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21231_end_0 = const()[name = tensor("op_21231_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_21231_end_mask_0 = const()[name = tensor("op_21231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21231_cast_fp16 = slice_by_index(begin = var_21231_begin_0, end = var_21231_end_0, end_mask = var_21231_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21231_cast_fp16")]; + tensor var_21235_begin_0 = const()[name = tensor("op_21235_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21235_end_0 = const()[name = tensor("op_21235_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_21235_end_mask_0 = const()[name = tensor("op_21235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21235_cast_fp16 = slice_by_index(begin = var_21235_begin_0, end = var_21235_end_0, end_mask = var_21235_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21235_cast_fp16")]; + tensor var_21239_begin_0 = const()[name = tensor("op_21239_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21239_end_0 = const()[name = tensor("op_21239_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_21239_end_mask_0 = const()[name = tensor("op_21239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21239_cast_fp16 = slice_by_index(begin = var_21239_begin_0, end = var_21239_end_0, end_mask = var_21239_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21239_cast_fp16")]; + tensor var_21243_begin_0 = const()[name = tensor("op_21243_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21243_end_0 = const()[name = tensor("op_21243_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_21243_end_mask_0 = const()[name = tensor("op_21243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21243_cast_fp16 = slice_by_index(begin = var_21243_begin_0, end = var_21243_end_0, end_mask = var_21243_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21243_cast_fp16")]; + tensor var_21247_begin_0 = const()[name = tensor("op_21247_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21247_end_0 = const()[name = tensor("op_21247_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_21247_end_mask_0 = const()[name = tensor("op_21247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21247_cast_fp16 = slice_by_index(begin = var_21247_begin_0, end = var_21247_end_0, end_mask = var_21247_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21247_cast_fp16")]; + tensor var_21251_begin_0 = const()[name = tensor("op_21251_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21251_end_0 = const()[name = tensor("op_21251_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_21251_end_mask_0 = const()[name = tensor("op_21251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21251_cast_fp16 = slice_by_index(begin = var_21251_begin_0, end = var_21251_end_0, end_mask = var_21251_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21251_cast_fp16")]; + tensor var_21255_begin_0 = const()[name = tensor("op_21255_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21255_end_0 = const()[name = tensor("op_21255_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_21255_end_mask_0 = const()[name = tensor("op_21255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21255_cast_fp16 = slice_by_index(begin = var_21255_begin_0, end = var_21255_end_0, end_mask = var_21255_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21255_cast_fp16")]; + tensor var_21259_begin_0 = const()[name = tensor("op_21259_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21259_end_0 = const()[name = tensor("op_21259_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_21259_end_mask_0 = const()[name = tensor("op_21259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21259_cast_fp16 = slice_by_index(begin = var_21259_begin_0, end = var_21259_end_0, end_mask = var_21259_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21259_cast_fp16")]; + tensor var_21263_begin_0 = const()[name = tensor("op_21263_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21263_end_0 = const()[name = tensor("op_21263_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_21263_end_mask_0 = const()[name = tensor("op_21263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21263_cast_fp16 = slice_by_index(begin = var_21263_begin_0, end = var_21263_end_0, end_mask = var_21263_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21263_cast_fp16")]; + tensor var_21267_begin_0 = const()[name = tensor("op_21267_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21267_end_0 = const()[name = tensor("op_21267_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_21267_end_mask_0 = const()[name = tensor("op_21267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21267_cast_fp16 = slice_by_index(begin = var_21267_begin_0, end = var_21267_end_0, end_mask = var_21267_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21267_cast_fp16")]; + tensor var_21271_begin_0 = const()[name = tensor("op_21271_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21271_end_0 = const()[name = tensor("op_21271_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21271_end_mask_0 = const()[name = tensor("op_21271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21271_cast_fp16 = slice_by_index(begin = var_21271_begin_0, end = var_21271_end_0, end_mask = var_21271_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21271_cast_fp16")]; + tensor var_21275_begin_0 = const()[name = tensor("op_21275_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21275_end_0 = const()[name = tensor("op_21275_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_21275_end_mask_0 = const()[name = tensor("op_21275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21275_cast_fp16 = slice_by_index(begin = var_21275_begin_0, end = var_21275_end_0, end_mask = var_21275_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21275_cast_fp16")]; + tensor var_21279_begin_0 = const()[name = tensor("op_21279_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21279_end_0 = const()[name = tensor("op_21279_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_21279_end_mask_0 = const()[name = tensor("op_21279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21279_cast_fp16 = slice_by_index(begin = var_21279_begin_0, end = var_21279_end_0, end_mask = var_21279_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21279_cast_fp16")]; + tensor var_21283_begin_0 = const()[name = tensor("op_21283_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21283_end_0 = const()[name = tensor("op_21283_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_21283_end_mask_0 = const()[name = tensor("op_21283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21283_cast_fp16 = slice_by_index(begin = var_21283_begin_0, end = var_21283_end_0, end_mask = var_21283_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21283_cast_fp16")]; + tensor var_21287_begin_0 = const()[name = tensor("op_21287_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21287_end_0 = const()[name = tensor("op_21287_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_21287_end_mask_0 = const()[name = tensor("op_21287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21287_cast_fp16 = slice_by_index(begin = var_21287_begin_0, end = var_21287_end_0, end_mask = var_21287_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21287_cast_fp16")]; + tensor k_195_perm_0 = const()[name = tensor("k_195_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_21294_begin_0 = const()[name = tensor("op_21294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21294_end_0 = const()[name = tensor("op_21294_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_21294_end_mask_0 = const()[name = tensor("op_21294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_195_cast_fp16 = transpose(perm = k_195_perm_0, x = k_193_cast_fp16)[name = tensor("transpose_19")]; + tensor var_21294_cast_fp16 = slice_by_index(begin = var_21294_begin_0, end = var_21294_end_0, end_mask = var_21294_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21294_cast_fp16")]; + tensor var_21298_begin_0 = const()[name = tensor("op_21298_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_21298_end_0 = const()[name = tensor("op_21298_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_21298_end_mask_0 = const()[name = tensor("op_21298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21298_cast_fp16 = slice_by_index(begin = var_21298_begin_0, end = var_21298_end_0, end_mask = var_21298_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21298_cast_fp16")]; + tensor var_21302_begin_0 = const()[name = tensor("op_21302_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_21302_end_0 = const()[name = tensor("op_21302_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_21302_end_mask_0 = const()[name = tensor("op_21302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21302_cast_fp16 = slice_by_index(begin = var_21302_begin_0, end = var_21302_end_0, end_mask = var_21302_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21302_cast_fp16")]; + tensor var_21306_begin_0 = const()[name = tensor("op_21306_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_21306_end_0 = const()[name = tensor("op_21306_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_21306_end_mask_0 = const()[name = tensor("op_21306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21306_cast_fp16 = slice_by_index(begin = var_21306_begin_0, end = var_21306_end_0, end_mask = var_21306_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21306_cast_fp16")]; + tensor var_21310_begin_0 = const()[name = tensor("op_21310_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_21310_end_0 = const()[name = tensor("op_21310_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_21310_end_mask_0 = const()[name = tensor("op_21310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21310_cast_fp16 = slice_by_index(begin = var_21310_begin_0, end = var_21310_end_0, end_mask = var_21310_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21310_cast_fp16")]; + tensor var_21314_begin_0 = const()[name = tensor("op_21314_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_21314_end_0 = const()[name = tensor("op_21314_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_21314_end_mask_0 = const()[name = tensor("op_21314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21314_cast_fp16 = slice_by_index(begin = var_21314_begin_0, end = var_21314_end_0, end_mask = var_21314_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21314_cast_fp16")]; + tensor var_21318_begin_0 = const()[name = tensor("op_21318_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_21318_end_0 = const()[name = tensor("op_21318_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_21318_end_mask_0 = const()[name = tensor("op_21318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21318_cast_fp16 = slice_by_index(begin = var_21318_begin_0, end = var_21318_end_0, end_mask = var_21318_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21318_cast_fp16")]; + tensor var_21322_begin_0 = const()[name = tensor("op_21322_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_21322_end_0 = const()[name = tensor("op_21322_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_21322_end_mask_0 = const()[name = tensor("op_21322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21322_cast_fp16 = slice_by_index(begin = var_21322_begin_0, end = var_21322_end_0, end_mask = var_21322_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21322_cast_fp16")]; + tensor var_21326_begin_0 = const()[name = tensor("op_21326_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_21326_end_0 = const()[name = tensor("op_21326_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_21326_end_mask_0 = const()[name = tensor("op_21326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21326_cast_fp16 = slice_by_index(begin = var_21326_begin_0, end = var_21326_end_0, end_mask = var_21326_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21326_cast_fp16")]; + tensor var_21330_begin_0 = const()[name = tensor("op_21330_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_21330_end_0 = const()[name = tensor("op_21330_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_21330_end_mask_0 = const()[name = tensor("op_21330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21330_cast_fp16 = slice_by_index(begin = var_21330_begin_0, end = var_21330_end_0, end_mask = var_21330_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21330_cast_fp16")]; + tensor var_21334_begin_0 = const()[name = tensor("op_21334_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_21334_end_0 = const()[name = tensor("op_21334_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_21334_end_mask_0 = const()[name = tensor("op_21334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21334_cast_fp16 = slice_by_index(begin = var_21334_begin_0, end = var_21334_end_0, end_mask = var_21334_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21334_cast_fp16")]; + tensor var_21338_begin_0 = const()[name = tensor("op_21338_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_21338_end_0 = const()[name = tensor("op_21338_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_21338_end_mask_0 = const()[name = tensor("op_21338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21338_cast_fp16 = slice_by_index(begin = var_21338_begin_0, end = var_21338_end_0, end_mask = var_21338_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21338_cast_fp16")]; + tensor var_21342_begin_0 = const()[name = tensor("op_21342_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_21342_end_0 = const()[name = tensor("op_21342_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_21342_end_mask_0 = const()[name = tensor("op_21342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21342_cast_fp16 = slice_by_index(begin = var_21342_begin_0, end = var_21342_end_0, end_mask = var_21342_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21342_cast_fp16")]; + tensor var_21346_begin_0 = const()[name = tensor("op_21346_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_21346_end_0 = const()[name = tensor("op_21346_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_21346_end_mask_0 = const()[name = tensor("op_21346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21346_cast_fp16 = slice_by_index(begin = var_21346_begin_0, end = var_21346_end_0, end_mask = var_21346_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21346_cast_fp16")]; + tensor var_21350_begin_0 = const()[name = tensor("op_21350_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_21350_end_0 = const()[name = tensor("op_21350_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_21350_end_mask_0 = const()[name = tensor("op_21350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21350_cast_fp16 = slice_by_index(begin = var_21350_begin_0, end = var_21350_end_0, end_mask = var_21350_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21350_cast_fp16")]; + tensor var_21354_begin_0 = const()[name = tensor("op_21354_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_21354_end_0 = const()[name = tensor("op_21354_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21354_end_mask_0 = const()[name = tensor("op_21354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21354_cast_fp16 = slice_by_index(begin = var_21354_begin_0, end = var_21354_end_0, end_mask = var_21354_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21354_cast_fp16")]; + tensor var_21358_begin_0 = const()[name = tensor("op_21358_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_21358_end_0 = const()[name = tensor("op_21358_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_21358_end_mask_0 = const()[name = tensor("op_21358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21358_cast_fp16 = slice_by_index(begin = var_21358_begin_0, end = var_21358_end_0, end_mask = var_21358_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21358_cast_fp16")]; + tensor var_21362_begin_0 = const()[name = tensor("op_21362_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_21362_end_0 = const()[name = tensor("op_21362_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_21362_end_mask_0 = const()[name = tensor("op_21362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21362_cast_fp16 = slice_by_index(begin = var_21362_begin_0, end = var_21362_end_0, end_mask = var_21362_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21362_cast_fp16")]; + tensor var_21366_begin_0 = const()[name = tensor("op_21366_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_21366_end_0 = const()[name = tensor("op_21366_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_21366_end_mask_0 = const()[name = tensor("op_21366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21366_cast_fp16 = slice_by_index(begin = var_21366_begin_0, end = var_21366_end_0, end_mask = var_21366_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21366_cast_fp16")]; + tensor var_21370_begin_0 = const()[name = tensor("op_21370_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_21370_end_0 = const()[name = tensor("op_21370_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_21370_end_mask_0 = const()[name = tensor("op_21370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21370_cast_fp16 = slice_by_index(begin = var_21370_begin_0, end = var_21370_end_0, end_mask = var_21370_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21370_cast_fp16")]; + tensor var_21372_begin_0 = const()[name = tensor("op_21372_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21372_end_0 = const()[name = tensor("op_21372_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_21372_end_mask_0 = const()[name = tensor("op_21372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21372_cast_fp16 = slice_by_index(begin = var_21372_begin_0, end = var_21372_end_0, end_mask = var_21372_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21372_cast_fp16")]; + tensor var_21376_begin_0 = const()[name = tensor("op_21376_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21376_end_0 = const()[name = tensor("op_21376_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_21376_end_mask_0 = const()[name = tensor("op_21376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21376_cast_fp16 = slice_by_index(begin = var_21376_begin_0, end = var_21376_end_0, end_mask = var_21376_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21376_cast_fp16")]; + tensor var_21380_begin_0 = const()[name = tensor("op_21380_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21380_end_0 = const()[name = tensor("op_21380_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_21380_end_mask_0 = const()[name = tensor("op_21380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21380_cast_fp16 = slice_by_index(begin = var_21380_begin_0, end = var_21380_end_0, end_mask = var_21380_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21380_cast_fp16")]; + tensor var_21384_begin_0 = const()[name = tensor("op_21384_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21384_end_0 = const()[name = tensor("op_21384_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_21384_end_mask_0 = const()[name = tensor("op_21384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21384_cast_fp16 = slice_by_index(begin = var_21384_begin_0, end = var_21384_end_0, end_mask = var_21384_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21384_cast_fp16")]; + tensor var_21388_begin_0 = const()[name = tensor("op_21388_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21388_end_0 = const()[name = tensor("op_21388_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_21388_end_mask_0 = const()[name = tensor("op_21388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21388_cast_fp16 = slice_by_index(begin = var_21388_begin_0, end = var_21388_end_0, end_mask = var_21388_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21388_cast_fp16")]; + tensor var_21392_begin_0 = const()[name = tensor("op_21392_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21392_end_0 = const()[name = tensor("op_21392_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_21392_end_mask_0 = const()[name = tensor("op_21392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21392_cast_fp16 = slice_by_index(begin = var_21392_begin_0, end = var_21392_end_0, end_mask = var_21392_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21392_cast_fp16")]; + tensor var_21396_begin_0 = const()[name = tensor("op_21396_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21396_end_0 = const()[name = tensor("op_21396_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_21396_end_mask_0 = const()[name = tensor("op_21396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21396_cast_fp16 = slice_by_index(begin = var_21396_begin_0, end = var_21396_end_0, end_mask = var_21396_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21396_cast_fp16")]; + tensor var_21400_begin_0 = const()[name = tensor("op_21400_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21400_end_0 = const()[name = tensor("op_21400_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_21400_end_mask_0 = const()[name = tensor("op_21400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21400_cast_fp16 = slice_by_index(begin = var_21400_begin_0, end = var_21400_end_0, end_mask = var_21400_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21400_cast_fp16")]; + tensor var_21404_begin_0 = const()[name = tensor("op_21404_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21404_end_0 = const()[name = tensor("op_21404_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_21404_end_mask_0 = const()[name = tensor("op_21404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21404_cast_fp16 = slice_by_index(begin = var_21404_begin_0, end = var_21404_end_0, end_mask = var_21404_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21404_cast_fp16")]; + tensor var_21408_begin_0 = const()[name = tensor("op_21408_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21408_end_0 = const()[name = tensor("op_21408_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_21408_end_mask_0 = const()[name = tensor("op_21408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21408_cast_fp16 = slice_by_index(begin = var_21408_begin_0, end = var_21408_end_0, end_mask = var_21408_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21408_cast_fp16")]; + tensor var_21412_begin_0 = const()[name = tensor("op_21412_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21412_end_0 = const()[name = tensor("op_21412_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_21412_end_mask_0 = const()[name = tensor("op_21412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21412_cast_fp16 = slice_by_index(begin = var_21412_begin_0, end = var_21412_end_0, end_mask = var_21412_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21412_cast_fp16")]; + tensor var_21416_begin_0 = const()[name = tensor("op_21416_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21416_end_0 = const()[name = tensor("op_21416_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_21416_end_mask_0 = const()[name = tensor("op_21416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21416_cast_fp16 = slice_by_index(begin = var_21416_begin_0, end = var_21416_end_0, end_mask = var_21416_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21416_cast_fp16")]; + tensor var_21420_begin_0 = const()[name = tensor("op_21420_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21420_end_0 = const()[name = tensor("op_21420_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_21420_end_mask_0 = const()[name = tensor("op_21420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21420_cast_fp16 = slice_by_index(begin = var_21420_begin_0, end = var_21420_end_0, end_mask = var_21420_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21420_cast_fp16")]; + tensor var_21424_begin_0 = const()[name = tensor("op_21424_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21424_end_0 = const()[name = tensor("op_21424_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_21424_end_mask_0 = const()[name = tensor("op_21424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21424_cast_fp16 = slice_by_index(begin = var_21424_begin_0, end = var_21424_end_0, end_mask = var_21424_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21424_cast_fp16")]; + tensor var_21428_begin_0 = const()[name = tensor("op_21428_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21428_end_0 = const()[name = tensor("op_21428_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_21428_end_mask_0 = const()[name = tensor("op_21428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21428_cast_fp16 = slice_by_index(begin = var_21428_begin_0, end = var_21428_end_0, end_mask = var_21428_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21428_cast_fp16")]; + tensor var_21432_begin_0 = const()[name = tensor("op_21432_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21432_end_0 = const()[name = tensor("op_21432_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21432_end_mask_0 = const()[name = tensor("op_21432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21432_cast_fp16 = slice_by_index(begin = var_21432_begin_0, end = var_21432_end_0, end_mask = var_21432_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21432_cast_fp16")]; + tensor var_21436_begin_0 = const()[name = tensor("op_21436_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21436_end_0 = const()[name = tensor("op_21436_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_21436_end_mask_0 = const()[name = tensor("op_21436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21436_cast_fp16 = slice_by_index(begin = var_21436_begin_0, end = var_21436_end_0, end_mask = var_21436_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21436_cast_fp16")]; + tensor var_21440_begin_0 = const()[name = tensor("op_21440_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21440_end_0 = const()[name = tensor("op_21440_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_21440_end_mask_0 = const()[name = tensor("op_21440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21440_cast_fp16 = slice_by_index(begin = var_21440_begin_0, end = var_21440_end_0, end_mask = var_21440_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21440_cast_fp16")]; + tensor var_21444_begin_0 = const()[name = tensor("op_21444_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21444_end_0 = const()[name = tensor("op_21444_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_21444_end_mask_0 = const()[name = tensor("op_21444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21444_cast_fp16 = slice_by_index(begin = var_21444_begin_0, end = var_21444_end_0, end_mask = var_21444_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21444_cast_fp16")]; + tensor var_21448_begin_0 = const()[name = tensor("op_21448_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21448_end_0 = const()[name = tensor("op_21448_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_21448_end_mask_0 = const()[name = tensor("op_21448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21448_cast_fp16 = slice_by_index(begin = var_21448_begin_0, end = var_21448_end_0, end_mask = var_21448_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21448_cast_fp16")]; + tensor var_21452_equation_0 = const()[name = tensor("op_21452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21452_cast_fp16 = einsum(equation = var_21452_equation_0, values = (var_21294_cast_fp16, var_21211_cast_fp16))[name = tensor("op_21452_cast_fp16")]; + tensor var_21453_to_fp16 = const()[name = tensor("op_21453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1761_cast_fp16 = mul(x = var_21452_cast_fp16, y = var_21453_to_fp16)[name = tensor("aw_1761_cast_fp16")]; + tensor var_21456_equation_0 = const()[name = tensor("op_21456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21456_cast_fp16 = einsum(equation = var_21456_equation_0, values = (var_21298_cast_fp16, var_21215_cast_fp16))[name = tensor("op_21456_cast_fp16")]; + tensor var_21457_to_fp16 = const()[name = tensor("op_21457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1763_cast_fp16 = mul(x = var_21456_cast_fp16, y = var_21457_to_fp16)[name = tensor("aw_1763_cast_fp16")]; + tensor var_21460_equation_0 = const()[name = tensor("op_21460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21460_cast_fp16 = einsum(equation = var_21460_equation_0, values = (var_21302_cast_fp16, var_21219_cast_fp16))[name = tensor("op_21460_cast_fp16")]; + tensor var_21461_to_fp16 = const()[name = tensor("op_21461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1765_cast_fp16 = mul(x = var_21460_cast_fp16, y = var_21461_to_fp16)[name = tensor("aw_1765_cast_fp16")]; + tensor var_21464_equation_0 = const()[name = tensor("op_21464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21464_cast_fp16 = einsum(equation = var_21464_equation_0, values = (var_21306_cast_fp16, var_21223_cast_fp16))[name = tensor("op_21464_cast_fp16")]; + tensor var_21465_to_fp16 = const()[name = tensor("op_21465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1767_cast_fp16 = mul(x = var_21464_cast_fp16, y = var_21465_to_fp16)[name = tensor("aw_1767_cast_fp16")]; + tensor var_21468_equation_0 = const()[name = tensor("op_21468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21468_cast_fp16 = einsum(equation = var_21468_equation_0, values = (var_21310_cast_fp16, var_21227_cast_fp16))[name = tensor("op_21468_cast_fp16")]; + tensor var_21469_to_fp16 = const()[name = tensor("op_21469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1769_cast_fp16 = mul(x = var_21468_cast_fp16, y = var_21469_to_fp16)[name = tensor("aw_1769_cast_fp16")]; + tensor var_21472_equation_0 = const()[name = tensor("op_21472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21472_cast_fp16 = einsum(equation = var_21472_equation_0, values = (var_21314_cast_fp16, var_21231_cast_fp16))[name = tensor("op_21472_cast_fp16")]; + tensor var_21473_to_fp16 = const()[name = tensor("op_21473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1771_cast_fp16 = mul(x = var_21472_cast_fp16, y = var_21473_to_fp16)[name = tensor("aw_1771_cast_fp16")]; + tensor var_21476_equation_0 = const()[name = tensor("op_21476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21476_cast_fp16 = einsum(equation = var_21476_equation_0, values = (var_21318_cast_fp16, var_21235_cast_fp16))[name = tensor("op_21476_cast_fp16")]; + tensor var_21477_to_fp16 = const()[name = tensor("op_21477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1773_cast_fp16 = mul(x = var_21476_cast_fp16, y = var_21477_to_fp16)[name = tensor("aw_1773_cast_fp16")]; + tensor var_21480_equation_0 = const()[name = tensor("op_21480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21480_cast_fp16 = einsum(equation = var_21480_equation_0, values = (var_21322_cast_fp16, var_21239_cast_fp16))[name = tensor("op_21480_cast_fp16")]; + tensor var_21481_to_fp16 = const()[name = tensor("op_21481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1775_cast_fp16 = mul(x = var_21480_cast_fp16, y = var_21481_to_fp16)[name = tensor("aw_1775_cast_fp16")]; + tensor var_21484_equation_0 = const()[name = tensor("op_21484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21484_cast_fp16 = einsum(equation = var_21484_equation_0, values = (var_21326_cast_fp16, var_21243_cast_fp16))[name = tensor("op_21484_cast_fp16")]; + tensor var_21485_to_fp16 = const()[name = tensor("op_21485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1777_cast_fp16 = mul(x = var_21484_cast_fp16, y = var_21485_to_fp16)[name = tensor("aw_1777_cast_fp16")]; + tensor var_21488_equation_0 = const()[name = tensor("op_21488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21488_cast_fp16 = einsum(equation = var_21488_equation_0, values = (var_21330_cast_fp16, var_21247_cast_fp16))[name = tensor("op_21488_cast_fp16")]; + tensor var_21489_to_fp16 = const()[name = tensor("op_21489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1779_cast_fp16 = mul(x = var_21488_cast_fp16, y = var_21489_to_fp16)[name = tensor("aw_1779_cast_fp16")]; + tensor var_21492_equation_0 = const()[name = tensor("op_21492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21492_cast_fp16 = einsum(equation = var_21492_equation_0, values = (var_21334_cast_fp16, var_21251_cast_fp16))[name = tensor("op_21492_cast_fp16")]; + tensor var_21493_to_fp16 = const()[name = tensor("op_21493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1781_cast_fp16 = mul(x = var_21492_cast_fp16, y = var_21493_to_fp16)[name = tensor("aw_1781_cast_fp16")]; + tensor var_21496_equation_0 = const()[name = tensor("op_21496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21496_cast_fp16 = einsum(equation = var_21496_equation_0, values = (var_21338_cast_fp16, var_21255_cast_fp16))[name = tensor("op_21496_cast_fp16")]; + tensor var_21497_to_fp16 = const()[name = tensor("op_21497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1783_cast_fp16 = mul(x = var_21496_cast_fp16, y = var_21497_to_fp16)[name = tensor("aw_1783_cast_fp16")]; + tensor var_21500_equation_0 = const()[name = tensor("op_21500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21500_cast_fp16 = einsum(equation = var_21500_equation_0, values = (var_21342_cast_fp16, var_21259_cast_fp16))[name = tensor("op_21500_cast_fp16")]; + tensor var_21501_to_fp16 = const()[name = tensor("op_21501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1785_cast_fp16 = mul(x = var_21500_cast_fp16, y = var_21501_to_fp16)[name = tensor("aw_1785_cast_fp16")]; + tensor var_21504_equation_0 = const()[name = tensor("op_21504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21504_cast_fp16 = einsum(equation = var_21504_equation_0, values = (var_21346_cast_fp16, var_21263_cast_fp16))[name = tensor("op_21504_cast_fp16")]; + tensor var_21505_to_fp16 = const()[name = tensor("op_21505_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1787_cast_fp16 = mul(x = var_21504_cast_fp16, y = var_21505_to_fp16)[name = tensor("aw_1787_cast_fp16")]; + tensor var_21508_equation_0 = const()[name = tensor("op_21508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21508_cast_fp16 = einsum(equation = var_21508_equation_0, values = (var_21350_cast_fp16, var_21267_cast_fp16))[name = tensor("op_21508_cast_fp16")]; + tensor var_21509_to_fp16 = const()[name = tensor("op_21509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1789_cast_fp16 = mul(x = var_21508_cast_fp16, y = var_21509_to_fp16)[name = tensor("aw_1789_cast_fp16")]; + tensor var_21512_equation_0 = const()[name = tensor("op_21512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21512_cast_fp16 = einsum(equation = var_21512_equation_0, values = (var_21354_cast_fp16, var_21271_cast_fp16))[name = tensor("op_21512_cast_fp16")]; + tensor var_21513_to_fp16 = const()[name = tensor("op_21513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1791_cast_fp16 = mul(x = var_21512_cast_fp16, y = var_21513_to_fp16)[name = tensor("aw_1791_cast_fp16")]; + tensor var_21516_equation_0 = const()[name = tensor("op_21516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21516_cast_fp16 = einsum(equation = var_21516_equation_0, values = (var_21358_cast_fp16, var_21275_cast_fp16))[name = tensor("op_21516_cast_fp16")]; + tensor var_21517_to_fp16 = const()[name = tensor("op_21517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1793_cast_fp16 = mul(x = var_21516_cast_fp16, y = var_21517_to_fp16)[name = tensor("aw_1793_cast_fp16")]; + tensor var_21520_equation_0 = const()[name = tensor("op_21520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21520_cast_fp16 = einsum(equation = var_21520_equation_0, values = (var_21362_cast_fp16, var_21279_cast_fp16))[name = tensor("op_21520_cast_fp16")]; + tensor var_21521_to_fp16 = const()[name = tensor("op_21521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1795_cast_fp16 = mul(x = var_21520_cast_fp16, y = var_21521_to_fp16)[name = tensor("aw_1795_cast_fp16")]; + tensor var_21524_equation_0 = const()[name = tensor("op_21524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21524_cast_fp16 = einsum(equation = var_21524_equation_0, values = (var_21366_cast_fp16, var_21283_cast_fp16))[name = tensor("op_21524_cast_fp16")]; + tensor var_21525_to_fp16 = const()[name = tensor("op_21525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1797_cast_fp16 = mul(x = var_21524_cast_fp16, y = var_21525_to_fp16)[name = tensor("aw_1797_cast_fp16")]; + tensor var_21528_equation_0 = const()[name = tensor("op_21528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21528_cast_fp16 = einsum(equation = var_21528_equation_0, values = (var_21370_cast_fp16, var_21287_cast_fp16))[name = tensor("op_21528_cast_fp16")]; + tensor var_21529_to_fp16 = const()[name = tensor("op_21529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1799_cast_fp16 = mul(x = var_21528_cast_fp16, y = var_21529_to_fp16)[name = tensor("aw_1799_cast_fp16")]; + tensor var_21531_cast_fp16 = softmax(axis = var_21077, x = aw_1761_cast_fp16)[name = tensor("op_21531_cast_fp16")]; + tensor var_21532_cast_fp16 = softmax(axis = var_21077, x = aw_1763_cast_fp16)[name = tensor("op_21532_cast_fp16")]; + tensor var_21533_cast_fp16 = softmax(axis = var_21077, x = aw_1765_cast_fp16)[name = tensor("op_21533_cast_fp16")]; + tensor var_21534_cast_fp16 = softmax(axis = var_21077, x = aw_1767_cast_fp16)[name = tensor("op_21534_cast_fp16")]; + tensor var_21535_cast_fp16 = softmax(axis = var_21077, x = aw_1769_cast_fp16)[name = tensor("op_21535_cast_fp16")]; + tensor var_21536_cast_fp16 = softmax(axis = var_21077, x = aw_1771_cast_fp16)[name = tensor("op_21536_cast_fp16")]; + tensor var_21537_cast_fp16 = softmax(axis = var_21077, x = aw_1773_cast_fp16)[name = tensor("op_21537_cast_fp16")]; + tensor var_21538_cast_fp16 = softmax(axis = var_21077, x = aw_1775_cast_fp16)[name = tensor("op_21538_cast_fp16")]; + tensor var_21539_cast_fp16 = softmax(axis = var_21077, x = aw_1777_cast_fp16)[name = tensor("op_21539_cast_fp16")]; + tensor var_21540_cast_fp16 = softmax(axis = var_21077, x = aw_1779_cast_fp16)[name = tensor("op_21540_cast_fp16")]; + tensor var_21541_cast_fp16 = softmax(axis = var_21077, x = aw_1781_cast_fp16)[name = tensor("op_21541_cast_fp16")]; + tensor var_21542_cast_fp16 = softmax(axis = var_21077, x = aw_1783_cast_fp16)[name = tensor("op_21542_cast_fp16")]; + tensor var_21543_cast_fp16 = softmax(axis = var_21077, x = aw_1785_cast_fp16)[name = tensor("op_21543_cast_fp16")]; + tensor var_21544_cast_fp16 = softmax(axis = var_21077, x = aw_1787_cast_fp16)[name = tensor("op_21544_cast_fp16")]; + tensor var_21545_cast_fp16 = softmax(axis = var_21077, x = aw_1789_cast_fp16)[name = tensor("op_21545_cast_fp16")]; + tensor var_21546_cast_fp16 = softmax(axis = var_21077, x = aw_1791_cast_fp16)[name = tensor("op_21546_cast_fp16")]; + tensor var_21547_cast_fp16 = softmax(axis = var_21077, x = aw_1793_cast_fp16)[name = tensor("op_21547_cast_fp16")]; + tensor var_21548_cast_fp16 = softmax(axis = var_21077, x = aw_1795_cast_fp16)[name = tensor("op_21548_cast_fp16")]; + tensor var_21549_cast_fp16 = softmax(axis = var_21077, x = aw_1797_cast_fp16)[name = tensor("op_21549_cast_fp16")]; + tensor var_21550_cast_fp16 = softmax(axis = var_21077, x = aw_1799_cast_fp16)[name = tensor("op_21550_cast_fp16")]; + tensor var_21552_equation_0 = const()[name = tensor("op_21552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21552_cast_fp16 = einsum(equation = var_21552_equation_0, values = (var_21372_cast_fp16, var_21531_cast_fp16))[name = tensor("op_21552_cast_fp16")]; + tensor var_21554_equation_0 = const()[name = tensor("op_21554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21554_cast_fp16 = einsum(equation = var_21554_equation_0, values = (var_21376_cast_fp16, var_21532_cast_fp16))[name = tensor("op_21554_cast_fp16")]; + tensor var_21556_equation_0 = const()[name = tensor("op_21556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21556_cast_fp16 = einsum(equation = var_21556_equation_0, values = (var_21380_cast_fp16, var_21533_cast_fp16))[name = tensor("op_21556_cast_fp16")]; + tensor var_21558_equation_0 = const()[name = tensor("op_21558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21558_cast_fp16 = einsum(equation = var_21558_equation_0, values = (var_21384_cast_fp16, var_21534_cast_fp16))[name = tensor("op_21558_cast_fp16")]; + tensor var_21560_equation_0 = const()[name = tensor("op_21560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21560_cast_fp16 = einsum(equation = var_21560_equation_0, values = (var_21388_cast_fp16, var_21535_cast_fp16))[name = tensor("op_21560_cast_fp16")]; + tensor var_21562_equation_0 = const()[name = tensor("op_21562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21562_cast_fp16 = einsum(equation = var_21562_equation_0, values = (var_21392_cast_fp16, var_21536_cast_fp16))[name = tensor("op_21562_cast_fp16")]; + tensor var_21564_equation_0 = const()[name = tensor("op_21564_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21564_cast_fp16 = einsum(equation = var_21564_equation_0, values = (var_21396_cast_fp16, var_21537_cast_fp16))[name = tensor("op_21564_cast_fp16")]; + tensor var_21566_equation_0 = const()[name = tensor("op_21566_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21566_cast_fp16 = einsum(equation = var_21566_equation_0, values = (var_21400_cast_fp16, var_21538_cast_fp16))[name = tensor("op_21566_cast_fp16")]; + tensor var_21568_equation_0 = const()[name = tensor("op_21568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21568_cast_fp16 = einsum(equation = var_21568_equation_0, values = (var_21404_cast_fp16, var_21539_cast_fp16))[name = tensor("op_21568_cast_fp16")]; + tensor var_21570_equation_0 = const()[name = tensor("op_21570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21570_cast_fp16 = einsum(equation = var_21570_equation_0, values = (var_21408_cast_fp16, var_21540_cast_fp16))[name = tensor("op_21570_cast_fp16")]; + tensor var_21572_equation_0 = const()[name = tensor("op_21572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21572_cast_fp16 = einsum(equation = var_21572_equation_0, values = (var_21412_cast_fp16, var_21541_cast_fp16))[name = tensor("op_21572_cast_fp16")]; + tensor var_21574_equation_0 = const()[name = tensor("op_21574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21574_cast_fp16 = einsum(equation = var_21574_equation_0, values = (var_21416_cast_fp16, var_21542_cast_fp16))[name = tensor("op_21574_cast_fp16")]; + tensor var_21576_equation_0 = const()[name = tensor("op_21576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21576_cast_fp16 = einsum(equation = var_21576_equation_0, values = (var_21420_cast_fp16, var_21543_cast_fp16))[name = tensor("op_21576_cast_fp16")]; + tensor var_21578_equation_0 = const()[name = tensor("op_21578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21578_cast_fp16 = einsum(equation = var_21578_equation_0, values = (var_21424_cast_fp16, var_21544_cast_fp16))[name = tensor("op_21578_cast_fp16")]; + tensor var_21580_equation_0 = const()[name = tensor("op_21580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21580_cast_fp16 = einsum(equation = var_21580_equation_0, values = (var_21428_cast_fp16, var_21545_cast_fp16))[name = tensor("op_21580_cast_fp16")]; + tensor var_21582_equation_0 = const()[name = tensor("op_21582_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21582_cast_fp16 = einsum(equation = var_21582_equation_0, values = (var_21432_cast_fp16, var_21546_cast_fp16))[name = tensor("op_21582_cast_fp16")]; + tensor var_21584_equation_0 = const()[name = tensor("op_21584_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21584_cast_fp16 = einsum(equation = var_21584_equation_0, values = (var_21436_cast_fp16, var_21547_cast_fp16))[name = tensor("op_21584_cast_fp16")]; + tensor var_21586_equation_0 = const()[name = tensor("op_21586_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21586_cast_fp16 = einsum(equation = var_21586_equation_0, values = (var_21440_cast_fp16, var_21548_cast_fp16))[name = tensor("op_21586_cast_fp16")]; + tensor var_21588_equation_0 = const()[name = tensor("op_21588_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21588_cast_fp16 = einsum(equation = var_21588_equation_0, values = (var_21444_cast_fp16, var_21549_cast_fp16))[name = tensor("op_21588_cast_fp16")]; + tensor var_21590_equation_0 = const()[name = tensor("op_21590_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21590_cast_fp16 = einsum(equation = var_21590_equation_0, values = (var_21448_cast_fp16, var_21550_cast_fp16))[name = tensor("op_21590_cast_fp16")]; + tensor input_327_interleave_0 = const()[name = tensor("input_327_interleave_0"), val = tensor(false)]; + tensor input_327_cast_fp16 = concat(axis = var_21077, interleave = input_327_interleave_0, values = (var_21552_cast_fp16, var_21554_cast_fp16, var_21556_cast_fp16, var_21558_cast_fp16, var_21560_cast_fp16, var_21562_cast_fp16, var_21564_cast_fp16, var_21566_cast_fp16, var_21568_cast_fp16, var_21570_cast_fp16, var_21572_cast_fp16, var_21574_cast_fp16, var_21576_cast_fp16, var_21578_cast_fp16, var_21580_cast_fp16, var_21582_cast_fp16, var_21584_cast_fp16, var_21586_cast_fp16, var_21588_cast_fp16, var_21590_cast_fp16))[name = tensor("input_327_cast_fp16")]; + tensor var_21600_pad_type_0 = const()[name = tensor("op_21600_pad_type_0"), val = tensor("valid")]; + tensor var_21600_strides_0 = const()[name = tensor("op_21600_strides_0"), val = tensor([1, 1])]; + tensor var_21600_pad_0 = const()[name = tensor("op_21600_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21600_dilations_0 = const()[name = tensor("op_21600_dilations_0"), val = tensor([1, 1])]; + tensor var_21600_groups_0 = const()[name = tensor("op_21600_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(651747840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(652976704))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(652976896)))]; + tensor var_21600_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_21600_dilations_0, groups = var_21600_groups_0, pad = var_21600_pad_0, pad_type = var_21600_pad_type_0, strides = var_21600_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_327_cast_fp16)[name = tensor("op_21600_cast_fp16")]; + tensor inputs_147_cast_fp16 = add(x = var_21600_cast_fp16, y = inputs_145_cast_fp16)[name = tensor("inputs_147_cast_fp16")]; + tensor hidden_states_211_axes_0 = const()[name = tensor("hidden_states_211_axes_0"), val = tensor([1])]; + tensor hidden_states_211_gamma_0_to_fp16 = const()[name = tensor("hidden_states_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(652979520)))]; + tensor hidden_states_211_beta_0_to_fp16 = const()[name = tensor("hidden_states_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(652982144)))]; + tensor var_21610_to_fp16 = const()[name = tensor("op_21610_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_211_cast_fp16 = layer_norm(axes = hidden_states_211_axes_0, beta = hidden_states_211_beta_0_to_fp16, epsilon = var_21610_to_fp16, gamma = hidden_states_211_gamma_0_to_fp16, x = inputs_147_cast_fp16)[name = tensor("hidden_states_211_cast_fp16")]; + tensor q_99_pad_type_0 = const()[name = tensor("q_99_pad_type_0"), val = tensor("valid")]; + tensor q_99_strides_0 = const()[name = tensor("q_99_strides_0"), val = tensor([1, 1])]; + tensor q_99_pad_0 = const()[name = tensor("q_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_99_dilations_0 = const()[name = tensor("q_99_dilations_0"), val = tensor([1, 1])]; + tensor q_99_groups_0 = const()[name = tensor("q_99_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(652984768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(654213632))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_99_cast_fp16 = conv(dilations = q_99_dilations_0, groups = q_99_groups_0, pad = q_99_pad_0, pad_type = q_99_pad_type_0, strides = q_99_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_211_cast_fp16)[name = tensor("q_99_cast_fp16")]; + tensor k_197_pad_type_0 = const()[name = tensor("k_197_pad_type_0"), val = tensor("valid")]; + tensor k_197_strides_0 = const()[name = tensor("k_197_strides_0"), val = tensor([1, 1])]; + tensor k_197_pad_0 = const()[name = tensor("k_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_197_dilations_0 = const()[name = tensor("k_197_dilations_0"), val = tensor([1, 1])]; + tensor k_197_groups_0 = const()[name = tensor("k_197_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(654213824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(656179968))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_197_cast_fp16 = conv(dilations = k_197_dilations_0, groups = k_197_groups_0, pad = k_197_pad_0, pad_type = k_197_pad_type_0, strides = k_197_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_197_cast_fp16")]; + tensor v_99_pad_type_0 = const()[name = tensor("v_99_pad_type_0"), val = tensor("valid")]; + tensor v_99_strides_0 = const()[name = tensor("v_99_strides_0"), val = tensor([1, 1])]; + tensor v_99_pad_0 = const()[name = tensor("v_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_99_dilations_0 = const()[name = tensor("v_99_dilations_0"), val = tensor([1, 1])]; + tensor v_99_groups_0 = const()[name = tensor("v_99_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(656180160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(658146304))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_99_cast_fp16 = conv(dilations = v_99_dilations_0, groups = v_99_groups_0, pad = v_99_pad_0, pad_type = v_99_pad_type_0, strides = v_99_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_99_cast_fp16")]; + tensor var_21643_begin_0 = const()[name = tensor("op_21643_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21643_end_0 = const()[name = tensor("op_21643_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_21643_end_mask_0 = const()[name = tensor("op_21643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21643_cast_fp16 = slice_by_index(begin = var_21643_begin_0, end = var_21643_end_0, end_mask = var_21643_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21643_cast_fp16")]; + tensor var_21647_begin_0 = const()[name = tensor("op_21647_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21647_end_0 = const()[name = tensor("op_21647_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_21647_end_mask_0 = const()[name = tensor("op_21647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21647_cast_fp16 = slice_by_index(begin = var_21647_begin_0, end = var_21647_end_0, end_mask = var_21647_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21647_cast_fp16")]; + tensor var_21651_begin_0 = const()[name = tensor("op_21651_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21651_end_0 = const()[name = tensor("op_21651_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_21651_end_mask_0 = const()[name = tensor("op_21651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21651_cast_fp16 = slice_by_index(begin = var_21651_begin_0, end = var_21651_end_0, end_mask = var_21651_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21651_cast_fp16")]; + tensor var_21655_begin_0 = const()[name = tensor("op_21655_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21655_end_0 = const()[name = tensor("op_21655_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_21655_end_mask_0 = const()[name = tensor("op_21655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21655_cast_fp16 = slice_by_index(begin = var_21655_begin_0, end = var_21655_end_0, end_mask = var_21655_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21655_cast_fp16")]; + tensor var_21659_begin_0 = const()[name = tensor("op_21659_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21659_end_0 = const()[name = tensor("op_21659_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_21659_end_mask_0 = const()[name = tensor("op_21659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21659_cast_fp16 = slice_by_index(begin = var_21659_begin_0, end = var_21659_end_0, end_mask = var_21659_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21659_cast_fp16")]; + tensor var_21663_begin_0 = const()[name = tensor("op_21663_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21663_end_0 = const()[name = tensor("op_21663_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_21663_end_mask_0 = const()[name = tensor("op_21663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21663_cast_fp16 = slice_by_index(begin = var_21663_begin_0, end = var_21663_end_0, end_mask = var_21663_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21663_cast_fp16")]; + tensor var_21667_begin_0 = const()[name = tensor("op_21667_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21667_end_0 = const()[name = tensor("op_21667_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_21667_end_mask_0 = const()[name = tensor("op_21667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21667_cast_fp16 = slice_by_index(begin = var_21667_begin_0, end = var_21667_end_0, end_mask = var_21667_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21667_cast_fp16")]; + tensor var_21671_begin_0 = const()[name = tensor("op_21671_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21671_end_0 = const()[name = tensor("op_21671_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_21671_end_mask_0 = const()[name = tensor("op_21671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21671_cast_fp16 = slice_by_index(begin = var_21671_begin_0, end = var_21671_end_0, end_mask = var_21671_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21671_cast_fp16")]; + tensor var_21675_begin_0 = const()[name = tensor("op_21675_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21675_end_0 = const()[name = tensor("op_21675_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_21675_end_mask_0 = const()[name = tensor("op_21675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21675_cast_fp16 = slice_by_index(begin = var_21675_begin_0, end = var_21675_end_0, end_mask = var_21675_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21675_cast_fp16")]; + tensor var_21679_begin_0 = const()[name = tensor("op_21679_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21679_end_0 = const()[name = tensor("op_21679_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_21679_end_mask_0 = const()[name = tensor("op_21679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21679_cast_fp16 = slice_by_index(begin = var_21679_begin_0, end = var_21679_end_0, end_mask = var_21679_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21679_cast_fp16")]; + tensor var_21683_begin_0 = const()[name = tensor("op_21683_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21683_end_0 = const()[name = tensor("op_21683_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_21683_end_mask_0 = const()[name = tensor("op_21683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21683_cast_fp16 = slice_by_index(begin = var_21683_begin_0, end = var_21683_end_0, end_mask = var_21683_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21683_cast_fp16")]; + tensor var_21687_begin_0 = const()[name = tensor("op_21687_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21687_end_0 = const()[name = tensor("op_21687_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_21687_end_mask_0 = const()[name = tensor("op_21687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21687_cast_fp16 = slice_by_index(begin = var_21687_begin_0, end = var_21687_end_0, end_mask = var_21687_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21687_cast_fp16")]; + tensor var_21691_begin_0 = const()[name = tensor("op_21691_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21691_end_0 = const()[name = tensor("op_21691_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_21691_end_mask_0 = const()[name = tensor("op_21691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21691_cast_fp16 = slice_by_index(begin = var_21691_begin_0, end = var_21691_end_0, end_mask = var_21691_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21691_cast_fp16")]; + tensor var_21695_begin_0 = const()[name = tensor("op_21695_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21695_end_0 = const()[name = tensor("op_21695_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_21695_end_mask_0 = const()[name = tensor("op_21695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21695_cast_fp16 = slice_by_index(begin = var_21695_begin_0, end = var_21695_end_0, end_mask = var_21695_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21695_cast_fp16")]; + tensor var_21699_begin_0 = const()[name = tensor("op_21699_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21699_end_0 = const()[name = tensor("op_21699_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_21699_end_mask_0 = const()[name = tensor("op_21699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21699_cast_fp16 = slice_by_index(begin = var_21699_begin_0, end = var_21699_end_0, end_mask = var_21699_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21699_cast_fp16")]; + tensor var_21703_begin_0 = const()[name = tensor("op_21703_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21703_end_0 = const()[name = tensor("op_21703_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21703_end_mask_0 = const()[name = tensor("op_21703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21703_cast_fp16 = slice_by_index(begin = var_21703_begin_0, end = var_21703_end_0, end_mask = var_21703_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21703_cast_fp16")]; + tensor var_21707_begin_0 = const()[name = tensor("op_21707_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21707_end_0 = const()[name = tensor("op_21707_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_21707_end_mask_0 = const()[name = tensor("op_21707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21707_cast_fp16 = slice_by_index(begin = var_21707_begin_0, end = var_21707_end_0, end_mask = var_21707_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21707_cast_fp16")]; + tensor var_21711_begin_0 = const()[name = tensor("op_21711_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21711_end_0 = const()[name = tensor("op_21711_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_21711_end_mask_0 = const()[name = tensor("op_21711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21711_cast_fp16 = slice_by_index(begin = var_21711_begin_0, end = var_21711_end_0, end_mask = var_21711_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21711_cast_fp16")]; + tensor var_21715_begin_0 = const()[name = tensor("op_21715_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21715_end_0 = const()[name = tensor("op_21715_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_21715_end_mask_0 = const()[name = tensor("op_21715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21715_cast_fp16 = slice_by_index(begin = var_21715_begin_0, end = var_21715_end_0, end_mask = var_21715_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21715_cast_fp16")]; + tensor var_21719_begin_0 = const()[name = tensor("op_21719_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21719_end_0 = const()[name = tensor("op_21719_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_21719_end_mask_0 = const()[name = tensor("op_21719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21719_cast_fp16 = slice_by_index(begin = var_21719_begin_0, end = var_21719_end_0, end_mask = var_21719_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21719_cast_fp16")]; + tensor k_199_perm_0 = const()[name = tensor("k_199_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_21726_begin_0 = const()[name = tensor("op_21726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21726_end_0 = const()[name = tensor("op_21726_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_21726_end_mask_0 = const()[name = tensor("op_21726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_199_cast_fp16 = transpose(perm = k_199_perm_0, x = k_197_cast_fp16)[name = tensor("transpose_18")]; + tensor var_21726_cast_fp16 = slice_by_index(begin = var_21726_begin_0, end = var_21726_end_0, end_mask = var_21726_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21726_cast_fp16")]; + tensor var_21730_begin_0 = const()[name = tensor("op_21730_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_21730_end_0 = const()[name = tensor("op_21730_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_21730_end_mask_0 = const()[name = tensor("op_21730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21730_cast_fp16 = slice_by_index(begin = var_21730_begin_0, end = var_21730_end_0, end_mask = var_21730_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21730_cast_fp16")]; + tensor var_21734_begin_0 = const()[name = tensor("op_21734_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_21734_end_0 = const()[name = tensor("op_21734_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_21734_end_mask_0 = const()[name = tensor("op_21734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21734_cast_fp16 = slice_by_index(begin = var_21734_begin_0, end = var_21734_end_0, end_mask = var_21734_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21734_cast_fp16")]; + tensor var_21738_begin_0 = const()[name = tensor("op_21738_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_21738_end_0 = const()[name = tensor("op_21738_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_21738_end_mask_0 = const()[name = tensor("op_21738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21738_cast_fp16 = slice_by_index(begin = var_21738_begin_0, end = var_21738_end_0, end_mask = var_21738_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21738_cast_fp16")]; + tensor var_21742_begin_0 = const()[name = tensor("op_21742_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_21742_end_0 = const()[name = tensor("op_21742_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_21742_end_mask_0 = const()[name = tensor("op_21742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21742_cast_fp16 = slice_by_index(begin = var_21742_begin_0, end = var_21742_end_0, end_mask = var_21742_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21742_cast_fp16")]; + tensor var_21746_begin_0 = const()[name = tensor("op_21746_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_21746_end_0 = const()[name = tensor("op_21746_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_21746_end_mask_0 = const()[name = tensor("op_21746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21746_cast_fp16 = slice_by_index(begin = var_21746_begin_0, end = var_21746_end_0, end_mask = var_21746_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21746_cast_fp16")]; + tensor var_21750_begin_0 = const()[name = tensor("op_21750_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_21750_end_0 = const()[name = tensor("op_21750_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_21750_end_mask_0 = const()[name = tensor("op_21750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21750_cast_fp16 = slice_by_index(begin = var_21750_begin_0, end = var_21750_end_0, end_mask = var_21750_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21750_cast_fp16")]; + tensor var_21754_begin_0 = const()[name = tensor("op_21754_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_21754_end_0 = const()[name = tensor("op_21754_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_21754_end_mask_0 = const()[name = tensor("op_21754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21754_cast_fp16 = slice_by_index(begin = var_21754_begin_0, end = var_21754_end_0, end_mask = var_21754_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21754_cast_fp16")]; + tensor var_21758_begin_0 = const()[name = tensor("op_21758_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_21758_end_0 = const()[name = tensor("op_21758_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_21758_end_mask_0 = const()[name = tensor("op_21758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21758_cast_fp16 = slice_by_index(begin = var_21758_begin_0, end = var_21758_end_0, end_mask = var_21758_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21758_cast_fp16")]; + tensor var_21762_begin_0 = const()[name = tensor("op_21762_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_21762_end_0 = const()[name = tensor("op_21762_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_21762_end_mask_0 = const()[name = tensor("op_21762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21762_cast_fp16 = slice_by_index(begin = var_21762_begin_0, end = var_21762_end_0, end_mask = var_21762_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21762_cast_fp16")]; + tensor var_21766_begin_0 = const()[name = tensor("op_21766_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_21766_end_0 = const()[name = tensor("op_21766_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_21766_end_mask_0 = const()[name = tensor("op_21766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21766_cast_fp16 = slice_by_index(begin = var_21766_begin_0, end = var_21766_end_0, end_mask = var_21766_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21766_cast_fp16")]; + tensor var_21770_begin_0 = const()[name = tensor("op_21770_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_21770_end_0 = const()[name = tensor("op_21770_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_21770_end_mask_0 = const()[name = tensor("op_21770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21770_cast_fp16 = slice_by_index(begin = var_21770_begin_0, end = var_21770_end_0, end_mask = var_21770_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21770_cast_fp16")]; + tensor var_21774_begin_0 = const()[name = tensor("op_21774_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_21774_end_0 = const()[name = tensor("op_21774_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_21774_end_mask_0 = const()[name = tensor("op_21774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21774_cast_fp16 = slice_by_index(begin = var_21774_begin_0, end = var_21774_end_0, end_mask = var_21774_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21774_cast_fp16")]; + tensor var_21778_begin_0 = const()[name = tensor("op_21778_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_21778_end_0 = const()[name = tensor("op_21778_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_21778_end_mask_0 = const()[name = tensor("op_21778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21778_cast_fp16 = slice_by_index(begin = var_21778_begin_0, end = var_21778_end_0, end_mask = var_21778_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21778_cast_fp16")]; + tensor var_21782_begin_0 = const()[name = tensor("op_21782_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_21782_end_0 = const()[name = tensor("op_21782_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_21782_end_mask_0 = const()[name = tensor("op_21782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21782_cast_fp16 = slice_by_index(begin = var_21782_begin_0, end = var_21782_end_0, end_mask = var_21782_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21782_cast_fp16")]; + tensor var_21786_begin_0 = const()[name = tensor("op_21786_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_21786_end_0 = const()[name = tensor("op_21786_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_21786_end_mask_0 = const()[name = tensor("op_21786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21786_cast_fp16 = slice_by_index(begin = var_21786_begin_0, end = var_21786_end_0, end_mask = var_21786_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21786_cast_fp16")]; + tensor var_21790_begin_0 = const()[name = tensor("op_21790_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_21790_end_0 = const()[name = tensor("op_21790_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_21790_end_mask_0 = const()[name = tensor("op_21790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21790_cast_fp16 = slice_by_index(begin = var_21790_begin_0, end = var_21790_end_0, end_mask = var_21790_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21790_cast_fp16")]; + tensor var_21794_begin_0 = const()[name = tensor("op_21794_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_21794_end_0 = const()[name = tensor("op_21794_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_21794_end_mask_0 = const()[name = tensor("op_21794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21794_cast_fp16 = slice_by_index(begin = var_21794_begin_0, end = var_21794_end_0, end_mask = var_21794_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21794_cast_fp16")]; + tensor var_21798_begin_0 = const()[name = tensor("op_21798_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_21798_end_0 = const()[name = tensor("op_21798_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_21798_end_mask_0 = const()[name = tensor("op_21798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21798_cast_fp16 = slice_by_index(begin = var_21798_begin_0, end = var_21798_end_0, end_mask = var_21798_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21798_cast_fp16")]; + tensor var_21802_begin_0 = const()[name = tensor("op_21802_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_21802_end_0 = const()[name = tensor("op_21802_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_21802_end_mask_0 = const()[name = tensor("op_21802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21802_cast_fp16 = slice_by_index(begin = var_21802_begin_0, end = var_21802_end_0, end_mask = var_21802_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21802_cast_fp16")]; + tensor var_21804_begin_0 = const()[name = tensor("op_21804_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21804_end_0 = const()[name = tensor("op_21804_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_21804_end_mask_0 = const()[name = tensor("op_21804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21804_cast_fp16 = slice_by_index(begin = var_21804_begin_0, end = var_21804_end_0, end_mask = var_21804_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21804_cast_fp16")]; + tensor var_21808_begin_0 = const()[name = tensor("op_21808_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21808_end_0 = const()[name = tensor("op_21808_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_21808_end_mask_0 = const()[name = tensor("op_21808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21808_cast_fp16 = slice_by_index(begin = var_21808_begin_0, end = var_21808_end_0, end_mask = var_21808_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21808_cast_fp16")]; + tensor var_21812_begin_0 = const()[name = tensor("op_21812_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21812_end_0 = const()[name = tensor("op_21812_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_21812_end_mask_0 = const()[name = tensor("op_21812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21812_cast_fp16 = slice_by_index(begin = var_21812_begin_0, end = var_21812_end_0, end_mask = var_21812_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21812_cast_fp16")]; + tensor var_21816_begin_0 = const()[name = tensor("op_21816_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21816_end_0 = const()[name = tensor("op_21816_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_21816_end_mask_0 = const()[name = tensor("op_21816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21816_cast_fp16 = slice_by_index(begin = var_21816_begin_0, end = var_21816_end_0, end_mask = var_21816_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21816_cast_fp16")]; + tensor var_21820_begin_0 = const()[name = tensor("op_21820_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21820_end_0 = const()[name = tensor("op_21820_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_21820_end_mask_0 = const()[name = tensor("op_21820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21820_cast_fp16 = slice_by_index(begin = var_21820_begin_0, end = var_21820_end_0, end_mask = var_21820_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21820_cast_fp16")]; + tensor var_21824_begin_0 = const()[name = tensor("op_21824_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21824_end_0 = const()[name = tensor("op_21824_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_21824_end_mask_0 = const()[name = tensor("op_21824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21824_cast_fp16 = slice_by_index(begin = var_21824_begin_0, end = var_21824_end_0, end_mask = var_21824_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21824_cast_fp16")]; + tensor var_21828_begin_0 = const()[name = tensor("op_21828_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21828_end_0 = const()[name = tensor("op_21828_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_21828_end_mask_0 = const()[name = tensor("op_21828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21828_cast_fp16 = slice_by_index(begin = var_21828_begin_0, end = var_21828_end_0, end_mask = var_21828_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21828_cast_fp16")]; + tensor var_21832_begin_0 = const()[name = tensor("op_21832_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21832_end_0 = const()[name = tensor("op_21832_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_21832_end_mask_0 = const()[name = tensor("op_21832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21832_cast_fp16 = slice_by_index(begin = var_21832_begin_0, end = var_21832_end_0, end_mask = var_21832_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21832_cast_fp16")]; + tensor var_21836_begin_0 = const()[name = tensor("op_21836_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21836_end_0 = const()[name = tensor("op_21836_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_21836_end_mask_0 = const()[name = tensor("op_21836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21836_cast_fp16 = slice_by_index(begin = var_21836_begin_0, end = var_21836_end_0, end_mask = var_21836_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21836_cast_fp16")]; + tensor var_21840_begin_0 = const()[name = tensor("op_21840_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21840_end_0 = const()[name = tensor("op_21840_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_21840_end_mask_0 = const()[name = tensor("op_21840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21840_cast_fp16 = slice_by_index(begin = var_21840_begin_0, end = var_21840_end_0, end_mask = var_21840_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21840_cast_fp16")]; + tensor var_21844_begin_0 = const()[name = tensor("op_21844_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21844_end_0 = const()[name = tensor("op_21844_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_21844_end_mask_0 = const()[name = tensor("op_21844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21844_cast_fp16 = slice_by_index(begin = var_21844_begin_0, end = var_21844_end_0, end_mask = var_21844_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21844_cast_fp16")]; + tensor var_21848_begin_0 = const()[name = tensor("op_21848_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21848_end_0 = const()[name = tensor("op_21848_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_21848_end_mask_0 = const()[name = tensor("op_21848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21848_cast_fp16 = slice_by_index(begin = var_21848_begin_0, end = var_21848_end_0, end_mask = var_21848_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21848_cast_fp16")]; + tensor var_21852_begin_0 = const()[name = tensor("op_21852_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21852_end_0 = const()[name = tensor("op_21852_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_21852_end_mask_0 = const()[name = tensor("op_21852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21852_cast_fp16 = slice_by_index(begin = var_21852_begin_0, end = var_21852_end_0, end_mask = var_21852_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21852_cast_fp16")]; + tensor var_21856_begin_0 = const()[name = tensor("op_21856_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21856_end_0 = const()[name = tensor("op_21856_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_21856_end_mask_0 = const()[name = tensor("op_21856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21856_cast_fp16 = slice_by_index(begin = var_21856_begin_0, end = var_21856_end_0, end_mask = var_21856_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21856_cast_fp16")]; + tensor var_21860_begin_0 = const()[name = tensor("op_21860_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21860_end_0 = const()[name = tensor("op_21860_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_21860_end_mask_0 = const()[name = tensor("op_21860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21860_cast_fp16 = slice_by_index(begin = var_21860_begin_0, end = var_21860_end_0, end_mask = var_21860_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21860_cast_fp16")]; + tensor var_21864_begin_0 = const()[name = tensor("op_21864_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21864_end_0 = const()[name = tensor("op_21864_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_21864_end_mask_0 = const()[name = tensor("op_21864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21864_cast_fp16 = slice_by_index(begin = var_21864_begin_0, end = var_21864_end_0, end_mask = var_21864_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21864_cast_fp16")]; + tensor var_21868_begin_0 = const()[name = tensor("op_21868_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21868_end_0 = const()[name = tensor("op_21868_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_21868_end_mask_0 = const()[name = tensor("op_21868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21868_cast_fp16 = slice_by_index(begin = var_21868_begin_0, end = var_21868_end_0, end_mask = var_21868_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21868_cast_fp16")]; + tensor var_21872_begin_0 = const()[name = tensor("op_21872_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21872_end_0 = const()[name = tensor("op_21872_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_21872_end_mask_0 = const()[name = tensor("op_21872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21872_cast_fp16 = slice_by_index(begin = var_21872_begin_0, end = var_21872_end_0, end_mask = var_21872_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21872_cast_fp16")]; + tensor var_21876_begin_0 = const()[name = tensor("op_21876_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21876_end_0 = const()[name = tensor("op_21876_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_21876_end_mask_0 = const()[name = tensor("op_21876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21876_cast_fp16 = slice_by_index(begin = var_21876_begin_0, end = var_21876_end_0, end_mask = var_21876_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21876_cast_fp16")]; + tensor var_21880_begin_0 = const()[name = tensor("op_21880_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21880_end_0 = const()[name = tensor("op_21880_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_21880_end_mask_0 = const()[name = tensor("op_21880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21880_cast_fp16 = slice_by_index(begin = var_21880_begin_0, end = var_21880_end_0, end_mask = var_21880_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21880_cast_fp16")]; + tensor var_21884_equation_0 = const()[name = tensor("op_21884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21884_cast_fp16 = einsum(equation = var_21884_equation_0, values = (var_21726_cast_fp16, var_21643_cast_fp16))[name = tensor("op_21884_cast_fp16")]; + tensor var_21885_to_fp16 = const()[name = tensor("op_21885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1801_cast_fp16 = mul(x = var_21884_cast_fp16, y = var_21885_to_fp16)[name = tensor("aw_1801_cast_fp16")]; + tensor var_21888_equation_0 = const()[name = tensor("op_21888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21888_cast_fp16 = einsum(equation = var_21888_equation_0, values = (var_21730_cast_fp16, var_21647_cast_fp16))[name = tensor("op_21888_cast_fp16")]; + tensor var_21889_to_fp16 = const()[name = tensor("op_21889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1803_cast_fp16 = mul(x = var_21888_cast_fp16, y = var_21889_to_fp16)[name = tensor("aw_1803_cast_fp16")]; + tensor var_21892_equation_0 = const()[name = tensor("op_21892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21892_cast_fp16 = einsum(equation = var_21892_equation_0, values = (var_21734_cast_fp16, var_21651_cast_fp16))[name = tensor("op_21892_cast_fp16")]; + tensor var_21893_to_fp16 = const()[name = tensor("op_21893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1805_cast_fp16 = mul(x = var_21892_cast_fp16, y = var_21893_to_fp16)[name = tensor("aw_1805_cast_fp16")]; + tensor var_21896_equation_0 = const()[name = tensor("op_21896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21896_cast_fp16 = einsum(equation = var_21896_equation_0, values = (var_21738_cast_fp16, var_21655_cast_fp16))[name = tensor("op_21896_cast_fp16")]; + tensor var_21897_to_fp16 = const()[name = tensor("op_21897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1807_cast_fp16 = mul(x = var_21896_cast_fp16, y = var_21897_to_fp16)[name = tensor("aw_1807_cast_fp16")]; + tensor var_21900_equation_0 = const()[name = tensor("op_21900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21900_cast_fp16 = einsum(equation = var_21900_equation_0, values = (var_21742_cast_fp16, var_21659_cast_fp16))[name = tensor("op_21900_cast_fp16")]; + tensor var_21901_to_fp16 = const()[name = tensor("op_21901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1809_cast_fp16 = mul(x = var_21900_cast_fp16, y = var_21901_to_fp16)[name = tensor("aw_1809_cast_fp16")]; + tensor var_21904_equation_0 = const()[name = tensor("op_21904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21904_cast_fp16 = einsum(equation = var_21904_equation_0, values = (var_21746_cast_fp16, var_21663_cast_fp16))[name = tensor("op_21904_cast_fp16")]; + tensor var_21905_to_fp16 = const()[name = tensor("op_21905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1811_cast_fp16 = mul(x = var_21904_cast_fp16, y = var_21905_to_fp16)[name = tensor("aw_1811_cast_fp16")]; + tensor var_21908_equation_0 = const()[name = tensor("op_21908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21908_cast_fp16 = einsum(equation = var_21908_equation_0, values = (var_21750_cast_fp16, var_21667_cast_fp16))[name = tensor("op_21908_cast_fp16")]; + tensor var_21909_to_fp16 = const()[name = tensor("op_21909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1813_cast_fp16 = mul(x = var_21908_cast_fp16, y = var_21909_to_fp16)[name = tensor("aw_1813_cast_fp16")]; + tensor var_21912_equation_0 = const()[name = tensor("op_21912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21912_cast_fp16 = einsum(equation = var_21912_equation_0, values = (var_21754_cast_fp16, var_21671_cast_fp16))[name = tensor("op_21912_cast_fp16")]; + tensor var_21913_to_fp16 = const()[name = tensor("op_21913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1815_cast_fp16 = mul(x = var_21912_cast_fp16, y = var_21913_to_fp16)[name = tensor("aw_1815_cast_fp16")]; + tensor var_21916_equation_0 = const()[name = tensor("op_21916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21916_cast_fp16 = einsum(equation = var_21916_equation_0, values = (var_21758_cast_fp16, var_21675_cast_fp16))[name = tensor("op_21916_cast_fp16")]; + tensor var_21917_to_fp16 = const()[name = tensor("op_21917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1817_cast_fp16 = mul(x = var_21916_cast_fp16, y = var_21917_to_fp16)[name = tensor("aw_1817_cast_fp16")]; + tensor var_21920_equation_0 = const()[name = tensor("op_21920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21920_cast_fp16 = einsum(equation = var_21920_equation_0, values = (var_21762_cast_fp16, var_21679_cast_fp16))[name = tensor("op_21920_cast_fp16")]; + tensor var_21921_to_fp16 = const()[name = tensor("op_21921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1819_cast_fp16 = mul(x = var_21920_cast_fp16, y = var_21921_to_fp16)[name = tensor("aw_1819_cast_fp16")]; + tensor var_21924_equation_0 = const()[name = tensor("op_21924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21924_cast_fp16 = einsum(equation = var_21924_equation_0, values = (var_21766_cast_fp16, var_21683_cast_fp16))[name = tensor("op_21924_cast_fp16")]; + tensor var_21925_to_fp16 = const()[name = tensor("op_21925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1821_cast_fp16 = mul(x = var_21924_cast_fp16, y = var_21925_to_fp16)[name = tensor("aw_1821_cast_fp16")]; + tensor var_21928_equation_0 = const()[name = tensor("op_21928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21928_cast_fp16 = einsum(equation = var_21928_equation_0, values = (var_21770_cast_fp16, var_21687_cast_fp16))[name = tensor("op_21928_cast_fp16")]; + tensor var_21929_to_fp16 = const()[name = tensor("op_21929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1823_cast_fp16 = mul(x = var_21928_cast_fp16, y = var_21929_to_fp16)[name = tensor("aw_1823_cast_fp16")]; + tensor var_21932_equation_0 = const()[name = tensor("op_21932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21932_cast_fp16 = einsum(equation = var_21932_equation_0, values = (var_21774_cast_fp16, var_21691_cast_fp16))[name = tensor("op_21932_cast_fp16")]; + tensor var_21933_to_fp16 = const()[name = tensor("op_21933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1825_cast_fp16 = mul(x = var_21932_cast_fp16, y = var_21933_to_fp16)[name = tensor("aw_1825_cast_fp16")]; + tensor var_21936_equation_0 = const()[name = tensor("op_21936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21936_cast_fp16 = einsum(equation = var_21936_equation_0, values = (var_21778_cast_fp16, var_21695_cast_fp16))[name = tensor("op_21936_cast_fp16")]; + tensor var_21937_to_fp16 = const()[name = tensor("op_21937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1827_cast_fp16 = mul(x = var_21936_cast_fp16, y = var_21937_to_fp16)[name = tensor("aw_1827_cast_fp16")]; + tensor var_21940_equation_0 = const()[name = tensor("op_21940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21940_cast_fp16 = einsum(equation = var_21940_equation_0, values = (var_21782_cast_fp16, var_21699_cast_fp16))[name = tensor("op_21940_cast_fp16")]; + tensor var_21941_to_fp16 = const()[name = tensor("op_21941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1829_cast_fp16 = mul(x = var_21940_cast_fp16, y = var_21941_to_fp16)[name = tensor("aw_1829_cast_fp16")]; + tensor var_21944_equation_0 = const()[name = tensor("op_21944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21944_cast_fp16 = einsum(equation = var_21944_equation_0, values = (var_21786_cast_fp16, var_21703_cast_fp16))[name = tensor("op_21944_cast_fp16")]; + tensor var_21945_to_fp16 = const()[name = tensor("op_21945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1831_cast_fp16 = mul(x = var_21944_cast_fp16, y = var_21945_to_fp16)[name = tensor("aw_1831_cast_fp16")]; + tensor var_21948_equation_0 = const()[name = tensor("op_21948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21948_cast_fp16 = einsum(equation = var_21948_equation_0, values = (var_21790_cast_fp16, var_21707_cast_fp16))[name = tensor("op_21948_cast_fp16")]; + tensor var_21949_to_fp16 = const()[name = tensor("op_21949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1833_cast_fp16 = mul(x = var_21948_cast_fp16, y = var_21949_to_fp16)[name = tensor("aw_1833_cast_fp16")]; + tensor var_21952_equation_0 = const()[name = tensor("op_21952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21952_cast_fp16 = einsum(equation = var_21952_equation_0, values = (var_21794_cast_fp16, var_21711_cast_fp16))[name = tensor("op_21952_cast_fp16")]; + tensor var_21953_to_fp16 = const()[name = tensor("op_21953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1835_cast_fp16 = mul(x = var_21952_cast_fp16, y = var_21953_to_fp16)[name = tensor("aw_1835_cast_fp16")]; + tensor var_21956_equation_0 = const()[name = tensor("op_21956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21956_cast_fp16 = einsum(equation = var_21956_equation_0, values = (var_21798_cast_fp16, var_21715_cast_fp16))[name = tensor("op_21956_cast_fp16")]; + tensor var_21957_to_fp16 = const()[name = tensor("op_21957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1837_cast_fp16 = mul(x = var_21956_cast_fp16, y = var_21957_to_fp16)[name = tensor("aw_1837_cast_fp16")]; + tensor var_21960_equation_0 = const()[name = tensor("op_21960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21960_cast_fp16 = einsum(equation = var_21960_equation_0, values = (var_21802_cast_fp16, var_21719_cast_fp16))[name = tensor("op_21960_cast_fp16")]; + tensor var_21961_to_fp16 = const()[name = tensor("op_21961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1839_cast_fp16 = mul(x = var_21960_cast_fp16, y = var_21961_to_fp16)[name = tensor("aw_1839_cast_fp16")]; + tensor var_21963_cast_fp16 = softmax(axis = var_21077, x = aw_1801_cast_fp16)[name = tensor("op_21963_cast_fp16")]; + tensor var_21964_cast_fp16 = softmax(axis = var_21077, x = aw_1803_cast_fp16)[name = tensor("op_21964_cast_fp16")]; + tensor var_21965_cast_fp16 = softmax(axis = var_21077, x = aw_1805_cast_fp16)[name = tensor("op_21965_cast_fp16")]; + tensor var_21966_cast_fp16 = softmax(axis = var_21077, x = aw_1807_cast_fp16)[name = tensor("op_21966_cast_fp16")]; + tensor var_21967_cast_fp16 = softmax(axis = var_21077, x = aw_1809_cast_fp16)[name = tensor("op_21967_cast_fp16")]; + tensor var_21968_cast_fp16 = softmax(axis = var_21077, x = aw_1811_cast_fp16)[name = tensor("op_21968_cast_fp16")]; + tensor var_21969_cast_fp16 = softmax(axis = var_21077, x = aw_1813_cast_fp16)[name = tensor("op_21969_cast_fp16")]; + tensor var_21970_cast_fp16 = softmax(axis = var_21077, x = aw_1815_cast_fp16)[name = tensor("op_21970_cast_fp16")]; + tensor var_21971_cast_fp16 = softmax(axis = var_21077, x = aw_1817_cast_fp16)[name = tensor("op_21971_cast_fp16")]; + tensor var_21972_cast_fp16 = softmax(axis = var_21077, x = aw_1819_cast_fp16)[name = tensor("op_21972_cast_fp16")]; + tensor var_21973_cast_fp16 = softmax(axis = var_21077, x = aw_1821_cast_fp16)[name = tensor("op_21973_cast_fp16")]; + tensor var_21974_cast_fp16 = softmax(axis = var_21077, x = aw_1823_cast_fp16)[name = tensor("op_21974_cast_fp16")]; + tensor var_21975_cast_fp16 = softmax(axis = var_21077, x = aw_1825_cast_fp16)[name = tensor("op_21975_cast_fp16")]; + tensor var_21976_cast_fp16 = softmax(axis = var_21077, x = aw_1827_cast_fp16)[name = tensor("op_21976_cast_fp16")]; + tensor var_21977_cast_fp16 = softmax(axis = var_21077, x = aw_1829_cast_fp16)[name = tensor("op_21977_cast_fp16")]; + tensor var_21978_cast_fp16 = softmax(axis = var_21077, x = aw_1831_cast_fp16)[name = tensor("op_21978_cast_fp16")]; + tensor var_21979_cast_fp16 = softmax(axis = var_21077, x = aw_1833_cast_fp16)[name = tensor("op_21979_cast_fp16")]; + tensor var_21980_cast_fp16 = softmax(axis = var_21077, x = aw_1835_cast_fp16)[name = tensor("op_21980_cast_fp16")]; + tensor var_21981_cast_fp16 = softmax(axis = var_21077, x = aw_1837_cast_fp16)[name = tensor("op_21981_cast_fp16")]; + tensor var_21982_cast_fp16 = softmax(axis = var_21077, x = aw_1839_cast_fp16)[name = tensor("op_21982_cast_fp16")]; + tensor var_21984_equation_0 = const()[name = tensor("op_21984_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21984_cast_fp16 = einsum(equation = var_21984_equation_0, values = (var_21804_cast_fp16, var_21963_cast_fp16))[name = tensor("op_21984_cast_fp16")]; + tensor var_21986_equation_0 = const()[name = tensor("op_21986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21986_cast_fp16 = einsum(equation = var_21986_equation_0, values = (var_21808_cast_fp16, var_21964_cast_fp16))[name = tensor("op_21986_cast_fp16")]; + tensor var_21988_equation_0 = const()[name = tensor("op_21988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21988_cast_fp16 = einsum(equation = var_21988_equation_0, values = (var_21812_cast_fp16, var_21965_cast_fp16))[name = tensor("op_21988_cast_fp16")]; + tensor var_21990_equation_0 = const()[name = tensor("op_21990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21990_cast_fp16 = einsum(equation = var_21990_equation_0, values = (var_21816_cast_fp16, var_21966_cast_fp16))[name = tensor("op_21990_cast_fp16")]; + tensor var_21992_equation_0 = const()[name = tensor("op_21992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21992_cast_fp16 = einsum(equation = var_21992_equation_0, values = (var_21820_cast_fp16, var_21967_cast_fp16))[name = tensor("op_21992_cast_fp16")]; + tensor var_21994_equation_0 = const()[name = tensor("op_21994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21994_cast_fp16 = einsum(equation = var_21994_equation_0, values = (var_21824_cast_fp16, var_21968_cast_fp16))[name = tensor("op_21994_cast_fp16")]; + tensor var_21996_equation_0 = const()[name = tensor("op_21996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21996_cast_fp16 = einsum(equation = var_21996_equation_0, values = (var_21828_cast_fp16, var_21969_cast_fp16))[name = tensor("op_21996_cast_fp16")]; + tensor var_21998_equation_0 = const()[name = tensor("op_21998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21998_cast_fp16 = einsum(equation = var_21998_equation_0, values = (var_21832_cast_fp16, var_21970_cast_fp16))[name = tensor("op_21998_cast_fp16")]; + tensor var_22000_equation_0 = const()[name = tensor("op_22000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22000_cast_fp16 = einsum(equation = var_22000_equation_0, values = (var_21836_cast_fp16, var_21971_cast_fp16))[name = tensor("op_22000_cast_fp16")]; + tensor var_22002_equation_0 = const()[name = tensor("op_22002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22002_cast_fp16 = einsum(equation = var_22002_equation_0, values = (var_21840_cast_fp16, var_21972_cast_fp16))[name = tensor("op_22002_cast_fp16")]; + tensor var_22004_equation_0 = const()[name = tensor("op_22004_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22004_cast_fp16 = einsum(equation = var_22004_equation_0, values = (var_21844_cast_fp16, var_21973_cast_fp16))[name = tensor("op_22004_cast_fp16")]; + tensor var_22006_equation_0 = const()[name = tensor("op_22006_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22006_cast_fp16 = einsum(equation = var_22006_equation_0, values = (var_21848_cast_fp16, var_21974_cast_fp16))[name = tensor("op_22006_cast_fp16")]; + tensor var_22008_equation_0 = const()[name = tensor("op_22008_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22008_cast_fp16 = einsum(equation = var_22008_equation_0, values = (var_21852_cast_fp16, var_21975_cast_fp16))[name = tensor("op_22008_cast_fp16")]; + tensor var_22010_equation_0 = const()[name = tensor("op_22010_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22010_cast_fp16 = einsum(equation = var_22010_equation_0, values = (var_21856_cast_fp16, var_21976_cast_fp16))[name = tensor("op_22010_cast_fp16")]; + tensor var_22012_equation_0 = const()[name = tensor("op_22012_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22012_cast_fp16 = einsum(equation = var_22012_equation_0, values = (var_21860_cast_fp16, var_21977_cast_fp16))[name = tensor("op_22012_cast_fp16")]; + tensor var_22014_equation_0 = const()[name = tensor("op_22014_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22014_cast_fp16 = einsum(equation = var_22014_equation_0, values = (var_21864_cast_fp16, var_21978_cast_fp16))[name = tensor("op_22014_cast_fp16")]; + tensor var_22016_equation_0 = const()[name = tensor("op_22016_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22016_cast_fp16 = einsum(equation = var_22016_equation_0, values = (var_21868_cast_fp16, var_21979_cast_fp16))[name = tensor("op_22016_cast_fp16")]; + tensor var_22018_equation_0 = const()[name = tensor("op_22018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22018_cast_fp16 = einsum(equation = var_22018_equation_0, values = (var_21872_cast_fp16, var_21980_cast_fp16))[name = tensor("op_22018_cast_fp16")]; + tensor var_22020_equation_0 = const()[name = tensor("op_22020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22020_cast_fp16 = einsum(equation = var_22020_equation_0, values = (var_21876_cast_fp16, var_21981_cast_fp16))[name = tensor("op_22020_cast_fp16")]; + tensor var_22022_equation_0 = const()[name = tensor("op_22022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22022_cast_fp16 = einsum(equation = var_22022_equation_0, values = (var_21880_cast_fp16, var_21982_cast_fp16))[name = tensor("op_22022_cast_fp16")]; + tensor input_329_interleave_0 = const()[name = tensor("input_329_interleave_0"), val = tensor(false)]; + tensor input_329_cast_fp16 = concat(axis = var_21077, interleave = input_329_interleave_0, values = (var_21984_cast_fp16, var_21986_cast_fp16, var_21988_cast_fp16, var_21990_cast_fp16, var_21992_cast_fp16, var_21994_cast_fp16, var_21996_cast_fp16, var_21998_cast_fp16, var_22000_cast_fp16, var_22002_cast_fp16, var_22004_cast_fp16, var_22006_cast_fp16, var_22008_cast_fp16, var_22010_cast_fp16, var_22012_cast_fp16, var_22014_cast_fp16, var_22016_cast_fp16, var_22018_cast_fp16, var_22020_cast_fp16, var_22022_cast_fp16))[name = tensor("input_329_cast_fp16")]; + tensor var_22032_pad_type_0 = const()[name = tensor("op_22032_pad_type_0"), val = tensor("valid")]; + tensor var_22032_strides_0 = const()[name = tensor("op_22032_strides_0"), val = tensor([1, 1])]; + tensor var_22032_pad_0 = const()[name = tensor("op_22032_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22032_dilations_0 = const()[name = tensor("op_22032_dilations_0"), val = tensor([1, 1])]; + tensor var_22032_groups_0 = const()[name = tensor("op_22032_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(658146496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(659375360))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(659375552)))]; + tensor var_22032_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_22032_dilations_0, groups = var_22032_groups_0, pad = var_22032_pad_0, pad_type = var_22032_pad_type_0, strides = var_22032_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_329_cast_fp16)[name = tensor("op_22032_cast_fp16")]; + tensor inputs_149_cast_fp16 = add(x = var_22032_cast_fp16, y = inputs_147_cast_fp16)[name = tensor("inputs_149_cast_fp16")]; + tensor input_331_axes_0 = const()[name = tensor("input_331_axes_0"), val = tensor([1])]; + tensor input_331_gamma_0_to_fp16 = const()[name = tensor("input_331_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(659378176)))]; + tensor input_331_beta_0_to_fp16 = const()[name = tensor("input_331_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(659380800)))]; + tensor var_22042_to_fp16 = const()[name = tensor("op_22042_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = input_331_beta_0_to_fp16, epsilon = var_22042_to_fp16, gamma = input_331_gamma_0_to_fp16, x = inputs_149_cast_fp16)[name = tensor("input_331_cast_fp16")]; + tensor var_22062_pad_type_0 = const()[name = tensor("op_22062_pad_type_0"), val = tensor("valid")]; + tensor var_22062_strides_0 = const()[name = tensor("op_22062_strides_0"), val = tensor([1, 1])]; + tensor var_22062_pad_0 = const()[name = tensor("op_22062_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22062_dilations_0 = const()[name = tensor("op_22062_dilations_0"), val = tensor([1, 1])]; + tensor var_22062_groups_0 = const()[name = tensor("op_22062_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(659383424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(669213888))), name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(669214080)))]; + tensor var_22062_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_22062_dilations_0, groups = var_22062_groups_0, pad = var_22062_pad_0, pad_type = var_22062_pad_type_0, strides = var_22062_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_331_cast_fp16)[name = tensor("op_22062_cast_fp16")]; + tensor var_22063_split_sizes_0 = const()[name = tensor("op_22063_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_22063_axis_0 = const()[name = tensor("op_22063_axis_0"), val = tensor(1)]; + tensor var_22063_cast_fp16_0, tensor var_22063_cast_fp16_1 = split(axis = var_22063_axis_0, split_sizes = var_22063_split_sizes_0, x = var_22062_cast_fp16)[name = tensor("op_22063_cast_fp16")]; + tensor var_22065_mode_0 = const()[name = tensor("op_22065_mode_0"), val = tensor("EXACT")]; + tensor var_22065_cast_fp16 = gelu(mode = var_22065_mode_0, x = var_22063_cast_fp16_1)[name = tensor("op_22065_cast_fp16")]; + tensor input_333_cast_fp16 = mul(x = var_22063_cast_fp16_0, y = var_22065_cast_fp16)[name = tensor("input_333_cast_fp16")]; + tensor var_22073_pad_type_0 = const()[name = tensor("op_22073_pad_type_0"), val = tensor("valid")]; + tensor var_22073_strides_0 = const()[name = tensor("op_22073_strides_0"), val = tensor([1, 1])]; + tensor var_22073_pad_0 = const()[name = tensor("op_22073_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22073_dilations_0 = const()[name = tensor("op_22073_dilations_0"), val = tensor([1, 1])]; + tensor var_22073_groups_0 = const()[name = tensor("op_22073_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(669234624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674149888))), name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674150080)))]; + tensor var_22073_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_22073_dilations_0, groups = var_22073_groups_0, pad = var_22073_pad_0, pad_type = var_22073_pad_type_0, strides = var_22073_strides_0, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_333_cast_fp16)[name = tensor("op_22073_cast_fp16")]; + tensor inputs_151_cast_fp16 = add(x = var_22073_cast_fp16, y = inputs_149_cast_fp16)[name = tensor("inputs_151_cast_fp16")]; + tensor hidden_states_215_axes_0 = const()[name = tensor("hidden_states_215_axes_0"), val = tensor([1])]; + tensor hidden_states_215_gamma_0_to_fp16 = const()[name = tensor("hidden_states_215_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674152704)))]; + tensor hidden_states_215_beta_0_to_fp16 = const()[name = tensor("hidden_states_215_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674155328)))]; + tensor var_22089_to_fp16 = const()[name = tensor("op_22089_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_215_cast_fp16 = layer_norm(axes = hidden_states_215_axes_0, beta = hidden_states_215_beta_0_to_fp16, epsilon = var_22089_to_fp16, gamma = hidden_states_215_gamma_0_to_fp16, x = inputs_151_cast_fp16)[name = tensor("hidden_states_215_cast_fp16")]; + tensor q_101_pad_type_0 = const()[name = tensor("q_101_pad_type_0"), val = tensor("valid")]; + tensor q_101_strides_0 = const()[name = tensor("q_101_strides_0"), val = tensor([1, 1])]; + tensor q_101_pad_0 = const()[name = tensor("q_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_101_dilations_0 = const()[name = tensor("q_101_dilations_0"), val = tensor([1, 1])]; + tensor q_101_groups_0 = const()[name = tensor("q_101_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(674157952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(675386816))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_101_cast_fp16 = conv(dilations = q_101_dilations_0, groups = q_101_groups_0, pad = q_101_pad_0, pad_type = q_101_pad_type_0, strides = q_101_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_215_cast_fp16)[name = tensor("q_101_cast_fp16")]; + tensor k_201_pad_type_0 = const()[name = tensor("k_201_pad_type_0"), val = tensor("valid")]; + tensor k_201_strides_0 = const()[name = tensor("k_201_strides_0"), val = tensor([1, 1])]; + tensor k_201_pad_0 = const()[name = tensor("k_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_201_dilations_0 = const()[name = tensor("k_201_dilations_0"), val = tensor([1, 1])]; + tensor k_201_groups_0 = const()[name = tensor("k_201_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(675387008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(676615872))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_201_cast_fp16 = conv(dilations = k_201_dilations_0, groups = k_201_groups_0, pad = k_201_pad_0, pad_type = k_201_pad_type_0, strides = k_201_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_215_cast_fp16)[name = tensor("k_201_cast_fp16")]; + tensor v_101_pad_type_0 = const()[name = tensor("v_101_pad_type_0"), val = tensor("valid")]; + tensor v_101_strides_0 = const()[name = tensor("v_101_strides_0"), val = tensor([1, 1])]; + tensor v_101_pad_0 = const()[name = tensor("v_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_101_dilations_0 = const()[name = tensor("v_101_dilations_0"), val = tensor([1, 1])]; + tensor v_101_groups_0 = const()[name = tensor("v_101_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(676616064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(677844928))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_101_cast_fp16 = conv(dilations = v_101_dilations_0, groups = v_101_groups_0, pad = v_101_pad_0, pad_type = v_101_pad_type_0, strides = v_101_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_215_cast_fp16)[name = tensor("v_101_cast_fp16")]; + tensor var_22122_begin_0 = const()[name = tensor("op_22122_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22122_end_0 = const()[name = tensor("op_22122_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_22122_end_mask_0 = const()[name = tensor("op_22122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22122_cast_fp16 = slice_by_index(begin = var_22122_begin_0, end = var_22122_end_0, end_mask = var_22122_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22122_cast_fp16")]; + tensor var_22126_begin_0 = const()[name = tensor("op_22126_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22126_end_0 = const()[name = tensor("op_22126_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_22126_end_mask_0 = const()[name = tensor("op_22126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22126_cast_fp16 = slice_by_index(begin = var_22126_begin_0, end = var_22126_end_0, end_mask = var_22126_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22126_cast_fp16")]; + tensor var_22130_begin_0 = const()[name = tensor("op_22130_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22130_end_0 = const()[name = tensor("op_22130_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_22130_end_mask_0 = const()[name = tensor("op_22130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22130_cast_fp16 = slice_by_index(begin = var_22130_begin_0, end = var_22130_end_0, end_mask = var_22130_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22130_cast_fp16")]; + tensor var_22134_begin_0 = const()[name = tensor("op_22134_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22134_end_0 = const()[name = tensor("op_22134_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_22134_end_mask_0 = const()[name = tensor("op_22134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22134_cast_fp16 = slice_by_index(begin = var_22134_begin_0, end = var_22134_end_0, end_mask = var_22134_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22134_cast_fp16")]; + tensor var_22138_begin_0 = const()[name = tensor("op_22138_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22138_end_0 = const()[name = tensor("op_22138_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_22138_end_mask_0 = const()[name = tensor("op_22138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22138_cast_fp16 = slice_by_index(begin = var_22138_begin_0, end = var_22138_end_0, end_mask = var_22138_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22138_cast_fp16")]; + tensor var_22142_begin_0 = const()[name = tensor("op_22142_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22142_end_0 = const()[name = tensor("op_22142_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_22142_end_mask_0 = const()[name = tensor("op_22142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22142_cast_fp16 = slice_by_index(begin = var_22142_begin_0, end = var_22142_end_0, end_mask = var_22142_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22142_cast_fp16")]; + tensor var_22146_begin_0 = const()[name = tensor("op_22146_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22146_end_0 = const()[name = tensor("op_22146_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_22146_end_mask_0 = const()[name = tensor("op_22146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22146_cast_fp16 = slice_by_index(begin = var_22146_begin_0, end = var_22146_end_0, end_mask = var_22146_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22146_cast_fp16")]; + tensor var_22150_begin_0 = const()[name = tensor("op_22150_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22150_end_0 = const()[name = tensor("op_22150_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_22150_end_mask_0 = const()[name = tensor("op_22150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22150_cast_fp16 = slice_by_index(begin = var_22150_begin_0, end = var_22150_end_0, end_mask = var_22150_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22150_cast_fp16")]; + tensor var_22154_begin_0 = const()[name = tensor("op_22154_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22154_end_0 = const()[name = tensor("op_22154_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_22154_end_mask_0 = const()[name = tensor("op_22154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22154_cast_fp16 = slice_by_index(begin = var_22154_begin_0, end = var_22154_end_0, end_mask = var_22154_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22154_cast_fp16")]; + tensor var_22158_begin_0 = const()[name = tensor("op_22158_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22158_end_0 = const()[name = tensor("op_22158_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_22158_end_mask_0 = const()[name = tensor("op_22158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22158_cast_fp16 = slice_by_index(begin = var_22158_begin_0, end = var_22158_end_0, end_mask = var_22158_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22158_cast_fp16")]; + tensor var_22162_begin_0 = const()[name = tensor("op_22162_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22162_end_0 = const()[name = tensor("op_22162_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_22162_end_mask_0 = const()[name = tensor("op_22162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22162_cast_fp16 = slice_by_index(begin = var_22162_begin_0, end = var_22162_end_0, end_mask = var_22162_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22162_cast_fp16")]; + tensor var_22166_begin_0 = const()[name = tensor("op_22166_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22166_end_0 = const()[name = tensor("op_22166_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_22166_end_mask_0 = const()[name = tensor("op_22166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22166_cast_fp16 = slice_by_index(begin = var_22166_begin_0, end = var_22166_end_0, end_mask = var_22166_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22166_cast_fp16")]; + tensor var_22170_begin_0 = const()[name = tensor("op_22170_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22170_end_0 = const()[name = tensor("op_22170_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_22170_end_mask_0 = const()[name = tensor("op_22170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22170_cast_fp16 = slice_by_index(begin = var_22170_begin_0, end = var_22170_end_0, end_mask = var_22170_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22170_cast_fp16")]; + tensor var_22174_begin_0 = const()[name = tensor("op_22174_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22174_end_0 = const()[name = tensor("op_22174_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_22174_end_mask_0 = const()[name = tensor("op_22174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22174_cast_fp16 = slice_by_index(begin = var_22174_begin_0, end = var_22174_end_0, end_mask = var_22174_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22174_cast_fp16")]; + tensor var_22178_begin_0 = const()[name = tensor("op_22178_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22178_end_0 = const()[name = tensor("op_22178_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_22178_end_mask_0 = const()[name = tensor("op_22178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22178_cast_fp16 = slice_by_index(begin = var_22178_begin_0, end = var_22178_end_0, end_mask = var_22178_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22178_cast_fp16")]; + tensor var_22182_begin_0 = const()[name = tensor("op_22182_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22182_end_0 = const()[name = tensor("op_22182_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22182_end_mask_0 = const()[name = tensor("op_22182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22182_cast_fp16 = slice_by_index(begin = var_22182_begin_0, end = var_22182_end_0, end_mask = var_22182_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22182_cast_fp16")]; + tensor var_22186_begin_0 = const()[name = tensor("op_22186_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22186_end_0 = const()[name = tensor("op_22186_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_22186_end_mask_0 = const()[name = tensor("op_22186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22186_cast_fp16 = slice_by_index(begin = var_22186_begin_0, end = var_22186_end_0, end_mask = var_22186_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22186_cast_fp16")]; + tensor var_22190_begin_0 = const()[name = tensor("op_22190_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22190_end_0 = const()[name = tensor("op_22190_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_22190_end_mask_0 = const()[name = tensor("op_22190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22190_cast_fp16 = slice_by_index(begin = var_22190_begin_0, end = var_22190_end_0, end_mask = var_22190_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22190_cast_fp16")]; + tensor var_22194_begin_0 = const()[name = tensor("op_22194_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22194_end_0 = const()[name = tensor("op_22194_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_22194_end_mask_0 = const()[name = tensor("op_22194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22194_cast_fp16 = slice_by_index(begin = var_22194_begin_0, end = var_22194_end_0, end_mask = var_22194_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22194_cast_fp16")]; + tensor var_22198_begin_0 = const()[name = tensor("op_22198_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22198_end_0 = const()[name = tensor("op_22198_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_22198_end_mask_0 = const()[name = tensor("op_22198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22198_cast_fp16 = slice_by_index(begin = var_22198_begin_0, end = var_22198_end_0, end_mask = var_22198_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22198_cast_fp16")]; + tensor k_203_perm_0 = const()[name = tensor("k_203_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_22205_begin_0 = const()[name = tensor("op_22205_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22205_end_0 = const()[name = tensor("op_22205_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_22205_end_mask_0 = const()[name = tensor("op_22205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_203_cast_fp16 = transpose(perm = k_203_perm_0, x = k_201_cast_fp16)[name = tensor("transpose_17")]; + tensor var_22205_cast_fp16 = slice_by_index(begin = var_22205_begin_0, end = var_22205_end_0, end_mask = var_22205_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22205_cast_fp16")]; + tensor var_22209_begin_0 = const()[name = tensor("op_22209_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_22209_end_0 = const()[name = tensor("op_22209_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_22209_end_mask_0 = const()[name = tensor("op_22209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22209_cast_fp16 = slice_by_index(begin = var_22209_begin_0, end = var_22209_end_0, end_mask = var_22209_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22209_cast_fp16")]; + tensor var_22213_begin_0 = const()[name = tensor("op_22213_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_22213_end_0 = const()[name = tensor("op_22213_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_22213_end_mask_0 = const()[name = tensor("op_22213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22213_cast_fp16 = slice_by_index(begin = var_22213_begin_0, end = var_22213_end_0, end_mask = var_22213_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22213_cast_fp16")]; + tensor var_22217_begin_0 = const()[name = tensor("op_22217_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_22217_end_0 = const()[name = tensor("op_22217_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_22217_end_mask_0 = const()[name = tensor("op_22217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22217_cast_fp16 = slice_by_index(begin = var_22217_begin_0, end = var_22217_end_0, end_mask = var_22217_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22217_cast_fp16")]; + tensor var_22221_begin_0 = const()[name = tensor("op_22221_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_22221_end_0 = const()[name = tensor("op_22221_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_22221_end_mask_0 = const()[name = tensor("op_22221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22221_cast_fp16 = slice_by_index(begin = var_22221_begin_0, end = var_22221_end_0, end_mask = var_22221_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22221_cast_fp16")]; + tensor var_22225_begin_0 = const()[name = tensor("op_22225_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_22225_end_0 = const()[name = tensor("op_22225_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_22225_end_mask_0 = const()[name = tensor("op_22225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22225_cast_fp16 = slice_by_index(begin = var_22225_begin_0, end = var_22225_end_0, end_mask = var_22225_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22225_cast_fp16")]; + tensor var_22229_begin_0 = const()[name = tensor("op_22229_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_22229_end_0 = const()[name = tensor("op_22229_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_22229_end_mask_0 = const()[name = tensor("op_22229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22229_cast_fp16 = slice_by_index(begin = var_22229_begin_0, end = var_22229_end_0, end_mask = var_22229_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22229_cast_fp16")]; + tensor var_22233_begin_0 = const()[name = tensor("op_22233_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_22233_end_0 = const()[name = tensor("op_22233_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_22233_end_mask_0 = const()[name = tensor("op_22233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22233_cast_fp16 = slice_by_index(begin = var_22233_begin_0, end = var_22233_end_0, end_mask = var_22233_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22233_cast_fp16")]; + tensor var_22237_begin_0 = const()[name = tensor("op_22237_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_22237_end_0 = const()[name = tensor("op_22237_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_22237_end_mask_0 = const()[name = tensor("op_22237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22237_cast_fp16 = slice_by_index(begin = var_22237_begin_0, end = var_22237_end_0, end_mask = var_22237_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22237_cast_fp16")]; + tensor var_22241_begin_0 = const()[name = tensor("op_22241_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_22241_end_0 = const()[name = tensor("op_22241_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_22241_end_mask_0 = const()[name = tensor("op_22241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22241_cast_fp16 = slice_by_index(begin = var_22241_begin_0, end = var_22241_end_0, end_mask = var_22241_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22241_cast_fp16")]; + tensor var_22245_begin_0 = const()[name = tensor("op_22245_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_22245_end_0 = const()[name = tensor("op_22245_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_22245_end_mask_0 = const()[name = tensor("op_22245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22245_cast_fp16 = slice_by_index(begin = var_22245_begin_0, end = var_22245_end_0, end_mask = var_22245_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22245_cast_fp16")]; + tensor var_22249_begin_0 = const()[name = tensor("op_22249_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_22249_end_0 = const()[name = tensor("op_22249_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_22249_end_mask_0 = const()[name = tensor("op_22249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22249_cast_fp16 = slice_by_index(begin = var_22249_begin_0, end = var_22249_end_0, end_mask = var_22249_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22249_cast_fp16")]; + tensor var_22253_begin_0 = const()[name = tensor("op_22253_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_22253_end_0 = const()[name = tensor("op_22253_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_22253_end_mask_0 = const()[name = tensor("op_22253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22253_cast_fp16 = slice_by_index(begin = var_22253_begin_0, end = var_22253_end_0, end_mask = var_22253_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22253_cast_fp16")]; + tensor var_22257_begin_0 = const()[name = tensor("op_22257_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_22257_end_0 = const()[name = tensor("op_22257_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_22257_end_mask_0 = const()[name = tensor("op_22257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22257_cast_fp16 = slice_by_index(begin = var_22257_begin_0, end = var_22257_end_0, end_mask = var_22257_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22257_cast_fp16")]; + tensor var_22261_begin_0 = const()[name = tensor("op_22261_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_22261_end_0 = const()[name = tensor("op_22261_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_22261_end_mask_0 = const()[name = tensor("op_22261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22261_cast_fp16 = slice_by_index(begin = var_22261_begin_0, end = var_22261_end_0, end_mask = var_22261_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22261_cast_fp16")]; + tensor var_22265_begin_0 = const()[name = tensor("op_22265_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_22265_end_0 = const()[name = tensor("op_22265_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22265_end_mask_0 = const()[name = tensor("op_22265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22265_cast_fp16 = slice_by_index(begin = var_22265_begin_0, end = var_22265_end_0, end_mask = var_22265_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22265_cast_fp16")]; + tensor var_22269_begin_0 = const()[name = tensor("op_22269_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_22269_end_0 = const()[name = tensor("op_22269_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_22269_end_mask_0 = const()[name = tensor("op_22269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22269_cast_fp16 = slice_by_index(begin = var_22269_begin_0, end = var_22269_end_0, end_mask = var_22269_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22269_cast_fp16")]; + tensor var_22273_begin_0 = const()[name = tensor("op_22273_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_22273_end_0 = const()[name = tensor("op_22273_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_22273_end_mask_0 = const()[name = tensor("op_22273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22273_cast_fp16 = slice_by_index(begin = var_22273_begin_0, end = var_22273_end_0, end_mask = var_22273_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22273_cast_fp16")]; + tensor var_22277_begin_0 = const()[name = tensor("op_22277_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_22277_end_0 = const()[name = tensor("op_22277_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_22277_end_mask_0 = const()[name = tensor("op_22277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22277_cast_fp16 = slice_by_index(begin = var_22277_begin_0, end = var_22277_end_0, end_mask = var_22277_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22277_cast_fp16")]; + tensor var_22281_begin_0 = const()[name = tensor("op_22281_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_22281_end_0 = const()[name = tensor("op_22281_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_22281_end_mask_0 = const()[name = tensor("op_22281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22281_cast_fp16 = slice_by_index(begin = var_22281_begin_0, end = var_22281_end_0, end_mask = var_22281_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22281_cast_fp16")]; + tensor var_22283_begin_0 = const()[name = tensor("op_22283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22283_end_0 = const()[name = tensor("op_22283_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_22283_end_mask_0 = const()[name = tensor("op_22283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22283_cast_fp16 = slice_by_index(begin = var_22283_begin_0, end = var_22283_end_0, end_mask = var_22283_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22283_cast_fp16")]; + tensor var_22287_begin_0 = const()[name = tensor("op_22287_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22287_end_0 = const()[name = tensor("op_22287_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_22287_end_mask_0 = const()[name = tensor("op_22287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22287_cast_fp16 = slice_by_index(begin = var_22287_begin_0, end = var_22287_end_0, end_mask = var_22287_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22287_cast_fp16")]; + tensor var_22291_begin_0 = const()[name = tensor("op_22291_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22291_end_0 = const()[name = tensor("op_22291_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_22291_end_mask_0 = const()[name = tensor("op_22291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22291_cast_fp16 = slice_by_index(begin = var_22291_begin_0, end = var_22291_end_0, end_mask = var_22291_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22291_cast_fp16")]; + tensor var_22295_begin_0 = const()[name = tensor("op_22295_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22295_end_0 = const()[name = tensor("op_22295_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_22295_end_mask_0 = const()[name = tensor("op_22295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22295_cast_fp16 = slice_by_index(begin = var_22295_begin_0, end = var_22295_end_0, end_mask = var_22295_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22295_cast_fp16")]; + tensor var_22299_begin_0 = const()[name = tensor("op_22299_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22299_end_0 = const()[name = tensor("op_22299_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_22299_end_mask_0 = const()[name = tensor("op_22299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22299_cast_fp16 = slice_by_index(begin = var_22299_begin_0, end = var_22299_end_0, end_mask = var_22299_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22299_cast_fp16")]; + tensor var_22303_begin_0 = const()[name = tensor("op_22303_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22303_end_0 = const()[name = tensor("op_22303_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_22303_end_mask_0 = const()[name = tensor("op_22303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22303_cast_fp16 = slice_by_index(begin = var_22303_begin_0, end = var_22303_end_0, end_mask = var_22303_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22303_cast_fp16")]; + tensor var_22307_begin_0 = const()[name = tensor("op_22307_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22307_end_0 = const()[name = tensor("op_22307_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_22307_end_mask_0 = const()[name = tensor("op_22307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22307_cast_fp16 = slice_by_index(begin = var_22307_begin_0, end = var_22307_end_0, end_mask = var_22307_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22307_cast_fp16")]; + tensor var_22311_begin_0 = const()[name = tensor("op_22311_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22311_end_0 = const()[name = tensor("op_22311_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_22311_end_mask_0 = const()[name = tensor("op_22311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22311_cast_fp16 = slice_by_index(begin = var_22311_begin_0, end = var_22311_end_0, end_mask = var_22311_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22311_cast_fp16")]; + tensor var_22315_begin_0 = const()[name = tensor("op_22315_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22315_end_0 = const()[name = tensor("op_22315_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_22315_end_mask_0 = const()[name = tensor("op_22315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22315_cast_fp16 = slice_by_index(begin = var_22315_begin_0, end = var_22315_end_0, end_mask = var_22315_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22315_cast_fp16")]; + tensor var_22319_begin_0 = const()[name = tensor("op_22319_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22319_end_0 = const()[name = tensor("op_22319_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_22319_end_mask_0 = const()[name = tensor("op_22319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22319_cast_fp16 = slice_by_index(begin = var_22319_begin_0, end = var_22319_end_0, end_mask = var_22319_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22319_cast_fp16")]; + tensor var_22323_begin_0 = const()[name = tensor("op_22323_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22323_end_0 = const()[name = tensor("op_22323_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_22323_end_mask_0 = const()[name = tensor("op_22323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22323_cast_fp16 = slice_by_index(begin = var_22323_begin_0, end = var_22323_end_0, end_mask = var_22323_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22323_cast_fp16")]; + tensor var_22327_begin_0 = const()[name = tensor("op_22327_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22327_end_0 = const()[name = tensor("op_22327_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_22327_end_mask_0 = const()[name = tensor("op_22327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22327_cast_fp16 = slice_by_index(begin = var_22327_begin_0, end = var_22327_end_0, end_mask = var_22327_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22327_cast_fp16")]; + tensor var_22331_begin_0 = const()[name = tensor("op_22331_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22331_end_0 = const()[name = tensor("op_22331_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_22331_end_mask_0 = const()[name = tensor("op_22331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22331_cast_fp16 = slice_by_index(begin = var_22331_begin_0, end = var_22331_end_0, end_mask = var_22331_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22331_cast_fp16")]; + tensor var_22335_begin_0 = const()[name = tensor("op_22335_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22335_end_0 = const()[name = tensor("op_22335_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_22335_end_mask_0 = const()[name = tensor("op_22335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22335_cast_fp16 = slice_by_index(begin = var_22335_begin_0, end = var_22335_end_0, end_mask = var_22335_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22335_cast_fp16")]; + tensor var_22339_begin_0 = const()[name = tensor("op_22339_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22339_end_0 = const()[name = tensor("op_22339_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_22339_end_mask_0 = const()[name = tensor("op_22339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22339_cast_fp16 = slice_by_index(begin = var_22339_begin_0, end = var_22339_end_0, end_mask = var_22339_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22339_cast_fp16")]; + tensor var_22343_begin_0 = const()[name = tensor("op_22343_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22343_end_0 = const()[name = tensor("op_22343_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22343_end_mask_0 = const()[name = tensor("op_22343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22343_cast_fp16 = slice_by_index(begin = var_22343_begin_0, end = var_22343_end_0, end_mask = var_22343_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22343_cast_fp16")]; + tensor var_22347_begin_0 = const()[name = tensor("op_22347_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22347_end_0 = const()[name = tensor("op_22347_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_22347_end_mask_0 = const()[name = tensor("op_22347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22347_cast_fp16 = slice_by_index(begin = var_22347_begin_0, end = var_22347_end_0, end_mask = var_22347_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22347_cast_fp16")]; + tensor var_22351_begin_0 = const()[name = tensor("op_22351_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22351_end_0 = const()[name = tensor("op_22351_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_22351_end_mask_0 = const()[name = tensor("op_22351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22351_cast_fp16 = slice_by_index(begin = var_22351_begin_0, end = var_22351_end_0, end_mask = var_22351_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22351_cast_fp16")]; + tensor var_22355_begin_0 = const()[name = tensor("op_22355_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22355_end_0 = const()[name = tensor("op_22355_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_22355_end_mask_0 = const()[name = tensor("op_22355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22355_cast_fp16 = slice_by_index(begin = var_22355_begin_0, end = var_22355_end_0, end_mask = var_22355_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22355_cast_fp16")]; + tensor var_22359_begin_0 = const()[name = tensor("op_22359_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22359_end_0 = const()[name = tensor("op_22359_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_22359_end_mask_0 = const()[name = tensor("op_22359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22359_cast_fp16 = slice_by_index(begin = var_22359_begin_0, end = var_22359_end_0, end_mask = var_22359_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22359_cast_fp16")]; + tensor var_22363_equation_0 = const()[name = tensor("op_22363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22363_cast_fp16 = einsum(equation = var_22363_equation_0, values = (var_22205_cast_fp16, var_22122_cast_fp16))[name = tensor("op_22363_cast_fp16")]; + tensor var_22364_to_fp16 = const()[name = tensor("op_22364_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1841_cast_fp16 = mul(x = var_22363_cast_fp16, y = var_22364_to_fp16)[name = tensor("aw_1841_cast_fp16")]; + tensor var_22367_equation_0 = const()[name = tensor("op_22367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22367_cast_fp16 = einsum(equation = var_22367_equation_0, values = (var_22209_cast_fp16, var_22126_cast_fp16))[name = tensor("op_22367_cast_fp16")]; + tensor var_22368_to_fp16 = const()[name = tensor("op_22368_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1843_cast_fp16 = mul(x = var_22367_cast_fp16, y = var_22368_to_fp16)[name = tensor("aw_1843_cast_fp16")]; + tensor var_22371_equation_0 = const()[name = tensor("op_22371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22371_cast_fp16 = einsum(equation = var_22371_equation_0, values = (var_22213_cast_fp16, var_22130_cast_fp16))[name = tensor("op_22371_cast_fp16")]; + tensor var_22372_to_fp16 = const()[name = tensor("op_22372_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1845_cast_fp16 = mul(x = var_22371_cast_fp16, y = var_22372_to_fp16)[name = tensor("aw_1845_cast_fp16")]; + tensor var_22375_equation_0 = const()[name = tensor("op_22375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22375_cast_fp16 = einsum(equation = var_22375_equation_0, values = (var_22217_cast_fp16, var_22134_cast_fp16))[name = tensor("op_22375_cast_fp16")]; + tensor var_22376_to_fp16 = const()[name = tensor("op_22376_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1847_cast_fp16 = mul(x = var_22375_cast_fp16, y = var_22376_to_fp16)[name = tensor("aw_1847_cast_fp16")]; + tensor var_22379_equation_0 = const()[name = tensor("op_22379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22379_cast_fp16 = einsum(equation = var_22379_equation_0, values = (var_22221_cast_fp16, var_22138_cast_fp16))[name = tensor("op_22379_cast_fp16")]; + tensor var_22380_to_fp16 = const()[name = tensor("op_22380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1849_cast_fp16 = mul(x = var_22379_cast_fp16, y = var_22380_to_fp16)[name = tensor("aw_1849_cast_fp16")]; + tensor var_22383_equation_0 = const()[name = tensor("op_22383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22383_cast_fp16 = einsum(equation = var_22383_equation_0, values = (var_22225_cast_fp16, var_22142_cast_fp16))[name = tensor("op_22383_cast_fp16")]; + tensor var_22384_to_fp16 = const()[name = tensor("op_22384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1851_cast_fp16 = mul(x = var_22383_cast_fp16, y = var_22384_to_fp16)[name = tensor("aw_1851_cast_fp16")]; + tensor var_22387_equation_0 = const()[name = tensor("op_22387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22387_cast_fp16 = einsum(equation = var_22387_equation_0, values = (var_22229_cast_fp16, var_22146_cast_fp16))[name = tensor("op_22387_cast_fp16")]; + tensor var_22388_to_fp16 = const()[name = tensor("op_22388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1853_cast_fp16 = mul(x = var_22387_cast_fp16, y = var_22388_to_fp16)[name = tensor("aw_1853_cast_fp16")]; + tensor var_22391_equation_0 = const()[name = tensor("op_22391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22391_cast_fp16 = einsum(equation = var_22391_equation_0, values = (var_22233_cast_fp16, var_22150_cast_fp16))[name = tensor("op_22391_cast_fp16")]; + tensor var_22392_to_fp16 = const()[name = tensor("op_22392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1855_cast_fp16 = mul(x = var_22391_cast_fp16, y = var_22392_to_fp16)[name = tensor("aw_1855_cast_fp16")]; + tensor var_22395_equation_0 = const()[name = tensor("op_22395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22395_cast_fp16 = einsum(equation = var_22395_equation_0, values = (var_22237_cast_fp16, var_22154_cast_fp16))[name = tensor("op_22395_cast_fp16")]; + tensor var_22396_to_fp16 = const()[name = tensor("op_22396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1857_cast_fp16 = mul(x = var_22395_cast_fp16, y = var_22396_to_fp16)[name = tensor("aw_1857_cast_fp16")]; + tensor var_22399_equation_0 = const()[name = tensor("op_22399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22399_cast_fp16 = einsum(equation = var_22399_equation_0, values = (var_22241_cast_fp16, var_22158_cast_fp16))[name = tensor("op_22399_cast_fp16")]; + tensor var_22400_to_fp16 = const()[name = tensor("op_22400_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1859_cast_fp16 = mul(x = var_22399_cast_fp16, y = var_22400_to_fp16)[name = tensor("aw_1859_cast_fp16")]; + tensor var_22403_equation_0 = const()[name = tensor("op_22403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22403_cast_fp16 = einsum(equation = var_22403_equation_0, values = (var_22245_cast_fp16, var_22162_cast_fp16))[name = tensor("op_22403_cast_fp16")]; + tensor var_22404_to_fp16 = const()[name = tensor("op_22404_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1861_cast_fp16 = mul(x = var_22403_cast_fp16, y = var_22404_to_fp16)[name = tensor("aw_1861_cast_fp16")]; + tensor var_22407_equation_0 = const()[name = tensor("op_22407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22407_cast_fp16 = einsum(equation = var_22407_equation_0, values = (var_22249_cast_fp16, var_22166_cast_fp16))[name = tensor("op_22407_cast_fp16")]; + tensor var_22408_to_fp16 = const()[name = tensor("op_22408_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1863_cast_fp16 = mul(x = var_22407_cast_fp16, y = var_22408_to_fp16)[name = tensor("aw_1863_cast_fp16")]; + tensor var_22411_equation_0 = const()[name = tensor("op_22411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22411_cast_fp16 = einsum(equation = var_22411_equation_0, values = (var_22253_cast_fp16, var_22170_cast_fp16))[name = tensor("op_22411_cast_fp16")]; + tensor var_22412_to_fp16 = const()[name = tensor("op_22412_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1865_cast_fp16 = mul(x = var_22411_cast_fp16, y = var_22412_to_fp16)[name = tensor("aw_1865_cast_fp16")]; + tensor var_22415_equation_0 = const()[name = tensor("op_22415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22415_cast_fp16 = einsum(equation = var_22415_equation_0, values = (var_22257_cast_fp16, var_22174_cast_fp16))[name = tensor("op_22415_cast_fp16")]; + tensor var_22416_to_fp16 = const()[name = tensor("op_22416_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1867_cast_fp16 = mul(x = var_22415_cast_fp16, y = var_22416_to_fp16)[name = tensor("aw_1867_cast_fp16")]; + tensor var_22419_equation_0 = const()[name = tensor("op_22419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22419_cast_fp16 = einsum(equation = var_22419_equation_0, values = (var_22261_cast_fp16, var_22178_cast_fp16))[name = tensor("op_22419_cast_fp16")]; + tensor var_22420_to_fp16 = const()[name = tensor("op_22420_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1869_cast_fp16 = mul(x = var_22419_cast_fp16, y = var_22420_to_fp16)[name = tensor("aw_1869_cast_fp16")]; + tensor var_22423_equation_0 = const()[name = tensor("op_22423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22423_cast_fp16 = einsum(equation = var_22423_equation_0, values = (var_22265_cast_fp16, var_22182_cast_fp16))[name = tensor("op_22423_cast_fp16")]; + tensor var_22424_to_fp16 = const()[name = tensor("op_22424_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1871_cast_fp16 = mul(x = var_22423_cast_fp16, y = var_22424_to_fp16)[name = tensor("aw_1871_cast_fp16")]; + tensor var_22427_equation_0 = const()[name = tensor("op_22427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22427_cast_fp16 = einsum(equation = var_22427_equation_0, values = (var_22269_cast_fp16, var_22186_cast_fp16))[name = tensor("op_22427_cast_fp16")]; + tensor var_22428_to_fp16 = const()[name = tensor("op_22428_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1873_cast_fp16 = mul(x = var_22427_cast_fp16, y = var_22428_to_fp16)[name = tensor("aw_1873_cast_fp16")]; + tensor var_22431_equation_0 = const()[name = tensor("op_22431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22431_cast_fp16 = einsum(equation = var_22431_equation_0, values = (var_22273_cast_fp16, var_22190_cast_fp16))[name = tensor("op_22431_cast_fp16")]; + tensor var_22432_to_fp16 = const()[name = tensor("op_22432_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1875_cast_fp16 = mul(x = var_22431_cast_fp16, y = var_22432_to_fp16)[name = tensor("aw_1875_cast_fp16")]; + tensor var_22435_equation_0 = const()[name = tensor("op_22435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22435_cast_fp16 = einsum(equation = var_22435_equation_0, values = (var_22277_cast_fp16, var_22194_cast_fp16))[name = tensor("op_22435_cast_fp16")]; + tensor var_22436_to_fp16 = const()[name = tensor("op_22436_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1877_cast_fp16 = mul(x = var_22435_cast_fp16, y = var_22436_to_fp16)[name = tensor("aw_1877_cast_fp16")]; + tensor var_22439_equation_0 = const()[name = tensor("op_22439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22439_cast_fp16 = einsum(equation = var_22439_equation_0, values = (var_22281_cast_fp16, var_22198_cast_fp16))[name = tensor("op_22439_cast_fp16")]; + tensor var_22440_to_fp16 = const()[name = tensor("op_22440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1879_cast_fp16 = mul(x = var_22439_cast_fp16, y = var_22440_to_fp16)[name = tensor("aw_1879_cast_fp16")]; + tensor var_22442_cast_fp16 = softmax(axis = var_21077, x = aw_1841_cast_fp16)[name = tensor("op_22442_cast_fp16")]; + tensor var_22443_cast_fp16 = softmax(axis = var_21077, x = aw_1843_cast_fp16)[name = tensor("op_22443_cast_fp16")]; + tensor var_22444_cast_fp16 = softmax(axis = var_21077, x = aw_1845_cast_fp16)[name = tensor("op_22444_cast_fp16")]; + tensor var_22445_cast_fp16 = softmax(axis = var_21077, x = aw_1847_cast_fp16)[name = tensor("op_22445_cast_fp16")]; + tensor var_22446_cast_fp16 = softmax(axis = var_21077, x = aw_1849_cast_fp16)[name = tensor("op_22446_cast_fp16")]; + tensor var_22447_cast_fp16 = softmax(axis = var_21077, x = aw_1851_cast_fp16)[name = tensor("op_22447_cast_fp16")]; + tensor var_22448_cast_fp16 = softmax(axis = var_21077, x = aw_1853_cast_fp16)[name = tensor("op_22448_cast_fp16")]; + tensor var_22449_cast_fp16 = softmax(axis = var_21077, x = aw_1855_cast_fp16)[name = tensor("op_22449_cast_fp16")]; + tensor var_22450_cast_fp16 = softmax(axis = var_21077, x = aw_1857_cast_fp16)[name = tensor("op_22450_cast_fp16")]; + tensor var_22451_cast_fp16 = softmax(axis = var_21077, x = aw_1859_cast_fp16)[name = tensor("op_22451_cast_fp16")]; + tensor var_22452_cast_fp16 = softmax(axis = var_21077, x = aw_1861_cast_fp16)[name = tensor("op_22452_cast_fp16")]; + tensor var_22453_cast_fp16 = softmax(axis = var_21077, x = aw_1863_cast_fp16)[name = tensor("op_22453_cast_fp16")]; + tensor var_22454_cast_fp16 = softmax(axis = var_21077, x = aw_1865_cast_fp16)[name = tensor("op_22454_cast_fp16")]; + tensor var_22455_cast_fp16 = softmax(axis = var_21077, x = aw_1867_cast_fp16)[name = tensor("op_22455_cast_fp16")]; + tensor var_22456_cast_fp16 = softmax(axis = var_21077, x = aw_1869_cast_fp16)[name = tensor("op_22456_cast_fp16")]; + tensor var_22457_cast_fp16 = softmax(axis = var_21077, x = aw_1871_cast_fp16)[name = tensor("op_22457_cast_fp16")]; + tensor var_22458_cast_fp16 = softmax(axis = var_21077, x = aw_1873_cast_fp16)[name = tensor("op_22458_cast_fp16")]; + tensor var_22459_cast_fp16 = softmax(axis = var_21077, x = aw_1875_cast_fp16)[name = tensor("op_22459_cast_fp16")]; + tensor var_22460_cast_fp16 = softmax(axis = var_21077, x = aw_1877_cast_fp16)[name = tensor("op_22460_cast_fp16")]; + tensor var_22461_cast_fp16 = softmax(axis = var_21077, x = aw_1879_cast_fp16)[name = tensor("op_22461_cast_fp16")]; + tensor var_22463_equation_0 = const()[name = tensor("op_22463_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22463_cast_fp16 = einsum(equation = var_22463_equation_0, values = (var_22283_cast_fp16, var_22442_cast_fp16))[name = tensor("op_22463_cast_fp16")]; + tensor var_22465_equation_0 = const()[name = tensor("op_22465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22465_cast_fp16 = einsum(equation = var_22465_equation_0, values = (var_22287_cast_fp16, var_22443_cast_fp16))[name = tensor("op_22465_cast_fp16")]; + tensor var_22467_equation_0 = const()[name = tensor("op_22467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22467_cast_fp16 = einsum(equation = var_22467_equation_0, values = (var_22291_cast_fp16, var_22444_cast_fp16))[name = tensor("op_22467_cast_fp16")]; + tensor var_22469_equation_0 = const()[name = tensor("op_22469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22469_cast_fp16 = einsum(equation = var_22469_equation_0, values = (var_22295_cast_fp16, var_22445_cast_fp16))[name = tensor("op_22469_cast_fp16")]; + tensor var_22471_equation_0 = const()[name = tensor("op_22471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22471_cast_fp16 = einsum(equation = var_22471_equation_0, values = (var_22299_cast_fp16, var_22446_cast_fp16))[name = tensor("op_22471_cast_fp16")]; + tensor var_22473_equation_0 = const()[name = tensor("op_22473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22473_cast_fp16 = einsum(equation = var_22473_equation_0, values = (var_22303_cast_fp16, var_22447_cast_fp16))[name = tensor("op_22473_cast_fp16")]; + tensor var_22475_equation_0 = const()[name = tensor("op_22475_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22475_cast_fp16 = einsum(equation = var_22475_equation_0, values = (var_22307_cast_fp16, var_22448_cast_fp16))[name = tensor("op_22475_cast_fp16")]; + tensor var_22477_equation_0 = const()[name = tensor("op_22477_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22477_cast_fp16 = einsum(equation = var_22477_equation_0, values = (var_22311_cast_fp16, var_22449_cast_fp16))[name = tensor("op_22477_cast_fp16")]; + tensor var_22479_equation_0 = const()[name = tensor("op_22479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22479_cast_fp16 = einsum(equation = var_22479_equation_0, values = (var_22315_cast_fp16, var_22450_cast_fp16))[name = tensor("op_22479_cast_fp16")]; + tensor var_22481_equation_0 = const()[name = tensor("op_22481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22481_cast_fp16 = einsum(equation = var_22481_equation_0, values = (var_22319_cast_fp16, var_22451_cast_fp16))[name = tensor("op_22481_cast_fp16")]; + tensor var_22483_equation_0 = const()[name = tensor("op_22483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22483_cast_fp16 = einsum(equation = var_22483_equation_0, values = (var_22323_cast_fp16, var_22452_cast_fp16))[name = tensor("op_22483_cast_fp16")]; + tensor var_22485_equation_0 = const()[name = tensor("op_22485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22485_cast_fp16 = einsum(equation = var_22485_equation_0, values = (var_22327_cast_fp16, var_22453_cast_fp16))[name = tensor("op_22485_cast_fp16")]; + tensor var_22487_equation_0 = const()[name = tensor("op_22487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22487_cast_fp16 = einsum(equation = var_22487_equation_0, values = (var_22331_cast_fp16, var_22454_cast_fp16))[name = tensor("op_22487_cast_fp16")]; + tensor var_22489_equation_0 = const()[name = tensor("op_22489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22489_cast_fp16 = einsum(equation = var_22489_equation_0, values = (var_22335_cast_fp16, var_22455_cast_fp16))[name = tensor("op_22489_cast_fp16")]; + tensor var_22491_equation_0 = const()[name = tensor("op_22491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22491_cast_fp16 = einsum(equation = var_22491_equation_0, values = (var_22339_cast_fp16, var_22456_cast_fp16))[name = tensor("op_22491_cast_fp16")]; + tensor var_22493_equation_0 = const()[name = tensor("op_22493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22493_cast_fp16 = einsum(equation = var_22493_equation_0, values = (var_22343_cast_fp16, var_22457_cast_fp16))[name = tensor("op_22493_cast_fp16")]; + tensor var_22495_equation_0 = const()[name = tensor("op_22495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22495_cast_fp16 = einsum(equation = var_22495_equation_0, values = (var_22347_cast_fp16, var_22458_cast_fp16))[name = tensor("op_22495_cast_fp16")]; + tensor var_22497_equation_0 = const()[name = tensor("op_22497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22497_cast_fp16 = einsum(equation = var_22497_equation_0, values = (var_22351_cast_fp16, var_22459_cast_fp16))[name = tensor("op_22497_cast_fp16")]; + tensor var_22499_equation_0 = const()[name = tensor("op_22499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22499_cast_fp16 = einsum(equation = var_22499_equation_0, values = (var_22355_cast_fp16, var_22460_cast_fp16))[name = tensor("op_22499_cast_fp16")]; + tensor var_22501_equation_0 = const()[name = tensor("op_22501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22501_cast_fp16 = einsum(equation = var_22501_equation_0, values = (var_22359_cast_fp16, var_22461_cast_fp16))[name = tensor("op_22501_cast_fp16")]; + tensor input_335_interleave_0 = const()[name = tensor("input_335_interleave_0"), val = tensor(false)]; + tensor input_335_cast_fp16 = concat(axis = var_21077, interleave = input_335_interleave_0, values = (var_22463_cast_fp16, var_22465_cast_fp16, var_22467_cast_fp16, var_22469_cast_fp16, var_22471_cast_fp16, var_22473_cast_fp16, var_22475_cast_fp16, var_22477_cast_fp16, var_22479_cast_fp16, var_22481_cast_fp16, var_22483_cast_fp16, var_22485_cast_fp16, var_22487_cast_fp16, var_22489_cast_fp16, var_22491_cast_fp16, var_22493_cast_fp16, var_22495_cast_fp16, var_22497_cast_fp16, var_22499_cast_fp16, var_22501_cast_fp16))[name = tensor("input_335_cast_fp16")]; + tensor var_22511_pad_type_0 = const()[name = tensor("op_22511_pad_type_0"), val = tensor("valid")]; + tensor var_22511_strides_0 = const()[name = tensor("op_22511_strides_0"), val = tensor([1, 1])]; + tensor var_22511_pad_0 = const()[name = tensor("op_22511_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22511_dilations_0 = const()[name = tensor("op_22511_dilations_0"), val = tensor([1, 1])]; + tensor var_22511_groups_0 = const()[name = tensor("op_22511_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(677845120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(679073984))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(679074176)))]; + tensor var_22511_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_22511_dilations_0, groups = var_22511_groups_0, pad = var_22511_pad_0, pad_type = var_22511_pad_type_0, strides = var_22511_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_335_cast_fp16)[name = tensor("op_22511_cast_fp16")]; + tensor inputs_153_cast_fp16 = add(x = var_22511_cast_fp16, y = inputs_151_cast_fp16)[name = tensor("inputs_153_cast_fp16")]; + tensor hidden_states_217_axes_0 = const()[name = tensor("hidden_states_217_axes_0"), val = tensor([1])]; + tensor hidden_states_217_gamma_0_to_fp16 = const()[name = tensor("hidden_states_217_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(679076800)))]; + tensor hidden_states_217_beta_0_to_fp16 = const()[name = tensor("hidden_states_217_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(679079424)))]; + tensor var_22521_to_fp16 = const()[name = tensor("op_22521_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_217_cast_fp16 = layer_norm(axes = hidden_states_217_axes_0, beta = hidden_states_217_beta_0_to_fp16, epsilon = var_22521_to_fp16, gamma = hidden_states_217_gamma_0_to_fp16, x = inputs_153_cast_fp16)[name = tensor("hidden_states_217_cast_fp16")]; + tensor q_103_pad_type_0 = const()[name = tensor("q_103_pad_type_0"), val = tensor("valid")]; + tensor q_103_strides_0 = const()[name = tensor("q_103_strides_0"), val = tensor([1, 1])]; + tensor q_103_pad_0 = const()[name = tensor("q_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_103_dilations_0 = const()[name = tensor("q_103_dilations_0"), val = tensor([1, 1])]; + tensor q_103_groups_0 = const()[name = tensor("q_103_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(679082048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(680310912))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_103_cast_fp16 = conv(dilations = q_103_dilations_0, groups = q_103_groups_0, pad = q_103_pad_0, pad_type = q_103_pad_type_0, strides = q_103_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_217_cast_fp16)[name = tensor("q_103_cast_fp16")]; + tensor k_205_pad_type_0 = const()[name = tensor("k_205_pad_type_0"), val = tensor("valid")]; + tensor k_205_strides_0 = const()[name = tensor("k_205_strides_0"), val = tensor([1, 1])]; + tensor k_205_pad_0 = const()[name = tensor("k_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_205_dilations_0 = const()[name = tensor("k_205_dilations_0"), val = tensor([1, 1])]; + tensor k_205_groups_0 = const()[name = tensor("k_205_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(680311104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(682277248))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_205_cast_fp16 = conv(dilations = k_205_dilations_0, groups = k_205_groups_0, pad = k_205_pad_0, pad_type = k_205_pad_type_0, strides = k_205_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_205_cast_fp16")]; + tensor v_103_pad_type_0 = const()[name = tensor("v_103_pad_type_0"), val = tensor("valid")]; + tensor v_103_strides_0 = const()[name = tensor("v_103_strides_0"), val = tensor([1, 1])]; + tensor v_103_pad_0 = const()[name = tensor("v_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_103_dilations_0 = const()[name = tensor("v_103_dilations_0"), val = tensor([1, 1])]; + tensor v_103_groups_0 = const()[name = tensor("v_103_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(682277440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684243584))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_103_cast_fp16 = conv(dilations = v_103_dilations_0, groups = v_103_groups_0, pad = v_103_pad_0, pad_type = v_103_pad_type_0, strides = v_103_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_103_cast_fp16")]; + tensor var_22554_begin_0 = const()[name = tensor("op_22554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22554_end_0 = const()[name = tensor("op_22554_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_22554_end_mask_0 = const()[name = tensor("op_22554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22554_cast_fp16 = slice_by_index(begin = var_22554_begin_0, end = var_22554_end_0, end_mask = var_22554_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22554_cast_fp16")]; + tensor var_22558_begin_0 = const()[name = tensor("op_22558_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22558_end_0 = const()[name = tensor("op_22558_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_22558_end_mask_0 = const()[name = tensor("op_22558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22558_cast_fp16 = slice_by_index(begin = var_22558_begin_0, end = var_22558_end_0, end_mask = var_22558_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22558_cast_fp16")]; + tensor var_22562_begin_0 = const()[name = tensor("op_22562_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22562_end_0 = const()[name = tensor("op_22562_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_22562_end_mask_0 = const()[name = tensor("op_22562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22562_cast_fp16 = slice_by_index(begin = var_22562_begin_0, end = var_22562_end_0, end_mask = var_22562_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22562_cast_fp16")]; + tensor var_22566_begin_0 = const()[name = tensor("op_22566_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22566_end_0 = const()[name = tensor("op_22566_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_22566_end_mask_0 = const()[name = tensor("op_22566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22566_cast_fp16 = slice_by_index(begin = var_22566_begin_0, end = var_22566_end_0, end_mask = var_22566_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22566_cast_fp16")]; + tensor var_22570_begin_0 = const()[name = tensor("op_22570_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22570_end_0 = const()[name = tensor("op_22570_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_22570_end_mask_0 = const()[name = tensor("op_22570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22570_cast_fp16 = slice_by_index(begin = var_22570_begin_0, end = var_22570_end_0, end_mask = var_22570_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22570_cast_fp16")]; + tensor var_22574_begin_0 = const()[name = tensor("op_22574_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22574_end_0 = const()[name = tensor("op_22574_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_22574_end_mask_0 = const()[name = tensor("op_22574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22574_cast_fp16 = slice_by_index(begin = var_22574_begin_0, end = var_22574_end_0, end_mask = var_22574_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22574_cast_fp16")]; + tensor var_22578_begin_0 = const()[name = tensor("op_22578_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22578_end_0 = const()[name = tensor("op_22578_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_22578_end_mask_0 = const()[name = tensor("op_22578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22578_cast_fp16 = slice_by_index(begin = var_22578_begin_0, end = var_22578_end_0, end_mask = var_22578_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22578_cast_fp16")]; + tensor var_22582_begin_0 = const()[name = tensor("op_22582_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22582_end_0 = const()[name = tensor("op_22582_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_22582_end_mask_0 = const()[name = tensor("op_22582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22582_cast_fp16 = slice_by_index(begin = var_22582_begin_0, end = var_22582_end_0, end_mask = var_22582_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22582_cast_fp16")]; + tensor var_22586_begin_0 = const()[name = tensor("op_22586_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22586_end_0 = const()[name = tensor("op_22586_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_22586_end_mask_0 = const()[name = tensor("op_22586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22586_cast_fp16 = slice_by_index(begin = var_22586_begin_0, end = var_22586_end_0, end_mask = var_22586_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22586_cast_fp16")]; + tensor var_22590_begin_0 = const()[name = tensor("op_22590_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22590_end_0 = const()[name = tensor("op_22590_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_22590_end_mask_0 = const()[name = tensor("op_22590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22590_cast_fp16 = slice_by_index(begin = var_22590_begin_0, end = var_22590_end_0, end_mask = var_22590_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22590_cast_fp16")]; + tensor var_22594_begin_0 = const()[name = tensor("op_22594_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22594_end_0 = const()[name = tensor("op_22594_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_22594_end_mask_0 = const()[name = tensor("op_22594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22594_cast_fp16 = slice_by_index(begin = var_22594_begin_0, end = var_22594_end_0, end_mask = var_22594_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22594_cast_fp16")]; + tensor var_22598_begin_0 = const()[name = tensor("op_22598_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22598_end_0 = const()[name = tensor("op_22598_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_22598_end_mask_0 = const()[name = tensor("op_22598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22598_cast_fp16 = slice_by_index(begin = var_22598_begin_0, end = var_22598_end_0, end_mask = var_22598_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22598_cast_fp16")]; + tensor var_22602_begin_0 = const()[name = tensor("op_22602_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22602_end_0 = const()[name = tensor("op_22602_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_22602_end_mask_0 = const()[name = tensor("op_22602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22602_cast_fp16 = slice_by_index(begin = var_22602_begin_0, end = var_22602_end_0, end_mask = var_22602_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22602_cast_fp16")]; + tensor var_22606_begin_0 = const()[name = tensor("op_22606_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22606_end_0 = const()[name = tensor("op_22606_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_22606_end_mask_0 = const()[name = tensor("op_22606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22606_cast_fp16 = slice_by_index(begin = var_22606_begin_0, end = var_22606_end_0, end_mask = var_22606_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22606_cast_fp16")]; + tensor var_22610_begin_0 = const()[name = tensor("op_22610_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22610_end_0 = const()[name = tensor("op_22610_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_22610_end_mask_0 = const()[name = tensor("op_22610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22610_cast_fp16 = slice_by_index(begin = var_22610_begin_0, end = var_22610_end_0, end_mask = var_22610_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22610_cast_fp16")]; + tensor var_22614_begin_0 = const()[name = tensor("op_22614_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22614_end_0 = const()[name = tensor("op_22614_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22614_end_mask_0 = const()[name = tensor("op_22614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22614_cast_fp16 = slice_by_index(begin = var_22614_begin_0, end = var_22614_end_0, end_mask = var_22614_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22614_cast_fp16")]; + tensor var_22618_begin_0 = const()[name = tensor("op_22618_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22618_end_0 = const()[name = tensor("op_22618_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_22618_end_mask_0 = const()[name = tensor("op_22618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22618_cast_fp16 = slice_by_index(begin = var_22618_begin_0, end = var_22618_end_0, end_mask = var_22618_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22618_cast_fp16")]; + tensor var_22622_begin_0 = const()[name = tensor("op_22622_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22622_end_0 = const()[name = tensor("op_22622_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_22622_end_mask_0 = const()[name = tensor("op_22622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22622_cast_fp16 = slice_by_index(begin = var_22622_begin_0, end = var_22622_end_0, end_mask = var_22622_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22622_cast_fp16")]; + tensor var_22626_begin_0 = const()[name = tensor("op_22626_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22626_end_0 = const()[name = tensor("op_22626_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_22626_end_mask_0 = const()[name = tensor("op_22626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22626_cast_fp16 = slice_by_index(begin = var_22626_begin_0, end = var_22626_end_0, end_mask = var_22626_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22626_cast_fp16")]; + tensor var_22630_begin_0 = const()[name = tensor("op_22630_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22630_end_0 = const()[name = tensor("op_22630_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_22630_end_mask_0 = const()[name = tensor("op_22630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22630_cast_fp16 = slice_by_index(begin = var_22630_begin_0, end = var_22630_end_0, end_mask = var_22630_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22630_cast_fp16")]; + tensor k_207_perm_0 = const()[name = tensor("k_207_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_22637_begin_0 = const()[name = tensor("op_22637_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22637_end_0 = const()[name = tensor("op_22637_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_22637_end_mask_0 = const()[name = tensor("op_22637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_207_cast_fp16 = transpose(perm = k_207_perm_0, x = k_205_cast_fp16)[name = tensor("transpose_16")]; + tensor var_22637_cast_fp16 = slice_by_index(begin = var_22637_begin_0, end = var_22637_end_0, end_mask = var_22637_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22637_cast_fp16")]; + tensor var_22641_begin_0 = const()[name = tensor("op_22641_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_22641_end_0 = const()[name = tensor("op_22641_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_22641_end_mask_0 = const()[name = tensor("op_22641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22641_cast_fp16 = slice_by_index(begin = var_22641_begin_0, end = var_22641_end_0, end_mask = var_22641_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22641_cast_fp16")]; + tensor var_22645_begin_0 = const()[name = tensor("op_22645_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_22645_end_0 = const()[name = tensor("op_22645_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_22645_end_mask_0 = const()[name = tensor("op_22645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22645_cast_fp16 = slice_by_index(begin = var_22645_begin_0, end = var_22645_end_0, end_mask = var_22645_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22645_cast_fp16")]; + tensor var_22649_begin_0 = const()[name = tensor("op_22649_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_22649_end_0 = const()[name = tensor("op_22649_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_22649_end_mask_0 = const()[name = tensor("op_22649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22649_cast_fp16 = slice_by_index(begin = var_22649_begin_0, end = var_22649_end_0, end_mask = var_22649_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22649_cast_fp16")]; + tensor var_22653_begin_0 = const()[name = tensor("op_22653_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_22653_end_0 = const()[name = tensor("op_22653_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_22653_end_mask_0 = const()[name = tensor("op_22653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22653_cast_fp16 = slice_by_index(begin = var_22653_begin_0, end = var_22653_end_0, end_mask = var_22653_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22653_cast_fp16")]; + tensor var_22657_begin_0 = const()[name = tensor("op_22657_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_22657_end_0 = const()[name = tensor("op_22657_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_22657_end_mask_0 = const()[name = tensor("op_22657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22657_cast_fp16 = slice_by_index(begin = var_22657_begin_0, end = var_22657_end_0, end_mask = var_22657_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22657_cast_fp16")]; + tensor var_22661_begin_0 = const()[name = tensor("op_22661_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_22661_end_0 = const()[name = tensor("op_22661_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_22661_end_mask_0 = const()[name = tensor("op_22661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22661_cast_fp16 = slice_by_index(begin = var_22661_begin_0, end = var_22661_end_0, end_mask = var_22661_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22661_cast_fp16")]; + tensor var_22665_begin_0 = const()[name = tensor("op_22665_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_22665_end_0 = const()[name = tensor("op_22665_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_22665_end_mask_0 = const()[name = tensor("op_22665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22665_cast_fp16 = slice_by_index(begin = var_22665_begin_0, end = var_22665_end_0, end_mask = var_22665_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22665_cast_fp16")]; + tensor var_22669_begin_0 = const()[name = tensor("op_22669_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_22669_end_0 = const()[name = tensor("op_22669_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_22669_end_mask_0 = const()[name = tensor("op_22669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22669_cast_fp16 = slice_by_index(begin = var_22669_begin_0, end = var_22669_end_0, end_mask = var_22669_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22669_cast_fp16")]; + tensor var_22673_begin_0 = const()[name = tensor("op_22673_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_22673_end_0 = const()[name = tensor("op_22673_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_22673_end_mask_0 = const()[name = tensor("op_22673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22673_cast_fp16 = slice_by_index(begin = var_22673_begin_0, end = var_22673_end_0, end_mask = var_22673_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22673_cast_fp16")]; + tensor var_22677_begin_0 = const()[name = tensor("op_22677_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_22677_end_0 = const()[name = tensor("op_22677_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_22677_end_mask_0 = const()[name = tensor("op_22677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22677_cast_fp16 = slice_by_index(begin = var_22677_begin_0, end = var_22677_end_0, end_mask = var_22677_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22677_cast_fp16")]; + tensor var_22681_begin_0 = const()[name = tensor("op_22681_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_22681_end_0 = const()[name = tensor("op_22681_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_22681_end_mask_0 = const()[name = tensor("op_22681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22681_cast_fp16 = slice_by_index(begin = var_22681_begin_0, end = var_22681_end_0, end_mask = var_22681_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22681_cast_fp16")]; + tensor var_22685_begin_0 = const()[name = tensor("op_22685_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_22685_end_0 = const()[name = tensor("op_22685_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_22685_end_mask_0 = const()[name = tensor("op_22685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22685_cast_fp16 = slice_by_index(begin = var_22685_begin_0, end = var_22685_end_0, end_mask = var_22685_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22685_cast_fp16")]; + tensor var_22689_begin_0 = const()[name = tensor("op_22689_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_22689_end_0 = const()[name = tensor("op_22689_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_22689_end_mask_0 = const()[name = tensor("op_22689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22689_cast_fp16 = slice_by_index(begin = var_22689_begin_0, end = var_22689_end_0, end_mask = var_22689_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22689_cast_fp16")]; + tensor var_22693_begin_0 = const()[name = tensor("op_22693_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_22693_end_0 = const()[name = tensor("op_22693_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_22693_end_mask_0 = const()[name = tensor("op_22693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22693_cast_fp16 = slice_by_index(begin = var_22693_begin_0, end = var_22693_end_0, end_mask = var_22693_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22693_cast_fp16")]; + tensor var_22697_begin_0 = const()[name = tensor("op_22697_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_22697_end_0 = const()[name = tensor("op_22697_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_22697_end_mask_0 = const()[name = tensor("op_22697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22697_cast_fp16 = slice_by_index(begin = var_22697_begin_0, end = var_22697_end_0, end_mask = var_22697_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22697_cast_fp16")]; + tensor var_22701_begin_0 = const()[name = tensor("op_22701_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_22701_end_0 = const()[name = tensor("op_22701_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_22701_end_mask_0 = const()[name = tensor("op_22701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22701_cast_fp16 = slice_by_index(begin = var_22701_begin_0, end = var_22701_end_0, end_mask = var_22701_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22701_cast_fp16")]; + tensor var_22705_begin_0 = const()[name = tensor("op_22705_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_22705_end_0 = const()[name = tensor("op_22705_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_22705_end_mask_0 = const()[name = tensor("op_22705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22705_cast_fp16 = slice_by_index(begin = var_22705_begin_0, end = var_22705_end_0, end_mask = var_22705_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22705_cast_fp16")]; + tensor var_22709_begin_0 = const()[name = tensor("op_22709_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_22709_end_0 = const()[name = tensor("op_22709_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_22709_end_mask_0 = const()[name = tensor("op_22709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22709_cast_fp16 = slice_by_index(begin = var_22709_begin_0, end = var_22709_end_0, end_mask = var_22709_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22709_cast_fp16")]; + tensor var_22713_begin_0 = const()[name = tensor("op_22713_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_22713_end_0 = const()[name = tensor("op_22713_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_22713_end_mask_0 = const()[name = tensor("op_22713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22713_cast_fp16 = slice_by_index(begin = var_22713_begin_0, end = var_22713_end_0, end_mask = var_22713_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22713_cast_fp16")]; + tensor var_22715_begin_0 = const()[name = tensor("op_22715_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22715_end_0 = const()[name = tensor("op_22715_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_22715_end_mask_0 = const()[name = tensor("op_22715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22715_cast_fp16 = slice_by_index(begin = var_22715_begin_0, end = var_22715_end_0, end_mask = var_22715_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22715_cast_fp16")]; + tensor var_22719_begin_0 = const()[name = tensor("op_22719_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22719_end_0 = const()[name = tensor("op_22719_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_22719_end_mask_0 = const()[name = tensor("op_22719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22719_cast_fp16 = slice_by_index(begin = var_22719_begin_0, end = var_22719_end_0, end_mask = var_22719_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22719_cast_fp16")]; + tensor var_22723_begin_0 = const()[name = tensor("op_22723_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22723_end_0 = const()[name = tensor("op_22723_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_22723_end_mask_0 = const()[name = tensor("op_22723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22723_cast_fp16 = slice_by_index(begin = var_22723_begin_0, end = var_22723_end_0, end_mask = var_22723_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22723_cast_fp16")]; + tensor var_22727_begin_0 = const()[name = tensor("op_22727_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22727_end_0 = const()[name = tensor("op_22727_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_22727_end_mask_0 = const()[name = tensor("op_22727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22727_cast_fp16 = slice_by_index(begin = var_22727_begin_0, end = var_22727_end_0, end_mask = var_22727_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22727_cast_fp16")]; + tensor var_22731_begin_0 = const()[name = tensor("op_22731_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22731_end_0 = const()[name = tensor("op_22731_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_22731_end_mask_0 = const()[name = tensor("op_22731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22731_cast_fp16 = slice_by_index(begin = var_22731_begin_0, end = var_22731_end_0, end_mask = var_22731_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22731_cast_fp16")]; + tensor var_22735_begin_0 = const()[name = tensor("op_22735_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22735_end_0 = const()[name = tensor("op_22735_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_22735_end_mask_0 = const()[name = tensor("op_22735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22735_cast_fp16 = slice_by_index(begin = var_22735_begin_0, end = var_22735_end_0, end_mask = var_22735_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22735_cast_fp16")]; + tensor var_22739_begin_0 = const()[name = tensor("op_22739_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22739_end_0 = const()[name = tensor("op_22739_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_22739_end_mask_0 = const()[name = tensor("op_22739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22739_cast_fp16 = slice_by_index(begin = var_22739_begin_0, end = var_22739_end_0, end_mask = var_22739_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22739_cast_fp16")]; + tensor var_22743_begin_0 = const()[name = tensor("op_22743_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22743_end_0 = const()[name = tensor("op_22743_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_22743_end_mask_0 = const()[name = tensor("op_22743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22743_cast_fp16 = slice_by_index(begin = var_22743_begin_0, end = var_22743_end_0, end_mask = var_22743_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22743_cast_fp16")]; + tensor var_22747_begin_0 = const()[name = tensor("op_22747_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22747_end_0 = const()[name = tensor("op_22747_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_22747_end_mask_0 = const()[name = tensor("op_22747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22747_cast_fp16 = slice_by_index(begin = var_22747_begin_0, end = var_22747_end_0, end_mask = var_22747_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22747_cast_fp16")]; + tensor var_22751_begin_0 = const()[name = tensor("op_22751_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22751_end_0 = const()[name = tensor("op_22751_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_22751_end_mask_0 = const()[name = tensor("op_22751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22751_cast_fp16 = slice_by_index(begin = var_22751_begin_0, end = var_22751_end_0, end_mask = var_22751_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22751_cast_fp16")]; + tensor var_22755_begin_0 = const()[name = tensor("op_22755_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22755_end_0 = const()[name = tensor("op_22755_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_22755_end_mask_0 = const()[name = tensor("op_22755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22755_cast_fp16 = slice_by_index(begin = var_22755_begin_0, end = var_22755_end_0, end_mask = var_22755_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22755_cast_fp16")]; + tensor var_22759_begin_0 = const()[name = tensor("op_22759_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22759_end_0 = const()[name = tensor("op_22759_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_22759_end_mask_0 = const()[name = tensor("op_22759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22759_cast_fp16 = slice_by_index(begin = var_22759_begin_0, end = var_22759_end_0, end_mask = var_22759_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22759_cast_fp16")]; + tensor var_22763_begin_0 = const()[name = tensor("op_22763_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22763_end_0 = const()[name = tensor("op_22763_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_22763_end_mask_0 = const()[name = tensor("op_22763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22763_cast_fp16 = slice_by_index(begin = var_22763_begin_0, end = var_22763_end_0, end_mask = var_22763_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22763_cast_fp16")]; + tensor var_22767_begin_0 = const()[name = tensor("op_22767_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22767_end_0 = const()[name = tensor("op_22767_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_22767_end_mask_0 = const()[name = tensor("op_22767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22767_cast_fp16 = slice_by_index(begin = var_22767_begin_0, end = var_22767_end_0, end_mask = var_22767_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22767_cast_fp16")]; + tensor var_22771_begin_0 = const()[name = tensor("op_22771_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22771_end_0 = const()[name = tensor("op_22771_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_22771_end_mask_0 = const()[name = tensor("op_22771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22771_cast_fp16 = slice_by_index(begin = var_22771_begin_0, end = var_22771_end_0, end_mask = var_22771_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22771_cast_fp16")]; + tensor var_22775_begin_0 = const()[name = tensor("op_22775_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22775_end_0 = const()[name = tensor("op_22775_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_22775_end_mask_0 = const()[name = tensor("op_22775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22775_cast_fp16 = slice_by_index(begin = var_22775_begin_0, end = var_22775_end_0, end_mask = var_22775_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22775_cast_fp16")]; + tensor var_22779_begin_0 = const()[name = tensor("op_22779_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22779_end_0 = const()[name = tensor("op_22779_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_22779_end_mask_0 = const()[name = tensor("op_22779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22779_cast_fp16 = slice_by_index(begin = var_22779_begin_0, end = var_22779_end_0, end_mask = var_22779_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22779_cast_fp16")]; + tensor var_22783_begin_0 = const()[name = tensor("op_22783_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22783_end_0 = const()[name = tensor("op_22783_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_22783_end_mask_0 = const()[name = tensor("op_22783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22783_cast_fp16 = slice_by_index(begin = var_22783_begin_0, end = var_22783_end_0, end_mask = var_22783_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22783_cast_fp16")]; + tensor var_22787_begin_0 = const()[name = tensor("op_22787_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22787_end_0 = const()[name = tensor("op_22787_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_22787_end_mask_0 = const()[name = tensor("op_22787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22787_cast_fp16 = slice_by_index(begin = var_22787_begin_0, end = var_22787_end_0, end_mask = var_22787_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22787_cast_fp16")]; + tensor var_22791_begin_0 = const()[name = tensor("op_22791_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22791_end_0 = const()[name = tensor("op_22791_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_22791_end_mask_0 = const()[name = tensor("op_22791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22791_cast_fp16 = slice_by_index(begin = var_22791_begin_0, end = var_22791_end_0, end_mask = var_22791_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22791_cast_fp16")]; + tensor var_22795_equation_0 = const()[name = tensor("op_22795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22795_cast_fp16 = einsum(equation = var_22795_equation_0, values = (var_22637_cast_fp16, var_22554_cast_fp16))[name = tensor("op_22795_cast_fp16")]; + tensor var_22796_to_fp16 = const()[name = tensor("op_22796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1881_cast_fp16 = mul(x = var_22795_cast_fp16, y = var_22796_to_fp16)[name = tensor("aw_1881_cast_fp16")]; + tensor var_22799_equation_0 = const()[name = tensor("op_22799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22799_cast_fp16 = einsum(equation = var_22799_equation_0, values = (var_22641_cast_fp16, var_22558_cast_fp16))[name = tensor("op_22799_cast_fp16")]; + tensor var_22800_to_fp16 = const()[name = tensor("op_22800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1883_cast_fp16 = mul(x = var_22799_cast_fp16, y = var_22800_to_fp16)[name = tensor("aw_1883_cast_fp16")]; + tensor var_22803_equation_0 = const()[name = tensor("op_22803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22803_cast_fp16 = einsum(equation = var_22803_equation_0, values = (var_22645_cast_fp16, var_22562_cast_fp16))[name = tensor("op_22803_cast_fp16")]; + tensor var_22804_to_fp16 = const()[name = tensor("op_22804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1885_cast_fp16 = mul(x = var_22803_cast_fp16, y = var_22804_to_fp16)[name = tensor("aw_1885_cast_fp16")]; + tensor var_22807_equation_0 = const()[name = tensor("op_22807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22807_cast_fp16 = einsum(equation = var_22807_equation_0, values = (var_22649_cast_fp16, var_22566_cast_fp16))[name = tensor("op_22807_cast_fp16")]; + tensor var_22808_to_fp16 = const()[name = tensor("op_22808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1887_cast_fp16 = mul(x = var_22807_cast_fp16, y = var_22808_to_fp16)[name = tensor("aw_1887_cast_fp16")]; + tensor var_22811_equation_0 = const()[name = tensor("op_22811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22811_cast_fp16 = einsum(equation = var_22811_equation_0, values = (var_22653_cast_fp16, var_22570_cast_fp16))[name = tensor("op_22811_cast_fp16")]; + tensor var_22812_to_fp16 = const()[name = tensor("op_22812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1889_cast_fp16 = mul(x = var_22811_cast_fp16, y = var_22812_to_fp16)[name = tensor("aw_1889_cast_fp16")]; + tensor var_22815_equation_0 = const()[name = tensor("op_22815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22815_cast_fp16 = einsum(equation = var_22815_equation_0, values = (var_22657_cast_fp16, var_22574_cast_fp16))[name = tensor("op_22815_cast_fp16")]; + tensor var_22816_to_fp16 = const()[name = tensor("op_22816_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1891_cast_fp16 = mul(x = var_22815_cast_fp16, y = var_22816_to_fp16)[name = tensor("aw_1891_cast_fp16")]; + tensor var_22819_equation_0 = const()[name = tensor("op_22819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22819_cast_fp16 = einsum(equation = var_22819_equation_0, values = (var_22661_cast_fp16, var_22578_cast_fp16))[name = tensor("op_22819_cast_fp16")]; + tensor var_22820_to_fp16 = const()[name = tensor("op_22820_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1893_cast_fp16 = mul(x = var_22819_cast_fp16, y = var_22820_to_fp16)[name = tensor("aw_1893_cast_fp16")]; + tensor var_22823_equation_0 = const()[name = tensor("op_22823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22823_cast_fp16 = einsum(equation = var_22823_equation_0, values = (var_22665_cast_fp16, var_22582_cast_fp16))[name = tensor("op_22823_cast_fp16")]; + tensor var_22824_to_fp16 = const()[name = tensor("op_22824_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1895_cast_fp16 = mul(x = var_22823_cast_fp16, y = var_22824_to_fp16)[name = tensor("aw_1895_cast_fp16")]; + tensor var_22827_equation_0 = const()[name = tensor("op_22827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22827_cast_fp16 = einsum(equation = var_22827_equation_0, values = (var_22669_cast_fp16, var_22586_cast_fp16))[name = tensor("op_22827_cast_fp16")]; + tensor var_22828_to_fp16 = const()[name = tensor("op_22828_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1897_cast_fp16 = mul(x = var_22827_cast_fp16, y = var_22828_to_fp16)[name = tensor("aw_1897_cast_fp16")]; + tensor var_22831_equation_0 = const()[name = tensor("op_22831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22831_cast_fp16 = einsum(equation = var_22831_equation_0, values = (var_22673_cast_fp16, var_22590_cast_fp16))[name = tensor("op_22831_cast_fp16")]; + tensor var_22832_to_fp16 = const()[name = tensor("op_22832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1899_cast_fp16 = mul(x = var_22831_cast_fp16, y = var_22832_to_fp16)[name = tensor("aw_1899_cast_fp16")]; + tensor var_22835_equation_0 = const()[name = tensor("op_22835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22835_cast_fp16 = einsum(equation = var_22835_equation_0, values = (var_22677_cast_fp16, var_22594_cast_fp16))[name = tensor("op_22835_cast_fp16")]; + tensor var_22836_to_fp16 = const()[name = tensor("op_22836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1901_cast_fp16 = mul(x = var_22835_cast_fp16, y = var_22836_to_fp16)[name = tensor("aw_1901_cast_fp16")]; + tensor var_22839_equation_0 = const()[name = tensor("op_22839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22839_cast_fp16 = einsum(equation = var_22839_equation_0, values = (var_22681_cast_fp16, var_22598_cast_fp16))[name = tensor("op_22839_cast_fp16")]; + tensor var_22840_to_fp16 = const()[name = tensor("op_22840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1903_cast_fp16 = mul(x = var_22839_cast_fp16, y = var_22840_to_fp16)[name = tensor("aw_1903_cast_fp16")]; + tensor var_22843_equation_0 = const()[name = tensor("op_22843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22843_cast_fp16 = einsum(equation = var_22843_equation_0, values = (var_22685_cast_fp16, var_22602_cast_fp16))[name = tensor("op_22843_cast_fp16")]; + tensor var_22844_to_fp16 = const()[name = tensor("op_22844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1905_cast_fp16 = mul(x = var_22843_cast_fp16, y = var_22844_to_fp16)[name = tensor("aw_1905_cast_fp16")]; + tensor var_22847_equation_0 = const()[name = tensor("op_22847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22847_cast_fp16 = einsum(equation = var_22847_equation_0, values = (var_22689_cast_fp16, var_22606_cast_fp16))[name = tensor("op_22847_cast_fp16")]; + tensor var_22848_to_fp16 = const()[name = tensor("op_22848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1907_cast_fp16 = mul(x = var_22847_cast_fp16, y = var_22848_to_fp16)[name = tensor("aw_1907_cast_fp16")]; + tensor var_22851_equation_0 = const()[name = tensor("op_22851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22851_cast_fp16 = einsum(equation = var_22851_equation_0, values = (var_22693_cast_fp16, var_22610_cast_fp16))[name = tensor("op_22851_cast_fp16")]; + tensor var_22852_to_fp16 = const()[name = tensor("op_22852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1909_cast_fp16 = mul(x = var_22851_cast_fp16, y = var_22852_to_fp16)[name = tensor("aw_1909_cast_fp16")]; + tensor var_22855_equation_0 = const()[name = tensor("op_22855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22855_cast_fp16 = einsum(equation = var_22855_equation_0, values = (var_22697_cast_fp16, var_22614_cast_fp16))[name = tensor("op_22855_cast_fp16")]; + tensor var_22856_to_fp16 = const()[name = tensor("op_22856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1911_cast_fp16 = mul(x = var_22855_cast_fp16, y = var_22856_to_fp16)[name = tensor("aw_1911_cast_fp16")]; + tensor var_22859_equation_0 = const()[name = tensor("op_22859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22859_cast_fp16 = einsum(equation = var_22859_equation_0, values = (var_22701_cast_fp16, var_22618_cast_fp16))[name = tensor("op_22859_cast_fp16")]; + tensor var_22860_to_fp16 = const()[name = tensor("op_22860_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1913_cast_fp16 = mul(x = var_22859_cast_fp16, y = var_22860_to_fp16)[name = tensor("aw_1913_cast_fp16")]; + tensor var_22863_equation_0 = const()[name = tensor("op_22863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22863_cast_fp16 = einsum(equation = var_22863_equation_0, values = (var_22705_cast_fp16, var_22622_cast_fp16))[name = tensor("op_22863_cast_fp16")]; + tensor var_22864_to_fp16 = const()[name = tensor("op_22864_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1915_cast_fp16 = mul(x = var_22863_cast_fp16, y = var_22864_to_fp16)[name = tensor("aw_1915_cast_fp16")]; + tensor var_22867_equation_0 = const()[name = tensor("op_22867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22867_cast_fp16 = einsum(equation = var_22867_equation_0, values = (var_22709_cast_fp16, var_22626_cast_fp16))[name = tensor("op_22867_cast_fp16")]; + tensor var_22868_to_fp16 = const()[name = tensor("op_22868_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1917_cast_fp16 = mul(x = var_22867_cast_fp16, y = var_22868_to_fp16)[name = tensor("aw_1917_cast_fp16")]; + tensor var_22871_equation_0 = const()[name = tensor("op_22871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22871_cast_fp16 = einsum(equation = var_22871_equation_0, values = (var_22713_cast_fp16, var_22630_cast_fp16))[name = tensor("op_22871_cast_fp16")]; + tensor var_22872_to_fp16 = const()[name = tensor("op_22872_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1919_cast_fp16 = mul(x = var_22871_cast_fp16, y = var_22872_to_fp16)[name = tensor("aw_1919_cast_fp16")]; + tensor var_22874_cast_fp16 = softmax(axis = var_21077, x = aw_1881_cast_fp16)[name = tensor("op_22874_cast_fp16")]; + tensor var_22875_cast_fp16 = softmax(axis = var_21077, x = aw_1883_cast_fp16)[name = tensor("op_22875_cast_fp16")]; + tensor var_22876_cast_fp16 = softmax(axis = var_21077, x = aw_1885_cast_fp16)[name = tensor("op_22876_cast_fp16")]; + tensor var_22877_cast_fp16 = softmax(axis = var_21077, x = aw_1887_cast_fp16)[name = tensor("op_22877_cast_fp16")]; + tensor var_22878_cast_fp16 = softmax(axis = var_21077, x = aw_1889_cast_fp16)[name = tensor("op_22878_cast_fp16")]; + tensor var_22879_cast_fp16 = softmax(axis = var_21077, x = aw_1891_cast_fp16)[name = tensor("op_22879_cast_fp16")]; + tensor var_22880_cast_fp16 = softmax(axis = var_21077, x = aw_1893_cast_fp16)[name = tensor("op_22880_cast_fp16")]; + tensor var_22881_cast_fp16 = softmax(axis = var_21077, x = aw_1895_cast_fp16)[name = tensor("op_22881_cast_fp16")]; + tensor var_22882_cast_fp16 = softmax(axis = var_21077, x = aw_1897_cast_fp16)[name = tensor("op_22882_cast_fp16")]; + tensor var_22883_cast_fp16 = softmax(axis = var_21077, x = aw_1899_cast_fp16)[name = tensor("op_22883_cast_fp16")]; + tensor var_22884_cast_fp16 = softmax(axis = var_21077, x = aw_1901_cast_fp16)[name = tensor("op_22884_cast_fp16")]; + tensor var_22885_cast_fp16 = softmax(axis = var_21077, x = aw_1903_cast_fp16)[name = tensor("op_22885_cast_fp16")]; + tensor var_22886_cast_fp16 = softmax(axis = var_21077, x = aw_1905_cast_fp16)[name = tensor("op_22886_cast_fp16")]; + tensor var_22887_cast_fp16 = softmax(axis = var_21077, x = aw_1907_cast_fp16)[name = tensor("op_22887_cast_fp16")]; + tensor var_22888_cast_fp16 = softmax(axis = var_21077, x = aw_1909_cast_fp16)[name = tensor("op_22888_cast_fp16")]; + tensor var_22889_cast_fp16 = softmax(axis = var_21077, x = aw_1911_cast_fp16)[name = tensor("op_22889_cast_fp16")]; + tensor var_22890_cast_fp16 = softmax(axis = var_21077, x = aw_1913_cast_fp16)[name = tensor("op_22890_cast_fp16")]; + tensor var_22891_cast_fp16 = softmax(axis = var_21077, x = aw_1915_cast_fp16)[name = tensor("op_22891_cast_fp16")]; + tensor var_22892_cast_fp16 = softmax(axis = var_21077, x = aw_1917_cast_fp16)[name = tensor("op_22892_cast_fp16")]; + tensor var_22893_cast_fp16 = softmax(axis = var_21077, x = aw_1919_cast_fp16)[name = tensor("op_22893_cast_fp16")]; + tensor var_22895_equation_0 = const()[name = tensor("op_22895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22895_cast_fp16 = einsum(equation = var_22895_equation_0, values = (var_22715_cast_fp16, var_22874_cast_fp16))[name = tensor("op_22895_cast_fp16")]; + tensor var_22897_equation_0 = const()[name = tensor("op_22897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22897_cast_fp16 = einsum(equation = var_22897_equation_0, values = (var_22719_cast_fp16, var_22875_cast_fp16))[name = tensor("op_22897_cast_fp16")]; + tensor var_22899_equation_0 = const()[name = tensor("op_22899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22899_cast_fp16 = einsum(equation = var_22899_equation_0, values = (var_22723_cast_fp16, var_22876_cast_fp16))[name = tensor("op_22899_cast_fp16")]; + tensor var_22901_equation_0 = const()[name = tensor("op_22901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22901_cast_fp16 = einsum(equation = var_22901_equation_0, values = (var_22727_cast_fp16, var_22877_cast_fp16))[name = tensor("op_22901_cast_fp16")]; + tensor var_22903_equation_0 = const()[name = tensor("op_22903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22903_cast_fp16 = einsum(equation = var_22903_equation_0, values = (var_22731_cast_fp16, var_22878_cast_fp16))[name = tensor("op_22903_cast_fp16")]; + tensor var_22905_equation_0 = const()[name = tensor("op_22905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22905_cast_fp16 = einsum(equation = var_22905_equation_0, values = (var_22735_cast_fp16, var_22879_cast_fp16))[name = tensor("op_22905_cast_fp16")]; + tensor var_22907_equation_0 = const()[name = tensor("op_22907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22907_cast_fp16 = einsum(equation = var_22907_equation_0, values = (var_22739_cast_fp16, var_22880_cast_fp16))[name = tensor("op_22907_cast_fp16")]; + tensor var_22909_equation_0 = const()[name = tensor("op_22909_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22909_cast_fp16 = einsum(equation = var_22909_equation_0, values = (var_22743_cast_fp16, var_22881_cast_fp16))[name = tensor("op_22909_cast_fp16")]; + tensor var_22911_equation_0 = const()[name = tensor("op_22911_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22911_cast_fp16 = einsum(equation = var_22911_equation_0, values = (var_22747_cast_fp16, var_22882_cast_fp16))[name = tensor("op_22911_cast_fp16")]; + tensor var_22913_equation_0 = const()[name = tensor("op_22913_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22913_cast_fp16 = einsum(equation = var_22913_equation_0, values = (var_22751_cast_fp16, var_22883_cast_fp16))[name = tensor("op_22913_cast_fp16")]; + tensor var_22915_equation_0 = const()[name = tensor("op_22915_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22915_cast_fp16 = einsum(equation = var_22915_equation_0, values = (var_22755_cast_fp16, var_22884_cast_fp16))[name = tensor("op_22915_cast_fp16")]; + tensor var_22917_equation_0 = const()[name = tensor("op_22917_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22917_cast_fp16 = einsum(equation = var_22917_equation_0, values = (var_22759_cast_fp16, var_22885_cast_fp16))[name = tensor("op_22917_cast_fp16")]; + tensor var_22919_equation_0 = const()[name = tensor("op_22919_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22919_cast_fp16 = einsum(equation = var_22919_equation_0, values = (var_22763_cast_fp16, var_22886_cast_fp16))[name = tensor("op_22919_cast_fp16")]; + tensor var_22921_equation_0 = const()[name = tensor("op_22921_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22921_cast_fp16 = einsum(equation = var_22921_equation_0, values = (var_22767_cast_fp16, var_22887_cast_fp16))[name = tensor("op_22921_cast_fp16")]; + tensor var_22923_equation_0 = const()[name = tensor("op_22923_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22923_cast_fp16 = einsum(equation = var_22923_equation_0, values = (var_22771_cast_fp16, var_22888_cast_fp16))[name = tensor("op_22923_cast_fp16")]; + tensor var_22925_equation_0 = const()[name = tensor("op_22925_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22925_cast_fp16 = einsum(equation = var_22925_equation_0, values = (var_22775_cast_fp16, var_22889_cast_fp16))[name = tensor("op_22925_cast_fp16")]; + tensor var_22927_equation_0 = const()[name = tensor("op_22927_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22927_cast_fp16 = einsum(equation = var_22927_equation_0, values = (var_22779_cast_fp16, var_22890_cast_fp16))[name = tensor("op_22927_cast_fp16")]; + tensor var_22929_equation_0 = const()[name = tensor("op_22929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22929_cast_fp16 = einsum(equation = var_22929_equation_0, values = (var_22783_cast_fp16, var_22891_cast_fp16))[name = tensor("op_22929_cast_fp16")]; + tensor var_22931_equation_0 = const()[name = tensor("op_22931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22931_cast_fp16 = einsum(equation = var_22931_equation_0, values = (var_22787_cast_fp16, var_22892_cast_fp16))[name = tensor("op_22931_cast_fp16")]; + tensor var_22933_equation_0 = const()[name = tensor("op_22933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22933_cast_fp16 = einsum(equation = var_22933_equation_0, values = (var_22791_cast_fp16, var_22893_cast_fp16))[name = tensor("op_22933_cast_fp16")]; + tensor input_337_interleave_0 = const()[name = tensor("input_337_interleave_0"), val = tensor(false)]; + tensor input_337_cast_fp16 = concat(axis = var_21077, interleave = input_337_interleave_0, values = (var_22895_cast_fp16, var_22897_cast_fp16, var_22899_cast_fp16, var_22901_cast_fp16, var_22903_cast_fp16, var_22905_cast_fp16, var_22907_cast_fp16, var_22909_cast_fp16, var_22911_cast_fp16, var_22913_cast_fp16, var_22915_cast_fp16, var_22917_cast_fp16, var_22919_cast_fp16, var_22921_cast_fp16, var_22923_cast_fp16, var_22925_cast_fp16, var_22927_cast_fp16, var_22929_cast_fp16, var_22931_cast_fp16, var_22933_cast_fp16))[name = tensor("input_337_cast_fp16")]; + tensor var_22943_pad_type_0 = const()[name = tensor("op_22943_pad_type_0"), val = tensor("valid")]; + tensor var_22943_strides_0 = const()[name = tensor("op_22943_strides_0"), val = tensor([1, 1])]; + tensor var_22943_pad_0 = const()[name = tensor("op_22943_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22943_dilations_0 = const()[name = tensor("op_22943_dilations_0"), val = tensor([1, 1])]; + tensor var_22943_groups_0 = const()[name = tensor("op_22943_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684243776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685472640))), name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685472832)))]; + tensor var_22943_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_22943_dilations_0, groups = var_22943_groups_0, pad = var_22943_pad_0, pad_type = var_22943_pad_type_0, strides = var_22943_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = tensor("op_22943_cast_fp16")]; + tensor inputs_155_cast_fp16 = add(x = var_22943_cast_fp16, y = inputs_153_cast_fp16)[name = tensor("inputs_155_cast_fp16")]; + tensor input_339_axes_0 = const()[name = tensor("input_339_axes_0"), val = tensor([1])]; + tensor input_339_gamma_0_to_fp16 = const()[name = tensor("input_339_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685475456)))]; + tensor input_339_beta_0_to_fp16 = const()[name = tensor("input_339_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685478080)))]; + tensor var_22953_to_fp16 = const()[name = tensor("op_22953_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_339_cast_fp16 = layer_norm(axes = input_339_axes_0, beta = input_339_beta_0_to_fp16, epsilon = var_22953_to_fp16, gamma = input_339_gamma_0_to_fp16, x = inputs_155_cast_fp16)[name = tensor("input_339_cast_fp16")]; + tensor var_22973_pad_type_0 = const()[name = tensor("op_22973_pad_type_0"), val = tensor("valid")]; + tensor var_22973_strides_0 = const()[name = tensor("op_22973_strides_0"), val = tensor([1, 1])]; + tensor var_22973_pad_0 = const()[name = tensor("op_22973_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22973_dilations_0 = const()[name = tensor("op_22973_dilations_0"), val = tensor([1, 1])]; + tensor var_22973_groups_0 = const()[name = tensor("op_22973_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(685480704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(695311168))), name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(695311360)))]; + tensor var_22973_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_22973_dilations_0, groups = var_22973_groups_0, pad = var_22973_pad_0, pad_type = var_22973_pad_type_0, strides = var_22973_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_339_cast_fp16)[name = tensor("op_22973_cast_fp16")]; + tensor var_22974_split_sizes_0 = const()[name = tensor("op_22974_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_22974_axis_0 = const()[name = tensor("op_22974_axis_0"), val = tensor(1)]; + tensor var_22974_cast_fp16_0, tensor var_22974_cast_fp16_1 = split(axis = var_22974_axis_0, split_sizes = var_22974_split_sizes_0, x = var_22973_cast_fp16)[name = tensor("op_22974_cast_fp16")]; + tensor var_22976_mode_0 = const()[name = tensor("op_22976_mode_0"), val = tensor("EXACT")]; + tensor var_22976_cast_fp16 = gelu(mode = var_22976_mode_0, x = var_22974_cast_fp16_1)[name = tensor("op_22976_cast_fp16")]; + tensor input_341_cast_fp16 = mul(x = var_22974_cast_fp16_0, y = var_22976_cast_fp16)[name = tensor("input_341_cast_fp16")]; + tensor var_22984_pad_type_0 = const()[name = tensor("op_22984_pad_type_0"), val = tensor("valid")]; + tensor var_22984_strides_0 = const()[name = tensor("op_22984_strides_0"), val = tensor([1, 1])]; + tensor var_22984_pad_0 = const()[name = tensor("op_22984_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22984_dilations_0 = const()[name = tensor("op_22984_dilations_0"), val = tensor([1, 1])]; + tensor var_22984_groups_0 = const()[name = tensor("op_22984_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(695331904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(700247168))), name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(700247360)))]; + tensor var_22984_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_22984_dilations_0, groups = var_22984_groups_0, pad = var_22984_pad_0, pad_type = var_22984_pad_type_0, strides = var_22984_strides_0, weight = mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_341_cast_fp16)[name = tensor("op_22984_cast_fp16")]; + tensor inputs_157_cast_fp16 = add(x = var_22984_cast_fp16, y = inputs_155_cast_fp16)[name = tensor("inputs_157_cast_fp16")]; + tensor hidden_states_221_axes_0 = const()[name = tensor("hidden_states_221_axes_0"), val = tensor([1])]; + tensor hidden_states_221_gamma_0_to_fp16 = const()[name = tensor("hidden_states_221_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(700249984)))]; + tensor hidden_states_221_beta_0_to_fp16 = const()[name = tensor("hidden_states_221_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(700252608)))]; + tensor var_23000_to_fp16 = const()[name = tensor("op_23000_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_221_cast_fp16 = layer_norm(axes = hidden_states_221_axes_0, beta = hidden_states_221_beta_0_to_fp16, epsilon = var_23000_to_fp16, gamma = hidden_states_221_gamma_0_to_fp16, x = inputs_157_cast_fp16)[name = tensor("hidden_states_221_cast_fp16")]; + tensor q_105_pad_type_0 = const()[name = tensor("q_105_pad_type_0"), val = tensor("valid")]; + tensor q_105_strides_0 = const()[name = tensor("q_105_strides_0"), val = tensor([1, 1])]; + tensor q_105_pad_0 = const()[name = tensor("q_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_105_dilations_0 = const()[name = tensor("q_105_dilations_0"), val = tensor([1, 1])]; + tensor q_105_groups_0 = const()[name = tensor("q_105_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(700255232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(701484096))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_105_cast_fp16 = conv(dilations = q_105_dilations_0, groups = q_105_groups_0, pad = q_105_pad_0, pad_type = q_105_pad_type_0, strides = q_105_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_221_cast_fp16)[name = tensor("q_105_cast_fp16")]; + tensor k_209_pad_type_0 = const()[name = tensor("k_209_pad_type_0"), val = tensor("valid")]; + tensor k_209_strides_0 = const()[name = tensor("k_209_strides_0"), val = tensor([1, 1])]; + tensor k_209_pad_0 = const()[name = tensor("k_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_209_dilations_0 = const()[name = tensor("k_209_dilations_0"), val = tensor([1, 1])]; + tensor k_209_groups_0 = const()[name = tensor("k_209_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(701484288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(702713152))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_209_cast_fp16 = conv(dilations = k_209_dilations_0, groups = k_209_groups_0, pad = k_209_pad_0, pad_type = k_209_pad_type_0, strides = k_209_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_221_cast_fp16)[name = tensor("k_209_cast_fp16")]; + tensor v_105_pad_type_0 = const()[name = tensor("v_105_pad_type_0"), val = tensor("valid")]; + tensor v_105_strides_0 = const()[name = tensor("v_105_strides_0"), val = tensor([1, 1])]; + tensor v_105_pad_0 = const()[name = tensor("v_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_105_dilations_0 = const()[name = tensor("v_105_dilations_0"), val = tensor([1, 1])]; + tensor v_105_groups_0 = const()[name = tensor("v_105_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(702713344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703942208))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_105_cast_fp16 = conv(dilations = v_105_dilations_0, groups = v_105_groups_0, pad = v_105_pad_0, pad_type = v_105_pad_type_0, strides = v_105_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_221_cast_fp16)[name = tensor("v_105_cast_fp16")]; + tensor var_23033_begin_0 = const()[name = tensor("op_23033_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23033_end_0 = const()[name = tensor("op_23033_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23033_end_mask_0 = const()[name = tensor("op_23033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23033_cast_fp16 = slice_by_index(begin = var_23033_begin_0, end = var_23033_end_0, end_mask = var_23033_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23033_cast_fp16")]; + tensor var_23037_begin_0 = const()[name = tensor("op_23037_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23037_end_0 = const()[name = tensor("op_23037_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23037_end_mask_0 = const()[name = tensor("op_23037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23037_cast_fp16 = slice_by_index(begin = var_23037_begin_0, end = var_23037_end_0, end_mask = var_23037_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23037_cast_fp16")]; + tensor var_23041_begin_0 = const()[name = tensor("op_23041_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23041_end_0 = const()[name = tensor("op_23041_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23041_end_mask_0 = const()[name = tensor("op_23041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23041_cast_fp16 = slice_by_index(begin = var_23041_begin_0, end = var_23041_end_0, end_mask = var_23041_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23041_cast_fp16")]; + tensor var_23045_begin_0 = const()[name = tensor("op_23045_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23045_end_0 = const()[name = tensor("op_23045_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23045_end_mask_0 = const()[name = tensor("op_23045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23045_cast_fp16 = slice_by_index(begin = var_23045_begin_0, end = var_23045_end_0, end_mask = var_23045_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23045_cast_fp16")]; + tensor var_23049_begin_0 = const()[name = tensor("op_23049_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23049_end_0 = const()[name = tensor("op_23049_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23049_end_mask_0 = const()[name = tensor("op_23049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23049_cast_fp16 = slice_by_index(begin = var_23049_begin_0, end = var_23049_end_0, end_mask = var_23049_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23049_cast_fp16")]; + tensor var_23053_begin_0 = const()[name = tensor("op_23053_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23053_end_0 = const()[name = tensor("op_23053_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23053_end_mask_0 = const()[name = tensor("op_23053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23053_cast_fp16 = slice_by_index(begin = var_23053_begin_0, end = var_23053_end_0, end_mask = var_23053_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23053_cast_fp16")]; + tensor var_23057_begin_0 = const()[name = tensor("op_23057_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23057_end_0 = const()[name = tensor("op_23057_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23057_end_mask_0 = const()[name = tensor("op_23057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23057_cast_fp16 = slice_by_index(begin = var_23057_begin_0, end = var_23057_end_0, end_mask = var_23057_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23057_cast_fp16")]; + tensor var_23061_begin_0 = const()[name = tensor("op_23061_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23061_end_0 = const()[name = tensor("op_23061_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23061_end_mask_0 = const()[name = tensor("op_23061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23061_cast_fp16 = slice_by_index(begin = var_23061_begin_0, end = var_23061_end_0, end_mask = var_23061_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23061_cast_fp16")]; + tensor var_23065_begin_0 = const()[name = tensor("op_23065_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23065_end_0 = const()[name = tensor("op_23065_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23065_end_mask_0 = const()[name = tensor("op_23065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23065_cast_fp16 = slice_by_index(begin = var_23065_begin_0, end = var_23065_end_0, end_mask = var_23065_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23065_cast_fp16")]; + tensor var_23069_begin_0 = const()[name = tensor("op_23069_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23069_end_0 = const()[name = tensor("op_23069_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23069_end_mask_0 = const()[name = tensor("op_23069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23069_cast_fp16 = slice_by_index(begin = var_23069_begin_0, end = var_23069_end_0, end_mask = var_23069_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23069_cast_fp16")]; + tensor var_23073_begin_0 = const()[name = tensor("op_23073_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23073_end_0 = const()[name = tensor("op_23073_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23073_end_mask_0 = const()[name = tensor("op_23073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23073_cast_fp16 = slice_by_index(begin = var_23073_begin_0, end = var_23073_end_0, end_mask = var_23073_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23073_cast_fp16")]; + tensor var_23077_begin_0 = const()[name = tensor("op_23077_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23077_end_0 = const()[name = tensor("op_23077_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23077_end_mask_0 = const()[name = tensor("op_23077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23077_cast_fp16 = slice_by_index(begin = var_23077_begin_0, end = var_23077_end_0, end_mask = var_23077_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23077_cast_fp16")]; + tensor var_23081_begin_0 = const()[name = tensor("op_23081_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23081_end_0 = const()[name = tensor("op_23081_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23081_end_mask_0 = const()[name = tensor("op_23081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23081_cast_fp16 = slice_by_index(begin = var_23081_begin_0, end = var_23081_end_0, end_mask = var_23081_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23081_cast_fp16")]; + tensor var_23085_begin_0 = const()[name = tensor("op_23085_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23085_end_0 = const()[name = tensor("op_23085_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23085_end_mask_0 = const()[name = tensor("op_23085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23085_cast_fp16 = slice_by_index(begin = var_23085_begin_0, end = var_23085_end_0, end_mask = var_23085_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23085_cast_fp16")]; + tensor var_23089_begin_0 = const()[name = tensor("op_23089_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23089_end_0 = const()[name = tensor("op_23089_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_23089_end_mask_0 = const()[name = tensor("op_23089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23089_cast_fp16 = slice_by_index(begin = var_23089_begin_0, end = var_23089_end_0, end_mask = var_23089_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23089_cast_fp16")]; + tensor var_23093_begin_0 = const()[name = tensor("op_23093_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23093_end_0 = const()[name = tensor("op_23093_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23093_end_mask_0 = const()[name = tensor("op_23093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23093_cast_fp16 = slice_by_index(begin = var_23093_begin_0, end = var_23093_end_0, end_mask = var_23093_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23093_cast_fp16")]; + tensor var_23097_begin_0 = const()[name = tensor("op_23097_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23097_end_0 = const()[name = tensor("op_23097_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_23097_end_mask_0 = const()[name = tensor("op_23097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23097_cast_fp16 = slice_by_index(begin = var_23097_begin_0, end = var_23097_end_0, end_mask = var_23097_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23097_cast_fp16")]; + tensor var_23101_begin_0 = const()[name = tensor("op_23101_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23101_end_0 = const()[name = tensor("op_23101_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_23101_end_mask_0 = const()[name = tensor("op_23101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23101_cast_fp16 = slice_by_index(begin = var_23101_begin_0, end = var_23101_end_0, end_mask = var_23101_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23101_cast_fp16")]; + tensor var_23105_begin_0 = const()[name = tensor("op_23105_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23105_end_0 = const()[name = tensor("op_23105_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_23105_end_mask_0 = const()[name = tensor("op_23105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23105_cast_fp16 = slice_by_index(begin = var_23105_begin_0, end = var_23105_end_0, end_mask = var_23105_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23105_cast_fp16")]; + tensor var_23109_begin_0 = const()[name = tensor("op_23109_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23109_end_0 = const()[name = tensor("op_23109_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_23109_end_mask_0 = const()[name = tensor("op_23109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23109_cast_fp16 = slice_by_index(begin = var_23109_begin_0, end = var_23109_end_0, end_mask = var_23109_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23109_cast_fp16")]; + tensor k_211_perm_0 = const()[name = tensor("k_211_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_23116_begin_0 = const()[name = tensor("op_23116_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23116_end_0 = const()[name = tensor("op_23116_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_23116_end_mask_0 = const()[name = tensor("op_23116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_211_cast_fp16 = transpose(perm = k_211_perm_0, x = k_209_cast_fp16)[name = tensor("transpose_15")]; + tensor var_23116_cast_fp16 = slice_by_index(begin = var_23116_begin_0, end = var_23116_end_0, end_mask = var_23116_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23116_cast_fp16")]; + tensor var_23120_begin_0 = const()[name = tensor("op_23120_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_23120_end_0 = const()[name = tensor("op_23120_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_23120_end_mask_0 = const()[name = tensor("op_23120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23120_cast_fp16 = slice_by_index(begin = var_23120_begin_0, end = var_23120_end_0, end_mask = var_23120_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23120_cast_fp16")]; + tensor var_23124_begin_0 = const()[name = tensor("op_23124_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_23124_end_0 = const()[name = tensor("op_23124_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_23124_end_mask_0 = const()[name = tensor("op_23124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23124_cast_fp16 = slice_by_index(begin = var_23124_begin_0, end = var_23124_end_0, end_mask = var_23124_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23124_cast_fp16")]; + tensor var_23128_begin_0 = const()[name = tensor("op_23128_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_23128_end_0 = const()[name = tensor("op_23128_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_23128_end_mask_0 = const()[name = tensor("op_23128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23128_cast_fp16 = slice_by_index(begin = var_23128_begin_0, end = var_23128_end_0, end_mask = var_23128_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23128_cast_fp16")]; + tensor var_23132_begin_0 = const()[name = tensor("op_23132_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_23132_end_0 = const()[name = tensor("op_23132_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_23132_end_mask_0 = const()[name = tensor("op_23132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23132_cast_fp16 = slice_by_index(begin = var_23132_begin_0, end = var_23132_end_0, end_mask = var_23132_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23132_cast_fp16")]; + tensor var_23136_begin_0 = const()[name = tensor("op_23136_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_23136_end_0 = const()[name = tensor("op_23136_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_23136_end_mask_0 = const()[name = tensor("op_23136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23136_cast_fp16 = slice_by_index(begin = var_23136_begin_0, end = var_23136_end_0, end_mask = var_23136_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23136_cast_fp16")]; + tensor var_23140_begin_0 = const()[name = tensor("op_23140_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_23140_end_0 = const()[name = tensor("op_23140_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_23140_end_mask_0 = const()[name = tensor("op_23140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23140_cast_fp16 = slice_by_index(begin = var_23140_begin_0, end = var_23140_end_0, end_mask = var_23140_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23140_cast_fp16")]; + tensor var_23144_begin_0 = const()[name = tensor("op_23144_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_23144_end_0 = const()[name = tensor("op_23144_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_23144_end_mask_0 = const()[name = tensor("op_23144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23144_cast_fp16 = slice_by_index(begin = var_23144_begin_0, end = var_23144_end_0, end_mask = var_23144_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23144_cast_fp16")]; + tensor var_23148_begin_0 = const()[name = tensor("op_23148_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_23148_end_0 = const()[name = tensor("op_23148_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_23148_end_mask_0 = const()[name = tensor("op_23148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23148_cast_fp16 = slice_by_index(begin = var_23148_begin_0, end = var_23148_end_0, end_mask = var_23148_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23148_cast_fp16")]; + tensor var_23152_begin_0 = const()[name = tensor("op_23152_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_23152_end_0 = const()[name = tensor("op_23152_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_23152_end_mask_0 = const()[name = tensor("op_23152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23152_cast_fp16 = slice_by_index(begin = var_23152_begin_0, end = var_23152_end_0, end_mask = var_23152_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23152_cast_fp16")]; + tensor var_23156_begin_0 = const()[name = tensor("op_23156_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_23156_end_0 = const()[name = tensor("op_23156_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_23156_end_mask_0 = const()[name = tensor("op_23156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23156_cast_fp16 = slice_by_index(begin = var_23156_begin_0, end = var_23156_end_0, end_mask = var_23156_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23156_cast_fp16")]; + tensor var_23160_begin_0 = const()[name = tensor("op_23160_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_23160_end_0 = const()[name = tensor("op_23160_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_23160_end_mask_0 = const()[name = tensor("op_23160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23160_cast_fp16 = slice_by_index(begin = var_23160_begin_0, end = var_23160_end_0, end_mask = var_23160_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23160_cast_fp16")]; + tensor var_23164_begin_0 = const()[name = tensor("op_23164_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_23164_end_0 = const()[name = tensor("op_23164_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_23164_end_mask_0 = const()[name = tensor("op_23164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23164_cast_fp16 = slice_by_index(begin = var_23164_begin_0, end = var_23164_end_0, end_mask = var_23164_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23164_cast_fp16")]; + tensor var_23168_begin_0 = const()[name = tensor("op_23168_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_23168_end_0 = const()[name = tensor("op_23168_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_23168_end_mask_0 = const()[name = tensor("op_23168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23168_cast_fp16 = slice_by_index(begin = var_23168_begin_0, end = var_23168_end_0, end_mask = var_23168_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23168_cast_fp16")]; + tensor var_23172_begin_0 = const()[name = tensor("op_23172_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_23172_end_0 = const()[name = tensor("op_23172_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_23172_end_mask_0 = const()[name = tensor("op_23172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23172_cast_fp16 = slice_by_index(begin = var_23172_begin_0, end = var_23172_end_0, end_mask = var_23172_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23172_cast_fp16")]; + tensor var_23176_begin_0 = const()[name = tensor("op_23176_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_23176_end_0 = const()[name = tensor("op_23176_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23176_end_mask_0 = const()[name = tensor("op_23176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23176_cast_fp16 = slice_by_index(begin = var_23176_begin_0, end = var_23176_end_0, end_mask = var_23176_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23176_cast_fp16")]; + tensor var_23180_begin_0 = const()[name = tensor("op_23180_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_23180_end_0 = const()[name = tensor("op_23180_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_23180_end_mask_0 = const()[name = tensor("op_23180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23180_cast_fp16 = slice_by_index(begin = var_23180_begin_0, end = var_23180_end_0, end_mask = var_23180_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23180_cast_fp16")]; + tensor var_23184_begin_0 = const()[name = tensor("op_23184_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_23184_end_0 = const()[name = tensor("op_23184_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_23184_end_mask_0 = const()[name = tensor("op_23184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23184_cast_fp16 = slice_by_index(begin = var_23184_begin_0, end = var_23184_end_0, end_mask = var_23184_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23184_cast_fp16")]; + tensor var_23188_begin_0 = const()[name = tensor("op_23188_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_23188_end_0 = const()[name = tensor("op_23188_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_23188_end_mask_0 = const()[name = tensor("op_23188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23188_cast_fp16 = slice_by_index(begin = var_23188_begin_0, end = var_23188_end_0, end_mask = var_23188_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23188_cast_fp16")]; + tensor var_23192_begin_0 = const()[name = tensor("op_23192_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_23192_end_0 = const()[name = tensor("op_23192_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_23192_end_mask_0 = const()[name = tensor("op_23192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23192_cast_fp16 = slice_by_index(begin = var_23192_begin_0, end = var_23192_end_0, end_mask = var_23192_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23192_cast_fp16")]; + tensor var_23194_begin_0 = const()[name = tensor("op_23194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23194_end_0 = const()[name = tensor("op_23194_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23194_end_mask_0 = const()[name = tensor("op_23194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23194_cast_fp16 = slice_by_index(begin = var_23194_begin_0, end = var_23194_end_0, end_mask = var_23194_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23194_cast_fp16")]; + tensor var_23198_begin_0 = const()[name = tensor("op_23198_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23198_end_0 = const()[name = tensor("op_23198_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23198_end_mask_0 = const()[name = tensor("op_23198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23198_cast_fp16 = slice_by_index(begin = var_23198_begin_0, end = var_23198_end_0, end_mask = var_23198_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23198_cast_fp16")]; + tensor var_23202_begin_0 = const()[name = tensor("op_23202_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23202_end_0 = const()[name = tensor("op_23202_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23202_end_mask_0 = const()[name = tensor("op_23202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23202_cast_fp16 = slice_by_index(begin = var_23202_begin_0, end = var_23202_end_0, end_mask = var_23202_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23202_cast_fp16")]; + tensor var_23206_begin_0 = const()[name = tensor("op_23206_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23206_end_0 = const()[name = tensor("op_23206_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23206_end_mask_0 = const()[name = tensor("op_23206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23206_cast_fp16 = slice_by_index(begin = var_23206_begin_0, end = var_23206_end_0, end_mask = var_23206_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23206_cast_fp16")]; + tensor var_23210_begin_0 = const()[name = tensor("op_23210_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23210_end_0 = const()[name = tensor("op_23210_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23210_end_mask_0 = const()[name = tensor("op_23210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23210_cast_fp16 = slice_by_index(begin = var_23210_begin_0, end = var_23210_end_0, end_mask = var_23210_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23210_cast_fp16")]; + tensor var_23214_begin_0 = const()[name = tensor("op_23214_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23214_end_0 = const()[name = tensor("op_23214_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23214_end_mask_0 = const()[name = tensor("op_23214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23214_cast_fp16 = slice_by_index(begin = var_23214_begin_0, end = var_23214_end_0, end_mask = var_23214_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23214_cast_fp16")]; + tensor var_23218_begin_0 = const()[name = tensor("op_23218_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23218_end_0 = const()[name = tensor("op_23218_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23218_end_mask_0 = const()[name = tensor("op_23218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23218_cast_fp16 = slice_by_index(begin = var_23218_begin_0, end = var_23218_end_0, end_mask = var_23218_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23218_cast_fp16")]; + tensor var_23222_begin_0 = const()[name = tensor("op_23222_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23222_end_0 = const()[name = tensor("op_23222_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23222_end_mask_0 = const()[name = tensor("op_23222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23222_cast_fp16 = slice_by_index(begin = var_23222_begin_0, end = var_23222_end_0, end_mask = var_23222_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23222_cast_fp16")]; + tensor var_23226_begin_0 = const()[name = tensor("op_23226_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23226_end_0 = const()[name = tensor("op_23226_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23226_end_mask_0 = const()[name = tensor("op_23226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23226_cast_fp16 = slice_by_index(begin = var_23226_begin_0, end = var_23226_end_0, end_mask = var_23226_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23226_cast_fp16")]; + tensor var_23230_begin_0 = const()[name = tensor("op_23230_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23230_end_0 = const()[name = tensor("op_23230_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23230_end_mask_0 = const()[name = tensor("op_23230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23230_cast_fp16 = slice_by_index(begin = var_23230_begin_0, end = var_23230_end_0, end_mask = var_23230_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23230_cast_fp16")]; + tensor var_23234_begin_0 = const()[name = tensor("op_23234_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23234_end_0 = const()[name = tensor("op_23234_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23234_end_mask_0 = const()[name = tensor("op_23234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23234_cast_fp16 = slice_by_index(begin = var_23234_begin_0, end = var_23234_end_0, end_mask = var_23234_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23234_cast_fp16")]; + tensor var_23238_begin_0 = const()[name = tensor("op_23238_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23238_end_0 = const()[name = tensor("op_23238_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23238_end_mask_0 = const()[name = tensor("op_23238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23238_cast_fp16 = slice_by_index(begin = var_23238_begin_0, end = var_23238_end_0, end_mask = var_23238_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23238_cast_fp16")]; + tensor var_23242_begin_0 = const()[name = tensor("op_23242_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23242_end_0 = const()[name = tensor("op_23242_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23242_end_mask_0 = const()[name = tensor("op_23242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23242_cast_fp16 = slice_by_index(begin = var_23242_begin_0, end = var_23242_end_0, end_mask = var_23242_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23242_cast_fp16")]; + tensor var_23246_begin_0 = const()[name = tensor("op_23246_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23246_end_0 = const()[name = tensor("op_23246_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23246_end_mask_0 = const()[name = tensor("op_23246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23246_cast_fp16 = slice_by_index(begin = var_23246_begin_0, end = var_23246_end_0, end_mask = var_23246_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23246_cast_fp16")]; + tensor var_23250_begin_0 = const()[name = tensor("op_23250_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23250_end_0 = const()[name = tensor("op_23250_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_23250_end_mask_0 = const()[name = tensor("op_23250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23250_cast_fp16 = slice_by_index(begin = var_23250_begin_0, end = var_23250_end_0, end_mask = var_23250_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23250_cast_fp16")]; + tensor var_23254_begin_0 = const()[name = tensor("op_23254_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23254_end_0 = const()[name = tensor("op_23254_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23254_end_mask_0 = const()[name = tensor("op_23254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23254_cast_fp16 = slice_by_index(begin = var_23254_begin_0, end = var_23254_end_0, end_mask = var_23254_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23254_cast_fp16")]; + tensor var_23258_begin_0 = const()[name = tensor("op_23258_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23258_end_0 = const()[name = tensor("op_23258_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_23258_end_mask_0 = const()[name = tensor("op_23258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23258_cast_fp16 = slice_by_index(begin = var_23258_begin_0, end = var_23258_end_0, end_mask = var_23258_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23258_cast_fp16")]; + tensor var_23262_begin_0 = const()[name = tensor("op_23262_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23262_end_0 = const()[name = tensor("op_23262_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_23262_end_mask_0 = const()[name = tensor("op_23262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23262_cast_fp16 = slice_by_index(begin = var_23262_begin_0, end = var_23262_end_0, end_mask = var_23262_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23262_cast_fp16")]; + tensor var_23266_begin_0 = const()[name = tensor("op_23266_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23266_end_0 = const()[name = tensor("op_23266_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_23266_end_mask_0 = const()[name = tensor("op_23266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23266_cast_fp16 = slice_by_index(begin = var_23266_begin_0, end = var_23266_end_0, end_mask = var_23266_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23266_cast_fp16")]; + tensor var_23270_begin_0 = const()[name = tensor("op_23270_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23270_end_0 = const()[name = tensor("op_23270_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_23270_end_mask_0 = const()[name = tensor("op_23270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23270_cast_fp16 = slice_by_index(begin = var_23270_begin_0, end = var_23270_end_0, end_mask = var_23270_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23270_cast_fp16")]; + tensor var_23274_equation_0 = const()[name = tensor("op_23274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23274_cast_fp16 = einsum(equation = var_23274_equation_0, values = (var_23116_cast_fp16, var_23033_cast_fp16))[name = tensor("op_23274_cast_fp16")]; + tensor var_23275_to_fp16 = const()[name = tensor("op_23275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1921_cast_fp16 = mul(x = var_23274_cast_fp16, y = var_23275_to_fp16)[name = tensor("aw_1921_cast_fp16")]; + tensor var_23278_equation_0 = const()[name = tensor("op_23278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23278_cast_fp16 = einsum(equation = var_23278_equation_0, values = (var_23120_cast_fp16, var_23037_cast_fp16))[name = tensor("op_23278_cast_fp16")]; + tensor var_23279_to_fp16 = const()[name = tensor("op_23279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1923_cast_fp16 = mul(x = var_23278_cast_fp16, y = var_23279_to_fp16)[name = tensor("aw_1923_cast_fp16")]; + tensor var_23282_equation_0 = const()[name = tensor("op_23282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23282_cast_fp16 = einsum(equation = var_23282_equation_0, values = (var_23124_cast_fp16, var_23041_cast_fp16))[name = tensor("op_23282_cast_fp16")]; + tensor var_23283_to_fp16 = const()[name = tensor("op_23283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1925_cast_fp16 = mul(x = var_23282_cast_fp16, y = var_23283_to_fp16)[name = tensor("aw_1925_cast_fp16")]; + tensor var_23286_equation_0 = const()[name = tensor("op_23286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23286_cast_fp16 = einsum(equation = var_23286_equation_0, values = (var_23128_cast_fp16, var_23045_cast_fp16))[name = tensor("op_23286_cast_fp16")]; + tensor var_23287_to_fp16 = const()[name = tensor("op_23287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1927_cast_fp16 = mul(x = var_23286_cast_fp16, y = var_23287_to_fp16)[name = tensor("aw_1927_cast_fp16")]; + tensor var_23290_equation_0 = const()[name = tensor("op_23290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23290_cast_fp16 = einsum(equation = var_23290_equation_0, values = (var_23132_cast_fp16, var_23049_cast_fp16))[name = tensor("op_23290_cast_fp16")]; + tensor var_23291_to_fp16 = const()[name = tensor("op_23291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1929_cast_fp16 = mul(x = var_23290_cast_fp16, y = var_23291_to_fp16)[name = tensor("aw_1929_cast_fp16")]; + tensor var_23294_equation_0 = const()[name = tensor("op_23294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23294_cast_fp16 = einsum(equation = var_23294_equation_0, values = (var_23136_cast_fp16, var_23053_cast_fp16))[name = tensor("op_23294_cast_fp16")]; + tensor var_23295_to_fp16 = const()[name = tensor("op_23295_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1931_cast_fp16 = mul(x = var_23294_cast_fp16, y = var_23295_to_fp16)[name = tensor("aw_1931_cast_fp16")]; + tensor var_23298_equation_0 = const()[name = tensor("op_23298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23298_cast_fp16 = einsum(equation = var_23298_equation_0, values = (var_23140_cast_fp16, var_23057_cast_fp16))[name = tensor("op_23298_cast_fp16")]; + tensor var_23299_to_fp16 = const()[name = tensor("op_23299_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1933_cast_fp16 = mul(x = var_23298_cast_fp16, y = var_23299_to_fp16)[name = tensor("aw_1933_cast_fp16")]; + tensor var_23302_equation_0 = const()[name = tensor("op_23302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23302_cast_fp16 = einsum(equation = var_23302_equation_0, values = (var_23144_cast_fp16, var_23061_cast_fp16))[name = tensor("op_23302_cast_fp16")]; + tensor var_23303_to_fp16 = const()[name = tensor("op_23303_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1935_cast_fp16 = mul(x = var_23302_cast_fp16, y = var_23303_to_fp16)[name = tensor("aw_1935_cast_fp16")]; + tensor var_23306_equation_0 = const()[name = tensor("op_23306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23306_cast_fp16 = einsum(equation = var_23306_equation_0, values = (var_23148_cast_fp16, var_23065_cast_fp16))[name = tensor("op_23306_cast_fp16")]; + tensor var_23307_to_fp16 = const()[name = tensor("op_23307_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1937_cast_fp16 = mul(x = var_23306_cast_fp16, y = var_23307_to_fp16)[name = tensor("aw_1937_cast_fp16")]; + tensor var_23310_equation_0 = const()[name = tensor("op_23310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23310_cast_fp16 = einsum(equation = var_23310_equation_0, values = (var_23152_cast_fp16, var_23069_cast_fp16))[name = tensor("op_23310_cast_fp16")]; + tensor var_23311_to_fp16 = const()[name = tensor("op_23311_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1939_cast_fp16 = mul(x = var_23310_cast_fp16, y = var_23311_to_fp16)[name = tensor("aw_1939_cast_fp16")]; + tensor var_23314_equation_0 = const()[name = tensor("op_23314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23314_cast_fp16 = einsum(equation = var_23314_equation_0, values = (var_23156_cast_fp16, var_23073_cast_fp16))[name = tensor("op_23314_cast_fp16")]; + tensor var_23315_to_fp16 = const()[name = tensor("op_23315_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1941_cast_fp16 = mul(x = var_23314_cast_fp16, y = var_23315_to_fp16)[name = tensor("aw_1941_cast_fp16")]; + tensor var_23318_equation_0 = const()[name = tensor("op_23318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23318_cast_fp16 = einsum(equation = var_23318_equation_0, values = (var_23160_cast_fp16, var_23077_cast_fp16))[name = tensor("op_23318_cast_fp16")]; + tensor var_23319_to_fp16 = const()[name = tensor("op_23319_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1943_cast_fp16 = mul(x = var_23318_cast_fp16, y = var_23319_to_fp16)[name = tensor("aw_1943_cast_fp16")]; + tensor var_23322_equation_0 = const()[name = tensor("op_23322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23322_cast_fp16 = einsum(equation = var_23322_equation_0, values = (var_23164_cast_fp16, var_23081_cast_fp16))[name = tensor("op_23322_cast_fp16")]; + tensor var_23323_to_fp16 = const()[name = tensor("op_23323_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1945_cast_fp16 = mul(x = var_23322_cast_fp16, y = var_23323_to_fp16)[name = tensor("aw_1945_cast_fp16")]; + tensor var_23326_equation_0 = const()[name = tensor("op_23326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23326_cast_fp16 = einsum(equation = var_23326_equation_0, values = (var_23168_cast_fp16, var_23085_cast_fp16))[name = tensor("op_23326_cast_fp16")]; + tensor var_23327_to_fp16 = const()[name = tensor("op_23327_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1947_cast_fp16 = mul(x = var_23326_cast_fp16, y = var_23327_to_fp16)[name = tensor("aw_1947_cast_fp16")]; + tensor var_23330_equation_0 = const()[name = tensor("op_23330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23330_cast_fp16 = einsum(equation = var_23330_equation_0, values = (var_23172_cast_fp16, var_23089_cast_fp16))[name = tensor("op_23330_cast_fp16")]; + tensor var_23331_to_fp16 = const()[name = tensor("op_23331_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1949_cast_fp16 = mul(x = var_23330_cast_fp16, y = var_23331_to_fp16)[name = tensor("aw_1949_cast_fp16")]; + tensor var_23334_equation_0 = const()[name = tensor("op_23334_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23334_cast_fp16 = einsum(equation = var_23334_equation_0, values = (var_23176_cast_fp16, var_23093_cast_fp16))[name = tensor("op_23334_cast_fp16")]; + tensor var_23335_to_fp16 = const()[name = tensor("op_23335_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1951_cast_fp16 = mul(x = var_23334_cast_fp16, y = var_23335_to_fp16)[name = tensor("aw_1951_cast_fp16")]; + tensor var_23338_equation_0 = const()[name = tensor("op_23338_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23338_cast_fp16 = einsum(equation = var_23338_equation_0, values = (var_23180_cast_fp16, var_23097_cast_fp16))[name = tensor("op_23338_cast_fp16")]; + tensor var_23339_to_fp16 = const()[name = tensor("op_23339_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1953_cast_fp16 = mul(x = var_23338_cast_fp16, y = var_23339_to_fp16)[name = tensor("aw_1953_cast_fp16")]; + tensor var_23342_equation_0 = const()[name = tensor("op_23342_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23342_cast_fp16 = einsum(equation = var_23342_equation_0, values = (var_23184_cast_fp16, var_23101_cast_fp16))[name = tensor("op_23342_cast_fp16")]; + tensor var_23343_to_fp16 = const()[name = tensor("op_23343_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1955_cast_fp16 = mul(x = var_23342_cast_fp16, y = var_23343_to_fp16)[name = tensor("aw_1955_cast_fp16")]; + tensor var_23346_equation_0 = const()[name = tensor("op_23346_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23346_cast_fp16 = einsum(equation = var_23346_equation_0, values = (var_23188_cast_fp16, var_23105_cast_fp16))[name = tensor("op_23346_cast_fp16")]; + tensor var_23347_to_fp16 = const()[name = tensor("op_23347_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1957_cast_fp16 = mul(x = var_23346_cast_fp16, y = var_23347_to_fp16)[name = tensor("aw_1957_cast_fp16")]; + tensor var_23350_equation_0 = const()[name = tensor("op_23350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23350_cast_fp16 = einsum(equation = var_23350_equation_0, values = (var_23192_cast_fp16, var_23109_cast_fp16))[name = tensor("op_23350_cast_fp16")]; + tensor var_23351_to_fp16 = const()[name = tensor("op_23351_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1959_cast_fp16 = mul(x = var_23350_cast_fp16, y = var_23351_to_fp16)[name = tensor("aw_1959_cast_fp16")]; + tensor var_23353_cast_fp16 = softmax(axis = var_21077, x = aw_1921_cast_fp16)[name = tensor("op_23353_cast_fp16")]; + tensor var_23354_cast_fp16 = softmax(axis = var_21077, x = aw_1923_cast_fp16)[name = tensor("op_23354_cast_fp16")]; + tensor var_23355_cast_fp16 = softmax(axis = var_21077, x = aw_1925_cast_fp16)[name = tensor("op_23355_cast_fp16")]; + tensor var_23356_cast_fp16 = softmax(axis = var_21077, x = aw_1927_cast_fp16)[name = tensor("op_23356_cast_fp16")]; + tensor var_23357_cast_fp16 = softmax(axis = var_21077, x = aw_1929_cast_fp16)[name = tensor("op_23357_cast_fp16")]; + tensor var_23358_cast_fp16 = softmax(axis = var_21077, x = aw_1931_cast_fp16)[name = tensor("op_23358_cast_fp16")]; + tensor var_23359_cast_fp16 = softmax(axis = var_21077, x = aw_1933_cast_fp16)[name = tensor("op_23359_cast_fp16")]; + tensor var_23360_cast_fp16 = softmax(axis = var_21077, x = aw_1935_cast_fp16)[name = tensor("op_23360_cast_fp16")]; + tensor var_23361_cast_fp16 = softmax(axis = var_21077, x = aw_1937_cast_fp16)[name = tensor("op_23361_cast_fp16")]; + tensor var_23362_cast_fp16 = softmax(axis = var_21077, x = aw_1939_cast_fp16)[name = tensor("op_23362_cast_fp16")]; + tensor var_23363_cast_fp16 = softmax(axis = var_21077, x = aw_1941_cast_fp16)[name = tensor("op_23363_cast_fp16")]; + tensor var_23364_cast_fp16 = softmax(axis = var_21077, x = aw_1943_cast_fp16)[name = tensor("op_23364_cast_fp16")]; + tensor var_23365_cast_fp16 = softmax(axis = var_21077, x = aw_1945_cast_fp16)[name = tensor("op_23365_cast_fp16")]; + tensor var_23366_cast_fp16 = softmax(axis = var_21077, x = aw_1947_cast_fp16)[name = tensor("op_23366_cast_fp16")]; + tensor var_23367_cast_fp16 = softmax(axis = var_21077, x = aw_1949_cast_fp16)[name = tensor("op_23367_cast_fp16")]; + tensor var_23368_cast_fp16 = softmax(axis = var_21077, x = aw_1951_cast_fp16)[name = tensor("op_23368_cast_fp16")]; + tensor var_23369_cast_fp16 = softmax(axis = var_21077, x = aw_1953_cast_fp16)[name = tensor("op_23369_cast_fp16")]; + tensor var_23370_cast_fp16 = softmax(axis = var_21077, x = aw_1955_cast_fp16)[name = tensor("op_23370_cast_fp16")]; + tensor var_23371_cast_fp16 = softmax(axis = var_21077, x = aw_1957_cast_fp16)[name = tensor("op_23371_cast_fp16")]; + tensor var_23372_cast_fp16 = softmax(axis = var_21077, x = aw_1959_cast_fp16)[name = tensor("op_23372_cast_fp16")]; + tensor var_23374_equation_0 = const()[name = tensor("op_23374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23374_cast_fp16 = einsum(equation = var_23374_equation_0, values = (var_23194_cast_fp16, var_23353_cast_fp16))[name = tensor("op_23374_cast_fp16")]; + tensor var_23376_equation_0 = const()[name = tensor("op_23376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23376_cast_fp16 = einsum(equation = var_23376_equation_0, values = (var_23198_cast_fp16, var_23354_cast_fp16))[name = tensor("op_23376_cast_fp16")]; + tensor var_23378_equation_0 = const()[name = tensor("op_23378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23378_cast_fp16 = einsum(equation = var_23378_equation_0, values = (var_23202_cast_fp16, var_23355_cast_fp16))[name = tensor("op_23378_cast_fp16")]; + tensor var_23380_equation_0 = const()[name = tensor("op_23380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23380_cast_fp16 = einsum(equation = var_23380_equation_0, values = (var_23206_cast_fp16, var_23356_cast_fp16))[name = tensor("op_23380_cast_fp16")]; + tensor var_23382_equation_0 = const()[name = tensor("op_23382_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23382_cast_fp16 = einsum(equation = var_23382_equation_0, values = (var_23210_cast_fp16, var_23357_cast_fp16))[name = tensor("op_23382_cast_fp16")]; + tensor var_23384_equation_0 = const()[name = tensor("op_23384_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23384_cast_fp16 = einsum(equation = var_23384_equation_0, values = (var_23214_cast_fp16, var_23358_cast_fp16))[name = tensor("op_23384_cast_fp16")]; + tensor var_23386_equation_0 = const()[name = tensor("op_23386_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23386_cast_fp16 = einsum(equation = var_23386_equation_0, values = (var_23218_cast_fp16, var_23359_cast_fp16))[name = tensor("op_23386_cast_fp16")]; + tensor var_23388_equation_0 = const()[name = tensor("op_23388_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23388_cast_fp16 = einsum(equation = var_23388_equation_0, values = (var_23222_cast_fp16, var_23360_cast_fp16))[name = tensor("op_23388_cast_fp16")]; + tensor var_23390_equation_0 = const()[name = tensor("op_23390_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23390_cast_fp16 = einsum(equation = var_23390_equation_0, values = (var_23226_cast_fp16, var_23361_cast_fp16))[name = tensor("op_23390_cast_fp16")]; + tensor var_23392_equation_0 = const()[name = tensor("op_23392_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23392_cast_fp16 = einsum(equation = var_23392_equation_0, values = (var_23230_cast_fp16, var_23362_cast_fp16))[name = tensor("op_23392_cast_fp16")]; + tensor var_23394_equation_0 = const()[name = tensor("op_23394_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23394_cast_fp16 = einsum(equation = var_23394_equation_0, values = (var_23234_cast_fp16, var_23363_cast_fp16))[name = tensor("op_23394_cast_fp16")]; + tensor var_23396_equation_0 = const()[name = tensor("op_23396_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23396_cast_fp16 = einsum(equation = var_23396_equation_0, values = (var_23238_cast_fp16, var_23364_cast_fp16))[name = tensor("op_23396_cast_fp16")]; + tensor var_23398_equation_0 = const()[name = tensor("op_23398_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23398_cast_fp16 = einsum(equation = var_23398_equation_0, values = (var_23242_cast_fp16, var_23365_cast_fp16))[name = tensor("op_23398_cast_fp16")]; + tensor var_23400_equation_0 = const()[name = tensor("op_23400_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23400_cast_fp16 = einsum(equation = var_23400_equation_0, values = (var_23246_cast_fp16, var_23366_cast_fp16))[name = tensor("op_23400_cast_fp16")]; + tensor var_23402_equation_0 = const()[name = tensor("op_23402_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23402_cast_fp16 = einsum(equation = var_23402_equation_0, values = (var_23250_cast_fp16, var_23367_cast_fp16))[name = tensor("op_23402_cast_fp16")]; + tensor var_23404_equation_0 = const()[name = tensor("op_23404_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23404_cast_fp16 = einsum(equation = var_23404_equation_0, values = (var_23254_cast_fp16, var_23368_cast_fp16))[name = tensor("op_23404_cast_fp16")]; + tensor var_23406_equation_0 = const()[name = tensor("op_23406_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23406_cast_fp16 = einsum(equation = var_23406_equation_0, values = (var_23258_cast_fp16, var_23369_cast_fp16))[name = tensor("op_23406_cast_fp16")]; + tensor var_23408_equation_0 = const()[name = tensor("op_23408_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23408_cast_fp16 = einsum(equation = var_23408_equation_0, values = (var_23262_cast_fp16, var_23370_cast_fp16))[name = tensor("op_23408_cast_fp16")]; + tensor var_23410_equation_0 = const()[name = tensor("op_23410_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23410_cast_fp16 = einsum(equation = var_23410_equation_0, values = (var_23266_cast_fp16, var_23371_cast_fp16))[name = tensor("op_23410_cast_fp16")]; + tensor var_23412_equation_0 = const()[name = tensor("op_23412_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23412_cast_fp16 = einsum(equation = var_23412_equation_0, values = (var_23270_cast_fp16, var_23372_cast_fp16))[name = tensor("op_23412_cast_fp16")]; + tensor input_343_interleave_0 = const()[name = tensor("input_343_interleave_0"), val = tensor(false)]; + tensor input_343_cast_fp16 = concat(axis = var_21077, interleave = input_343_interleave_0, values = (var_23374_cast_fp16, var_23376_cast_fp16, var_23378_cast_fp16, var_23380_cast_fp16, var_23382_cast_fp16, var_23384_cast_fp16, var_23386_cast_fp16, var_23388_cast_fp16, var_23390_cast_fp16, var_23392_cast_fp16, var_23394_cast_fp16, var_23396_cast_fp16, var_23398_cast_fp16, var_23400_cast_fp16, var_23402_cast_fp16, var_23404_cast_fp16, var_23406_cast_fp16, var_23408_cast_fp16, var_23410_cast_fp16, var_23412_cast_fp16))[name = tensor("input_343_cast_fp16")]; + tensor var_23422_pad_type_0 = const()[name = tensor("op_23422_pad_type_0"), val = tensor("valid")]; + tensor var_23422_strides_0 = const()[name = tensor("op_23422_strides_0"), val = tensor([1, 1])]; + tensor var_23422_pad_0 = const()[name = tensor("op_23422_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23422_dilations_0 = const()[name = tensor("op_23422_dilations_0"), val = tensor([1, 1])]; + tensor var_23422_groups_0 = const()[name = tensor("op_23422_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703942400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(705171264))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(705171456)))]; + tensor var_23422_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_23422_dilations_0, groups = var_23422_groups_0, pad = var_23422_pad_0, pad_type = var_23422_pad_type_0, strides = var_23422_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized, x = input_343_cast_fp16)[name = tensor("op_23422_cast_fp16")]; + tensor inputs_159_cast_fp16 = add(x = var_23422_cast_fp16, y = inputs_157_cast_fp16)[name = tensor("inputs_159_cast_fp16")]; + tensor hidden_states_223_axes_0 = const()[name = tensor("hidden_states_223_axes_0"), val = tensor([1])]; + tensor hidden_states_223_gamma_0_to_fp16 = const()[name = tensor("hidden_states_223_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(705174080)))]; + tensor hidden_states_223_beta_0_to_fp16 = const()[name = tensor("hidden_states_223_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(705176704)))]; + tensor var_23432_to_fp16 = const()[name = tensor("op_23432_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_223_cast_fp16 = layer_norm(axes = hidden_states_223_axes_0, beta = hidden_states_223_beta_0_to_fp16, epsilon = var_23432_to_fp16, gamma = hidden_states_223_gamma_0_to_fp16, x = inputs_159_cast_fp16)[name = tensor("hidden_states_223_cast_fp16")]; + tensor q_107_pad_type_0 = const()[name = tensor("q_107_pad_type_0"), val = tensor("valid")]; + tensor q_107_strides_0 = const()[name = tensor("q_107_strides_0"), val = tensor([1, 1])]; + tensor q_107_pad_0 = const()[name = tensor("q_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_107_dilations_0 = const()[name = tensor("q_107_dilations_0"), val = tensor([1, 1])]; + tensor q_107_groups_0 = const()[name = tensor("q_107_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(705179328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(706408192))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_107_cast_fp16 = conv(dilations = q_107_dilations_0, groups = q_107_groups_0, pad = q_107_pad_0, pad_type = q_107_pad_type_0, strides = q_107_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_223_cast_fp16)[name = tensor("q_107_cast_fp16")]; + tensor k_213_pad_type_0 = const()[name = tensor("k_213_pad_type_0"), val = tensor("valid")]; + tensor k_213_strides_0 = const()[name = tensor("k_213_strides_0"), val = tensor([1, 1])]; + tensor k_213_pad_0 = const()[name = tensor("k_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_213_dilations_0 = const()[name = tensor("k_213_dilations_0"), val = tensor([1, 1])]; + tensor k_213_groups_0 = const()[name = tensor("k_213_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(706408384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(708374528))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_213_cast_fp16 = conv(dilations = k_213_dilations_0, groups = k_213_groups_0, pad = k_213_pad_0, pad_type = k_213_pad_type_0, strides = k_213_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_213_cast_fp16")]; + tensor v_107_pad_type_0 = const()[name = tensor("v_107_pad_type_0"), val = tensor("valid")]; + tensor v_107_strides_0 = const()[name = tensor("v_107_strides_0"), val = tensor([1, 1])]; + tensor v_107_pad_0 = const()[name = tensor("v_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_107_dilations_0 = const()[name = tensor("v_107_dilations_0"), val = tensor([1, 1])]; + tensor v_107_groups_0 = const()[name = tensor("v_107_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(708374720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(710340864))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_107_cast_fp16 = conv(dilations = v_107_dilations_0, groups = v_107_groups_0, pad = v_107_pad_0, pad_type = v_107_pad_type_0, strides = v_107_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_107_cast_fp16")]; + tensor var_23465_begin_0 = const()[name = tensor("op_23465_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23465_end_0 = const()[name = tensor("op_23465_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23465_end_mask_0 = const()[name = tensor("op_23465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23465_cast_fp16 = slice_by_index(begin = var_23465_begin_0, end = var_23465_end_0, end_mask = var_23465_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23465_cast_fp16")]; + tensor var_23469_begin_0 = const()[name = tensor("op_23469_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23469_end_0 = const()[name = tensor("op_23469_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23469_end_mask_0 = const()[name = tensor("op_23469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23469_cast_fp16 = slice_by_index(begin = var_23469_begin_0, end = var_23469_end_0, end_mask = var_23469_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23469_cast_fp16")]; + tensor var_23473_begin_0 = const()[name = tensor("op_23473_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23473_end_0 = const()[name = tensor("op_23473_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23473_end_mask_0 = const()[name = tensor("op_23473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23473_cast_fp16 = slice_by_index(begin = var_23473_begin_0, end = var_23473_end_0, end_mask = var_23473_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23473_cast_fp16")]; + tensor var_23477_begin_0 = const()[name = tensor("op_23477_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23477_end_0 = const()[name = tensor("op_23477_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23477_end_mask_0 = const()[name = tensor("op_23477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23477_cast_fp16 = slice_by_index(begin = var_23477_begin_0, end = var_23477_end_0, end_mask = var_23477_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23477_cast_fp16")]; + tensor var_23481_begin_0 = const()[name = tensor("op_23481_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23481_end_0 = const()[name = tensor("op_23481_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23481_end_mask_0 = const()[name = tensor("op_23481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23481_cast_fp16 = slice_by_index(begin = var_23481_begin_0, end = var_23481_end_0, end_mask = var_23481_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23481_cast_fp16")]; + tensor var_23485_begin_0 = const()[name = tensor("op_23485_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23485_end_0 = const()[name = tensor("op_23485_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23485_end_mask_0 = const()[name = tensor("op_23485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23485_cast_fp16 = slice_by_index(begin = var_23485_begin_0, end = var_23485_end_0, end_mask = var_23485_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23485_cast_fp16")]; + tensor var_23489_begin_0 = const()[name = tensor("op_23489_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23489_end_0 = const()[name = tensor("op_23489_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23489_end_mask_0 = const()[name = tensor("op_23489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23489_cast_fp16 = slice_by_index(begin = var_23489_begin_0, end = var_23489_end_0, end_mask = var_23489_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23489_cast_fp16")]; + tensor var_23493_begin_0 = const()[name = tensor("op_23493_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23493_end_0 = const()[name = tensor("op_23493_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23493_end_mask_0 = const()[name = tensor("op_23493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23493_cast_fp16 = slice_by_index(begin = var_23493_begin_0, end = var_23493_end_0, end_mask = var_23493_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23493_cast_fp16")]; + tensor var_23497_begin_0 = const()[name = tensor("op_23497_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23497_end_0 = const()[name = tensor("op_23497_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23497_end_mask_0 = const()[name = tensor("op_23497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23497_cast_fp16 = slice_by_index(begin = var_23497_begin_0, end = var_23497_end_0, end_mask = var_23497_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23497_cast_fp16")]; + tensor var_23501_begin_0 = const()[name = tensor("op_23501_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23501_end_0 = const()[name = tensor("op_23501_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23501_end_mask_0 = const()[name = tensor("op_23501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23501_cast_fp16 = slice_by_index(begin = var_23501_begin_0, end = var_23501_end_0, end_mask = var_23501_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23501_cast_fp16")]; + tensor var_23505_begin_0 = const()[name = tensor("op_23505_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23505_end_0 = const()[name = tensor("op_23505_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23505_end_mask_0 = const()[name = tensor("op_23505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23505_cast_fp16 = slice_by_index(begin = var_23505_begin_0, end = var_23505_end_0, end_mask = var_23505_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23505_cast_fp16")]; + tensor var_23509_begin_0 = const()[name = tensor("op_23509_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23509_end_0 = const()[name = tensor("op_23509_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23509_end_mask_0 = const()[name = tensor("op_23509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23509_cast_fp16 = slice_by_index(begin = var_23509_begin_0, end = var_23509_end_0, end_mask = var_23509_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23509_cast_fp16")]; + tensor var_23513_begin_0 = const()[name = tensor("op_23513_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23513_end_0 = const()[name = tensor("op_23513_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23513_end_mask_0 = const()[name = tensor("op_23513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23513_cast_fp16 = slice_by_index(begin = var_23513_begin_0, end = var_23513_end_0, end_mask = var_23513_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23513_cast_fp16")]; + tensor var_23517_begin_0 = const()[name = tensor("op_23517_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23517_end_0 = const()[name = tensor("op_23517_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23517_end_mask_0 = const()[name = tensor("op_23517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23517_cast_fp16 = slice_by_index(begin = var_23517_begin_0, end = var_23517_end_0, end_mask = var_23517_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23517_cast_fp16")]; + tensor var_23521_begin_0 = const()[name = tensor("op_23521_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23521_end_0 = const()[name = tensor("op_23521_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_23521_end_mask_0 = const()[name = tensor("op_23521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23521_cast_fp16 = slice_by_index(begin = var_23521_begin_0, end = var_23521_end_0, end_mask = var_23521_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23521_cast_fp16")]; + tensor var_23525_begin_0 = const()[name = tensor("op_23525_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23525_end_0 = const()[name = tensor("op_23525_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23525_end_mask_0 = const()[name = tensor("op_23525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23525_cast_fp16 = slice_by_index(begin = var_23525_begin_0, end = var_23525_end_0, end_mask = var_23525_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23525_cast_fp16")]; + tensor var_23529_begin_0 = const()[name = tensor("op_23529_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23529_end_0 = const()[name = tensor("op_23529_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_23529_end_mask_0 = const()[name = tensor("op_23529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23529_cast_fp16 = slice_by_index(begin = var_23529_begin_0, end = var_23529_end_0, end_mask = var_23529_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23529_cast_fp16")]; + tensor var_23533_begin_0 = const()[name = tensor("op_23533_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23533_end_0 = const()[name = tensor("op_23533_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_23533_end_mask_0 = const()[name = tensor("op_23533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23533_cast_fp16 = slice_by_index(begin = var_23533_begin_0, end = var_23533_end_0, end_mask = var_23533_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23533_cast_fp16")]; + tensor var_23537_begin_0 = const()[name = tensor("op_23537_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23537_end_0 = const()[name = tensor("op_23537_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_23537_end_mask_0 = const()[name = tensor("op_23537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23537_cast_fp16 = slice_by_index(begin = var_23537_begin_0, end = var_23537_end_0, end_mask = var_23537_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23537_cast_fp16")]; + tensor var_23541_begin_0 = const()[name = tensor("op_23541_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23541_end_0 = const()[name = tensor("op_23541_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_23541_end_mask_0 = const()[name = tensor("op_23541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23541_cast_fp16 = slice_by_index(begin = var_23541_begin_0, end = var_23541_end_0, end_mask = var_23541_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23541_cast_fp16")]; + tensor k_215_perm_0 = const()[name = tensor("k_215_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_23548_begin_0 = const()[name = tensor("op_23548_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23548_end_0 = const()[name = tensor("op_23548_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_23548_end_mask_0 = const()[name = tensor("op_23548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_215_cast_fp16 = transpose(perm = k_215_perm_0, x = k_213_cast_fp16)[name = tensor("transpose_14")]; + tensor var_23548_cast_fp16 = slice_by_index(begin = var_23548_begin_0, end = var_23548_end_0, end_mask = var_23548_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23548_cast_fp16")]; + tensor var_23552_begin_0 = const()[name = tensor("op_23552_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_23552_end_0 = const()[name = tensor("op_23552_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_23552_end_mask_0 = const()[name = tensor("op_23552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23552_cast_fp16 = slice_by_index(begin = var_23552_begin_0, end = var_23552_end_0, end_mask = var_23552_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23552_cast_fp16")]; + tensor var_23556_begin_0 = const()[name = tensor("op_23556_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_23556_end_0 = const()[name = tensor("op_23556_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_23556_end_mask_0 = const()[name = tensor("op_23556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23556_cast_fp16 = slice_by_index(begin = var_23556_begin_0, end = var_23556_end_0, end_mask = var_23556_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23556_cast_fp16")]; + tensor var_23560_begin_0 = const()[name = tensor("op_23560_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_23560_end_0 = const()[name = tensor("op_23560_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_23560_end_mask_0 = const()[name = tensor("op_23560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23560_cast_fp16 = slice_by_index(begin = var_23560_begin_0, end = var_23560_end_0, end_mask = var_23560_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23560_cast_fp16")]; + tensor var_23564_begin_0 = const()[name = tensor("op_23564_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_23564_end_0 = const()[name = tensor("op_23564_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_23564_end_mask_0 = const()[name = tensor("op_23564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23564_cast_fp16 = slice_by_index(begin = var_23564_begin_0, end = var_23564_end_0, end_mask = var_23564_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23564_cast_fp16")]; + tensor var_23568_begin_0 = const()[name = tensor("op_23568_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_23568_end_0 = const()[name = tensor("op_23568_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_23568_end_mask_0 = const()[name = tensor("op_23568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23568_cast_fp16 = slice_by_index(begin = var_23568_begin_0, end = var_23568_end_0, end_mask = var_23568_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23568_cast_fp16")]; + tensor var_23572_begin_0 = const()[name = tensor("op_23572_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_23572_end_0 = const()[name = tensor("op_23572_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_23572_end_mask_0 = const()[name = tensor("op_23572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23572_cast_fp16 = slice_by_index(begin = var_23572_begin_0, end = var_23572_end_0, end_mask = var_23572_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23572_cast_fp16")]; + tensor var_23576_begin_0 = const()[name = tensor("op_23576_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_23576_end_0 = const()[name = tensor("op_23576_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_23576_end_mask_0 = const()[name = tensor("op_23576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23576_cast_fp16 = slice_by_index(begin = var_23576_begin_0, end = var_23576_end_0, end_mask = var_23576_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23576_cast_fp16")]; + tensor var_23580_begin_0 = const()[name = tensor("op_23580_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_23580_end_0 = const()[name = tensor("op_23580_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_23580_end_mask_0 = const()[name = tensor("op_23580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23580_cast_fp16 = slice_by_index(begin = var_23580_begin_0, end = var_23580_end_0, end_mask = var_23580_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23580_cast_fp16")]; + tensor var_23584_begin_0 = const()[name = tensor("op_23584_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_23584_end_0 = const()[name = tensor("op_23584_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_23584_end_mask_0 = const()[name = tensor("op_23584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23584_cast_fp16 = slice_by_index(begin = var_23584_begin_0, end = var_23584_end_0, end_mask = var_23584_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23584_cast_fp16")]; + tensor var_23588_begin_0 = const()[name = tensor("op_23588_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_23588_end_0 = const()[name = tensor("op_23588_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_23588_end_mask_0 = const()[name = tensor("op_23588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23588_cast_fp16 = slice_by_index(begin = var_23588_begin_0, end = var_23588_end_0, end_mask = var_23588_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23588_cast_fp16")]; + tensor var_23592_begin_0 = const()[name = tensor("op_23592_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_23592_end_0 = const()[name = tensor("op_23592_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_23592_end_mask_0 = const()[name = tensor("op_23592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23592_cast_fp16 = slice_by_index(begin = var_23592_begin_0, end = var_23592_end_0, end_mask = var_23592_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23592_cast_fp16")]; + tensor var_23596_begin_0 = const()[name = tensor("op_23596_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_23596_end_0 = const()[name = tensor("op_23596_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_23596_end_mask_0 = const()[name = tensor("op_23596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23596_cast_fp16 = slice_by_index(begin = var_23596_begin_0, end = var_23596_end_0, end_mask = var_23596_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23596_cast_fp16")]; + tensor var_23600_begin_0 = const()[name = tensor("op_23600_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_23600_end_0 = const()[name = tensor("op_23600_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_23600_end_mask_0 = const()[name = tensor("op_23600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23600_cast_fp16 = slice_by_index(begin = var_23600_begin_0, end = var_23600_end_0, end_mask = var_23600_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23600_cast_fp16")]; + tensor var_23604_begin_0 = const()[name = tensor("op_23604_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_23604_end_0 = const()[name = tensor("op_23604_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_23604_end_mask_0 = const()[name = tensor("op_23604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23604_cast_fp16 = slice_by_index(begin = var_23604_begin_0, end = var_23604_end_0, end_mask = var_23604_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23604_cast_fp16")]; + tensor var_23608_begin_0 = const()[name = tensor("op_23608_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_23608_end_0 = const()[name = tensor("op_23608_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_23608_end_mask_0 = const()[name = tensor("op_23608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23608_cast_fp16 = slice_by_index(begin = var_23608_begin_0, end = var_23608_end_0, end_mask = var_23608_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23608_cast_fp16")]; + tensor var_23612_begin_0 = const()[name = tensor("op_23612_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_23612_end_0 = const()[name = tensor("op_23612_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_23612_end_mask_0 = const()[name = tensor("op_23612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23612_cast_fp16 = slice_by_index(begin = var_23612_begin_0, end = var_23612_end_0, end_mask = var_23612_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23612_cast_fp16")]; + tensor var_23616_begin_0 = const()[name = tensor("op_23616_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_23616_end_0 = const()[name = tensor("op_23616_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_23616_end_mask_0 = const()[name = tensor("op_23616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23616_cast_fp16 = slice_by_index(begin = var_23616_begin_0, end = var_23616_end_0, end_mask = var_23616_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23616_cast_fp16")]; + tensor var_23620_begin_0 = const()[name = tensor("op_23620_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_23620_end_0 = const()[name = tensor("op_23620_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_23620_end_mask_0 = const()[name = tensor("op_23620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23620_cast_fp16 = slice_by_index(begin = var_23620_begin_0, end = var_23620_end_0, end_mask = var_23620_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23620_cast_fp16")]; + tensor var_23624_begin_0 = const()[name = tensor("op_23624_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_23624_end_0 = const()[name = tensor("op_23624_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_23624_end_mask_0 = const()[name = tensor("op_23624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23624_cast_fp16 = slice_by_index(begin = var_23624_begin_0, end = var_23624_end_0, end_mask = var_23624_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23624_cast_fp16")]; + tensor var_23626_begin_0 = const()[name = tensor("op_23626_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23626_end_0 = const()[name = tensor("op_23626_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_23626_end_mask_0 = const()[name = tensor("op_23626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23626_cast_fp16 = slice_by_index(begin = var_23626_begin_0, end = var_23626_end_0, end_mask = var_23626_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23626_cast_fp16")]; + tensor var_23630_begin_0 = const()[name = tensor("op_23630_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23630_end_0 = const()[name = tensor("op_23630_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_23630_end_mask_0 = const()[name = tensor("op_23630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23630_cast_fp16 = slice_by_index(begin = var_23630_begin_0, end = var_23630_end_0, end_mask = var_23630_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23630_cast_fp16")]; + tensor var_23634_begin_0 = const()[name = tensor("op_23634_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23634_end_0 = const()[name = tensor("op_23634_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_23634_end_mask_0 = const()[name = tensor("op_23634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23634_cast_fp16 = slice_by_index(begin = var_23634_begin_0, end = var_23634_end_0, end_mask = var_23634_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23634_cast_fp16")]; + tensor var_23638_begin_0 = const()[name = tensor("op_23638_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23638_end_0 = const()[name = tensor("op_23638_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_23638_end_mask_0 = const()[name = tensor("op_23638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23638_cast_fp16 = slice_by_index(begin = var_23638_begin_0, end = var_23638_end_0, end_mask = var_23638_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23638_cast_fp16")]; + tensor var_23642_begin_0 = const()[name = tensor("op_23642_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23642_end_0 = const()[name = tensor("op_23642_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_23642_end_mask_0 = const()[name = tensor("op_23642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23642_cast_fp16 = slice_by_index(begin = var_23642_begin_0, end = var_23642_end_0, end_mask = var_23642_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23642_cast_fp16")]; + tensor var_23646_begin_0 = const()[name = tensor("op_23646_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23646_end_0 = const()[name = tensor("op_23646_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_23646_end_mask_0 = const()[name = tensor("op_23646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23646_cast_fp16 = slice_by_index(begin = var_23646_begin_0, end = var_23646_end_0, end_mask = var_23646_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23646_cast_fp16")]; + tensor var_23650_begin_0 = const()[name = tensor("op_23650_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23650_end_0 = const()[name = tensor("op_23650_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_23650_end_mask_0 = const()[name = tensor("op_23650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23650_cast_fp16 = slice_by_index(begin = var_23650_begin_0, end = var_23650_end_0, end_mask = var_23650_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23650_cast_fp16")]; + tensor var_23654_begin_0 = const()[name = tensor("op_23654_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23654_end_0 = const()[name = tensor("op_23654_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_23654_end_mask_0 = const()[name = tensor("op_23654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23654_cast_fp16 = slice_by_index(begin = var_23654_begin_0, end = var_23654_end_0, end_mask = var_23654_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23654_cast_fp16")]; + tensor var_23658_begin_0 = const()[name = tensor("op_23658_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23658_end_0 = const()[name = tensor("op_23658_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_23658_end_mask_0 = const()[name = tensor("op_23658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23658_cast_fp16 = slice_by_index(begin = var_23658_begin_0, end = var_23658_end_0, end_mask = var_23658_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23658_cast_fp16")]; + tensor var_23662_begin_0 = const()[name = tensor("op_23662_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23662_end_0 = const()[name = tensor("op_23662_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_23662_end_mask_0 = const()[name = tensor("op_23662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23662_cast_fp16 = slice_by_index(begin = var_23662_begin_0, end = var_23662_end_0, end_mask = var_23662_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23662_cast_fp16")]; + tensor var_23666_begin_0 = const()[name = tensor("op_23666_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23666_end_0 = const()[name = tensor("op_23666_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_23666_end_mask_0 = const()[name = tensor("op_23666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23666_cast_fp16 = slice_by_index(begin = var_23666_begin_0, end = var_23666_end_0, end_mask = var_23666_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23666_cast_fp16")]; + tensor var_23670_begin_0 = const()[name = tensor("op_23670_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23670_end_0 = const()[name = tensor("op_23670_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_23670_end_mask_0 = const()[name = tensor("op_23670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23670_cast_fp16 = slice_by_index(begin = var_23670_begin_0, end = var_23670_end_0, end_mask = var_23670_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23670_cast_fp16")]; + tensor var_23674_begin_0 = const()[name = tensor("op_23674_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23674_end_0 = const()[name = tensor("op_23674_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_23674_end_mask_0 = const()[name = tensor("op_23674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23674_cast_fp16 = slice_by_index(begin = var_23674_begin_0, end = var_23674_end_0, end_mask = var_23674_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23674_cast_fp16")]; + tensor var_23678_begin_0 = const()[name = tensor("op_23678_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23678_end_0 = const()[name = tensor("op_23678_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_23678_end_mask_0 = const()[name = tensor("op_23678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23678_cast_fp16 = slice_by_index(begin = var_23678_begin_0, end = var_23678_end_0, end_mask = var_23678_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23678_cast_fp16")]; + tensor var_23682_begin_0 = const()[name = tensor("op_23682_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23682_end_0 = const()[name = tensor("op_23682_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_23682_end_mask_0 = const()[name = tensor("op_23682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23682_cast_fp16 = slice_by_index(begin = var_23682_begin_0, end = var_23682_end_0, end_mask = var_23682_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23682_cast_fp16")]; + tensor var_23686_begin_0 = const()[name = tensor("op_23686_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23686_end_0 = const()[name = tensor("op_23686_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_23686_end_mask_0 = const()[name = tensor("op_23686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23686_cast_fp16 = slice_by_index(begin = var_23686_begin_0, end = var_23686_end_0, end_mask = var_23686_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23686_cast_fp16")]; + tensor var_23690_begin_0 = const()[name = tensor("op_23690_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23690_end_0 = const()[name = tensor("op_23690_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_23690_end_mask_0 = const()[name = tensor("op_23690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23690_cast_fp16 = slice_by_index(begin = var_23690_begin_0, end = var_23690_end_0, end_mask = var_23690_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23690_cast_fp16")]; + tensor var_23694_begin_0 = const()[name = tensor("op_23694_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23694_end_0 = const()[name = tensor("op_23694_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_23694_end_mask_0 = const()[name = tensor("op_23694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23694_cast_fp16 = slice_by_index(begin = var_23694_begin_0, end = var_23694_end_0, end_mask = var_23694_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23694_cast_fp16")]; + tensor var_23698_begin_0 = const()[name = tensor("op_23698_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23698_end_0 = const()[name = tensor("op_23698_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_23698_end_mask_0 = const()[name = tensor("op_23698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23698_cast_fp16 = slice_by_index(begin = var_23698_begin_0, end = var_23698_end_0, end_mask = var_23698_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23698_cast_fp16")]; + tensor var_23702_begin_0 = const()[name = tensor("op_23702_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23702_end_0 = const()[name = tensor("op_23702_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_23702_end_mask_0 = const()[name = tensor("op_23702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23702_cast_fp16 = slice_by_index(begin = var_23702_begin_0, end = var_23702_end_0, end_mask = var_23702_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23702_cast_fp16")]; + tensor var_23706_equation_0 = const()[name = tensor("op_23706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23706_cast_fp16 = einsum(equation = var_23706_equation_0, values = (var_23548_cast_fp16, var_23465_cast_fp16))[name = tensor("op_23706_cast_fp16")]; + tensor var_23707_to_fp16 = const()[name = tensor("op_23707_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1961_cast_fp16 = mul(x = var_23706_cast_fp16, y = var_23707_to_fp16)[name = tensor("aw_1961_cast_fp16")]; + tensor var_23710_equation_0 = const()[name = tensor("op_23710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23710_cast_fp16 = einsum(equation = var_23710_equation_0, values = (var_23552_cast_fp16, var_23469_cast_fp16))[name = tensor("op_23710_cast_fp16")]; + tensor var_23711_to_fp16 = const()[name = tensor("op_23711_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1963_cast_fp16 = mul(x = var_23710_cast_fp16, y = var_23711_to_fp16)[name = tensor("aw_1963_cast_fp16")]; + tensor var_23714_equation_0 = const()[name = tensor("op_23714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23714_cast_fp16 = einsum(equation = var_23714_equation_0, values = (var_23556_cast_fp16, var_23473_cast_fp16))[name = tensor("op_23714_cast_fp16")]; + tensor var_23715_to_fp16 = const()[name = tensor("op_23715_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1965_cast_fp16 = mul(x = var_23714_cast_fp16, y = var_23715_to_fp16)[name = tensor("aw_1965_cast_fp16")]; + tensor var_23718_equation_0 = const()[name = tensor("op_23718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23718_cast_fp16 = einsum(equation = var_23718_equation_0, values = (var_23560_cast_fp16, var_23477_cast_fp16))[name = tensor("op_23718_cast_fp16")]; + tensor var_23719_to_fp16 = const()[name = tensor("op_23719_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1967_cast_fp16 = mul(x = var_23718_cast_fp16, y = var_23719_to_fp16)[name = tensor("aw_1967_cast_fp16")]; + tensor var_23722_equation_0 = const()[name = tensor("op_23722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23722_cast_fp16 = einsum(equation = var_23722_equation_0, values = (var_23564_cast_fp16, var_23481_cast_fp16))[name = tensor("op_23722_cast_fp16")]; + tensor var_23723_to_fp16 = const()[name = tensor("op_23723_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1969_cast_fp16 = mul(x = var_23722_cast_fp16, y = var_23723_to_fp16)[name = tensor("aw_1969_cast_fp16")]; + tensor var_23726_equation_0 = const()[name = tensor("op_23726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23726_cast_fp16 = einsum(equation = var_23726_equation_0, values = (var_23568_cast_fp16, var_23485_cast_fp16))[name = tensor("op_23726_cast_fp16")]; + tensor var_23727_to_fp16 = const()[name = tensor("op_23727_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1971_cast_fp16 = mul(x = var_23726_cast_fp16, y = var_23727_to_fp16)[name = tensor("aw_1971_cast_fp16")]; + tensor var_23730_equation_0 = const()[name = tensor("op_23730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23730_cast_fp16 = einsum(equation = var_23730_equation_0, values = (var_23572_cast_fp16, var_23489_cast_fp16))[name = tensor("op_23730_cast_fp16")]; + tensor var_23731_to_fp16 = const()[name = tensor("op_23731_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1973_cast_fp16 = mul(x = var_23730_cast_fp16, y = var_23731_to_fp16)[name = tensor("aw_1973_cast_fp16")]; + tensor var_23734_equation_0 = const()[name = tensor("op_23734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23734_cast_fp16 = einsum(equation = var_23734_equation_0, values = (var_23576_cast_fp16, var_23493_cast_fp16))[name = tensor("op_23734_cast_fp16")]; + tensor var_23735_to_fp16 = const()[name = tensor("op_23735_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1975_cast_fp16 = mul(x = var_23734_cast_fp16, y = var_23735_to_fp16)[name = tensor("aw_1975_cast_fp16")]; + tensor var_23738_equation_0 = const()[name = tensor("op_23738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23738_cast_fp16 = einsum(equation = var_23738_equation_0, values = (var_23580_cast_fp16, var_23497_cast_fp16))[name = tensor("op_23738_cast_fp16")]; + tensor var_23739_to_fp16 = const()[name = tensor("op_23739_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1977_cast_fp16 = mul(x = var_23738_cast_fp16, y = var_23739_to_fp16)[name = tensor("aw_1977_cast_fp16")]; + tensor var_23742_equation_0 = const()[name = tensor("op_23742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23742_cast_fp16 = einsum(equation = var_23742_equation_0, values = (var_23584_cast_fp16, var_23501_cast_fp16))[name = tensor("op_23742_cast_fp16")]; + tensor var_23743_to_fp16 = const()[name = tensor("op_23743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1979_cast_fp16 = mul(x = var_23742_cast_fp16, y = var_23743_to_fp16)[name = tensor("aw_1979_cast_fp16")]; + tensor var_23746_equation_0 = const()[name = tensor("op_23746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23746_cast_fp16 = einsum(equation = var_23746_equation_0, values = (var_23588_cast_fp16, var_23505_cast_fp16))[name = tensor("op_23746_cast_fp16")]; + tensor var_23747_to_fp16 = const()[name = tensor("op_23747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1981_cast_fp16 = mul(x = var_23746_cast_fp16, y = var_23747_to_fp16)[name = tensor("aw_1981_cast_fp16")]; + tensor var_23750_equation_0 = const()[name = tensor("op_23750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23750_cast_fp16 = einsum(equation = var_23750_equation_0, values = (var_23592_cast_fp16, var_23509_cast_fp16))[name = tensor("op_23750_cast_fp16")]; + tensor var_23751_to_fp16 = const()[name = tensor("op_23751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1983_cast_fp16 = mul(x = var_23750_cast_fp16, y = var_23751_to_fp16)[name = tensor("aw_1983_cast_fp16")]; + tensor var_23754_equation_0 = const()[name = tensor("op_23754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23754_cast_fp16 = einsum(equation = var_23754_equation_0, values = (var_23596_cast_fp16, var_23513_cast_fp16))[name = tensor("op_23754_cast_fp16")]; + tensor var_23755_to_fp16 = const()[name = tensor("op_23755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1985_cast_fp16 = mul(x = var_23754_cast_fp16, y = var_23755_to_fp16)[name = tensor("aw_1985_cast_fp16")]; + tensor var_23758_equation_0 = const()[name = tensor("op_23758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23758_cast_fp16 = einsum(equation = var_23758_equation_0, values = (var_23600_cast_fp16, var_23517_cast_fp16))[name = tensor("op_23758_cast_fp16")]; + tensor var_23759_to_fp16 = const()[name = tensor("op_23759_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1987_cast_fp16 = mul(x = var_23758_cast_fp16, y = var_23759_to_fp16)[name = tensor("aw_1987_cast_fp16")]; + tensor var_23762_equation_0 = const()[name = tensor("op_23762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23762_cast_fp16 = einsum(equation = var_23762_equation_0, values = (var_23604_cast_fp16, var_23521_cast_fp16))[name = tensor("op_23762_cast_fp16")]; + tensor var_23763_to_fp16 = const()[name = tensor("op_23763_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1989_cast_fp16 = mul(x = var_23762_cast_fp16, y = var_23763_to_fp16)[name = tensor("aw_1989_cast_fp16")]; + tensor var_23766_equation_0 = const()[name = tensor("op_23766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23766_cast_fp16 = einsum(equation = var_23766_equation_0, values = (var_23608_cast_fp16, var_23525_cast_fp16))[name = tensor("op_23766_cast_fp16")]; + tensor var_23767_to_fp16 = const()[name = tensor("op_23767_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1991_cast_fp16 = mul(x = var_23766_cast_fp16, y = var_23767_to_fp16)[name = tensor("aw_1991_cast_fp16")]; + tensor var_23770_equation_0 = const()[name = tensor("op_23770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23770_cast_fp16 = einsum(equation = var_23770_equation_0, values = (var_23612_cast_fp16, var_23529_cast_fp16))[name = tensor("op_23770_cast_fp16")]; + tensor var_23771_to_fp16 = const()[name = tensor("op_23771_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1993_cast_fp16 = mul(x = var_23770_cast_fp16, y = var_23771_to_fp16)[name = tensor("aw_1993_cast_fp16")]; + tensor var_23774_equation_0 = const()[name = tensor("op_23774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23774_cast_fp16 = einsum(equation = var_23774_equation_0, values = (var_23616_cast_fp16, var_23533_cast_fp16))[name = tensor("op_23774_cast_fp16")]; + tensor var_23775_to_fp16 = const()[name = tensor("op_23775_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1995_cast_fp16 = mul(x = var_23774_cast_fp16, y = var_23775_to_fp16)[name = tensor("aw_1995_cast_fp16")]; + tensor var_23778_equation_0 = const()[name = tensor("op_23778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23778_cast_fp16 = einsum(equation = var_23778_equation_0, values = (var_23620_cast_fp16, var_23537_cast_fp16))[name = tensor("op_23778_cast_fp16")]; + tensor var_23779_to_fp16 = const()[name = tensor("op_23779_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1997_cast_fp16 = mul(x = var_23778_cast_fp16, y = var_23779_to_fp16)[name = tensor("aw_1997_cast_fp16")]; + tensor var_23782_equation_0 = const()[name = tensor("op_23782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23782_cast_fp16 = einsum(equation = var_23782_equation_0, values = (var_23624_cast_fp16, var_23541_cast_fp16))[name = tensor("op_23782_cast_fp16")]; + tensor var_23783_to_fp16 = const()[name = tensor("op_23783_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1999_cast_fp16 = mul(x = var_23782_cast_fp16, y = var_23783_to_fp16)[name = tensor("aw_1999_cast_fp16")]; + tensor var_23785_cast_fp16 = softmax(axis = var_21077, x = aw_1961_cast_fp16)[name = tensor("op_23785_cast_fp16")]; + tensor var_23786_cast_fp16 = softmax(axis = var_21077, x = aw_1963_cast_fp16)[name = tensor("op_23786_cast_fp16")]; + tensor var_23787_cast_fp16 = softmax(axis = var_21077, x = aw_1965_cast_fp16)[name = tensor("op_23787_cast_fp16")]; + tensor var_23788_cast_fp16 = softmax(axis = var_21077, x = aw_1967_cast_fp16)[name = tensor("op_23788_cast_fp16")]; + tensor var_23789_cast_fp16 = softmax(axis = var_21077, x = aw_1969_cast_fp16)[name = tensor("op_23789_cast_fp16")]; + tensor var_23790_cast_fp16 = softmax(axis = var_21077, x = aw_1971_cast_fp16)[name = tensor("op_23790_cast_fp16")]; + tensor var_23791_cast_fp16 = softmax(axis = var_21077, x = aw_1973_cast_fp16)[name = tensor("op_23791_cast_fp16")]; + tensor var_23792_cast_fp16 = softmax(axis = var_21077, x = aw_1975_cast_fp16)[name = tensor("op_23792_cast_fp16")]; + tensor var_23793_cast_fp16 = softmax(axis = var_21077, x = aw_1977_cast_fp16)[name = tensor("op_23793_cast_fp16")]; + tensor var_23794_cast_fp16 = softmax(axis = var_21077, x = aw_1979_cast_fp16)[name = tensor("op_23794_cast_fp16")]; + tensor var_23795_cast_fp16 = softmax(axis = var_21077, x = aw_1981_cast_fp16)[name = tensor("op_23795_cast_fp16")]; + tensor var_23796_cast_fp16 = softmax(axis = var_21077, x = aw_1983_cast_fp16)[name = tensor("op_23796_cast_fp16")]; + tensor var_23797_cast_fp16 = softmax(axis = var_21077, x = aw_1985_cast_fp16)[name = tensor("op_23797_cast_fp16")]; + tensor var_23798_cast_fp16 = softmax(axis = var_21077, x = aw_1987_cast_fp16)[name = tensor("op_23798_cast_fp16")]; + tensor var_23799_cast_fp16 = softmax(axis = var_21077, x = aw_1989_cast_fp16)[name = tensor("op_23799_cast_fp16")]; + tensor var_23800_cast_fp16 = softmax(axis = var_21077, x = aw_1991_cast_fp16)[name = tensor("op_23800_cast_fp16")]; + tensor var_23801_cast_fp16 = softmax(axis = var_21077, x = aw_1993_cast_fp16)[name = tensor("op_23801_cast_fp16")]; + tensor var_23802_cast_fp16 = softmax(axis = var_21077, x = aw_1995_cast_fp16)[name = tensor("op_23802_cast_fp16")]; + tensor var_23803_cast_fp16 = softmax(axis = var_21077, x = aw_1997_cast_fp16)[name = tensor("op_23803_cast_fp16")]; + tensor var_23804_cast_fp16 = softmax(axis = var_21077, x = aw_1999_cast_fp16)[name = tensor("op_23804_cast_fp16")]; + tensor var_23806_equation_0 = const()[name = tensor("op_23806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23806_cast_fp16 = einsum(equation = var_23806_equation_0, values = (var_23626_cast_fp16, var_23785_cast_fp16))[name = tensor("op_23806_cast_fp16")]; + tensor var_23808_equation_0 = const()[name = tensor("op_23808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23808_cast_fp16 = einsum(equation = var_23808_equation_0, values = (var_23630_cast_fp16, var_23786_cast_fp16))[name = tensor("op_23808_cast_fp16")]; + tensor var_23810_equation_0 = const()[name = tensor("op_23810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23810_cast_fp16 = einsum(equation = var_23810_equation_0, values = (var_23634_cast_fp16, var_23787_cast_fp16))[name = tensor("op_23810_cast_fp16")]; + tensor var_23812_equation_0 = const()[name = tensor("op_23812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23812_cast_fp16 = einsum(equation = var_23812_equation_0, values = (var_23638_cast_fp16, var_23788_cast_fp16))[name = tensor("op_23812_cast_fp16")]; + tensor var_23814_equation_0 = const()[name = tensor("op_23814_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23814_cast_fp16 = einsum(equation = var_23814_equation_0, values = (var_23642_cast_fp16, var_23789_cast_fp16))[name = tensor("op_23814_cast_fp16")]; + tensor var_23816_equation_0 = const()[name = tensor("op_23816_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23816_cast_fp16 = einsum(equation = var_23816_equation_0, values = (var_23646_cast_fp16, var_23790_cast_fp16))[name = tensor("op_23816_cast_fp16")]; + tensor var_23818_equation_0 = const()[name = tensor("op_23818_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23818_cast_fp16 = einsum(equation = var_23818_equation_0, values = (var_23650_cast_fp16, var_23791_cast_fp16))[name = tensor("op_23818_cast_fp16")]; + tensor var_23820_equation_0 = const()[name = tensor("op_23820_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23820_cast_fp16 = einsum(equation = var_23820_equation_0, values = (var_23654_cast_fp16, var_23792_cast_fp16))[name = tensor("op_23820_cast_fp16")]; + tensor var_23822_equation_0 = const()[name = tensor("op_23822_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23822_cast_fp16 = einsum(equation = var_23822_equation_0, values = (var_23658_cast_fp16, var_23793_cast_fp16))[name = tensor("op_23822_cast_fp16")]; + tensor var_23824_equation_0 = const()[name = tensor("op_23824_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23824_cast_fp16 = einsum(equation = var_23824_equation_0, values = (var_23662_cast_fp16, var_23794_cast_fp16))[name = tensor("op_23824_cast_fp16")]; + tensor var_23826_equation_0 = const()[name = tensor("op_23826_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23826_cast_fp16 = einsum(equation = var_23826_equation_0, values = (var_23666_cast_fp16, var_23795_cast_fp16))[name = tensor("op_23826_cast_fp16")]; + tensor var_23828_equation_0 = const()[name = tensor("op_23828_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23828_cast_fp16 = einsum(equation = var_23828_equation_0, values = (var_23670_cast_fp16, var_23796_cast_fp16))[name = tensor("op_23828_cast_fp16")]; + tensor var_23830_equation_0 = const()[name = tensor("op_23830_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23830_cast_fp16 = einsum(equation = var_23830_equation_0, values = (var_23674_cast_fp16, var_23797_cast_fp16))[name = tensor("op_23830_cast_fp16")]; + tensor var_23832_equation_0 = const()[name = tensor("op_23832_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23832_cast_fp16 = einsum(equation = var_23832_equation_0, values = (var_23678_cast_fp16, var_23798_cast_fp16))[name = tensor("op_23832_cast_fp16")]; + tensor var_23834_equation_0 = const()[name = tensor("op_23834_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23834_cast_fp16 = einsum(equation = var_23834_equation_0, values = (var_23682_cast_fp16, var_23799_cast_fp16))[name = tensor("op_23834_cast_fp16")]; + tensor var_23836_equation_0 = const()[name = tensor("op_23836_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23836_cast_fp16 = einsum(equation = var_23836_equation_0, values = (var_23686_cast_fp16, var_23800_cast_fp16))[name = tensor("op_23836_cast_fp16")]; + tensor var_23838_equation_0 = const()[name = tensor("op_23838_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23838_cast_fp16 = einsum(equation = var_23838_equation_0, values = (var_23690_cast_fp16, var_23801_cast_fp16))[name = tensor("op_23838_cast_fp16")]; + tensor var_23840_equation_0 = const()[name = tensor("op_23840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23840_cast_fp16 = einsum(equation = var_23840_equation_0, values = (var_23694_cast_fp16, var_23802_cast_fp16))[name = tensor("op_23840_cast_fp16")]; + tensor var_23842_equation_0 = const()[name = tensor("op_23842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23842_cast_fp16 = einsum(equation = var_23842_equation_0, values = (var_23698_cast_fp16, var_23803_cast_fp16))[name = tensor("op_23842_cast_fp16")]; + tensor var_23844_equation_0 = const()[name = tensor("op_23844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23844_cast_fp16 = einsum(equation = var_23844_equation_0, values = (var_23702_cast_fp16, var_23804_cast_fp16))[name = tensor("op_23844_cast_fp16")]; + tensor input_345_interleave_0 = const()[name = tensor("input_345_interleave_0"), val = tensor(false)]; + tensor input_345_cast_fp16 = concat(axis = var_21077, interleave = input_345_interleave_0, values = (var_23806_cast_fp16, var_23808_cast_fp16, var_23810_cast_fp16, var_23812_cast_fp16, var_23814_cast_fp16, var_23816_cast_fp16, var_23818_cast_fp16, var_23820_cast_fp16, var_23822_cast_fp16, var_23824_cast_fp16, var_23826_cast_fp16, var_23828_cast_fp16, var_23830_cast_fp16, var_23832_cast_fp16, var_23834_cast_fp16, var_23836_cast_fp16, var_23838_cast_fp16, var_23840_cast_fp16, var_23842_cast_fp16, var_23844_cast_fp16))[name = tensor("input_345_cast_fp16")]; + tensor var_23854_pad_type_0 = const()[name = tensor("op_23854_pad_type_0"), val = tensor("valid")]; + tensor var_23854_strides_0 = const()[name = tensor("op_23854_strides_0"), val = tensor([1, 1])]; + tensor var_23854_pad_0 = const()[name = tensor("op_23854_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23854_dilations_0 = const()[name = tensor("op_23854_dilations_0"), val = tensor([1, 1])]; + tensor var_23854_groups_0 = const()[name = tensor("op_23854_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(710341056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711569920))), name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711570112)))]; + tensor var_23854_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_23854_dilations_0, groups = var_23854_groups_0, pad = var_23854_pad_0, pad_type = var_23854_pad_type_0, strides = var_23854_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized, x = input_345_cast_fp16)[name = tensor("op_23854_cast_fp16")]; + tensor inputs_161_cast_fp16 = add(x = var_23854_cast_fp16, y = inputs_159_cast_fp16)[name = tensor("inputs_161_cast_fp16")]; + tensor input_347_axes_0 = const()[name = tensor("input_347_axes_0"), val = tensor([1])]; + tensor input_347_gamma_0_to_fp16 = const()[name = tensor("input_347_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711572736)))]; + tensor input_347_beta_0_to_fp16 = const()[name = tensor("input_347_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711575360)))]; + tensor var_23864_to_fp16 = const()[name = tensor("op_23864_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_347_cast_fp16 = layer_norm(axes = input_347_axes_0, beta = input_347_beta_0_to_fp16, epsilon = var_23864_to_fp16, gamma = input_347_gamma_0_to_fp16, x = inputs_161_cast_fp16)[name = tensor("input_347_cast_fp16")]; + tensor var_23884_pad_type_0 = const()[name = tensor("op_23884_pad_type_0"), val = tensor("valid")]; + tensor var_23884_strides_0 = const()[name = tensor("op_23884_strides_0"), val = tensor([1, 1])]; + tensor var_23884_pad_0 = const()[name = tensor("op_23884_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23884_dilations_0 = const()[name = tensor("op_23884_dilations_0"), val = tensor([1, 1])]; + tensor var_23884_groups_0 = const()[name = tensor("op_23884_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711577984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(721408448))), name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(721408640)))]; + tensor var_23884_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_23884_dilations_0, groups = var_23884_groups_0, pad = var_23884_pad_0, pad_type = var_23884_pad_type_0, strides = var_23884_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized, x = input_347_cast_fp16)[name = tensor("op_23884_cast_fp16")]; + tensor var_23885_split_sizes_0 = const()[name = tensor("op_23885_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_23885_axis_0 = const()[name = tensor("op_23885_axis_0"), val = tensor(1)]; + tensor var_23885_cast_fp16_0, tensor var_23885_cast_fp16_1 = split(axis = var_23885_axis_0, split_sizes = var_23885_split_sizes_0, x = var_23884_cast_fp16)[name = tensor("op_23885_cast_fp16")]; + tensor var_23887_mode_0 = const()[name = tensor("op_23887_mode_0"), val = tensor("EXACT")]; + tensor var_23887_cast_fp16 = gelu(mode = var_23887_mode_0, x = var_23885_cast_fp16_1)[name = tensor("op_23887_cast_fp16")]; + tensor input_349_cast_fp16 = mul(x = var_23885_cast_fp16_0, y = var_23887_cast_fp16)[name = tensor("input_349_cast_fp16")]; + tensor var_23895_pad_type_0 = const()[name = tensor("op_23895_pad_type_0"), val = tensor("valid")]; + tensor var_23895_strides_0 = const()[name = tensor("op_23895_strides_0"), val = tensor([1, 1])]; + tensor var_23895_pad_0 = const()[name = tensor("op_23895_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23895_dilations_0 = const()[name = tensor("op_23895_dilations_0"), val = tensor([1, 1])]; + tensor var_23895_groups_0 = const()[name = tensor("op_23895_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(721429184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726344448))), name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726344640)))]; + tensor var_23895_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_23895_dilations_0, groups = var_23895_groups_0, pad = var_23895_pad_0, pad_type = var_23895_pad_type_0, strides = var_23895_strides_0, weight = mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized, x = input_349_cast_fp16)[name = tensor("op_23895_cast_fp16")]; + tensor inputs_163_cast_fp16 = add(x = var_23895_cast_fp16, y = inputs_161_cast_fp16)[name = tensor("inputs_163_cast_fp16")]; + tensor hidden_states_227_axes_0 = const()[name = tensor("hidden_states_227_axes_0"), val = tensor([1])]; + tensor hidden_states_227_gamma_0_to_fp16 = const()[name = tensor("hidden_states_227_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726347264)))]; + tensor hidden_states_227_beta_0_to_fp16 = const()[name = tensor("hidden_states_227_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726349888)))]; + tensor var_23911_to_fp16 = const()[name = tensor("op_23911_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_227_cast_fp16 = layer_norm(axes = hidden_states_227_axes_0, beta = hidden_states_227_beta_0_to_fp16, epsilon = var_23911_to_fp16, gamma = hidden_states_227_gamma_0_to_fp16, x = inputs_163_cast_fp16)[name = tensor("hidden_states_227_cast_fp16")]; + tensor q_109_pad_type_0 = const()[name = tensor("q_109_pad_type_0"), val = tensor("valid")]; + tensor q_109_strides_0 = const()[name = tensor("q_109_strides_0"), val = tensor([1, 1])]; + tensor q_109_pad_0 = const()[name = tensor("q_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_109_dilations_0 = const()[name = tensor("q_109_dilations_0"), val = tensor([1, 1])]; + tensor q_109_groups_0 = const()[name = tensor("q_109_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726352512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(727581376))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_109_cast_fp16 = conv(dilations = q_109_dilations_0, groups = q_109_groups_0, pad = q_109_pad_0, pad_type = q_109_pad_type_0, strides = q_109_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_227_cast_fp16)[name = tensor("q_109_cast_fp16")]; + tensor k_217_pad_type_0 = const()[name = tensor("k_217_pad_type_0"), val = tensor("valid")]; + tensor k_217_strides_0 = const()[name = tensor("k_217_strides_0"), val = tensor([1, 1])]; + tensor k_217_pad_0 = const()[name = tensor("k_217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_217_dilations_0 = const()[name = tensor("k_217_dilations_0"), val = tensor([1, 1])]; + tensor k_217_groups_0 = const()[name = tensor("k_217_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(727581568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728810432))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_217_cast_fp16 = conv(dilations = k_217_dilations_0, groups = k_217_groups_0, pad = k_217_pad_0, pad_type = k_217_pad_type_0, strides = k_217_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_227_cast_fp16)[name = tensor("k_217_cast_fp16")]; + tensor v_109_pad_type_0 = const()[name = tensor("v_109_pad_type_0"), val = tensor("valid")]; + tensor v_109_strides_0 = const()[name = tensor("v_109_strides_0"), val = tensor([1, 1])]; + tensor v_109_pad_0 = const()[name = tensor("v_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_109_dilations_0 = const()[name = tensor("v_109_dilations_0"), val = tensor([1, 1])]; + tensor v_109_groups_0 = const()[name = tensor("v_109_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728810624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(730039488))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_109_cast_fp16 = conv(dilations = v_109_dilations_0, groups = v_109_groups_0, pad = v_109_pad_0, pad_type = v_109_pad_type_0, strides = v_109_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_227_cast_fp16)[name = tensor("v_109_cast_fp16")]; + tensor var_23944_begin_0 = const()[name = tensor("op_23944_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23944_end_0 = const()[name = tensor("op_23944_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23944_end_mask_0 = const()[name = tensor("op_23944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23944_cast_fp16 = slice_by_index(begin = var_23944_begin_0, end = var_23944_end_0, end_mask = var_23944_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23944_cast_fp16")]; + tensor var_23948_begin_0 = const()[name = tensor("op_23948_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23948_end_0 = const()[name = tensor("op_23948_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23948_end_mask_0 = const()[name = tensor("op_23948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23948_cast_fp16 = slice_by_index(begin = var_23948_begin_0, end = var_23948_end_0, end_mask = var_23948_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23948_cast_fp16")]; + tensor var_23952_begin_0 = const()[name = tensor("op_23952_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23952_end_0 = const()[name = tensor("op_23952_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23952_end_mask_0 = const()[name = tensor("op_23952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23952_cast_fp16 = slice_by_index(begin = var_23952_begin_0, end = var_23952_end_0, end_mask = var_23952_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23952_cast_fp16")]; + tensor var_23956_begin_0 = const()[name = tensor("op_23956_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23956_end_0 = const()[name = tensor("op_23956_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23956_end_mask_0 = const()[name = tensor("op_23956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23956_cast_fp16 = slice_by_index(begin = var_23956_begin_0, end = var_23956_end_0, end_mask = var_23956_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23956_cast_fp16")]; + tensor var_23960_begin_0 = const()[name = tensor("op_23960_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23960_end_0 = const()[name = tensor("op_23960_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23960_end_mask_0 = const()[name = tensor("op_23960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23960_cast_fp16 = slice_by_index(begin = var_23960_begin_0, end = var_23960_end_0, end_mask = var_23960_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23960_cast_fp16")]; + tensor var_23964_begin_0 = const()[name = tensor("op_23964_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23964_end_0 = const()[name = tensor("op_23964_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23964_end_mask_0 = const()[name = tensor("op_23964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23964_cast_fp16 = slice_by_index(begin = var_23964_begin_0, end = var_23964_end_0, end_mask = var_23964_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23964_cast_fp16")]; + tensor var_23968_begin_0 = const()[name = tensor("op_23968_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23968_end_0 = const()[name = tensor("op_23968_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23968_end_mask_0 = const()[name = tensor("op_23968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23968_cast_fp16 = slice_by_index(begin = var_23968_begin_0, end = var_23968_end_0, end_mask = var_23968_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23968_cast_fp16")]; + tensor var_23972_begin_0 = const()[name = tensor("op_23972_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23972_end_0 = const()[name = tensor("op_23972_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23972_end_mask_0 = const()[name = tensor("op_23972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23972_cast_fp16 = slice_by_index(begin = var_23972_begin_0, end = var_23972_end_0, end_mask = var_23972_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23972_cast_fp16")]; + tensor var_23976_begin_0 = const()[name = tensor("op_23976_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23976_end_0 = const()[name = tensor("op_23976_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23976_end_mask_0 = const()[name = tensor("op_23976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23976_cast_fp16 = slice_by_index(begin = var_23976_begin_0, end = var_23976_end_0, end_mask = var_23976_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23976_cast_fp16")]; + tensor var_23980_begin_0 = const()[name = tensor("op_23980_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23980_end_0 = const()[name = tensor("op_23980_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23980_end_mask_0 = const()[name = tensor("op_23980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23980_cast_fp16 = slice_by_index(begin = var_23980_begin_0, end = var_23980_end_0, end_mask = var_23980_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23980_cast_fp16")]; + tensor var_23984_begin_0 = const()[name = tensor("op_23984_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23984_end_0 = const()[name = tensor("op_23984_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23984_end_mask_0 = const()[name = tensor("op_23984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23984_cast_fp16 = slice_by_index(begin = var_23984_begin_0, end = var_23984_end_0, end_mask = var_23984_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23984_cast_fp16")]; + tensor var_23988_begin_0 = const()[name = tensor("op_23988_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23988_end_0 = const()[name = tensor("op_23988_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23988_end_mask_0 = const()[name = tensor("op_23988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23988_cast_fp16 = slice_by_index(begin = var_23988_begin_0, end = var_23988_end_0, end_mask = var_23988_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23988_cast_fp16")]; + tensor var_23992_begin_0 = const()[name = tensor("op_23992_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23992_end_0 = const()[name = tensor("op_23992_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23992_end_mask_0 = const()[name = tensor("op_23992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23992_cast_fp16 = slice_by_index(begin = var_23992_begin_0, end = var_23992_end_0, end_mask = var_23992_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23992_cast_fp16")]; + tensor var_23996_begin_0 = const()[name = tensor("op_23996_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23996_end_0 = const()[name = tensor("op_23996_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23996_end_mask_0 = const()[name = tensor("op_23996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23996_cast_fp16 = slice_by_index(begin = var_23996_begin_0, end = var_23996_end_0, end_mask = var_23996_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23996_cast_fp16")]; + tensor var_24000_begin_0 = const()[name = tensor("op_24000_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24000_end_0 = const()[name = tensor("op_24000_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24000_end_mask_0 = const()[name = tensor("op_24000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24000_cast_fp16 = slice_by_index(begin = var_24000_begin_0, end = var_24000_end_0, end_mask = var_24000_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24000_cast_fp16")]; + tensor var_24004_begin_0 = const()[name = tensor("op_24004_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24004_end_0 = const()[name = tensor("op_24004_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24004_end_mask_0 = const()[name = tensor("op_24004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24004_cast_fp16 = slice_by_index(begin = var_24004_begin_0, end = var_24004_end_0, end_mask = var_24004_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24004_cast_fp16")]; + tensor var_24008_begin_0 = const()[name = tensor("op_24008_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24008_end_0 = const()[name = tensor("op_24008_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24008_end_mask_0 = const()[name = tensor("op_24008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24008_cast_fp16 = slice_by_index(begin = var_24008_begin_0, end = var_24008_end_0, end_mask = var_24008_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24008_cast_fp16")]; + tensor var_24012_begin_0 = const()[name = tensor("op_24012_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24012_end_0 = const()[name = tensor("op_24012_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24012_end_mask_0 = const()[name = tensor("op_24012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24012_cast_fp16 = slice_by_index(begin = var_24012_begin_0, end = var_24012_end_0, end_mask = var_24012_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24012_cast_fp16")]; + tensor var_24016_begin_0 = const()[name = tensor("op_24016_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24016_end_0 = const()[name = tensor("op_24016_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24016_end_mask_0 = const()[name = tensor("op_24016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24016_cast_fp16 = slice_by_index(begin = var_24016_begin_0, end = var_24016_end_0, end_mask = var_24016_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24016_cast_fp16")]; + tensor var_24020_begin_0 = const()[name = tensor("op_24020_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24020_end_0 = const()[name = tensor("op_24020_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24020_end_mask_0 = const()[name = tensor("op_24020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24020_cast_fp16 = slice_by_index(begin = var_24020_begin_0, end = var_24020_end_0, end_mask = var_24020_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24020_cast_fp16")]; + tensor k_219_perm_0 = const()[name = tensor("k_219_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_24027_begin_0 = const()[name = tensor("op_24027_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24027_end_0 = const()[name = tensor("op_24027_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_24027_end_mask_0 = const()[name = tensor("op_24027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_219_cast_fp16 = transpose(perm = k_219_perm_0, x = k_217_cast_fp16)[name = tensor("transpose_13")]; + tensor var_24027_cast_fp16 = slice_by_index(begin = var_24027_begin_0, end = var_24027_end_0, end_mask = var_24027_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24027_cast_fp16")]; + tensor var_24031_begin_0 = const()[name = tensor("op_24031_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_24031_end_0 = const()[name = tensor("op_24031_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_24031_end_mask_0 = const()[name = tensor("op_24031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24031_cast_fp16 = slice_by_index(begin = var_24031_begin_0, end = var_24031_end_0, end_mask = var_24031_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24031_cast_fp16")]; + tensor var_24035_begin_0 = const()[name = tensor("op_24035_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_24035_end_0 = const()[name = tensor("op_24035_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_24035_end_mask_0 = const()[name = tensor("op_24035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24035_cast_fp16 = slice_by_index(begin = var_24035_begin_0, end = var_24035_end_0, end_mask = var_24035_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24035_cast_fp16")]; + tensor var_24039_begin_0 = const()[name = tensor("op_24039_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_24039_end_0 = const()[name = tensor("op_24039_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_24039_end_mask_0 = const()[name = tensor("op_24039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24039_cast_fp16 = slice_by_index(begin = var_24039_begin_0, end = var_24039_end_0, end_mask = var_24039_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24039_cast_fp16")]; + tensor var_24043_begin_0 = const()[name = tensor("op_24043_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_24043_end_0 = const()[name = tensor("op_24043_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_24043_end_mask_0 = const()[name = tensor("op_24043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24043_cast_fp16 = slice_by_index(begin = var_24043_begin_0, end = var_24043_end_0, end_mask = var_24043_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24043_cast_fp16")]; + tensor var_24047_begin_0 = const()[name = tensor("op_24047_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_24047_end_0 = const()[name = tensor("op_24047_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_24047_end_mask_0 = const()[name = tensor("op_24047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24047_cast_fp16 = slice_by_index(begin = var_24047_begin_0, end = var_24047_end_0, end_mask = var_24047_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24047_cast_fp16")]; + tensor var_24051_begin_0 = const()[name = tensor("op_24051_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_24051_end_0 = const()[name = tensor("op_24051_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_24051_end_mask_0 = const()[name = tensor("op_24051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24051_cast_fp16 = slice_by_index(begin = var_24051_begin_0, end = var_24051_end_0, end_mask = var_24051_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24051_cast_fp16")]; + tensor var_24055_begin_0 = const()[name = tensor("op_24055_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_24055_end_0 = const()[name = tensor("op_24055_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_24055_end_mask_0 = const()[name = tensor("op_24055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24055_cast_fp16 = slice_by_index(begin = var_24055_begin_0, end = var_24055_end_0, end_mask = var_24055_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24055_cast_fp16")]; + tensor var_24059_begin_0 = const()[name = tensor("op_24059_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_24059_end_0 = const()[name = tensor("op_24059_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_24059_end_mask_0 = const()[name = tensor("op_24059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24059_cast_fp16 = slice_by_index(begin = var_24059_begin_0, end = var_24059_end_0, end_mask = var_24059_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24059_cast_fp16")]; + tensor var_24063_begin_0 = const()[name = tensor("op_24063_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_24063_end_0 = const()[name = tensor("op_24063_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_24063_end_mask_0 = const()[name = tensor("op_24063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24063_cast_fp16 = slice_by_index(begin = var_24063_begin_0, end = var_24063_end_0, end_mask = var_24063_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24063_cast_fp16")]; + tensor var_24067_begin_0 = const()[name = tensor("op_24067_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_24067_end_0 = const()[name = tensor("op_24067_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_24067_end_mask_0 = const()[name = tensor("op_24067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24067_cast_fp16 = slice_by_index(begin = var_24067_begin_0, end = var_24067_end_0, end_mask = var_24067_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24067_cast_fp16")]; + tensor var_24071_begin_0 = const()[name = tensor("op_24071_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_24071_end_0 = const()[name = tensor("op_24071_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_24071_end_mask_0 = const()[name = tensor("op_24071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24071_cast_fp16 = slice_by_index(begin = var_24071_begin_0, end = var_24071_end_0, end_mask = var_24071_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24071_cast_fp16")]; + tensor var_24075_begin_0 = const()[name = tensor("op_24075_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_24075_end_0 = const()[name = tensor("op_24075_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_24075_end_mask_0 = const()[name = tensor("op_24075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24075_cast_fp16 = slice_by_index(begin = var_24075_begin_0, end = var_24075_end_0, end_mask = var_24075_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24075_cast_fp16")]; + tensor var_24079_begin_0 = const()[name = tensor("op_24079_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_24079_end_0 = const()[name = tensor("op_24079_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_24079_end_mask_0 = const()[name = tensor("op_24079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24079_cast_fp16 = slice_by_index(begin = var_24079_begin_0, end = var_24079_end_0, end_mask = var_24079_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24079_cast_fp16")]; + tensor var_24083_begin_0 = const()[name = tensor("op_24083_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_24083_end_0 = const()[name = tensor("op_24083_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_24083_end_mask_0 = const()[name = tensor("op_24083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24083_cast_fp16 = slice_by_index(begin = var_24083_begin_0, end = var_24083_end_0, end_mask = var_24083_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24083_cast_fp16")]; + tensor var_24087_begin_0 = const()[name = tensor("op_24087_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_24087_end_0 = const()[name = tensor("op_24087_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24087_end_mask_0 = const()[name = tensor("op_24087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24087_cast_fp16 = slice_by_index(begin = var_24087_begin_0, end = var_24087_end_0, end_mask = var_24087_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24087_cast_fp16")]; + tensor var_24091_begin_0 = const()[name = tensor("op_24091_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_24091_end_0 = const()[name = tensor("op_24091_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_24091_end_mask_0 = const()[name = tensor("op_24091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24091_cast_fp16 = slice_by_index(begin = var_24091_begin_0, end = var_24091_end_0, end_mask = var_24091_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24091_cast_fp16")]; + tensor var_24095_begin_0 = const()[name = tensor("op_24095_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_24095_end_0 = const()[name = tensor("op_24095_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_24095_end_mask_0 = const()[name = tensor("op_24095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24095_cast_fp16 = slice_by_index(begin = var_24095_begin_0, end = var_24095_end_0, end_mask = var_24095_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24095_cast_fp16")]; + tensor var_24099_begin_0 = const()[name = tensor("op_24099_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_24099_end_0 = const()[name = tensor("op_24099_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_24099_end_mask_0 = const()[name = tensor("op_24099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24099_cast_fp16 = slice_by_index(begin = var_24099_begin_0, end = var_24099_end_0, end_mask = var_24099_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24099_cast_fp16")]; + tensor var_24103_begin_0 = const()[name = tensor("op_24103_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_24103_end_0 = const()[name = tensor("op_24103_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_24103_end_mask_0 = const()[name = tensor("op_24103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24103_cast_fp16 = slice_by_index(begin = var_24103_begin_0, end = var_24103_end_0, end_mask = var_24103_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24103_cast_fp16")]; + tensor var_24105_begin_0 = const()[name = tensor("op_24105_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24105_end_0 = const()[name = tensor("op_24105_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_24105_end_mask_0 = const()[name = tensor("op_24105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24105_cast_fp16 = slice_by_index(begin = var_24105_begin_0, end = var_24105_end_0, end_mask = var_24105_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24105_cast_fp16")]; + tensor var_24109_begin_0 = const()[name = tensor("op_24109_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24109_end_0 = const()[name = tensor("op_24109_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_24109_end_mask_0 = const()[name = tensor("op_24109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24109_cast_fp16 = slice_by_index(begin = var_24109_begin_0, end = var_24109_end_0, end_mask = var_24109_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24109_cast_fp16")]; + tensor var_24113_begin_0 = const()[name = tensor("op_24113_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24113_end_0 = const()[name = tensor("op_24113_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_24113_end_mask_0 = const()[name = tensor("op_24113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24113_cast_fp16 = slice_by_index(begin = var_24113_begin_0, end = var_24113_end_0, end_mask = var_24113_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24113_cast_fp16")]; + tensor var_24117_begin_0 = const()[name = tensor("op_24117_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24117_end_0 = const()[name = tensor("op_24117_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_24117_end_mask_0 = const()[name = tensor("op_24117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24117_cast_fp16 = slice_by_index(begin = var_24117_begin_0, end = var_24117_end_0, end_mask = var_24117_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24117_cast_fp16")]; + tensor var_24121_begin_0 = const()[name = tensor("op_24121_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24121_end_0 = const()[name = tensor("op_24121_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_24121_end_mask_0 = const()[name = tensor("op_24121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24121_cast_fp16 = slice_by_index(begin = var_24121_begin_0, end = var_24121_end_0, end_mask = var_24121_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24121_cast_fp16")]; + tensor var_24125_begin_0 = const()[name = tensor("op_24125_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24125_end_0 = const()[name = tensor("op_24125_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_24125_end_mask_0 = const()[name = tensor("op_24125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24125_cast_fp16 = slice_by_index(begin = var_24125_begin_0, end = var_24125_end_0, end_mask = var_24125_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24125_cast_fp16")]; + tensor var_24129_begin_0 = const()[name = tensor("op_24129_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24129_end_0 = const()[name = tensor("op_24129_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_24129_end_mask_0 = const()[name = tensor("op_24129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24129_cast_fp16 = slice_by_index(begin = var_24129_begin_0, end = var_24129_end_0, end_mask = var_24129_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24129_cast_fp16")]; + tensor var_24133_begin_0 = const()[name = tensor("op_24133_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24133_end_0 = const()[name = tensor("op_24133_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_24133_end_mask_0 = const()[name = tensor("op_24133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24133_cast_fp16 = slice_by_index(begin = var_24133_begin_0, end = var_24133_end_0, end_mask = var_24133_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24133_cast_fp16")]; + tensor var_24137_begin_0 = const()[name = tensor("op_24137_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24137_end_0 = const()[name = tensor("op_24137_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_24137_end_mask_0 = const()[name = tensor("op_24137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24137_cast_fp16 = slice_by_index(begin = var_24137_begin_0, end = var_24137_end_0, end_mask = var_24137_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24137_cast_fp16")]; + tensor var_24141_begin_0 = const()[name = tensor("op_24141_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24141_end_0 = const()[name = tensor("op_24141_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_24141_end_mask_0 = const()[name = tensor("op_24141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24141_cast_fp16 = slice_by_index(begin = var_24141_begin_0, end = var_24141_end_0, end_mask = var_24141_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24141_cast_fp16")]; + tensor var_24145_begin_0 = const()[name = tensor("op_24145_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24145_end_0 = const()[name = tensor("op_24145_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_24145_end_mask_0 = const()[name = tensor("op_24145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24145_cast_fp16 = slice_by_index(begin = var_24145_begin_0, end = var_24145_end_0, end_mask = var_24145_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24145_cast_fp16")]; + tensor var_24149_begin_0 = const()[name = tensor("op_24149_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24149_end_0 = const()[name = tensor("op_24149_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_24149_end_mask_0 = const()[name = tensor("op_24149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24149_cast_fp16 = slice_by_index(begin = var_24149_begin_0, end = var_24149_end_0, end_mask = var_24149_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24149_cast_fp16")]; + tensor var_24153_begin_0 = const()[name = tensor("op_24153_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24153_end_0 = const()[name = tensor("op_24153_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_24153_end_mask_0 = const()[name = tensor("op_24153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24153_cast_fp16 = slice_by_index(begin = var_24153_begin_0, end = var_24153_end_0, end_mask = var_24153_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24153_cast_fp16")]; + tensor var_24157_begin_0 = const()[name = tensor("op_24157_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24157_end_0 = const()[name = tensor("op_24157_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_24157_end_mask_0 = const()[name = tensor("op_24157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24157_cast_fp16 = slice_by_index(begin = var_24157_begin_0, end = var_24157_end_0, end_mask = var_24157_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24157_cast_fp16")]; + tensor var_24161_begin_0 = const()[name = tensor("op_24161_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24161_end_0 = const()[name = tensor("op_24161_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24161_end_mask_0 = const()[name = tensor("op_24161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24161_cast_fp16 = slice_by_index(begin = var_24161_begin_0, end = var_24161_end_0, end_mask = var_24161_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24161_cast_fp16")]; + tensor var_24165_begin_0 = const()[name = tensor("op_24165_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24165_end_0 = const()[name = tensor("op_24165_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24165_end_mask_0 = const()[name = tensor("op_24165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24165_cast_fp16 = slice_by_index(begin = var_24165_begin_0, end = var_24165_end_0, end_mask = var_24165_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24165_cast_fp16")]; + tensor var_24169_begin_0 = const()[name = tensor("op_24169_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24169_end_0 = const()[name = tensor("op_24169_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24169_end_mask_0 = const()[name = tensor("op_24169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24169_cast_fp16 = slice_by_index(begin = var_24169_begin_0, end = var_24169_end_0, end_mask = var_24169_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24169_cast_fp16")]; + tensor var_24173_begin_0 = const()[name = tensor("op_24173_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24173_end_0 = const()[name = tensor("op_24173_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24173_end_mask_0 = const()[name = tensor("op_24173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24173_cast_fp16 = slice_by_index(begin = var_24173_begin_0, end = var_24173_end_0, end_mask = var_24173_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24173_cast_fp16")]; + tensor var_24177_begin_0 = const()[name = tensor("op_24177_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24177_end_0 = const()[name = tensor("op_24177_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24177_end_mask_0 = const()[name = tensor("op_24177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24177_cast_fp16 = slice_by_index(begin = var_24177_begin_0, end = var_24177_end_0, end_mask = var_24177_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24177_cast_fp16")]; + tensor var_24181_begin_0 = const()[name = tensor("op_24181_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24181_end_0 = const()[name = tensor("op_24181_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24181_end_mask_0 = const()[name = tensor("op_24181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24181_cast_fp16 = slice_by_index(begin = var_24181_begin_0, end = var_24181_end_0, end_mask = var_24181_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24181_cast_fp16")]; + tensor var_24185_equation_0 = const()[name = tensor("op_24185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24185_cast_fp16 = einsum(equation = var_24185_equation_0, values = (var_24027_cast_fp16, var_23944_cast_fp16))[name = tensor("op_24185_cast_fp16")]; + tensor var_24186_to_fp16 = const()[name = tensor("op_24186_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2001_cast_fp16 = mul(x = var_24185_cast_fp16, y = var_24186_to_fp16)[name = tensor("aw_2001_cast_fp16")]; + tensor var_24189_equation_0 = const()[name = tensor("op_24189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24189_cast_fp16 = einsum(equation = var_24189_equation_0, values = (var_24031_cast_fp16, var_23948_cast_fp16))[name = tensor("op_24189_cast_fp16")]; + tensor var_24190_to_fp16 = const()[name = tensor("op_24190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2003_cast_fp16 = mul(x = var_24189_cast_fp16, y = var_24190_to_fp16)[name = tensor("aw_2003_cast_fp16")]; + tensor var_24193_equation_0 = const()[name = tensor("op_24193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24193_cast_fp16 = einsum(equation = var_24193_equation_0, values = (var_24035_cast_fp16, var_23952_cast_fp16))[name = tensor("op_24193_cast_fp16")]; + tensor var_24194_to_fp16 = const()[name = tensor("op_24194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2005_cast_fp16 = mul(x = var_24193_cast_fp16, y = var_24194_to_fp16)[name = tensor("aw_2005_cast_fp16")]; + tensor var_24197_equation_0 = const()[name = tensor("op_24197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24197_cast_fp16 = einsum(equation = var_24197_equation_0, values = (var_24039_cast_fp16, var_23956_cast_fp16))[name = tensor("op_24197_cast_fp16")]; + tensor var_24198_to_fp16 = const()[name = tensor("op_24198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2007_cast_fp16 = mul(x = var_24197_cast_fp16, y = var_24198_to_fp16)[name = tensor("aw_2007_cast_fp16")]; + tensor var_24201_equation_0 = const()[name = tensor("op_24201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24201_cast_fp16 = einsum(equation = var_24201_equation_0, values = (var_24043_cast_fp16, var_23960_cast_fp16))[name = tensor("op_24201_cast_fp16")]; + tensor var_24202_to_fp16 = const()[name = tensor("op_24202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2009_cast_fp16 = mul(x = var_24201_cast_fp16, y = var_24202_to_fp16)[name = tensor("aw_2009_cast_fp16")]; + tensor var_24205_equation_0 = const()[name = tensor("op_24205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24205_cast_fp16 = einsum(equation = var_24205_equation_0, values = (var_24047_cast_fp16, var_23964_cast_fp16))[name = tensor("op_24205_cast_fp16")]; + tensor var_24206_to_fp16 = const()[name = tensor("op_24206_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2011_cast_fp16 = mul(x = var_24205_cast_fp16, y = var_24206_to_fp16)[name = tensor("aw_2011_cast_fp16")]; + tensor var_24209_equation_0 = const()[name = tensor("op_24209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24209_cast_fp16 = einsum(equation = var_24209_equation_0, values = (var_24051_cast_fp16, var_23968_cast_fp16))[name = tensor("op_24209_cast_fp16")]; + tensor var_24210_to_fp16 = const()[name = tensor("op_24210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2013_cast_fp16 = mul(x = var_24209_cast_fp16, y = var_24210_to_fp16)[name = tensor("aw_2013_cast_fp16")]; + tensor var_24213_equation_0 = const()[name = tensor("op_24213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24213_cast_fp16 = einsum(equation = var_24213_equation_0, values = (var_24055_cast_fp16, var_23972_cast_fp16))[name = tensor("op_24213_cast_fp16")]; + tensor var_24214_to_fp16 = const()[name = tensor("op_24214_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2015_cast_fp16 = mul(x = var_24213_cast_fp16, y = var_24214_to_fp16)[name = tensor("aw_2015_cast_fp16")]; + tensor var_24217_equation_0 = const()[name = tensor("op_24217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24217_cast_fp16 = einsum(equation = var_24217_equation_0, values = (var_24059_cast_fp16, var_23976_cast_fp16))[name = tensor("op_24217_cast_fp16")]; + tensor var_24218_to_fp16 = const()[name = tensor("op_24218_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2017_cast_fp16 = mul(x = var_24217_cast_fp16, y = var_24218_to_fp16)[name = tensor("aw_2017_cast_fp16")]; + tensor var_24221_equation_0 = const()[name = tensor("op_24221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24221_cast_fp16 = einsum(equation = var_24221_equation_0, values = (var_24063_cast_fp16, var_23980_cast_fp16))[name = tensor("op_24221_cast_fp16")]; + tensor var_24222_to_fp16 = const()[name = tensor("op_24222_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2019_cast_fp16 = mul(x = var_24221_cast_fp16, y = var_24222_to_fp16)[name = tensor("aw_2019_cast_fp16")]; + tensor var_24225_equation_0 = const()[name = tensor("op_24225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24225_cast_fp16 = einsum(equation = var_24225_equation_0, values = (var_24067_cast_fp16, var_23984_cast_fp16))[name = tensor("op_24225_cast_fp16")]; + tensor var_24226_to_fp16 = const()[name = tensor("op_24226_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2021_cast_fp16 = mul(x = var_24225_cast_fp16, y = var_24226_to_fp16)[name = tensor("aw_2021_cast_fp16")]; + tensor var_24229_equation_0 = const()[name = tensor("op_24229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24229_cast_fp16 = einsum(equation = var_24229_equation_0, values = (var_24071_cast_fp16, var_23988_cast_fp16))[name = tensor("op_24229_cast_fp16")]; + tensor var_24230_to_fp16 = const()[name = tensor("op_24230_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2023_cast_fp16 = mul(x = var_24229_cast_fp16, y = var_24230_to_fp16)[name = tensor("aw_2023_cast_fp16")]; + tensor var_24233_equation_0 = const()[name = tensor("op_24233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24233_cast_fp16 = einsum(equation = var_24233_equation_0, values = (var_24075_cast_fp16, var_23992_cast_fp16))[name = tensor("op_24233_cast_fp16")]; + tensor var_24234_to_fp16 = const()[name = tensor("op_24234_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2025_cast_fp16 = mul(x = var_24233_cast_fp16, y = var_24234_to_fp16)[name = tensor("aw_2025_cast_fp16")]; + tensor var_24237_equation_0 = const()[name = tensor("op_24237_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24237_cast_fp16 = einsum(equation = var_24237_equation_0, values = (var_24079_cast_fp16, var_23996_cast_fp16))[name = tensor("op_24237_cast_fp16")]; + tensor var_24238_to_fp16 = const()[name = tensor("op_24238_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2027_cast_fp16 = mul(x = var_24237_cast_fp16, y = var_24238_to_fp16)[name = tensor("aw_2027_cast_fp16")]; + tensor var_24241_equation_0 = const()[name = tensor("op_24241_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24241_cast_fp16 = einsum(equation = var_24241_equation_0, values = (var_24083_cast_fp16, var_24000_cast_fp16))[name = tensor("op_24241_cast_fp16")]; + tensor var_24242_to_fp16 = const()[name = tensor("op_24242_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2029_cast_fp16 = mul(x = var_24241_cast_fp16, y = var_24242_to_fp16)[name = tensor("aw_2029_cast_fp16")]; + tensor var_24245_equation_0 = const()[name = tensor("op_24245_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24245_cast_fp16 = einsum(equation = var_24245_equation_0, values = (var_24087_cast_fp16, var_24004_cast_fp16))[name = tensor("op_24245_cast_fp16")]; + tensor var_24246_to_fp16 = const()[name = tensor("op_24246_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2031_cast_fp16 = mul(x = var_24245_cast_fp16, y = var_24246_to_fp16)[name = tensor("aw_2031_cast_fp16")]; + tensor var_24249_equation_0 = const()[name = tensor("op_24249_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24249_cast_fp16 = einsum(equation = var_24249_equation_0, values = (var_24091_cast_fp16, var_24008_cast_fp16))[name = tensor("op_24249_cast_fp16")]; + tensor var_24250_to_fp16 = const()[name = tensor("op_24250_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2033_cast_fp16 = mul(x = var_24249_cast_fp16, y = var_24250_to_fp16)[name = tensor("aw_2033_cast_fp16")]; + tensor var_24253_equation_0 = const()[name = tensor("op_24253_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24253_cast_fp16 = einsum(equation = var_24253_equation_0, values = (var_24095_cast_fp16, var_24012_cast_fp16))[name = tensor("op_24253_cast_fp16")]; + tensor var_24254_to_fp16 = const()[name = tensor("op_24254_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2035_cast_fp16 = mul(x = var_24253_cast_fp16, y = var_24254_to_fp16)[name = tensor("aw_2035_cast_fp16")]; + tensor var_24257_equation_0 = const()[name = tensor("op_24257_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24257_cast_fp16 = einsum(equation = var_24257_equation_0, values = (var_24099_cast_fp16, var_24016_cast_fp16))[name = tensor("op_24257_cast_fp16")]; + tensor var_24258_to_fp16 = const()[name = tensor("op_24258_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2037_cast_fp16 = mul(x = var_24257_cast_fp16, y = var_24258_to_fp16)[name = tensor("aw_2037_cast_fp16")]; + tensor var_24261_equation_0 = const()[name = tensor("op_24261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24261_cast_fp16 = einsum(equation = var_24261_equation_0, values = (var_24103_cast_fp16, var_24020_cast_fp16))[name = tensor("op_24261_cast_fp16")]; + tensor var_24262_to_fp16 = const()[name = tensor("op_24262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2039_cast_fp16 = mul(x = var_24261_cast_fp16, y = var_24262_to_fp16)[name = tensor("aw_2039_cast_fp16")]; + tensor var_24264_cast_fp16 = softmax(axis = var_21077, x = aw_2001_cast_fp16)[name = tensor("op_24264_cast_fp16")]; + tensor var_24265_cast_fp16 = softmax(axis = var_21077, x = aw_2003_cast_fp16)[name = tensor("op_24265_cast_fp16")]; + tensor var_24266_cast_fp16 = softmax(axis = var_21077, x = aw_2005_cast_fp16)[name = tensor("op_24266_cast_fp16")]; + tensor var_24267_cast_fp16 = softmax(axis = var_21077, x = aw_2007_cast_fp16)[name = tensor("op_24267_cast_fp16")]; + tensor var_24268_cast_fp16 = softmax(axis = var_21077, x = aw_2009_cast_fp16)[name = tensor("op_24268_cast_fp16")]; + tensor var_24269_cast_fp16 = softmax(axis = var_21077, x = aw_2011_cast_fp16)[name = tensor("op_24269_cast_fp16")]; + tensor var_24270_cast_fp16 = softmax(axis = var_21077, x = aw_2013_cast_fp16)[name = tensor("op_24270_cast_fp16")]; + tensor var_24271_cast_fp16 = softmax(axis = var_21077, x = aw_2015_cast_fp16)[name = tensor("op_24271_cast_fp16")]; + tensor var_24272_cast_fp16 = softmax(axis = var_21077, x = aw_2017_cast_fp16)[name = tensor("op_24272_cast_fp16")]; + tensor var_24273_cast_fp16 = softmax(axis = var_21077, x = aw_2019_cast_fp16)[name = tensor("op_24273_cast_fp16")]; + tensor var_24274_cast_fp16 = softmax(axis = var_21077, x = aw_2021_cast_fp16)[name = tensor("op_24274_cast_fp16")]; + tensor var_24275_cast_fp16 = softmax(axis = var_21077, x = aw_2023_cast_fp16)[name = tensor("op_24275_cast_fp16")]; + tensor var_24276_cast_fp16 = softmax(axis = var_21077, x = aw_2025_cast_fp16)[name = tensor("op_24276_cast_fp16")]; + tensor var_24277_cast_fp16 = softmax(axis = var_21077, x = aw_2027_cast_fp16)[name = tensor("op_24277_cast_fp16")]; + tensor var_24278_cast_fp16 = softmax(axis = var_21077, x = aw_2029_cast_fp16)[name = tensor("op_24278_cast_fp16")]; + tensor var_24279_cast_fp16 = softmax(axis = var_21077, x = aw_2031_cast_fp16)[name = tensor("op_24279_cast_fp16")]; + tensor var_24280_cast_fp16 = softmax(axis = var_21077, x = aw_2033_cast_fp16)[name = tensor("op_24280_cast_fp16")]; + tensor var_24281_cast_fp16 = softmax(axis = var_21077, x = aw_2035_cast_fp16)[name = tensor("op_24281_cast_fp16")]; + tensor var_24282_cast_fp16 = softmax(axis = var_21077, x = aw_2037_cast_fp16)[name = tensor("op_24282_cast_fp16")]; + tensor var_24283_cast_fp16 = softmax(axis = var_21077, x = aw_2039_cast_fp16)[name = tensor("op_24283_cast_fp16")]; + tensor var_24285_equation_0 = const()[name = tensor("op_24285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24285_cast_fp16 = einsum(equation = var_24285_equation_0, values = (var_24105_cast_fp16, var_24264_cast_fp16))[name = tensor("op_24285_cast_fp16")]; + tensor var_24287_equation_0 = const()[name = tensor("op_24287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24287_cast_fp16 = einsum(equation = var_24287_equation_0, values = (var_24109_cast_fp16, var_24265_cast_fp16))[name = tensor("op_24287_cast_fp16")]; + tensor var_24289_equation_0 = const()[name = tensor("op_24289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24289_cast_fp16 = einsum(equation = var_24289_equation_0, values = (var_24113_cast_fp16, var_24266_cast_fp16))[name = tensor("op_24289_cast_fp16")]; + tensor var_24291_equation_0 = const()[name = tensor("op_24291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24291_cast_fp16 = einsum(equation = var_24291_equation_0, values = (var_24117_cast_fp16, var_24267_cast_fp16))[name = tensor("op_24291_cast_fp16")]; + tensor var_24293_equation_0 = const()[name = tensor("op_24293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24293_cast_fp16 = einsum(equation = var_24293_equation_0, values = (var_24121_cast_fp16, var_24268_cast_fp16))[name = tensor("op_24293_cast_fp16")]; + tensor var_24295_equation_0 = const()[name = tensor("op_24295_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24295_cast_fp16 = einsum(equation = var_24295_equation_0, values = (var_24125_cast_fp16, var_24269_cast_fp16))[name = tensor("op_24295_cast_fp16")]; + tensor var_24297_equation_0 = const()[name = tensor("op_24297_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24297_cast_fp16 = einsum(equation = var_24297_equation_0, values = (var_24129_cast_fp16, var_24270_cast_fp16))[name = tensor("op_24297_cast_fp16")]; + tensor var_24299_equation_0 = const()[name = tensor("op_24299_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24299_cast_fp16 = einsum(equation = var_24299_equation_0, values = (var_24133_cast_fp16, var_24271_cast_fp16))[name = tensor("op_24299_cast_fp16")]; + tensor var_24301_equation_0 = const()[name = tensor("op_24301_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24301_cast_fp16 = einsum(equation = var_24301_equation_0, values = (var_24137_cast_fp16, var_24272_cast_fp16))[name = tensor("op_24301_cast_fp16")]; + tensor var_24303_equation_0 = const()[name = tensor("op_24303_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24303_cast_fp16 = einsum(equation = var_24303_equation_0, values = (var_24141_cast_fp16, var_24273_cast_fp16))[name = tensor("op_24303_cast_fp16")]; + tensor var_24305_equation_0 = const()[name = tensor("op_24305_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24305_cast_fp16 = einsum(equation = var_24305_equation_0, values = (var_24145_cast_fp16, var_24274_cast_fp16))[name = tensor("op_24305_cast_fp16")]; + tensor var_24307_equation_0 = const()[name = tensor("op_24307_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24307_cast_fp16 = einsum(equation = var_24307_equation_0, values = (var_24149_cast_fp16, var_24275_cast_fp16))[name = tensor("op_24307_cast_fp16")]; + tensor var_24309_equation_0 = const()[name = tensor("op_24309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24309_cast_fp16 = einsum(equation = var_24309_equation_0, values = (var_24153_cast_fp16, var_24276_cast_fp16))[name = tensor("op_24309_cast_fp16")]; + tensor var_24311_equation_0 = const()[name = tensor("op_24311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24311_cast_fp16 = einsum(equation = var_24311_equation_0, values = (var_24157_cast_fp16, var_24277_cast_fp16))[name = tensor("op_24311_cast_fp16")]; + tensor var_24313_equation_0 = const()[name = tensor("op_24313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24313_cast_fp16 = einsum(equation = var_24313_equation_0, values = (var_24161_cast_fp16, var_24278_cast_fp16))[name = tensor("op_24313_cast_fp16")]; + tensor var_24315_equation_0 = const()[name = tensor("op_24315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24315_cast_fp16 = einsum(equation = var_24315_equation_0, values = (var_24165_cast_fp16, var_24279_cast_fp16))[name = tensor("op_24315_cast_fp16")]; + tensor var_24317_equation_0 = const()[name = tensor("op_24317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24317_cast_fp16 = einsum(equation = var_24317_equation_0, values = (var_24169_cast_fp16, var_24280_cast_fp16))[name = tensor("op_24317_cast_fp16")]; + tensor var_24319_equation_0 = const()[name = tensor("op_24319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24319_cast_fp16 = einsum(equation = var_24319_equation_0, values = (var_24173_cast_fp16, var_24281_cast_fp16))[name = tensor("op_24319_cast_fp16")]; + tensor var_24321_equation_0 = const()[name = tensor("op_24321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24321_cast_fp16 = einsum(equation = var_24321_equation_0, values = (var_24177_cast_fp16, var_24282_cast_fp16))[name = tensor("op_24321_cast_fp16")]; + tensor var_24323_equation_0 = const()[name = tensor("op_24323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24323_cast_fp16 = einsum(equation = var_24323_equation_0, values = (var_24181_cast_fp16, var_24283_cast_fp16))[name = tensor("op_24323_cast_fp16")]; + tensor input_351_interleave_0 = const()[name = tensor("input_351_interleave_0"), val = tensor(false)]; + tensor input_351_cast_fp16 = concat(axis = var_21077, interleave = input_351_interleave_0, values = (var_24285_cast_fp16, var_24287_cast_fp16, var_24289_cast_fp16, var_24291_cast_fp16, var_24293_cast_fp16, var_24295_cast_fp16, var_24297_cast_fp16, var_24299_cast_fp16, var_24301_cast_fp16, var_24303_cast_fp16, var_24305_cast_fp16, var_24307_cast_fp16, var_24309_cast_fp16, var_24311_cast_fp16, var_24313_cast_fp16, var_24315_cast_fp16, var_24317_cast_fp16, var_24319_cast_fp16, var_24321_cast_fp16, var_24323_cast_fp16))[name = tensor("input_351_cast_fp16")]; + tensor var_24333_pad_type_0 = const()[name = tensor("op_24333_pad_type_0"), val = tensor("valid")]; + tensor var_24333_strides_0 = const()[name = tensor("op_24333_strides_0"), val = tensor([1, 1])]; + tensor var_24333_pad_0 = const()[name = tensor("op_24333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24333_dilations_0 = const()[name = tensor("op_24333_dilations_0"), val = tensor([1, 1])]; + tensor var_24333_groups_0 = const()[name = tensor("op_24333_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(730039680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731268544))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731268736)))]; + tensor var_24333_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_24333_dilations_0, groups = var_24333_groups_0, pad = var_24333_pad_0, pad_type = var_24333_pad_type_0, strides = var_24333_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized, x = input_351_cast_fp16)[name = tensor("op_24333_cast_fp16")]; + tensor inputs_165_cast_fp16 = add(x = var_24333_cast_fp16, y = inputs_163_cast_fp16)[name = tensor("inputs_165_cast_fp16")]; + tensor hidden_states_229_axes_0 = const()[name = tensor("hidden_states_229_axes_0"), val = tensor([1])]; + tensor hidden_states_229_gamma_0_to_fp16 = const()[name = tensor("hidden_states_229_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731271360)))]; + tensor hidden_states_229_beta_0_to_fp16 = const()[name = tensor("hidden_states_229_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731273984)))]; + tensor var_24343_to_fp16 = const()[name = tensor("op_24343_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_229_cast_fp16 = layer_norm(axes = hidden_states_229_axes_0, beta = hidden_states_229_beta_0_to_fp16, epsilon = var_24343_to_fp16, gamma = hidden_states_229_gamma_0_to_fp16, x = inputs_165_cast_fp16)[name = tensor("hidden_states_229_cast_fp16")]; + tensor q_111_pad_type_0 = const()[name = tensor("q_111_pad_type_0"), val = tensor("valid")]; + tensor q_111_strides_0 = const()[name = tensor("q_111_strides_0"), val = tensor([1, 1])]; + tensor q_111_pad_0 = const()[name = tensor("q_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_111_dilations_0 = const()[name = tensor("q_111_dilations_0"), val = tensor([1, 1])]; + tensor q_111_groups_0 = const()[name = tensor("q_111_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731276608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732505472))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_111_cast_fp16 = conv(dilations = q_111_dilations_0, groups = q_111_groups_0, pad = q_111_pad_0, pad_type = q_111_pad_type_0, strides = q_111_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_229_cast_fp16)[name = tensor("q_111_cast_fp16")]; + tensor k_221_pad_type_0 = const()[name = tensor("k_221_pad_type_0"), val = tensor("valid")]; + tensor k_221_strides_0 = const()[name = tensor("k_221_strides_0"), val = tensor([1, 1])]; + tensor k_221_pad_0 = const()[name = tensor("k_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_221_dilations_0 = const()[name = tensor("k_221_dilations_0"), val = tensor([1, 1])]; + tensor k_221_groups_0 = const()[name = tensor("k_221_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732505664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(734471808))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_221_cast_fp16 = conv(dilations = k_221_dilations_0, groups = k_221_groups_0, pad = k_221_pad_0, pad_type = k_221_pad_type_0, strides = k_221_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_221_cast_fp16")]; + tensor v_111_pad_type_0 = const()[name = tensor("v_111_pad_type_0"), val = tensor("valid")]; + tensor v_111_strides_0 = const()[name = tensor("v_111_strides_0"), val = tensor([1, 1])]; + tensor v_111_pad_0 = const()[name = tensor("v_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_111_dilations_0 = const()[name = tensor("v_111_dilations_0"), val = tensor([1, 1])]; + tensor v_111_groups_0 = const()[name = tensor("v_111_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(734472000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736438144))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_111_cast_fp16 = conv(dilations = v_111_dilations_0, groups = v_111_groups_0, pad = v_111_pad_0, pad_type = v_111_pad_type_0, strides = v_111_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_111_cast_fp16")]; + tensor var_24376_begin_0 = const()[name = tensor("op_24376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24376_end_0 = const()[name = tensor("op_24376_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_24376_end_mask_0 = const()[name = tensor("op_24376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24376_cast_fp16 = slice_by_index(begin = var_24376_begin_0, end = var_24376_end_0, end_mask = var_24376_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24376_cast_fp16")]; + tensor var_24380_begin_0 = const()[name = tensor("op_24380_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24380_end_0 = const()[name = tensor("op_24380_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_24380_end_mask_0 = const()[name = tensor("op_24380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24380_cast_fp16 = slice_by_index(begin = var_24380_begin_0, end = var_24380_end_0, end_mask = var_24380_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24380_cast_fp16")]; + tensor var_24384_begin_0 = const()[name = tensor("op_24384_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24384_end_0 = const()[name = tensor("op_24384_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_24384_end_mask_0 = const()[name = tensor("op_24384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24384_cast_fp16 = slice_by_index(begin = var_24384_begin_0, end = var_24384_end_0, end_mask = var_24384_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24384_cast_fp16")]; + tensor var_24388_begin_0 = const()[name = tensor("op_24388_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24388_end_0 = const()[name = tensor("op_24388_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_24388_end_mask_0 = const()[name = tensor("op_24388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24388_cast_fp16 = slice_by_index(begin = var_24388_begin_0, end = var_24388_end_0, end_mask = var_24388_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24388_cast_fp16")]; + tensor var_24392_begin_0 = const()[name = tensor("op_24392_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24392_end_0 = const()[name = tensor("op_24392_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_24392_end_mask_0 = const()[name = tensor("op_24392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24392_cast_fp16 = slice_by_index(begin = var_24392_begin_0, end = var_24392_end_0, end_mask = var_24392_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24392_cast_fp16")]; + tensor var_24396_begin_0 = const()[name = tensor("op_24396_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24396_end_0 = const()[name = tensor("op_24396_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_24396_end_mask_0 = const()[name = tensor("op_24396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24396_cast_fp16 = slice_by_index(begin = var_24396_begin_0, end = var_24396_end_0, end_mask = var_24396_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24396_cast_fp16")]; + tensor var_24400_begin_0 = const()[name = tensor("op_24400_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24400_end_0 = const()[name = tensor("op_24400_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_24400_end_mask_0 = const()[name = tensor("op_24400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24400_cast_fp16 = slice_by_index(begin = var_24400_begin_0, end = var_24400_end_0, end_mask = var_24400_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24400_cast_fp16")]; + tensor var_24404_begin_0 = const()[name = tensor("op_24404_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24404_end_0 = const()[name = tensor("op_24404_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_24404_end_mask_0 = const()[name = tensor("op_24404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24404_cast_fp16 = slice_by_index(begin = var_24404_begin_0, end = var_24404_end_0, end_mask = var_24404_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24404_cast_fp16")]; + tensor var_24408_begin_0 = const()[name = tensor("op_24408_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24408_end_0 = const()[name = tensor("op_24408_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_24408_end_mask_0 = const()[name = tensor("op_24408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24408_cast_fp16 = slice_by_index(begin = var_24408_begin_0, end = var_24408_end_0, end_mask = var_24408_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24408_cast_fp16")]; + tensor var_24412_begin_0 = const()[name = tensor("op_24412_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24412_end_0 = const()[name = tensor("op_24412_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_24412_end_mask_0 = const()[name = tensor("op_24412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24412_cast_fp16 = slice_by_index(begin = var_24412_begin_0, end = var_24412_end_0, end_mask = var_24412_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24412_cast_fp16")]; + tensor var_24416_begin_0 = const()[name = tensor("op_24416_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24416_end_0 = const()[name = tensor("op_24416_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_24416_end_mask_0 = const()[name = tensor("op_24416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24416_cast_fp16 = slice_by_index(begin = var_24416_begin_0, end = var_24416_end_0, end_mask = var_24416_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24416_cast_fp16")]; + tensor var_24420_begin_0 = const()[name = tensor("op_24420_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24420_end_0 = const()[name = tensor("op_24420_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_24420_end_mask_0 = const()[name = tensor("op_24420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24420_cast_fp16 = slice_by_index(begin = var_24420_begin_0, end = var_24420_end_0, end_mask = var_24420_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24420_cast_fp16")]; + tensor var_24424_begin_0 = const()[name = tensor("op_24424_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24424_end_0 = const()[name = tensor("op_24424_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_24424_end_mask_0 = const()[name = tensor("op_24424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24424_cast_fp16 = slice_by_index(begin = var_24424_begin_0, end = var_24424_end_0, end_mask = var_24424_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24424_cast_fp16")]; + tensor var_24428_begin_0 = const()[name = tensor("op_24428_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24428_end_0 = const()[name = tensor("op_24428_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_24428_end_mask_0 = const()[name = tensor("op_24428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24428_cast_fp16 = slice_by_index(begin = var_24428_begin_0, end = var_24428_end_0, end_mask = var_24428_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24428_cast_fp16")]; + tensor var_24432_begin_0 = const()[name = tensor("op_24432_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24432_end_0 = const()[name = tensor("op_24432_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24432_end_mask_0 = const()[name = tensor("op_24432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24432_cast_fp16 = slice_by_index(begin = var_24432_begin_0, end = var_24432_end_0, end_mask = var_24432_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24432_cast_fp16")]; + tensor var_24436_begin_0 = const()[name = tensor("op_24436_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24436_end_0 = const()[name = tensor("op_24436_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24436_end_mask_0 = const()[name = tensor("op_24436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24436_cast_fp16 = slice_by_index(begin = var_24436_begin_0, end = var_24436_end_0, end_mask = var_24436_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24436_cast_fp16")]; + tensor var_24440_begin_0 = const()[name = tensor("op_24440_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24440_end_0 = const()[name = tensor("op_24440_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24440_end_mask_0 = const()[name = tensor("op_24440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24440_cast_fp16 = slice_by_index(begin = var_24440_begin_0, end = var_24440_end_0, end_mask = var_24440_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24440_cast_fp16")]; + tensor var_24444_begin_0 = const()[name = tensor("op_24444_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24444_end_0 = const()[name = tensor("op_24444_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24444_end_mask_0 = const()[name = tensor("op_24444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24444_cast_fp16 = slice_by_index(begin = var_24444_begin_0, end = var_24444_end_0, end_mask = var_24444_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24444_cast_fp16")]; + tensor var_24448_begin_0 = const()[name = tensor("op_24448_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24448_end_0 = const()[name = tensor("op_24448_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24448_end_mask_0 = const()[name = tensor("op_24448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24448_cast_fp16 = slice_by_index(begin = var_24448_begin_0, end = var_24448_end_0, end_mask = var_24448_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24448_cast_fp16")]; + tensor var_24452_begin_0 = const()[name = tensor("op_24452_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24452_end_0 = const()[name = tensor("op_24452_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24452_end_mask_0 = const()[name = tensor("op_24452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24452_cast_fp16 = slice_by_index(begin = var_24452_begin_0, end = var_24452_end_0, end_mask = var_24452_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24452_cast_fp16")]; + tensor k_223_perm_0 = const()[name = tensor("k_223_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_24459_begin_0 = const()[name = tensor("op_24459_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24459_end_0 = const()[name = tensor("op_24459_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_24459_end_mask_0 = const()[name = tensor("op_24459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_223_cast_fp16 = transpose(perm = k_223_perm_0, x = k_221_cast_fp16)[name = tensor("transpose_12")]; + tensor var_24459_cast_fp16 = slice_by_index(begin = var_24459_begin_0, end = var_24459_end_0, end_mask = var_24459_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24459_cast_fp16")]; + tensor var_24463_begin_0 = const()[name = tensor("op_24463_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_24463_end_0 = const()[name = tensor("op_24463_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_24463_end_mask_0 = const()[name = tensor("op_24463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24463_cast_fp16 = slice_by_index(begin = var_24463_begin_0, end = var_24463_end_0, end_mask = var_24463_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24463_cast_fp16")]; + tensor var_24467_begin_0 = const()[name = tensor("op_24467_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_24467_end_0 = const()[name = tensor("op_24467_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_24467_end_mask_0 = const()[name = tensor("op_24467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24467_cast_fp16 = slice_by_index(begin = var_24467_begin_0, end = var_24467_end_0, end_mask = var_24467_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24467_cast_fp16")]; + tensor var_24471_begin_0 = const()[name = tensor("op_24471_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_24471_end_0 = const()[name = tensor("op_24471_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_24471_end_mask_0 = const()[name = tensor("op_24471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24471_cast_fp16 = slice_by_index(begin = var_24471_begin_0, end = var_24471_end_0, end_mask = var_24471_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24471_cast_fp16")]; + tensor var_24475_begin_0 = const()[name = tensor("op_24475_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_24475_end_0 = const()[name = tensor("op_24475_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_24475_end_mask_0 = const()[name = tensor("op_24475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24475_cast_fp16 = slice_by_index(begin = var_24475_begin_0, end = var_24475_end_0, end_mask = var_24475_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24475_cast_fp16")]; + tensor var_24479_begin_0 = const()[name = tensor("op_24479_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_24479_end_0 = const()[name = tensor("op_24479_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_24479_end_mask_0 = const()[name = tensor("op_24479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24479_cast_fp16 = slice_by_index(begin = var_24479_begin_0, end = var_24479_end_0, end_mask = var_24479_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24479_cast_fp16")]; + tensor var_24483_begin_0 = const()[name = tensor("op_24483_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_24483_end_0 = const()[name = tensor("op_24483_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_24483_end_mask_0 = const()[name = tensor("op_24483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24483_cast_fp16 = slice_by_index(begin = var_24483_begin_0, end = var_24483_end_0, end_mask = var_24483_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24483_cast_fp16")]; + tensor var_24487_begin_0 = const()[name = tensor("op_24487_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_24487_end_0 = const()[name = tensor("op_24487_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_24487_end_mask_0 = const()[name = tensor("op_24487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24487_cast_fp16 = slice_by_index(begin = var_24487_begin_0, end = var_24487_end_0, end_mask = var_24487_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24487_cast_fp16")]; + tensor var_24491_begin_0 = const()[name = tensor("op_24491_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_24491_end_0 = const()[name = tensor("op_24491_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_24491_end_mask_0 = const()[name = tensor("op_24491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24491_cast_fp16 = slice_by_index(begin = var_24491_begin_0, end = var_24491_end_0, end_mask = var_24491_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24491_cast_fp16")]; + tensor var_24495_begin_0 = const()[name = tensor("op_24495_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_24495_end_0 = const()[name = tensor("op_24495_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_24495_end_mask_0 = const()[name = tensor("op_24495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24495_cast_fp16 = slice_by_index(begin = var_24495_begin_0, end = var_24495_end_0, end_mask = var_24495_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24495_cast_fp16")]; + tensor var_24499_begin_0 = const()[name = tensor("op_24499_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_24499_end_0 = const()[name = tensor("op_24499_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_24499_end_mask_0 = const()[name = tensor("op_24499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24499_cast_fp16 = slice_by_index(begin = var_24499_begin_0, end = var_24499_end_0, end_mask = var_24499_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24499_cast_fp16")]; + tensor var_24503_begin_0 = const()[name = tensor("op_24503_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_24503_end_0 = const()[name = tensor("op_24503_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_24503_end_mask_0 = const()[name = tensor("op_24503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24503_cast_fp16 = slice_by_index(begin = var_24503_begin_0, end = var_24503_end_0, end_mask = var_24503_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24503_cast_fp16")]; + tensor var_24507_begin_0 = const()[name = tensor("op_24507_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_24507_end_0 = const()[name = tensor("op_24507_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_24507_end_mask_0 = const()[name = tensor("op_24507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24507_cast_fp16 = slice_by_index(begin = var_24507_begin_0, end = var_24507_end_0, end_mask = var_24507_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24507_cast_fp16")]; + tensor var_24511_begin_0 = const()[name = tensor("op_24511_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_24511_end_0 = const()[name = tensor("op_24511_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_24511_end_mask_0 = const()[name = tensor("op_24511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24511_cast_fp16 = slice_by_index(begin = var_24511_begin_0, end = var_24511_end_0, end_mask = var_24511_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24511_cast_fp16")]; + tensor var_24515_begin_0 = const()[name = tensor("op_24515_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_24515_end_0 = const()[name = tensor("op_24515_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_24515_end_mask_0 = const()[name = tensor("op_24515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24515_cast_fp16 = slice_by_index(begin = var_24515_begin_0, end = var_24515_end_0, end_mask = var_24515_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24515_cast_fp16")]; + tensor var_24519_begin_0 = const()[name = tensor("op_24519_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_24519_end_0 = const()[name = tensor("op_24519_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_24519_end_mask_0 = const()[name = tensor("op_24519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24519_cast_fp16 = slice_by_index(begin = var_24519_begin_0, end = var_24519_end_0, end_mask = var_24519_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24519_cast_fp16")]; + tensor var_24523_begin_0 = const()[name = tensor("op_24523_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_24523_end_0 = const()[name = tensor("op_24523_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_24523_end_mask_0 = const()[name = tensor("op_24523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24523_cast_fp16 = slice_by_index(begin = var_24523_begin_0, end = var_24523_end_0, end_mask = var_24523_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24523_cast_fp16")]; + tensor var_24527_begin_0 = const()[name = tensor("op_24527_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_24527_end_0 = const()[name = tensor("op_24527_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_24527_end_mask_0 = const()[name = tensor("op_24527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24527_cast_fp16 = slice_by_index(begin = var_24527_begin_0, end = var_24527_end_0, end_mask = var_24527_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24527_cast_fp16")]; + tensor var_24531_begin_0 = const()[name = tensor("op_24531_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_24531_end_0 = const()[name = tensor("op_24531_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_24531_end_mask_0 = const()[name = tensor("op_24531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24531_cast_fp16 = slice_by_index(begin = var_24531_begin_0, end = var_24531_end_0, end_mask = var_24531_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24531_cast_fp16")]; + tensor var_24535_begin_0 = const()[name = tensor("op_24535_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_24535_end_0 = const()[name = tensor("op_24535_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_24535_end_mask_0 = const()[name = tensor("op_24535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24535_cast_fp16 = slice_by_index(begin = var_24535_begin_0, end = var_24535_end_0, end_mask = var_24535_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24535_cast_fp16")]; + tensor var_24537_begin_0 = const()[name = tensor("op_24537_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24537_end_0 = const()[name = tensor("op_24537_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_24537_end_mask_0 = const()[name = tensor("op_24537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24537_cast_fp16 = slice_by_index(begin = var_24537_begin_0, end = var_24537_end_0, end_mask = var_24537_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24537_cast_fp16")]; + tensor var_24541_begin_0 = const()[name = tensor("op_24541_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24541_end_0 = const()[name = tensor("op_24541_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_24541_end_mask_0 = const()[name = tensor("op_24541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24541_cast_fp16 = slice_by_index(begin = var_24541_begin_0, end = var_24541_end_0, end_mask = var_24541_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24541_cast_fp16")]; + tensor var_24545_begin_0 = const()[name = tensor("op_24545_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24545_end_0 = const()[name = tensor("op_24545_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_24545_end_mask_0 = const()[name = tensor("op_24545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24545_cast_fp16 = slice_by_index(begin = var_24545_begin_0, end = var_24545_end_0, end_mask = var_24545_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24545_cast_fp16")]; + tensor var_24549_begin_0 = const()[name = tensor("op_24549_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24549_end_0 = const()[name = tensor("op_24549_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_24549_end_mask_0 = const()[name = tensor("op_24549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24549_cast_fp16 = slice_by_index(begin = var_24549_begin_0, end = var_24549_end_0, end_mask = var_24549_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24549_cast_fp16")]; + tensor var_24553_begin_0 = const()[name = tensor("op_24553_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24553_end_0 = const()[name = tensor("op_24553_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_24553_end_mask_0 = const()[name = tensor("op_24553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24553_cast_fp16 = slice_by_index(begin = var_24553_begin_0, end = var_24553_end_0, end_mask = var_24553_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24553_cast_fp16")]; + tensor var_24557_begin_0 = const()[name = tensor("op_24557_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24557_end_0 = const()[name = tensor("op_24557_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_24557_end_mask_0 = const()[name = tensor("op_24557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24557_cast_fp16 = slice_by_index(begin = var_24557_begin_0, end = var_24557_end_0, end_mask = var_24557_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24557_cast_fp16")]; + tensor var_24561_begin_0 = const()[name = tensor("op_24561_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24561_end_0 = const()[name = tensor("op_24561_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_24561_end_mask_0 = const()[name = tensor("op_24561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24561_cast_fp16 = slice_by_index(begin = var_24561_begin_0, end = var_24561_end_0, end_mask = var_24561_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24561_cast_fp16")]; + tensor var_24565_begin_0 = const()[name = tensor("op_24565_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24565_end_0 = const()[name = tensor("op_24565_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_24565_end_mask_0 = const()[name = tensor("op_24565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24565_cast_fp16 = slice_by_index(begin = var_24565_begin_0, end = var_24565_end_0, end_mask = var_24565_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24565_cast_fp16")]; + tensor var_24569_begin_0 = const()[name = tensor("op_24569_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24569_end_0 = const()[name = tensor("op_24569_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_24569_end_mask_0 = const()[name = tensor("op_24569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24569_cast_fp16 = slice_by_index(begin = var_24569_begin_0, end = var_24569_end_0, end_mask = var_24569_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24569_cast_fp16")]; + tensor var_24573_begin_0 = const()[name = tensor("op_24573_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24573_end_0 = const()[name = tensor("op_24573_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_24573_end_mask_0 = const()[name = tensor("op_24573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24573_cast_fp16 = slice_by_index(begin = var_24573_begin_0, end = var_24573_end_0, end_mask = var_24573_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24573_cast_fp16")]; + tensor var_24577_begin_0 = const()[name = tensor("op_24577_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24577_end_0 = const()[name = tensor("op_24577_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_24577_end_mask_0 = const()[name = tensor("op_24577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24577_cast_fp16 = slice_by_index(begin = var_24577_begin_0, end = var_24577_end_0, end_mask = var_24577_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24577_cast_fp16")]; + tensor var_24581_begin_0 = const()[name = tensor("op_24581_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24581_end_0 = const()[name = tensor("op_24581_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_24581_end_mask_0 = const()[name = tensor("op_24581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24581_cast_fp16 = slice_by_index(begin = var_24581_begin_0, end = var_24581_end_0, end_mask = var_24581_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24581_cast_fp16")]; + tensor var_24585_begin_0 = const()[name = tensor("op_24585_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24585_end_0 = const()[name = tensor("op_24585_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_24585_end_mask_0 = const()[name = tensor("op_24585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24585_cast_fp16 = slice_by_index(begin = var_24585_begin_0, end = var_24585_end_0, end_mask = var_24585_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24585_cast_fp16")]; + tensor var_24589_begin_0 = const()[name = tensor("op_24589_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24589_end_0 = const()[name = tensor("op_24589_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_24589_end_mask_0 = const()[name = tensor("op_24589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24589_cast_fp16 = slice_by_index(begin = var_24589_begin_0, end = var_24589_end_0, end_mask = var_24589_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24589_cast_fp16")]; + tensor var_24593_begin_0 = const()[name = tensor("op_24593_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24593_end_0 = const()[name = tensor("op_24593_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_24593_end_mask_0 = const()[name = tensor("op_24593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24593_cast_fp16 = slice_by_index(begin = var_24593_begin_0, end = var_24593_end_0, end_mask = var_24593_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24593_cast_fp16")]; + tensor var_24597_begin_0 = const()[name = tensor("op_24597_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24597_end_0 = const()[name = tensor("op_24597_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_24597_end_mask_0 = const()[name = tensor("op_24597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24597_cast_fp16 = slice_by_index(begin = var_24597_begin_0, end = var_24597_end_0, end_mask = var_24597_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24597_cast_fp16")]; + tensor var_24601_begin_0 = const()[name = tensor("op_24601_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24601_end_0 = const()[name = tensor("op_24601_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_24601_end_mask_0 = const()[name = tensor("op_24601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24601_cast_fp16 = slice_by_index(begin = var_24601_begin_0, end = var_24601_end_0, end_mask = var_24601_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24601_cast_fp16")]; + tensor var_24605_begin_0 = const()[name = tensor("op_24605_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24605_end_0 = const()[name = tensor("op_24605_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_24605_end_mask_0 = const()[name = tensor("op_24605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24605_cast_fp16 = slice_by_index(begin = var_24605_begin_0, end = var_24605_end_0, end_mask = var_24605_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24605_cast_fp16")]; + tensor var_24609_begin_0 = const()[name = tensor("op_24609_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24609_end_0 = const()[name = tensor("op_24609_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_24609_end_mask_0 = const()[name = tensor("op_24609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24609_cast_fp16 = slice_by_index(begin = var_24609_begin_0, end = var_24609_end_0, end_mask = var_24609_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24609_cast_fp16")]; + tensor var_24613_begin_0 = const()[name = tensor("op_24613_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24613_end_0 = const()[name = tensor("op_24613_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_24613_end_mask_0 = const()[name = tensor("op_24613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24613_cast_fp16 = slice_by_index(begin = var_24613_begin_0, end = var_24613_end_0, end_mask = var_24613_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24613_cast_fp16")]; + tensor var_24617_equation_0 = const()[name = tensor("op_24617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24617_cast_fp16 = einsum(equation = var_24617_equation_0, values = (var_24459_cast_fp16, var_24376_cast_fp16))[name = tensor("op_24617_cast_fp16")]; + tensor var_24618_to_fp16 = const()[name = tensor("op_24618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2041_cast_fp16 = mul(x = var_24617_cast_fp16, y = var_24618_to_fp16)[name = tensor("aw_2041_cast_fp16")]; + tensor var_24621_equation_0 = const()[name = tensor("op_24621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24621_cast_fp16 = einsum(equation = var_24621_equation_0, values = (var_24463_cast_fp16, var_24380_cast_fp16))[name = tensor("op_24621_cast_fp16")]; + tensor var_24622_to_fp16 = const()[name = tensor("op_24622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2043_cast_fp16 = mul(x = var_24621_cast_fp16, y = var_24622_to_fp16)[name = tensor("aw_2043_cast_fp16")]; + tensor var_24625_equation_0 = const()[name = tensor("op_24625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24625_cast_fp16 = einsum(equation = var_24625_equation_0, values = (var_24467_cast_fp16, var_24384_cast_fp16))[name = tensor("op_24625_cast_fp16")]; + tensor var_24626_to_fp16 = const()[name = tensor("op_24626_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2045_cast_fp16 = mul(x = var_24625_cast_fp16, y = var_24626_to_fp16)[name = tensor("aw_2045_cast_fp16")]; + tensor var_24629_equation_0 = const()[name = tensor("op_24629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24629_cast_fp16 = einsum(equation = var_24629_equation_0, values = (var_24471_cast_fp16, var_24388_cast_fp16))[name = tensor("op_24629_cast_fp16")]; + tensor var_24630_to_fp16 = const()[name = tensor("op_24630_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2047_cast_fp16 = mul(x = var_24629_cast_fp16, y = var_24630_to_fp16)[name = tensor("aw_2047_cast_fp16")]; + tensor var_24633_equation_0 = const()[name = tensor("op_24633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24633_cast_fp16 = einsum(equation = var_24633_equation_0, values = (var_24475_cast_fp16, var_24392_cast_fp16))[name = tensor("op_24633_cast_fp16")]; + tensor var_24634_to_fp16 = const()[name = tensor("op_24634_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2049_cast_fp16 = mul(x = var_24633_cast_fp16, y = var_24634_to_fp16)[name = tensor("aw_2049_cast_fp16")]; + tensor var_24637_equation_0 = const()[name = tensor("op_24637_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24637_cast_fp16 = einsum(equation = var_24637_equation_0, values = (var_24479_cast_fp16, var_24396_cast_fp16))[name = tensor("op_24637_cast_fp16")]; + tensor var_24638_to_fp16 = const()[name = tensor("op_24638_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2051_cast_fp16 = mul(x = var_24637_cast_fp16, y = var_24638_to_fp16)[name = tensor("aw_2051_cast_fp16")]; + tensor var_24641_equation_0 = const()[name = tensor("op_24641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24641_cast_fp16 = einsum(equation = var_24641_equation_0, values = (var_24483_cast_fp16, var_24400_cast_fp16))[name = tensor("op_24641_cast_fp16")]; + tensor var_24642_to_fp16 = const()[name = tensor("op_24642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2053_cast_fp16 = mul(x = var_24641_cast_fp16, y = var_24642_to_fp16)[name = tensor("aw_2053_cast_fp16")]; + tensor var_24645_equation_0 = const()[name = tensor("op_24645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24645_cast_fp16 = einsum(equation = var_24645_equation_0, values = (var_24487_cast_fp16, var_24404_cast_fp16))[name = tensor("op_24645_cast_fp16")]; + tensor var_24646_to_fp16 = const()[name = tensor("op_24646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2055_cast_fp16 = mul(x = var_24645_cast_fp16, y = var_24646_to_fp16)[name = tensor("aw_2055_cast_fp16")]; + tensor var_24649_equation_0 = const()[name = tensor("op_24649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24649_cast_fp16 = einsum(equation = var_24649_equation_0, values = (var_24491_cast_fp16, var_24408_cast_fp16))[name = tensor("op_24649_cast_fp16")]; + tensor var_24650_to_fp16 = const()[name = tensor("op_24650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2057_cast_fp16 = mul(x = var_24649_cast_fp16, y = var_24650_to_fp16)[name = tensor("aw_2057_cast_fp16")]; + tensor var_24653_equation_0 = const()[name = tensor("op_24653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24653_cast_fp16 = einsum(equation = var_24653_equation_0, values = (var_24495_cast_fp16, var_24412_cast_fp16))[name = tensor("op_24653_cast_fp16")]; + tensor var_24654_to_fp16 = const()[name = tensor("op_24654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2059_cast_fp16 = mul(x = var_24653_cast_fp16, y = var_24654_to_fp16)[name = tensor("aw_2059_cast_fp16")]; + tensor var_24657_equation_0 = const()[name = tensor("op_24657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24657_cast_fp16 = einsum(equation = var_24657_equation_0, values = (var_24499_cast_fp16, var_24416_cast_fp16))[name = tensor("op_24657_cast_fp16")]; + tensor var_24658_to_fp16 = const()[name = tensor("op_24658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2061_cast_fp16 = mul(x = var_24657_cast_fp16, y = var_24658_to_fp16)[name = tensor("aw_2061_cast_fp16")]; + tensor var_24661_equation_0 = const()[name = tensor("op_24661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24661_cast_fp16 = einsum(equation = var_24661_equation_0, values = (var_24503_cast_fp16, var_24420_cast_fp16))[name = tensor("op_24661_cast_fp16")]; + tensor var_24662_to_fp16 = const()[name = tensor("op_24662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2063_cast_fp16 = mul(x = var_24661_cast_fp16, y = var_24662_to_fp16)[name = tensor("aw_2063_cast_fp16")]; + tensor var_24665_equation_0 = const()[name = tensor("op_24665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24665_cast_fp16 = einsum(equation = var_24665_equation_0, values = (var_24507_cast_fp16, var_24424_cast_fp16))[name = tensor("op_24665_cast_fp16")]; + tensor var_24666_to_fp16 = const()[name = tensor("op_24666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2065_cast_fp16 = mul(x = var_24665_cast_fp16, y = var_24666_to_fp16)[name = tensor("aw_2065_cast_fp16")]; + tensor var_24669_equation_0 = const()[name = tensor("op_24669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24669_cast_fp16 = einsum(equation = var_24669_equation_0, values = (var_24511_cast_fp16, var_24428_cast_fp16))[name = tensor("op_24669_cast_fp16")]; + tensor var_24670_to_fp16 = const()[name = tensor("op_24670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2067_cast_fp16 = mul(x = var_24669_cast_fp16, y = var_24670_to_fp16)[name = tensor("aw_2067_cast_fp16")]; + tensor var_24673_equation_0 = const()[name = tensor("op_24673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24673_cast_fp16 = einsum(equation = var_24673_equation_0, values = (var_24515_cast_fp16, var_24432_cast_fp16))[name = tensor("op_24673_cast_fp16")]; + tensor var_24674_to_fp16 = const()[name = tensor("op_24674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2069_cast_fp16 = mul(x = var_24673_cast_fp16, y = var_24674_to_fp16)[name = tensor("aw_2069_cast_fp16")]; + tensor var_24677_equation_0 = const()[name = tensor("op_24677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24677_cast_fp16 = einsum(equation = var_24677_equation_0, values = (var_24519_cast_fp16, var_24436_cast_fp16))[name = tensor("op_24677_cast_fp16")]; + tensor var_24678_to_fp16 = const()[name = tensor("op_24678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2071_cast_fp16 = mul(x = var_24677_cast_fp16, y = var_24678_to_fp16)[name = tensor("aw_2071_cast_fp16")]; + tensor var_24681_equation_0 = const()[name = tensor("op_24681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24681_cast_fp16 = einsum(equation = var_24681_equation_0, values = (var_24523_cast_fp16, var_24440_cast_fp16))[name = tensor("op_24681_cast_fp16")]; + tensor var_24682_to_fp16 = const()[name = tensor("op_24682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2073_cast_fp16 = mul(x = var_24681_cast_fp16, y = var_24682_to_fp16)[name = tensor("aw_2073_cast_fp16")]; + tensor var_24685_equation_0 = const()[name = tensor("op_24685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24685_cast_fp16 = einsum(equation = var_24685_equation_0, values = (var_24527_cast_fp16, var_24444_cast_fp16))[name = tensor("op_24685_cast_fp16")]; + tensor var_24686_to_fp16 = const()[name = tensor("op_24686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2075_cast_fp16 = mul(x = var_24685_cast_fp16, y = var_24686_to_fp16)[name = tensor("aw_2075_cast_fp16")]; + tensor var_24689_equation_0 = const()[name = tensor("op_24689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24689_cast_fp16 = einsum(equation = var_24689_equation_0, values = (var_24531_cast_fp16, var_24448_cast_fp16))[name = tensor("op_24689_cast_fp16")]; + tensor var_24690_to_fp16 = const()[name = tensor("op_24690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2077_cast_fp16 = mul(x = var_24689_cast_fp16, y = var_24690_to_fp16)[name = tensor("aw_2077_cast_fp16")]; + tensor var_24693_equation_0 = const()[name = tensor("op_24693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24693_cast_fp16 = einsum(equation = var_24693_equation_0, values = (var_24535_cast_fp16, var_24452_cast_fp16))[name = tensor("op_24693_cast_fp16")]; + tensor var_24694_to_fp16 = const()[name = tensor("op_24694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2079_cast_fp16 = mul(x = var_24693_cast_fp16, y = var_24694_to_fp16)[name = tensor("aw_2079_cast_fp16")]; + tensor var_24696_cast_fp16 = softmax(axis = var_21077, x = aw_2041_cast_fp16)[name = tensor("op_24696_cast_fp16")]; + tensor var_24697_cast_fp16 = softmax(axis = var_21077, x = aw_2043_cast_fp16)[name = tensor("op_24697_cast_fp16")]; + tensor var_24698_cast_fp16 = softmax(axis = var_21077, x = aw_2045_cast_fp16)[name = tensor("op_24698_cast_fp16")]; + tensor var_24699_cast_fp16 = softmax(axis = var_21077, x = aw_2047_cast_fp16)[name = tensor("op_24699_cast_fp16")]; + tensor var_24700_cast_fp16 = softmax(axis = var_21077, x = aw_2049_cast_fp16)[name = tensor("op_24700_cast_fp16")]; + tensor var_24701_cast_fp16 = softmax(axis = var_21077, x = aw_2051_cast_fp16)[name = tensor("op_24701_cast_fp16")]; + tensor var_24702_cast_fp16 = softmax(axis = var_21077, x = aw_2053_cast_fp16)[name = tensor("op_24702_cast_fp16")]; + tensor var_24703_cast_fp16 = softmax(axis = var_21077, x = aw_2055_cast_fp16)[name = tensor("op_24703_cast_fp16")]; + tensor var_24704_cast_fp16 = softmax(axis = var_21077, x = aw_2057_cast_fp16)[name = tensor("op_24704_cast_fp16")]; + tensor var_24705_cast_fp16 = softmax(axis = var_21077, x = aw_2059_cast_fp16)[name = tensor("op_24705_cast_fp16")]; + tensor var_24706_cast_fp16 = softmax(axis = var_21077, x = aw_2061_cast_fp16)[name = tensor("op_24706_cast_fp16")]; + tensor var_24707_cast_fp16 = softmax(axis = var_21077, x = aw_2063_cast_fp16)[name = tensor("op_24707_cast_fp16")]; + tensor var_24708_cast_fp16 = softmax(axis = var_21077, x = aw_2065_cast_fp16)[name = tensor("op_24708_cast_fp16")]; + tensor var_24709_cast_fp16 = softmax(axis = var_21077, x = aw_2067_cast_fp16)[name = tensor("op_24709_cast_fp16")]; + tensor var_24710_cast_fp16 = softmax(axis = var_21077, x = aw_2069_cast_fp16)[name = tensor("op_24710_cast_fp16")]; + tensor var_24711_cast_fp16 = softmax(axis = var_21077, x = aw_2071_cast_fp16)[name = tensor("op_24711_cast_fp16")]; + tensor var_24712_cast_fp16 = softmax(axis = var_21077, x = aw_2073_cast_fp16)[name = tensor("op_24712_cast_fp16")]; + tensor var_24713_cast_fp16 = softmax(axis = var_21077, x = aw_2075_cast_fp16)[name = tensor("op_24713_cast_fp16")]; + tensor var_24714_cast_fp16 = softmax(axis = var_21077, x = aw_2077_cast_fp16)[name = tensor("op_24714_cast_fp16")]; + tensor var_24715_cast_fp16 = softmax(axis = var_21077, x = aw_2079_cast_fp16)[name = tensor("op_24715_cast_fp16")]; + tensor var_24717_equation_0 = const()[name = tensor("op_24717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24717_cast_fp16 = einsum(equation = var_24717_equation_0, values = (var_24537_cast_fp16, var_24696_cast_fp16))[name = tensor("op_24717_cast_fp16")]; + tensor var_24719_equation_0 = const()[name = tensor("op_24719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24719_cast_fp16 = einsum(equation = var_24719_equation_0, values = (var_24541_cast_fp16, var_24697_cast_fp16))[name = tensor("op_24719_cast_fp16")]; + tensor var_24721_equation_0 = const()[name = tensor("op_24721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24721_cast_fp16 = einsum(equation = var_24721_equation_0, values = (var_24545_cast_fp16, var_24698_cast_fp16))[name = tensor("op_24721_cast_fp16")]; + tensor var_24723_equation_0 = const()[name = tensor("op_24723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24723_cast_fp16 = einsum(equation = var_24723_equation_0, values = (var_24549_cast_fp16, var_24699_cast_fp16))[name = tensor("op_24723_cast_fp16")]; + tensor var_24725_equation_0 = const()[name = tensor("op_24725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24725_cast_fp16 = einsum(equation = var_24725_equation_0, values = (var_24553_cast_fp16, var_24700_cast_fp16))[name = tensor("op_24725_cast_fp16")]; + tensor var_24727_equation_0 = const()[name = tensor("op_24727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24727_cast_fp16 = einsum(equation = var_24727_equation_0, values = (var_24557_cast_fp16, var_24701_cast_fp16))[name = tensor("op_24727_cast_fp16")]; + tensor var_24729_equation_0 = const()[name = tensor("op_24729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24729_cast_fp16 = einsum(equation = var_24729_equation_0, values = (var_24561_cast_fp16, var_24702_cast_fp16))[name = tensor("op_24729_cast_fp16")]; + tensor var_24731_equation_0 = const()[name = tensor("op_24731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24731_cast_fp16 = einsum(equation = var_24731_equation_0, values = (var_24565_cast_fp16, var_24703_cast_fp16))[name = tensor("op_24731_cast_fp16")]; + tensor var_24733_equation_0 = const()[name = tensor("op_24733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24733_cast_fp16 = einsum(equation = var_24733_equation_0, values = (var_24569_cast_fp16, var_24704_cast_fp16))[name = tensor("op_24733_cast_fp16")]; + tensor var_24735_equation_0 = const()[name = tensor("op_24735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24735_cast_fp16 = einsum(equation = var_24735_equation_0, values = (var_24573_cast_fp16, var_24705_cast_fp16))[name = tensor("op_24735_cast_fp16")]; + tensor var_24737_equation_0 = const()[name = tensor("op_24737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24737_cast_fp16 = einsum(equation = var_24737_equation_0, values = (var_24577_cast_fp16, var_24706_cast_fp16))[name = tensor("op_24737_cast_fp16")]; + tensor var_24739_equation_0 = const()[name = tensor("op_24739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24739_cast_fp16 = einsum(equation = var_24739_equation_0, values = (var_24581_cast_fp16, var_24707_cast_fp16))[name = tensor("op_24739_cast_fp16")]; + tensor var_24741_equation_0 = const()[name = tensor("op_24741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24741_cast_fp16 = einsum(equation = var_24741_equation_0, values = (var_24585_cast_fp16, var_24708_cast_fp16))[name = tensor("op_24741_cast_fp16")]; + tensor var_24743_equation_0 = const()[name = tensor("op_24743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24743_cast_fp16 = einsum(equation = var_24743_equation_0, values = (var_24589_cast_fp16, var_24709_cast_fp16))[name = tensor("op_24743_cast_fp16")]; + tensor var_24745_equation_0 = const()[name = tensor("op_24745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24745_cast_fp16 = einsum(equation = var_24745_equation_0, values = (var_24593_cast_fp16, var_24710_cast_fp16))[name = tensor("op_24745_cast_fp16")]; + tensor var_24747_equation_0 = const()[name = tensor("op_24747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24747_cast_fp16 = einsum(equation = var_24747_equation_0, values = (var_24597_cast_fp16, var_24711_cast_fp16))[name = tensor("op_24747_cast_fp16")]; + tensor var_24749_equation_0 = const()[name = tensor("op_24749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24749_cast_fp16 = einsum(equation = var_24749_equation_0, values = (var_24601_cast_fp16, var_24712_cast_fp16))[name = tensor("op_24749_cast_fp16")]; + tensor var_24751_equation_0 = const()[name = tensor("op_24751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24751_cast_fp16 = einsum(equation = var_24751_equation_0, values = (var_24605_cast_fp16, var_24713_cast_fp16))[name = tensor("op_24751_cast_fp16")]; + tensor var_24753_equation_0 = const()[name = tensor("op_24753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24753_cast_fp16 = einsum(equation = var_24753_equation_0, values = (var_24609_cast_fp16, var_24714_cast_fp16))[name = tensor("op_24753_cast_fp16")]; + tensor var_24755_equation_0 = const()[name = tensor("op_24755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24755_cast_fp16 = einsum(equation = var_24755_equation_0, values = (var_24613_cast_fp16, var_24715_cast_fp16))[name = tensor("op_24755_cast_fp16")]; + tensor input_353_interleave_0 = const()[name = tensor("input_353_interleave_0"), val = tensor(false)]; + tensor input_353_cast_fp16 = concat(axis = var_21077, interleave = input_353_interleave_0, values = (var_24717_cast_fp16, var_24719_cast_fp16, var_24721_cast_fp16, var_24723_cast_fp16, var_24725_cast_fp16, var_24727_cast_fp16, var_24729_cast_fp16, var_24731_cast_fp16, var_24733_cast_fp16, var_24735_cast_fp16, var_24737_cast_fp16, var_24739_cast_fp16, var_24741_cast_fp16, var_24743_cast_fp16, var_24745_cast_fp16, var_24747_cast_fp16, var_24749_cast_fp16, var_24751_cast_fp16, var_24753_cast_fp16, var_24755_cast_fp16))[name = tensor("input_353_cast_fp16")]; + tensor var_24765_pad_type_0 = const()[name = tensor("op_24765_pad_type_0"), val = tensor("valid")]; + tensor var_24765_strides_0 = const()[name = tensor("op_24765_strides_0"), val = tensor([1, 1])]; + tensor var_24765_pad_0 = const()[name = tensor("op_24765_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24765_dilations_0 = const()[name = tensor("op_24765_dilations_0"), val = tensor([1, 1])]; + tensor var_24765_groups_0 = const()[name = tensor("op_24765_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736438336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737667200))), name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737667392)))]; + tensor var_24765_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_24765_dilations_0, groups = var_24765_groups_0, pad = var_24765_pad_0, pad_type = var_24765_pad_type_0, strides = var_24765_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized, x = input_353_cast_fp16)[name = tensor("op_24765_cast_fp16")]; + tensor inputs_167_cast_fp16 = add(x = var_24765_cast_fp16, y = inputs_165_cast_fp16)[name = tensor("inputs_167_cast_fp16")]; + tensor input_355_axes_0 = const()[name = tensor("input_355_axes_0"), val = tensor([1])]; + tensor input_355_gamma_0_to_fp16 = const()[name = tensor("input_355_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737670016)))]; + tensor input_355_beta_0_to_fp16 = const()[name = tensor("input_355_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737672640)))]; + tensor var_24775_to_fp16 = const()[name = tensor("op_24775_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_355_cast_fp16 = layer_norm(axes = input_355_axes_0, beta = input_355_beta_0_to_fp16, epsilon = var_24775_to_fp16, gamma = input_355_gamma_0_to_fp16, x = inputs_167_cast_fp16)[name = tensor("input_355_cast_fp16")]; + tensor var_24795_pad_type_0 = const()[name = tensor("op_24795_pad_type_0"), val = tensor("valid")]; + tensor var_24795_strides_0 = const()[name = tensor("op_24795_strides_0"), val = tensor([1, 1])]; + tensor var_24795_pad_0 = const()[name = tensor("op_24795_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24795_dilations_0 = const()[name = tensor("op_24795_dilations_0"), val = tensor([1, 1])]; + tensor var_24795_groups_0 = const()[name = tensor("op_24795_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737675264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(747505728))), name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(747505920)))]; + tensor var_24795_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_24795_dilations_0, groups = var_24795_groups_0, pad = var_24795_pad_0, pad_type = var_24795_pad_type_0, strides = var_24795_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized, x = input_355_cast_fp16)[name = tensor("op_24795_cast_fp16")]; + tensor var_24796_split_sizes_0 = const()[name = tensor("op_24796_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_24796_axis_0 = const()[name = tensor("op_24796_axis_0"), val = tensor(1)]; + tensor var_24796_cast_fp16_0, tensor var_24796_cast_fp16_1 = split(axis = var_24796_axis_0, split_sizes = var_24796_split_sizes_0, x = var_24795_cast_fp16)[name = tensor("op_24796_cast_fp16")]; + tensor var_24798_mode_0 = const()[name = tensor("op_24798_mode_0"), val = tensor("EXACT")]; + tensor var_24798_cast_fp16 = gelu(mode = var_24798_mode_0, x = var_24796_cast_fp16_1)[name = tensor("op_24798_cast_fp16")]; + tensor input_357_cast_fp16 = mul(x = var_24796_cast_fp16_0, y = var_24798_cast_fp16)[name = tensor("input_357_cast_fp16")]; + tensor var_24806_pad_type_0 = const()[name = tensor("op_24806_pad_type_0"), val = tensor("valid")]; + tensor var_24806_strides_0 = const()[name = tensor("op_24806_strides_0"), val = tensor([1, 1])]; + tensor var_24806_pad_0 = const()[name = tensor("op_24806_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24806_dilations_0 = const()[name = tensor("op_24806_dilations_0"), val = tensor([1, 1])]; + tensor var_24806_groups_0 = const()[name = tensor("op_24806_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(747526464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752441728))), name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752441920)))]; + tensor var_24806_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_24806_dilations_0, groups = var_24806_groups_0, pad = var_24806_pad_0, pad_type = var_24806_pad_type_0, strides = var_24806_strides_0, weight = mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized, x = input_357_cast_fp16)[name = tensor("op_24806_cast_fp16")]; + tensor inputs_169_cast_fp16 = add(x = var_24806_cast_fp16, y = inputs_167_cast_fp16)[name = tensor("inputs_169_cast_fp16")]; + tensor hidden_states_233_axes_0 = const()[name = tensor("hidden_states_233_axes_0"), val = tensor([1])]; + tensor hidden_states_233_gamma_0_to_fp16 = const()[name = tensor("hidden_states_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752444544)))]; + tensor hidden_states_233_beta_0_to_fp16 = const()[name = tensor("hidden_states_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752447168)))]; + tensor var_24822_to_fp16 = const()[name = tensor("op_24822_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_233_cast_fp16 = layer_norm(axes = hidden_states_233_axes_0, beta = hidden_states_233_beta_0_to_fp16, epsilon = var_24822_to_fp16, gamma = hidden_states_233_gamma_0_to_fp16, x = inputs_169_cast_fp16)[name = tensor("hidden_states_233_cast_fp16")]; + tensor q_113_pad_type_0 = const()[name = tensor("q_113_pad_type_0"), val = tensor("valid")]; + tensor q_113_strides_0 = const()[name = tensor("q_113_strides_0"), val = tensor([1, 1])]; + tensor q_113_pad_0 = const()[name = tensor("q_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_113_dilations_0 = const()[name = tensor("q_113_dilations_0"), val = tensor([1, 1])]; + tensor q_113_groups_0 = const()[name = tensor("q_113_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752449792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(753678656))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_113_cast_fp16 = conv(dilations = q_113_dilations_0, groups = q_113_groups_0, pad = q_113_pad_0, pad_type = q_113_pad_type_0, strides = q_113_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_233_cast_fp16)[name = tensor("q_113_cast_fp16")]; + tensor k_225_pad_type_0 = const()[name = tensor("k_225_pad_type_0"), val = tensor("valid")]; + tensor k_225_strides_0 = const()[name = tensor("k_225_strides_0"), val = tensor([1, 1])]; + tensor k_225_pad_0 = const()[name = tensor("k_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_225_dilations_0 = const()[name = tensor("k_225_dilations_0"), val = tensor([1, 1])]; + tensor k_225_groups_0 = const()[name = tensor("k_225_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(753678848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(754907712))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_225_cast_fp16 = conv(dilations = k_225_dilations_0, groups = k_225_groups_0, pad = k_225_pad_0, pad_type = k_225_pad_type_0, strides = k_225_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_233_cast_fp16)[name = tensor("k_225_cast_fp16")]; + tensor v_113_pad_type_0 = const()[name = tensor("v_113_pad_type_0"), val = tensor("valid")]; + tensor v_113_strides_0 = const()[name = tensor("v_113_strides_0"), val = tensor([1, 1])]; + tensor v_113_pad_0 = const()[name = tensor("v_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_113_dilations_0 = const()[name = tensor("v_113_dilations_0"), val = tensor([1, 1])]; + tensor v_113_groups_0 = const()[name = tensor("v_113_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(754907904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(756136768))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_113_cast_fp16 = conv(dilations = v_113_dilations_0, groups = v_113_groups_0, pad = v_113_pad_0, pad_type = v_113_pad_type_0, strides = v_113_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_233_cast_fp16)[name = tensor("v_113_cast_fp16")]; + tensor var_24855_begin_0 = const()[name = tensor("op_24855_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24855_end_0 = const()[name = tensor("op_24855_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_24855_end_mask_0 = const()[name = tensor("op_24855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24855_cast_fp16 = slice_by_index(begin = var_24855_begin_0, end = var_24855_end_0, end_mask = var_24855_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24855_cast_fp16")]; + tensor var_24859_begin_0 = const()[name = tensor("op_24859_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24859_end_0 = const()[name = tensor("op_24859_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_24859_end_mask_0 = const()[name = tensor("op_24859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24859_cast_fp16 = slice_by_index(begin = var_24859_begin_0, end = var_24859_end_0, end_mask = var_24859_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24859_cast_fp16")]; + tensor var_24863_begin_0 = const()[name = tensor("op_24863_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24863_end_0 = const()[name = tensor("op_24863_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_24863_end_mask_0 = const()[name = tensor("op_24863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24863_cast_fp16 = slice_by_index(begin = var_24863_begin_0, end = var_24863_end_0, end_mask = var_24863_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24863_cast_fp16")]; + tensor var_24867_begin_0 = const()[name = tensor("op_24867_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24867_end_0 = const()[name = tensor("op_24867_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_24867_end_mask_0 = const()[name = tensor("op_24867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24867_cast_fp16 = slice_by_index(begin = var_24867_begin_0, end = var_24867_end_0, end_mask = var_24867_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24867_cast_fp16")]; + tensor var_24871_begin_0 = const()[name = tensor("op_24871_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24871_end_0 = const()[name = tensor("op_24871_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_24871_end_mask_0 = const()[name = tensor("op_24871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24871_cast_fp16 = slice_by_index(begin = var_24871_begin_0, end = var_24871_end_0, end_mask = var_24871_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24871_cast_fp16")]; + tensor var_24875_begin_0 = const()[name = tensor("op_24875_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24875_end_0 = const()[name = tensor("op_24875_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_24875_end_mask_0 = const()[name = tensor("op_24875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24875_cast_fp16 = slice_by_index(begin = var_24875_begin_0, end = var_24875_end_0, end_mask = var_24875_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24875_cast_fp16")]; + tensor var_24879_begin_0 = const()[name = tensor("op_24879_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24879_end_0 = const()[name = tensor("op_24879_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_24879_end_mask_0 = const()[name = tensor("op_24879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24879_cast_fp16 = slice_by_index(begin = var_24879_begin_0, end = var_24879_end_0, end_mask = var_24879_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24879_cast_fp16")]; + tensor var_24883_begin_0 = const()[name = tensor("op_24883_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24883_end_0 = const()[name = tensor("op_24883_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_24883_end_mask_0 = const()[name = tensor("op_24883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24883_cast_fp16 = slice_by_index(begin = var_24883_begin_0, end = var_24883_end_0, end_mask = var_24883_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24883_cast_fp16")]; + tensor var_24887_begin_0 = const()[name = tensor("op_24887_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24887_end_0 = const()[name = tensor("op_24887_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_24887_end_mask_0 = const()[name = tensor("op_24887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24887_cast_fp16 = slice_by_index(begin = var_24887_begin_0, end = var_24887_end_0, end_mask = var_24887_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24887_cast_fp16")]; + tensor var_24891_begin_0 = const()[name = tensor("op_24891_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24891_end_0 = const()[name = tensor("op_24891_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_24891_end_mask_0 = const()[name = tensor("op_24891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24891_cast_fp16 = slice_by_index(begin = var_24891_begin_0, end = var_24891_end_0, end_mask = var_24891_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24891_cast_fp16")]; + tensor var_24895_begin_0 = const()[name = tensor("op_24895_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24895_end_0 = const()[name = tensor("op_24895_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_24895_end_mask_0 = const()[name = tensor("op_24895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24895_cast_fp16 = slice_by_index(begin = var_24895_begin_0, end = var_24895_end_0, end_mask = var_24895_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24895_cast_fp16")]; + tensor var_24899_begin_0 = const()[name = tensor("op_24899_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24899_end_0 = const()[name = tensor("op_24899_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_24899_end_mask_0 = const()[name = tensor("op_24899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24899_cast_fp16 = slice_by_index(begin = var_24899_begin_0, end = var_24899_end_0, end_mask = var_24899_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24899_cast_fp16")]; + tensor var_24903_begin_0 = const()[name = tensor("op_24903_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24903_end_0 = const()[name = tensor("op_24903_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_24903_end_mask_0 = const()[name = tensor("op_24903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24903_cast_fp16 = slice_by_index(begin = var_24903_begin_0, end = var_24903_end_0, end_mask = var_24903_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24903_cast_fp16")]; + tensor var_24907_begin_0 = const()[name = tensor("op_24907_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24907_end_0 = const()[name = tensor("op_24907_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_24907_end_mask_0 = const()[name = tensor("op_24907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24907_cast_fp16 = slice_by_index(begin = var_24907_begin_0, end = var_24907_end_0, end_mask = var_24907_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24907_cast_fp16")]; + tensor var_24911_begin_0 = const()[name = tensor("op_24911_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24911_end_0 = const()[name = tensor("op_24911_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24911_end_mask_0 = const()[name = tensor("op_24911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24911_cast_fp16 = slice_by_index(begin = var_24911_begin_0, end = var_24911_end_0, end_mask = var_24911_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24911_cast_fp16")]; + tensor var_24915_begin_0 = const()[name = tensor("op_24915_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24915_end_0 = const()[name = tensor("op_24915_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24915_end_mask_0 = const()[name = tensor("op_24915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24915_cast_fp16 = slice_by_index(begin = var_24915_begin_0, end = var_24915_end_0, end_mask = var_24915_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24915_cast_fp16")]; + tensor var_24919_begin_0 = const()[name = tensor("op_24919_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24919_end_0 = const()[name = tensor("op_24919_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24919_end_mask_0 = const()[name = tensor("op_24919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24919_cast_fp16 = slice_by_index(begin = var_24919_begin_0, end = var_24919_end_0, end_mask = var_24919_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24919_cast_fp16")]; + tensor var_24923_begin_0 = const()[name = tensor("op_24923_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24923_end_0 = const()[name = tensor("op_24923_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24923_end_mask_0 = const()[name = tensor("op_24923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24923_cast_fp16 = slice_by_index(begin = var_24923_begin_0, end = var_24923_end_0, end_mask = var_24923_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24923_cast_fp16")]; + tensor var_24927_begin_0 = const()[name = tensor("op_24927_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24927_end_0 = const()[name = tensor("op_24927_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24927_end_mask_0 = const()[name = tensor("op_24927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24927_cast_fp16 = slice_by_index(begin = var_24927_begin_0, end = var_24927_end_0, end_mask = var_24927_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24927_cast_fp16")]; + tensor var_24931_begin_0 = const()[name = tensor("op_24931_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24931_end_0 = const()[name = tensor("op_24931_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24931_end_mask_0 = const()[name = tensor("op_24931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24931_cast_fp16 = slice_by_index(begin = var_24931_begin_0, end = var_24931_end_0, end_mask = var_24931_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24931_cast_fp16")]; + tensor k_227_perm_0 = const()[name = tensor("k_227_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_24938_begin_0 = const()[name = tensor("op_24938_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24938_end_0 = const()[name = tensor("op_24938_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_24938_end_mask_0 = const()[name = tensor("op_24938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_227_cast_fp16 = transpose(perm = k_227_perm_0, x = k_225_cast_fp16)[name = tensor("transpose_11")]; + tensor var_24938_cast_fp16 = slice_by_index(begin = var_24938_begin_0, end = var_24938_end_0, end_mask = var_24938_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24938_cast_fp16")]; + tensor var_24942_begin_0 = const()[name = tensor("op_24942_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_24942_end_0 = const()[name = tensor("op_24942_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_24942_end_mask_0 = const()[name = tensor("op_24942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24942_cast_fp16 = slice_by_index(begin = var_24942_begin_0, end = var_24942_end_0, end_mask = var_24942_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24942_cast_fp16")]; + tensor var_24946_begin_0 = const()[name = tensor("op_24946_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_24946_end_0 = const()[name = tensor("op_24946_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_24946_end_mask_0 = const()[name = tensor("op_24946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24946_cast_fp16 = slice_by_index(begin = var_24946_begin_0, end = var_24946_end_0, end_mask = var_24946_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24946_cast_fp16")]; + tensor var_24950_begin_0 = const()[name = tensor("op_24950_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_24950_end_0 = const()[name = tensor("op_24950_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_24950_end_mask_0 = const()[name = tensor("op_24950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24950_cast_fp16 = slice_by_index(begin = var_24950_begin_0, end = var_24950_end_0, end_mask = var_24950_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24950_cast_fp16")]; + tensor var_24954_begin_0 = const()[name = tensor("op_24954_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_24954_end_0 = const()[name = tensor("op_24954_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_24954_end_mask_0 = const()[name = tensor("op_24954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24954_cast_fp16 = slice_by_index(begin = var_24954_begin_0, end = var_24954_end_0, end_mask = var_24954_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24954_cast_fp16")]; + tensor var_24958_begin_0 = const()[name = tensor("op_24958_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_24958_end_0 = const()[name = tensor("op_24958_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_24958_end_mask_0 = const()[name = tensor("op_24958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24958_cast_fp16 = slice_by_index(begin = var_24958_begin_0, end = var_24958_end_0, end_mask = var_24958_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24958_cast_fp16")]; + tensor var_24962_begin_0 = const()[name = tensor("op_24962_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_24962_end_0 = const()[name = tensor("op_24962_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_24962_end_mask_0 = const()[name = tensor("op_24962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24962_cast_fp16 = slice_by_index(begin = var_24962_begin_0, end = var_24962_end_0, end_mask = var_24962_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24962_cast_fp16")]; + tensor var_24966_begin_0 = const()[name = tensor("op_24966_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_24966_end_0 = const()[name = tensor("op_24966_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_24966_end_mask_0 = const()[name = tensor("op_24966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24966_cast_fp16 = slice_by_index(begin = var_24966_begin_0, end = var_24966_end_0, end_mask = var_24966_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24966_cast_fp16")]; + tensor var_24970_begin_0 = const()[name = tensor("op_24970_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_24970_end_0 = const()[name = tensor("op_24970_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_24970_end_mask_0 = const()[name = tensor("op_24970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24970_cast_fp16 = slice_by_index(begin = var_24970_begin_0, end = var_24970_end_0, end_mask = var_24970_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24970_cast_fp16")]; + tensor var_24974_begin_0 = const()[name = tensor("op_24974_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_24974_end_0 = const()[name = tensor("op_24974_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_24974_end_mask_0 = const()[name = tensor("op_24974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24974_cast_fp16 = slice_by_index(begin = var_24974_begin_0, end = var_24974_end_0, end_mask = var_24974_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24974_cast_fp16")]; + tensor var_24978_begin_0 = const()[name = tensor("op_24978_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_24978_end_0 = const()[name = tensor("op_24978_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_24978_end_mask_0 = const()[name = tensor("op_24978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24978_cast_fp16 = slice_by_index(begin = var_24978_begin_0, end = var_24978_end_0, end_mask = var_24978_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24978_cast_fp16")]; + tensor var_24982_begin_0 = const()[name = tensor("op_24982_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_24982_end_0 = const()[name = tensor("op_24982_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_24982_end_mask_0 = const()[name = tensor("op_24982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24982_cast_fp16 = slice_by_index(begin = var_24982_begin_0, end = var_24982_end_0, end_mask = var_24982_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24982_cast_fp16")]; + tensor var_24986_begin_0 = const()[name = tensor("op_24986_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_24986_end_0 = const()[name = tensor("op_24986_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_24986_end_mask_0 = const()[name = tensor("op_24986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24986_cast_fp16 = slice_by_index(begin = var_24986_begin_0, end = var_24986_end_0, end_mask = var_24986_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24986_cast_fp16")]; + tensor var_24990_begin_0 = const()[name = tensor("op_24990_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_24990_end_0 = const()[name = tensor("op_24990_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_24990_end_mask_0 = const()[name = tensor("op_24990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24990_cast_fp16 = slice_by_index(begin = var_24990_begin_0, end = var_24990_end_0, end_mask = var_24990_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24990_cast_fp16")]; + tensor var_24994_begin_0 = const()[name = tensor("op_24994_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_24994_end_0 = const()[name = tensor("op_24994_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_24994_end_mask_0 = const()[name = tensor("op_24994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24994_cast_fp16 = slice_by_index(begin = var_24994_begin_0, end = var_24994_end_0, end_mask = var_24994_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24994_cast_fp16")]; + tensor var_24998_begin_0 = const()[name = tensor("op_24998_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_24998_end_0 = const()[name = tensor("op_24998_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24998_end_mask_0 = const()[name = tensor("op_24998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24998_cast_fp16 = slice_by_index(begin = var_24998_begin_0, end = var_24998_end_0, end_mask = var_24998_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24998_cast_fp16")]; + tensor var_25002_begin_0 = const()[name = tensor("op_25002_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25002_end_0 = const()[name = tensor("op_25002_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_25002_end_mask_0 = const()[name = tensor("op_25002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25002_cast_fp16 = slice_by_index(begin = var_25002_begin_0, end = var_25002_end_0, end_mask = var_25002_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25002_cast_fp16")]; + tensor var_25006_begin_0 = const()[name = tensor("op_25006_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25006_end_0 = const()[name = tensor("op_25006_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_25006_end_mask_0 = const()[name = tensor("op_25006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25006_cast_fp16 = slice_by_index(begin = var_25006_begin_0, end = var_25006_end_0, end_mask = var_25006_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25006_cast_fp16")]; + tensor var_25010_begin_0 = const()[name = tensor("op_25010_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25010_end_0 = const()[name = tensor("op_25010_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_25010_end_mask_0 = const()[name = tensor("op_25010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25010_cast_fp16 = slice_by_index(begin = var_25010_begin_0, end = var_25010_end_0, end_mask = var_25010_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25010_cast_fp16")]; + tensor var_25014_begin_0 = const()[name = tensor("op_25014_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25014_end_0 = const()[name = tensor("op_25014_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_25014_end_mask_0 = const()[name = tensor("op_25014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25014_cast_fp16 = slice_by_index(begin = var_25014_begin_0, end = var_25014_end_0, end_mask = var_25014_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25014_cast_fp16")]; + tensor var_25016_begin_0 = const()[name = tensor("op_25016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25016_end_0 = const()[name = tensor("op_25016_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25016_end_mask_0 = const()[name = tensor("op_25016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25016_cast_fp16 = slice_by_index(begin = var_25016_begin_0, end = var_25016_end_0, end_mask = var_25016_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25016_cast_fp16")]; + tensor var_25020_begin_0 = const()[name = tensor("op_25020_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25020_end_0 = const()[name = tensor("op_25020_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25020_end_mask_0 = const()[name = tensor("op_25020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25020_cast_fp16 = slice_by_index(begin = var_25020_begin_0, end = var_25020_end_0, end_mask = var_25020_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25020_cast_fp16")]; + tensor var_25024_begin_0 = const()[name = tensor("op_25024_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25024_end_0 = const()[name = tensor("op_25024_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25024_end_mask_0 = const()[name = tensor("op_25024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25024_cast_fp16 = slice_by_index(begin = var_25024_begin_0, end = var_25024_end_0, end_mask = var_25024_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25024_cast_fp16")]; + tensor var_25028_begin_0 = const()[name = tensor("op_25028_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25028_end_0 = const()[name = tensor("op_25028_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25028_end_mask_0 = const()[name = tensor("op_25028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25028_cast_fp16 = slice_by_index(begin = var_25028_begin_0, end = var_25028_end_0, end_mask = var_25028_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25028_cast_fp16")]; + tensor var_25032_begin_0 = const()[name = tensor("op_25032_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25032_end_0 = const()[name = tensor("op_25032_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25032_end_mask_0 = const()[name = tensor("op_25032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25032_cast_fp16 = slice_by_index(begin = var_25032_begin_0, end = var_25032_end_0, end_mask = var_25032_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25032_cast_fp16")]; + tensor var_25036_begin_0 = const()[name = tensor("op_25036_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25036_end_0 = const()[name = tensor("op_25036_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25036_end_mask_0 = const()[name = tensor("op_25036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25036_cast_fp16 = slice_by_index(begin = var_25036_begin_0, end = var_25036_end_0, end_mask = var_25036_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25036_cast_fp16")]; + tensor var_25040_begin_0 = const()[name = tensor("op_25040_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25040_end_0 = const()[name = tensor("op_25040_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25040_end_mask_0 = const()[name = tensor("op_25040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25040_cast_fp16 = slice_by_index(begin = var_25040_begin_0, end = var_25040_end_0, end_mask = var_25040_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25040_cast_fp16")]; + tensor var_25044_begin_0 = const()[name = tensor("op_25044_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25044_end_0 = const()[name = tensor("op_25044_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25044_end_mask_0 = const()[name = tensor("op_25044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25044_cast_fp16 = slice_by_index(begin = var_25044_begin_0, end = var_25044_end_0, end_mask = var_25044_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25044_cast_fp16")]; + tensor var_25048_begin_0 = const()[name = tensor("op_25048_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25048_end_0 = const()[name = tensor("op_25048_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25048_end_mask_0 = const()[name = tensor("op_25048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25048_cast_fp16 = slice_by_index(begin = var_25048_begin_0, end = var_25048_end_0, end_mask = var_25048_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25048_cast_fp16")]; + tensor var_25052_begin_0 = const()[name = tensor("op_25052_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25052_end_0 = const()[name = tensor("op_25052_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25052_end_mask_0 = const()[name = tensor("op_25052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25052_cast_fp16 = slice_by_index(begin = var_25052_begin_0, end = var_25052_end_0, end_mask = var_25052_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25052_cast_fp16")]; + tensor var_25056_begin_0 = const()[name = tensor("op_25056_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25056_end_0 = const()[name = tensor("op_25056_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25056_end_mask_0 = const()[name = tensor("op_25056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25056_cast_fp16 = slice_by_index(begin = var_25056_begin_0, end = var_25056_end_0, end_mask = var_25056_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25056_cast_fp16")]; + tensor var_25060_begin_0 = const()[name = tensor("op_25060_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25060_end_0 = const()[name = tensor("op_25060_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25060_end_mask_0 = const()[name = tensor("op_25060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25060_cast_fp16 = slice_by_index(begin = var_25060_begin_0, end = var_25060_end_0, end_mask = var_25060_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25060_cast_fp16")]; + tensor var_25064_begin_0 = const()[name = tensor("op_25064_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25064_end_0 = const()[name = tensor("op_25064_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25064_end_mask_0 = const()[name = tensor("op_25064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25064_cast_fp16 = slice_by_index(begin = var_25064_begin_0, end = var_25064_end_0, end_mask = var_25064_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25064_cast_fp16")]; + tensor var_25068_begin_0 = const()[name = tensor("op_25068_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25068_end_0 = const()[name = tensor("op_25068_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25068_end_mask_0 = const()[name = tensor("op_25068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25068_cast_fp16 = slice_by_index(begin = var_25068_begin_0, end = var_25068_end_0, end_mask = var_25068_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25068_cast_fp16")]; + tensor var_25072_begin_0 = const()[name = tensor("op_25072_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25072_end_0 = const()[name = tensor("op_25072_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25072_end_mask_0 = const()[name = tensor("op_25072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25072_cast_fp16 = slice_by_index(begin = var_25072_begin_0, end = var_25072_end_0, end_mask = var_25072_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25072_cast_fp16")]; + tensor var_25076_begin_0 = const()[name = tensor("op_25076_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25076_end_0 = const()[name = tensor("op_25076_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25076_end_mask_0 = const()[name = tensor("op_25076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25076_cast_fp16 = slice_by_index(begin = var_25076_begin_0, end = var_25076_end_0, end_mask = var_25076_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25076_cast_fp16")]; + tensor var_25080_begin_0 = const()[name = tensor("op_25080_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25080_end_0 = const()[name = tensor("op_25080_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25080_end_mask_0 = const()[name = tensor("op_25080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25080_cast_fp16 = slice_by_index(begin = var_25080_begin_0, end = var_25080_end_0, end_mask = var_25080_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25080_cast_fp16")]; + tensor var_25084_begin_0 = const()[name = tensor("op_25084_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25084_end_0 = const()[name = tensor("op_25084_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25084_end_mask_0 = const()[name = tensor("op_25084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25084_cast_fp16 = slice_by_index(begin = var_25084_begin_0, end = var_25084_end_0, end_mask = var_25084_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25084_cast_fp16")]; + tensor var_25088_begin_0 = const()[name = tensor("op_25088_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25088_end_0 = const()[name = tensor("op_25088_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25088_end_mask_0 = const()[name = tensor("op_25088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25088_cast_fp16 = slice_by_index(begin = var_25088_begin_0, end = var_25088_end_0, end_mask = var_25088_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25088_cast_fp16")]; + tensor var_25092_begin_0 = const()[name = tensor("op_25092_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25092_end_0 = const()[name = tensor("op_25092_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_25092_end_mask_0 = const()[name = tensor("op_25092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25092_cast_fp16 = slice_by_index(begin = var_25092_begin_0, end = var_25092_end_0, end_mask = var_25092_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25092_cast_fp16")]; + tensor var_25096_equation_0 = const()[name = tensor("op_25096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25096_cast_fp16 = einsum(equation = var_25096_equation_0, values = (var_24938_cast_fp16, var_24855_cast_fp16))[name = tensor("op_25096_cast_fp16")]; + tensor var_25097_to_fp16 = const()[name = tensor("op_25097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2081_cast_fp16 = mul(x = var_25096_cast_fp16, y = var_25097_to_fp16)[name = tensor("aw_2081_cast_fp16")]; + tensor var_25100_equation_0 = const()[name = tensor("op_25100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25100_cast_fp16 = einsum(equation = var_25100_equation_0, values = (var_24942_cast_fp16, var_24859_cast_fp16))[name = tensor("op_25100_cast_fp16")]; + tensor var_25101_to_fp16 = const()[name = tensor("op_25101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2083_cast_fp16 = mul(x = var_25100_cast_fp16, y = var_25101_to_fp16)[name = tensor("aw_2083_cast_fp16")]; + tensor var_25104_equation_0 = const()[name = tensor("op_25104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25104_cast_fp16 = einsum(equation = var_25104_equation_0, values = (var_24946_cast_fp16, var_24863_cast_fp16))[name = tensor("op_25104_cast_fp16")]; + tensor var_25105_to_fp16 = const()[name = tensor("op_25105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2085_cast_fp16 = mul(x = var_25104_cast_fp16, y = var_25105_to_fp16)[name = tensor("aw_2085_cast_fp16")]; + tensor var_25108_equation_0 = const()[name = tensor("op_25108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25108_cast_fp16 = einsum(equation = var_25108_equation_0, values = (var_24950_cast_fp16, var_24867_cast_fp16))[name = tensor("op_25108_cast_fp16")]; + tensor var_25109_to_fp16 = const()[name = tensor("op_25109_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2087_cast_fp16 = mul(x = var_25108_cast_fp16, y = var_25109_to_fp16)[name = tensor("aw_2087_cast_fp16")]; + tensor var_25112_equation_0 = const()[name = tensor("op_25112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25112_cast_fp16 = einsum(equation = var_25112_equation_0, values = (var_24954_cast_fp16, var_24871_cast_fp16))[name = tensor("op_25112_cast_fp16")]; + tensor var_25113_to_fp16 = const()[name = tensor("op_25113_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2089_cast_fp16 = mul(x = var_25112_cast_fp16, y = var_25113_to_fp16)[name = tensor("aw_2089_cast_fp16")]; + tensor var_25116_equation_0 = const()[name = tensor("op_25116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25116_cast_fp16 = einsum(equation = var_25116_equation_0, values = (var_24958_cast_fp16, var_24875_cast_fp16))[name = tensor("op_25116_cast_fp16")]; + tensor var_25117_to_fp16 = const()[name = tensor("op_25117_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2091_cast_fp16 = mul(x = var_25116_cast_fp16, y = var_25117_to_fp16)[name = tensor("aw_2091_cast_fp16")]; + tensor var_25120_equation_0 = const()[name = tensor("op_25120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25120_cast_fp16 = einsum(equation = var_25120_equation_0, values = (var_24962_cast_fp16, var_24879_cast_fp16))[name = tensor("op_25120_cast_fp16")]; + tensor var_25121_to_fp16 = const()[name = tensor("op_25121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2093_cast_fp16 = mul(x = var_25120_cast_fp16, y = var_25121_to_fp16)[name = tensor("aw_2093_cast_fp16")]; + tensor var_25124_equation_0 = const()[name = tensor("op_25124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25124_cast_fp16 = einsum(equation = var_25124_equation_0, values = (var_24966_cast_fp16, var_24883_cast_fp16))[name = tensor("op_25124_cast_fp16")]; + tensor var_25125_to_fp16 = const()[name = tensor("op_25125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2095_cast_fp16 = mul(x = var_25124_cast_fp16, y = var_25125_to_fp16)[name = tensor("aw_2095_cast_fp16")]; + tensor var_25128_equation_0 = const()[name = tensor("op_25128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25128_cast_fp16 = einsum(equation = var_25128_equation_0, values = (var_24970_cast_fp16, var_24887_cast_fp16))[name = tensor("op_25128_cast_fp16")]; + tensor var_25129_to_fp16 = const()[name = tensor("op_25129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2097_cast_fp16 = mul(x = var_25128_cast_fp16, y = var_25129_to_fp16)[name = tensor("aw_2097_cast_fp16")]; + tensor var_25132_equation_0 = const()[name = tensor("op_25132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25132_cast_fp16 = einsum(equation = var_25132_equation_0, values = (var_24974_cast_fp16, var_24891_cast_fp16))[name = tensor("op_25132_cast_fp16")]; + tensor var_25133_to_fp16 = const()[name = tensor("op_25133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2099_cast_fp16 = mul(x = var_25132_cast_fp16, y = var_25133_to_fp16)[name = tensor("aw_2099_cast_fp16")]; + tensor var_25136_equation_0 = const()[name = tensor("op_25136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25136_cast_fp16 = einsum(equation = var_25136_equation_0, values = (var_24978_cast_fp16, var_24895_cast_fp16))[name = tensor("op_25136_cast_fp16")]; + tensor var_25137_to_fp16 = const()[name = tensor("op_25137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2101_cast_fp16 = mul(x = var_25136_cast_fp16, y = var_25137_to_fp16)[name = tensor("aw_2101_cast_fp16")]; + tensor var_25140_equation_0 = const()[name = tensor("op_25140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25140_cast_fp16 = einsum(equation = var_25140_equation_0, values = (var_24982_cast_fp16, var_24899_cast_fp16))[name = tensor("op_25140_cast_fp16")]; + tensor var_25141_to_fp16 = const()[name = tensor("op_25141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2103_cast_fp16 = mul(x = var_25140_cast_fp16, y = var_25141_to_fp16)[name = tensor("aw_2103_cast_fp16")]; + tensor var_25144_equation_0 = const()[name = tensor("op_25144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25144_cast_fp16 = einsum(equation = var_25144_equation_0, values = (var_24986_cast_fp16, var_24903_cast_fp16))[name = tensor("op_25144_cast_fp16")]; + tensor var_25145_to_fp16 = const()[name = tensor("op_25145_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2105_cast_fp16 = mul(x = var_25144_cast_fp16, y = var_25145_to_fp16)[name = tensor("aw_2105_cast_fp16")]; + tensor var_25148_equation_0 = const()[name = tensor("op_25148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25148_cast_fp16 = einsum(equation = var_25148_equation_0, values = (var_24990_cast_fp16, var_24907_cast_fp16))[name = tensor("op_25148_cast_fp16")]; + tensor var_25149_to_fp16 = const()[name = tensor("op_25149_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2107_cast_fp16 = mul(x = var_25148_cast_fp16, y = var_25149_to_fp16)[name = tensor("aw_2107_cast_fp16")]; + tensor var_25152_equation_0 = const()[name = tensor("op_25152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25152_cast_fp16 = einsum(equation = var_25152_equation_0, values = (var_24994_cast_fp16, var_24911_cast_fp16))[name = tensor("op_25152_cast_fp16")]; + tensor var_25153_to_fp16 = const()[name = tensor("op_25153_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2109_cast_fp16 = mul(x = var_25152_cast_fp16, y = var_25153_to_fp16)[name = tensor("aw_2109_cast_fp16")]; + tensor var_25156_equation_0 = const()[name = tensor("op_25156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25156_cast_fp16 = einsum(equation = var_25156_equation_0, values = (var_24998_cast_fp16, var_24915_cast_fp16))[name = tensor("op_25156_cast_fp16")]; + tensor var_25157_to_fp16 = const()[name = tensor("op_25157_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2111_cast_fp16 = mul(x = var_25156_cast_fp16, y = var_25157_to_fp16)[name = tensor("aw_2111_cast_fp16")]; + tensor var_25160_equation_0 = const()[name = tensor("op_25160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25160_cast_fp16 = einsum(equation = var_25160_equation_0, values = (var_25002_cast_fp16, var_24919_cast_fp16))[name = tensor("op_25160_cast_fp16")]; + tensor var_25161_to_fp16 = const()[name = tensor("op_25161_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2113_cast_fp16 = mul(x = var_25160_cast_fp16, y = var_25161_to_fp16)[name = tensor("aw_2113_cast_fp16")]; + tensor var_25164_equation_0 = const()[name = tensor("op_25164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25164_cast_fp16 = einsum(equation = var_25164_equation_0, values = (var_25006_cast_fp16, var_24923_cast_fp16))[name = tensor("op_25164_cast_fp16")]; + tensor var_25165_to_fp16 = const()[name = tensor("op_25165_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2115_cast_fp16 = mul(x = var_25164_cast_fp16, y = var_25165_to_fp16)[name = tensor("aw_2115_cast_fp16")]; + tensor var_25168_equation_0 = const()[name = tensor("op_25168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25168_cast_fp16 = einsum(equation = var_25168_equation_0, values = (var_25010_cast_fp16, var_24927_cast_fp16))[name = tensor("op_25168_cast_fp16")]; + tensor var_25169_to_fp16 = const()[name = tensor("op_25169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2117_cast_fp16 = mul(x = var_25168_cast_fp16, y = var_25169_to_fp16)[name = tensor("aw_2117_cast_fp16")]; + tensor var_25172_equation_0 = const()[name = tensor("op_25172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25172_cast_fp16 = einsum(equation = var_25172_equation_0, values = (var_25014_cast_fp16, var_24931_cast_fp16))[name = tensor("op_25172_cast_fp16")]; + tensor var_25173_to_fp16 = const()[name = tensor("op_25173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2119_cast_fp16 = mul(x = var_25172_cast_fp16, y = var_25173_to_fp16)[name = tensor("aw_2119_cast_fp16")]; + tensor var_25175_cast_fp16 = softmax(axis = var_21077, x = aw_2081_cast_fp16)[name = tensor("op_25175_cast_fp16")]; + tensor var_25176_cast_fp16 = softmax(axis = var_21077, x = aw_2083_cast_fp16)[name = tensor("op_25176_cast_fp16")]; + tensor var_25177_cast_fp16 = softmax(axis = var_21077, x = aw_2085_cast_fp16)[name = tensor("op_25177_cast_fp16")]; + tensor var_25178_cast_fp16 = softmax(axis = var_21077, x = aw_2087_cast_fp16)[name = tensor("op_25178_cast_fp16")]; + tensor var_25179_cast_fp16 = softmax(axis = var_21077, x = aw_2089_cast_fp16)[name = tensor("op_25179_cast_fp16")]; + tensor var_25180_cast_fp16 = softmax(axis = var_21077, x = aw_2091_cast_fp16)[name = tensor("op_25180_cast_fp16")]; + tensor var_25181_cast_fp16 = softmax(axis = var_21077, x = aw_2093_cast_fp16)[name = tensor("op_25181_cast_fp16")]; + tensor var_25182_cast_fp16 = softmax(axis = var_21077, x = aw_2095_cast_fp16)[name = tensor("op_25182_cast_fp16")]; + tensor var_25183_cast_fp16 = softmax(axis = var_21077, x = aw_2097_cast_fp16)[name = tensor("op_25183_cast_fp16")]; + tensor var_25184_cast_fp16 = softmax(axis = var_21077, x = aw_2099_cast_fp16)[name = tensor("op_25184_cast_fp16")]; + tensor var_25185_cast_fp16 = softmax(axis = var_21077, x = aw_2101_cast_fp16)[name = tensor("op_25185_cast_fp16")]; + tensor var_25186_cast_fp16 = softmax(axis = var_21077, x = aw_2103_cast_fp16)[name = tensor("op_25186_cast_fp16")]; + tensor var_25187_cast_fp16 = softmax(axis = var_21077, x = aw_2105_cast_fp16)[name = tensor("op_25187_cast_fp16")]; + tensor var_25188_cast_fp16 = softmax(axis = var_21077, x = aw_2107_cast_fp16)[name = tensor("op_25188_cast_fp16")]; + tensor var_25189_cast_fp16 = softmax(axis = var_21077, x = aw_2109_cast_fp16)[name = tensor("op_25189_cast_fp16")]; + tensor var_25190_cast_fp16 = softmax(axis = var_21077, x = aw_2111_cast_fp16)[name = tensor("op_25190_cast_fp16")]; + tensor var_25191_cast_fp16 = softmax(axis = var_21077, x = aw_2113_cast_fp16)[name = tensor("op_25191_cast_fp16")]; + tensor var_25192_cast_fp16 = softmax(axis = var_21077, x = aw_2115_cast_fp16)[name = tensor("op_25192_cast_fp16")]; + tensor var_25193_cast_fp16 = softmax(axis = var_21077, x = aw_2117_cast_fp16)[name = tensor("op_25193_cast_fp16")]; + tensor var_25194_cast_fp16 = softmax(axis = var_21077, x = aw_2119_cast_fp16)[name = tensor("op_25194_cast_fp16")]; + tensor var_25196_equation_0 = const()[name = tensor("op_25196_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25196_cast_fp16 = einsum(equation = var_25196_equation_0, values = (var_25016_cast_fp16, var_25175_cast_fp16))[name = tensor("op_25196_cast_fp16")]; + tensor var_25198_equation_0 = const()[name = tensor("op_25198_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25198_cast_fp16 = einsum(equation = var_25198_equation_0, values = (var_25020_cast_fp16, var_25176_cast_fp16))[name = tensor("op_25198_cast_fp16")]; + tensor var_25200_equation_0 = const()[name = tensor("op_25200_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25200_cast_fp16 = einsum(equation = var_25200_equation_0, values = (var_25024_cast_fp16, var_25177_cast_fp16))[name = tensor("op_25200_cast_fp16")]; + tensor var_25202_equation_0 = const()[name = tensor("op_25202_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25202_cast_fp16 = einsum(equation = var_25202_equation_0, values = (var_25028_cast_fp16, var_25178_cast_fp16))[name = tensor("op_25202_cast_fp16")]; + tensor var_25204_equation_0 = const()[name = tensor("op_25204_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25204_cast_fp16 = einsum(equation = var_25204_equation_0, values = (var_25032_cast_fp16, var_25179_cast_fp16))[name = tensor("op_25204_cast_fp16")]; + tensor var_25206_equation_0 = const()[name = tensor("op_25206_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25206_cast_fp16 = einsum(equation = var_25206_equation_0, values = (var_25036_cast_fp16, var_25180_cast_fp16))[name = tensor("op_25206_cast_fp16")]; + tensor var_25208_equation_0 = const()[name = tensor("op_25208_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25208_cast_fp16 = einsum(equation = var_25208_equation_0, values = (var_25040_cast_fp16, var_25181_cast_fp16))[name = tensor("op_25208_cast_fp16")]; + tensor var_25210_equation_0 = const()[name = tensor("op_25210_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25210_cast_fp16 = einsum(equation = var_25210_equation_0, values = (var_25044_cast_fp16, var_25182_cast_fp16))[name = tensor("op_25210_cast_fp16")]; + tensor var_25212_equation_0 = const()[name = tensor("op_25212_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25212_cast_fp16 = einsum(equation = var_25212_equation_0, values = (var_25048_cast_fp16, var_25183_cast_fp16))[name = tensor("op_25212_cast_fp16")]; + tensor var_25214_equation_0 = const()[name = tensor("op_25214_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25214_cast_fp16 = einsum(equation = var_25214_equation_0, values = (var_25052_cast_fp16, var_25184_cast_fp16))[name = tensor("op_25214_cast_fp16")]; + tensor var_25216_equation_0 = const()[name = tensor("op_25216_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25216_cast_fp16 = einsum(equation = var_25216_equation_0, values = (var_25056_cast_fp16, var_25185_cast_fp16))[name = tensor("op_25216_cast_fp16")]; + tensor var_25218_equation_0 = const()[name = tensor("op_25218_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25218_cast_fp16 = einsum(equation = var_25218_equation_0, values = (var_25060_cast_fp16, var_25186_cast_fp16))[name = tensor("op_25218_cast_fp16")]; + tensor var_25220_equation_0 = const()[name = tensor("op_25220_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25220_cast_fp16 = einsum(equation = var_25220_equation_0, values = (var_25064_cast_fp16, var_25187_cast_fp16))[name = tensor("op_25220_cast_fp16")]; + tensor var_25222_equation_0 = const()[name = tensor("op_25222_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25222_cast_fp16 = einsum(equation = var_25222_equation_0, values = (var_25068_cast_fp16, var_25188_cast_fp16))[name = tensor("op_25222_cast_fp16")]; + tensor var_25224_equation_0 = const()[name = tensor("op_25224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25224_cast_fp16 = einsum(equation = var_25224_equation_0, values = (var_25072_cast_fp16, var_25189_cast_fp16))[name = tensor("op_25224_cast_fp16")]; + tensor var_25226_equation_0 = const()[name = tensor("op_25226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25226_cast_fp16 = einsum(equation = var_25226_equation_0, values = (var_25076_cast_fp16, var_25190_cast_fp16))[name = tensor("op_25226_cast_fp16")]; + tensor var_25228_equation_0 = const()[name = tensor("op_25228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25228_cast_fp16 = einsum(equation = var_25228_equation_0, values = (var_25080_cast_fp16, var_25191_cast_fp16))[name = tensor("op_25228_cast_fp16")]; + tensor var_25230_equation_0 = const()[name = tensor("op_25230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25230_cast_fp16 = einsum(equation = var_25230_equation_0, values = (var_25084_cast_fp16, var_25192_cast_fp16))[name = tensor("op_25230_cast_fp16")]; + tensor var_25232_equation_0 = const()[name = tensor("op_25232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25232_cast_fp16 = einsum(equation = var_25232_equation_0, values = (var_25088_cast_fp16, var_25193_cast_fp16))[name = tensor("op_25232_cast_fp16")]; + tensor var_25234_equation_0 = const()[name = tensor("op_25234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25234_cast_fp16 = einsum(equation = var_25234_equation_0, values = (var_25092_cast_fp16, var_25194_cast_fp16))[name = tensor("op_25234_cast_fp16")]; + tensor input_359_interleave_0 = const()[name = tensor("input_359_interleave_0"), val = tensor(false)]; + tensor input_359_cast_fp16 = concat(axis = var_21077, interleave = input_359_interleave_0, values = (var_25196_cast_fp16, var_25198_cast_fp16, var_25200_cast_fp16, var_25202_cast_fp16, var_25204_cast_fp16, var_25206_cast_fp16, var_25208_cast_fp16, var_25210_cast_fp16, var_25212_cast_fp16, var_25214_cast_fp16, var_25216_cast_fp16, var_25218_cast_fp16, var_25220_cast_fp16, var_25222_cast_fp16, var_25224_cast_fp16, var_25226_cast_fp16, var_25228_cast_fp16, var_25230_cast_fp16, var_25232_cast_fp16, var_25234_cast_fp16))[name = tensor("input_359_cast_fp16")]; + tensor var_25244_pad_type_0 = const()[name = tensor("op_25244_pad_type_0"), val = tensor("valid")]; + tensor var_25244_strides_0 = const()[name = tensor("op_25244_strides_0"), val = tensor([1, 1])]; + tensor var_25244_pad_0 = const()[name = tensor("op_25244_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25244_dilations_0 = const()[name = tensor("op_25244_dilations_0"), val = tensor([1, 1])]; + tensor var_25244_groups_0 = const()[name = tensor("op_25244_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(756136960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757365824))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757366016)))]; + tensor var_25244_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_25244_dilations_0, groups = var_25244_groups_0, pad = var_25244_pad_0, pad_type = var_25244_pad_type_0, strides = var_25244_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized, x = input_359_cast_fp16)[name = tensor("op_25244_cast_fp16")]; + tensor inputs_171_cast_fp16 = add(x = var_25244_cast_fp16, y = inputs_169_cast_fp16)[name = tensor("inputs_171_cast_fp16")]; + tensor hidden_states_235_axes_0 = const()[name = tensor("hidden_states_235_axes_0"), val = tensor([1])]; + tensor hidden_states_235_gamma_0_to_fp16 = const()[name = tensor("hidden_states_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757368640)))]; + tensor hidden_states_235_beta_0_to_fp16 = const()[name = tensor("hidden_states_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757371264)))]; + tensor var_25254_to_fp16 = const()[name = tensor("op_25254_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_235_cast_fp16 = layer_norm(axes = hidden_states_235_axes_0, beta = hidden_states_235_beta_0_to_fp16, epsilon = var_25254_to_fp16, gamma = hidden_states_235_gamma_0_to_fp16, x = inputs_171_cast_fp16)[name = tensor("hidden_states_235_cast_fp16")]; + tensor q_115_pad_type_0 = const()[name = tensor("q_115_pad_type_0"), val = tensor("valid")]; + tensor q_115_strides_0 = const()[name = tensor("q_115_strides_0"), val = tensor([1, 1])]; + tensor q_115_pad_0 = const()[name = tensor("q_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_115_dilations_0 = const()[name = tensor("q_115_dilations_0"), val = tensor([1, 1])]; + tensor q_115_groups_0 = const()[name = tensor("q_115_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757373888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(758602752))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_115_cast_fp16 = conv(dilations = q_115_dilations_0, groups = q_115_groups_0, pad = q_115_pad_0, pad_type = q_115_pad_type_0, strides = q_115_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_235_cast_fp16)[name = tensor("q_115_cast_fp16")]; + tensor k_229_pad_type_0 = const()[name = tensor("k_229_pad_type_0"), val = tensor("valid")]; + tensor k_229_strides_0 = const()[name = tensor("k_229_strides_0"), val = tensor([1, 1])]; + tensor k_229_pad_0 = const()[name = tensor("k_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_229_dilations_0 = const()[name = tensor("k_229_dilations_0"), val = tensor([1, 1])]; + tensor k_229_groups_0 = const()[name = tensor("k_229_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(758602944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(760569088))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_229_cast_fp16 = conv(dilations = k_229_dilations_0, groups = k_229_groups_0, pad = k_229_pad_0, pad_type = k_229_pad_type_0, strides = k_229_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_229_cast_fp16")]; + tensor v_115_pad_type_0 = const()[name = tensor("v_115_pad_type_0"), val = tensor("valid")]; + tensor v_115_strides_0 = const()[name = tensor("v_115_strides_0"), val = tensor([1, 1])]; + tensor v_115_pad_0 = const()[name = tensor("v_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_115_dilations_0 = const()[name = tensor("v_115_dilations_0"), val = tensor([1, 1])]; + tensor v_115_groups_0 = const()[name = tensor("v_115_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(760569280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762535424))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_115_cast_fp16 = conv(dilations = v_115_dilations_0, groups = v_115_groups_0, pad = v_115_pad_0, pad_type = v_115_pad_type_0, strides = v_115_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_115_cast_fp16")]; + tensor var_25287_begin_0 = const()[name = tensor("op_25287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25287_end_0 = const()[name = tensor("op_25287_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25287_end_mask_0 = const()[name = tensor("op_25287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25287_cast_fp16 = slice_by_index(begin = var_25287_begin_0, end = var_25287_end_0, end_mask = var_25287_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25287_cast_fp16")]; + tensor var_25291_begin_0 = const()[name = tensor("op_25291_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25291_end_0 = const()[name = tensor("op_25291_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25291_end_mask_0 = const()[name = tensor("op_25291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25291_cast_fp16 = slice_by_index(begin = var_25291_begin_0, end = var_25291_end_0, end_mask = var_25291_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25291_cast_fp16")]; + tensor var_25295_begin_0 = const()[name = tensor("op_25295_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25295_end_0 = const()[name = tensor("op_25295_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25295_end_mask_0 = const()[name = tensor("op_25295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25295_cast_fp16 = slice_by_index(begin = var_25295_begin_0, end = var_25295_end_0, end_mask = var_25295_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25295_cast_fp16")]; + tensor var_25299_begin_0 = const()[name = tensor("op_25299_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25299_end_0 = const()[name = tensor("op_25299_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25299_end_mask_0 = const()[name = tensor("op_25299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25299_cast_fp16 = slice_by_index(begin = var_25299_begin_0, end = var_25299_end_0, end_mask = var_25299_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25299_cast_fp16")]; + tensor var_25303_begin_0 = const()[name = tensor("op_25303_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25303_end_0 = const()[name = tensor("op_25303_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25303_end_mask_0 = const()[name = tensor("op_25303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25303_cast_fp16 = slice_by_index(begin = var_25303_begin_0, end = var_25303_end_0, end_mask = var_25303_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25303_cast_fp16")]; + tensor var_25307_begin_0 = const()[name = tensor("op_25307_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25307_end_0 = const()[name = tensor("op_25307_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25307_end_mask_0 = const()[name = tensor("op_25307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25307_cast_fp16 = slice_by_index(begin = var_25307_begin_0, end = var_25307_end_0, end_mask = var_25307_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25307_cast_fp16")]; + tensor var_25311_begin_0 = const()[name = tensor("op_25311_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25311_end_0 = const()[name = tensor("op_25311_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25311_end_mask_0 = const()[name = tensor("op_25311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25311_cast_fp16 = slice_by_index(begin = var_25311_begin_0, end = var_25311_end_0, end_mask = var_25311_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25311_cast_fp16")]; + tensor var_25315_begin_0 = const()[name = tensor("op_25315_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25315_end_0 = const()[name = tensor("op_25315_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25315_end_mask_0 = const()[name = tensor("op_25315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25315_cast_fp16 = slice_by_index(begin = var_25315_begin_0, end = var_25315_end_0, end_mask = var_25315_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25315_cast_fp16")]; + tensor var_25319_begin_0 = const()[name = tensor("op_25319_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25319_end_0 = const()[name = tensor("op_25319_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25319_end_mask_0 = const()[name = tensor("op_25319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25319_cast_fp16 = slice_by_index(begin = var_25319_begin_0, end = var_25319_end_0, end_mask = var_25319_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25319_cast_fp16")]; + tensor var_25323_begin_0 = const()[name = tensor("op_25323_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25323_end_0 = const()[name = tensor("op_25323_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25323_end_mask_0 = const()[name = tensor("op_25323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25323_cast_fp16 = slice_by_index(begin = var_25323_begin_0, end = var_25323_end_0, end_mask = var_25323_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25323_cast_fp16")]; + tensor var_25327_begin_0 = const()[name = tensor("op_25327_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25327_end_0 = const()[name = tensor("op_25327_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25327_end_mask_0 = const()[name = tensor("op_25327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25327_cast_fp16 = slice_by_index(begin = var_25327_begin_0, end = var_25327_end_0, end_mask = var_25327_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25327_cast_fp16")]; + tensor var_25331_begin_0 = const()[name = tensor("op_25331_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25331_end_0 = const()[name = tensor("op_25331_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25331_end_mask_0 = const()[name = tensor("op_25331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25331_cast_fp16 = slice_by_index(begin = var_25331_begin_0, end = var_25331_end_0, end_mask = var_25331_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25331_cast_fp16")]; + tensor var_25335_begin_0 = const()[name = tensor("op_25335_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25335_end_0 = const()[name = tensor("op_25335_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25335_end_mask_0 = const()[name = tensor("op_25335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25335_cast_fp16 = slice_by_index(begin = var_25335_begin_0, end = var_25335_end_0, end_mask = var_25335_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25335_cast_fp16")]; + tensor var_25339_begin_0 = const()[name = tensor("op_25339_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25339_end_0 = const()[name = tensor("op_25339_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25339_end_mask_0 = const()[name = tensor("op_25339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25339_cast_fp16 = slice_by_index(begin = var_25339_begin_0, end = var_25339_end_0, end_mask = var_25339_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25339_cast_fp16")]; + tensor var_25343_begin_0 = const()[name = tensor("op_25343_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25343_end_0 = const()[name = tensor("op_25343_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25343_end_mask_0 = const()[name = tensor("op_25343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25343_cast_fp16 = slice_by_index(begin = var_25343_begin_0, end = var_25343_end_0, end_mask = var_25343_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25343_cast_fp16")]; + tensor var_25347_begin_0 = const()[name = tensor("op_25347_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25347_end_0 = const()[name = tensor("op_25347_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25347_end_mask_0 = const()[name = tensor("op_25347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25347_cast_fp16 = slice_by_index(begin = var_25347_begin_0, end = var_25347_end_0, end_mask = var_25347_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25347_cast_fp16")]; + tensor var_25351_begin_0 = const()[name = tensor("op_25351_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25351_end_0 = const()[name = tensor("op_25351_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25351_end_mask_0 = const()[name = tensor("op_25351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25351_cast_fp16 = slice_by_index(begin = var_25351_begin_0, end = var_25351_end_0, end_mask = var_25351_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25351_cast_fp16")]; + tensor var_25355_begin_0 = const()[name = tensor("op_25355_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25355_end_0 = const()[name = tensor("op_25355_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25355_end_mask_0 = const()[name = tensor("op_25355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25355_cast_fp16 = slice_by_index(begin = var_25355_begin_0, end = var_25355_end_0, end_mask = var_25355_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25355_cast_fp16")]; + tensor var_25359_begin_0 = const()[name = tensor("op_25359_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25359_end_0 = const()[name = tensor("op_25359_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25359_end_mask_0 = const()[name = tensor("op_25359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25359_cast_fp16 = slice_by_index(begin = var_25359_begin_0, end = var_25359_end_0, end_mask = var_25359_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25359_cast_fp16")]; + tensor var_25363_begin_0 = const()[name = tensor("op_25363_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25363_end_0 = const()[name = tensor("op_25363_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_25363_end_mask_0 = const()[name = tensor("op_25363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25363_cast_fp16 = slice_by_index(begin = var_25363_begin_0, end = var_25363_end_0, end_mask = var_25363_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25363_cast_fp16")]; + tensor k_231_perm_0 = const()[name = tensor("k_231_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_25370_begin_0 = const()[name = tensor("op_25370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25370_end_0 = const()[name = tensor("op_25370_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_25370_end_mask_0 = const()[name = tensor("op_25370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_231_cast_fp16 = transpose(perm = k_231_perm_0, x = k_229_cast_fp16)[name = tensor("transpose_10")]; + tensor var_25370_cast_fp16 = slice_by_index(begin = var_25370_begin_0, end = var_25370_end_0, end_mask = var_25370_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25370_cast_fp16")]; + tensor var_25374_begin_0 = const()[name = tensor("op_25374_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_25374_end_0 = const()[name = tensor("op_25374_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_25374_end_mask_0 = const()[name = tensor("op_25374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25374_cast_fp16 = slice_by_index(begin = var_25374_begin_0, end = var_25374_end_0, end_mask = var_25374_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25374_cast_fp16")]; + tensor var_25378_begin_0 = const()[name = tensor("op_25378_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_25378_end_0 = const()[name = tensor("op_25378_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_25378_end_mask_0 = const()[name = tensor("op_25378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25378_cast_fp16 = slice_by_index(begin = var_25378_begin_0, end = var_25378_end_0, end_mask = var_25378_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25378_cast_fp16")]; + tensor var_25382_begin_0 = const()[name = tensor("op_25382_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_25382_end_0 = const()[name = tensor("op_25382_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_25382_end_mask_0 = const()[name = tensor("op_25382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25382_cast_fp16 = slice_by_index(begin = var_25382_begin_0, end = var_25382_end_0, end_mask = var_25382_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25382_cast_fp16")]; + tensor var_25386_begin_0 = const()[name = tensor("op_25386_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_25386_end_0 = const()[name = tensor("op_25386_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_25386_end_mask_0 = const()[name = tensor("op_25386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25386_cast_fp16 = slice_by_index(begin = var_25386_begin_0, end = var_25386_end_0, end_mask = var_25386_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25386_cast_fp16")]; + tensor var_25390_begin_0 = const()[name = tensor("op_25390_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_25390_end_0 = const()[name = tensor("op_25390_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_25390_end_mask_0 = const()[name = tensor("op_25390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25390_cast_fp16 = slice_by_index(begin = var_25390_begin_0, end = var_25390_end_0, end_mask = var_25390_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25390_cast_fp16")]; + tensor var_25394_begin_0 = const()[name = tensor("op_25394_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_25394_end_0 = const()[name = tensor("op_25394_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_25394_end_mask_0 = const()[name = tensor("op_25394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25394_cast_fp16 = slice_by_index(begin = var_25394_begin_0, end = var_25394_end_0, end_mask = var_25394_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25394_cast_fp16")]; + tensor var_25398_begin_0 = const()[name = tensor("op_25398_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_25398_end_0 = const()[name = tensor("op_25398_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_25398_end_mask_0 = const()[name = tensor("op_25398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25398_cast_fp16 = slice_by_index(begin = var_25398_begin_0, end = var_25398_end_0, end_mask = var_25398_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25398_cast_fp16")]; + tensor var_25402_begin_0 = const()[name = tensor("op_25402_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_25402_end_0 = const()[name = tensor("op_25402_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_25402_end_mask_0 = const()[name = tensor("op_25402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25402_cast_fp16 = slice_by_index(begin = var_25402_begin_0, end = var_25402_end_0, end_mask = var_25402_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25402_cast_fp16")]; + tensor var_25406_begin_0 = const()[name = tensor("op_25406_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_25406_end_0 = const()[name = tensor("op_25406_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_25406_end_mask_0 = const()[name = tensor("op_25406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25406_cast_fp16 = slice_by_index(begin = var_25406_begin_0, end = var_25406_end_0, end_mask = var_25406_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25406_cast_fp16")]; + tensor var_25410_begin_0 = const()[name = tensor("op_25410_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_25410_end_0 = const()[name = tensor("op_25410_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_25410_end_mask_0 = const()[name = tensor("op_25410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25410_cast_fp16 = slice_by_index(begin = var_25410_begin_0, end = var_25410_end_0, end_mask = var_25410_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25410_cast_fp16")]; + tensor var_25414_begin_0 = const()[name = tensor("op_25414_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_25414_end_0 = const()[name = tensor("op_25414_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_25414_end_mask_0 = const()[name = tensor("op_25414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25414_cast_fp16 = slice_by_index(begin = var_25414_begin_0, end = var_25414_end_0, end_mask = var_25414_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25414_cast_fp16")]; + tensor var_25418_begin_0 = const()[name = tensor("op_25418_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_25418_end_0 = const()[name = tensor("op_25418_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_25418_end_mask_0 = const()[name = tensor("op_25418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25418_cast_fp16 = slice_by_index(begin = var_25418_begin_0, end = var_25418_end_0, end_mask = var_25418_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25418_cast_fp16")]; + tensor var_25422_begin_0 = const()[name = tensor("op_25422_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_25422_end_0 = const()[name = tensor("op_25422_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_25422_end_mask_0 = const()[name = tensor("op_25422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25422_cast_fp16 = slice_by_index(begin = var_25422_begin_0, end = var_25422_end_0, end_mask = var_25422_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25422_cast_fp16")]; + tensor var_25426_begin_0 = const()[name = tensor("op_25426_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_25426_end_0 = const()[name = tensor("op_25426_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_25426_end_mask_0 = const()[name = tensor("op_25426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25426_cast_fp16 = slice_by_index(begin = var_25426_begin_0, end = var_25426_end_0, end_mask = var_25426_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25426_cast_fp16")]; + tensor var_25430_begin_0 = const()[name = tensor("op_25430_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_25430_end_0 = const()[name = tensor("op_25430_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_25430_end_mask_0 = const()[name = tensor("op_25430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25430_cast_fp16 = slice_by_index(begin = var_25430_begin_0, end = var_25430_end_0, end_mask = var_25430_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25430_cast_fp16")]; + tensor var_25434_begin_0 = const()[name = tensor("op_25434_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25434_end_0 = const()[name = tensor("op_25434_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_25434_end_mask_0 = const()[name = tensor("op_25434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25434_cast_fp16 = slice_by_index(begin = var_25434_begin_0, end = var_25434_end_0, end_mask = var_25434_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25434_cast_fp16")]; + tensor var_25438_begin_0 = const()[name = tensor("op_25438_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25438_end_0 = const()[name = tensor("op_25438_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_25438_end_mask_0 = const()[name = tensor("op_25438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25438_cast_fp16 = slice_by_index(begin = var_25438_begin_0, end = var_25438_end_0, end_mask = var_25438_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25438_cast_fp16")]; + tensor var_25442_begin_0 = const()[name = tensor("op_25442_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25442_end_0 = const()[name = tensor("op_25442_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_25442_end_mask_0 = const()[name = tensor("op_25442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25442_cast_fp16 = slice_by_index(begin = var_25442_begin_0, end = var_25442_end_0, end_mask = var_25442_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25442_cast_fp16")]; + tensor var_25446_begin_0 = const()[name = tensor("op_25446_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25446_end_0 = const()[name = tensor("op_25446_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_25446_end_mask_0 = const()[name = tensor("op_25446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25446_cast_fp16 = slice_by_index(begin = var_25446_begin_0, end = var_25446_end_0, end_mask = var_25446_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25446_cast_fp16")]; + tensor var_25448_begin_0 = const()[name = tensor("op_25448_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25448_end_0 = const()[name = tensor("op_25448_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_25448_end_mask_0 = const()[name = tensor("op_25448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25448_cast_fp16 = slice_by_index(begin = var_25448_begin_0, end = var_25448_end_0, end_mask = var_25448_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25448_cast_fp16")]; + tensor var_25452_begin_0 = const()[name = tensor("op_25452_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25452_end_0 = const()[name = tensor("op_25452_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_25452_end_mask_0 = const()[name = tensor("op_25452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25452_cast_fp16 = slice_by_index(begin = var_25452_begin_0, end = var_25452_end_0, end_mask = var_25452_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25452_cast_fp16")]; + tensor var_25456_begin_0 = const()[name = tensor("op_25456_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25456_end_0 = const()[name = tensor("op_25456_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_25456_end_mask_0 = const()[name = tensor("op_25456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25456_cast_fp16 = slice_by_index(begin = var_25456_begin_0, end = var_25456_end_0, end_mask = var_25456_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25456_cast_fp16")]; + tensor var_25460_begin_0 = const()[name = tensor("op_25460_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25460_end_0 = const()[name = tensor("op_25460_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_25460_end_mask_0 = const()[name = tensor("op_25460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25460_cast_fp16 = slice_by_index(begin = var_25460_begin_0, end = var_25460_end_0, end_mask = var_25460_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25460_cast_fp16")]; + tensor var_25464_begin_0 = const()[name = tensor("op_25464_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25464_end_0 = const()[name = tensor("op_25464_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_25464_end_mask_0 = const()[name = tensor("op_25464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25464_cast_fp16 = slice_by_index(begin = var_25464_begin_0, end = var_25464_end_0, end_mask = var_25464_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25464_cast_fp16")]; + tensor var_25468_begin_0 = const()[name = tensor("op_25468_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25468_end_0 = const()[name = tensor("op_25468_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_25468_end_mask_0 = const()[name = tensor("op_25468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25468_cast_fp16 = slice_by_index(begin = var_25468_begin_0, end = var_25468_end_0, end_mask = var_25468_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25468_cast_fp16")]; + tensor var_25472_begin_0 = const()[name = tensor("op_25472_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25472_end_0 = const()[name = tensor("op_25472_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_25472_end_mask_0 = const()[name = tensor("op_25472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25472_cast_fp16 = slice_by_index(begin = var_25472_begin_0, end = var_25472_end_0, end_mask = var_25472_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25472_cast_fp16")]; + tensor var_25476_begin_0 = const()[name = tensor("op_25476_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25476_end_0 = const()[name = tensor("op_25476_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_25476_end_mask_0 = const()[name = tensor("op_25476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25476_cast_fp16 = slice_by_index(begin = var_25476_begin_0, end = var_25476_end_0, end_mask = var_25476_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25476_cast_fp16")]; + tensor var_25480_begin_0 = const()[name = tensor("op_25480_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25480_end_0 = const()[name = tensor("op_25480_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_25480_end_mask_0 = const()[name = tensor("op_25480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25480_cast_fp16 = slice_by_index(begin = var_25480_begin_0, end = var_25480_end_0, end_mask = var_25480_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25480_cast_fp16")]; + tensor var_25484_begin_0 = const()[name = tensor("op_25484_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25484_end_0 = const()[name = tensor("op_25484_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_25484_end_mask_0 = const()[name = tensor("op_25484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25484_cast_fp16 = slice_by_index(begin = var_25484_begin_0, end = var_25484_end_0, end_mask = var_25484_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25484_cast_fp16")]; + tensor var_25488_begin_0 = const()[name = tensor("op_25488_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25488_end_0 = const()[name = tensor("op_25488_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_25488_end_mask_0 = const()[name = tensor("op_25488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25488_cast_fp16 = slice_by_index(begin = var_25488_begin_0, end = var_25488_end_0, end_mask = var_25488_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25488_cast_fp16")]; + tensor var_25492_begin_0 = const()[name = tensor("op_25492_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25492_end_0 = const()[name = tensor("op_25492_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_25492_end_mask_0 = const()[name = tensor("op_25492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25492_cast_fp16 = slice_by_index(begin = var_25492_begin_0, end = var_25492_end_0, end_mask = var_25492_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25492_cast_fp16")]; + tensor var_25496_begin_0 = const()[name = tensor("op_25496_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25496_end_0 = const()[name = tensor("op_25496_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_25496_end_mask_0 = const()[name = tensor("op_25496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25496_cast_fp16 = slice_by_index(begin = var_25496_begin_0, end = var_25496_end_0, end_mask = var_25496_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25496_cast_fp16")]; + tensor var_25500_begin_0 = const()[name = tensor("op_25500_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25500_end_0 = const()[name = tensor("op_25500_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_25500_end_mask_0 = const()[name = tensor("op_25500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25500_cast_fp16 = slice_by_index(begin = var_25500_begin_0, end = var_25500_end_0, end_mask = var_25500_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25500_cast_fp16")]; + tensor var_25504_begin_0 = const()[name = tensor("op_25504_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25504_end_0 = const()[name = tensor("op_25504_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_25504_end_mask_0 = const()[name = tensor("op_25504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25504_cast_fp16 = slice_by_index(begin = var_25504_begin_0, end = var_25504_end_0, end_mask = var_25504_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25504_cast_fp16")]; + tensor var_25508_begin_0 = const()[name = tensor("op_25508_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25508_end_0 = const()[name = tensor("op_25508_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_25508_end_mask_0 = const()[name = tensor("op_25508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25508_cast_fp16 = slice_by_index(begin = var_25508_begin_0, end = var_25508_end_0, end_mask = var_25508_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25508_cast_fp16")]; + tensor var_25512_begin_0 = const()[name = tensor("op_25512_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25512_end_0 = const()[name = tensor("op_25512_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_25512_end_mask_0 = const()[name = tensor("op_25512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25512_cast_fp16 = slice_by_index(begin = var_25512_begin_0, end = var_25512_end_0, end_mask = var_25512_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25512_cast_fp16")]; + tensor var_25516_begin_0 = const()[name = tensor("op_25516_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25516_end_0 = const()[name = tensor("op_25516_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_25516_end_mask_0 = const()[name = tensor("op_25516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25516_cast_fp16 = slice_by_index(begin = var_25516_begin_0, end = var_25516_end_0, end_mask = var_25516_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25516_cast_fp16")]; + tensor var_25520_begin_0 = const()[name = tensor("op_25520_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25520_end_0 = const()[name = tensor("op_25520_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_25520_end_mask_0 = const()[name = tensor("op_25520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25520_cast_fp16 = slice_by_index(begin = var_25520_begin_0, end = var_25520_end_0, end_mask = var_25520_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25520_cast_fp16")]; + tensor var_25524_begin_0 = const()[name = tensor("op_25524_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25524_end_0 = const()[name = tensor("op_25524_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_25524_end_mask_0 = const()[name = tensor("op_25524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25524_cast_fp16 = slice_by_index(begin = var_25524_begin_0, end = var_25524_end_0, end_mask = var_25524_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25524_cast_fp16")]; + tensor var_25528_equation_0 = const()[name = tensor("op_25528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25528_cast_fp16 = einsum(equation = var_25528_equation_0, values = (var_25370_cast_fp16, var_25287_cast_fp16))[name = tensor("op_25528_cast_fp16")]; + tensor var_25529_to_fp16 = const()[name = tensor("op_25529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2121_cast_fp16 = mul(x = var_25528_cast_fp16, y = var_25529_to_fp16)[name = tensor("aw_2121_cast_fp16")]; + tensor var_25532_equation_0 = const()[name = tensor("op_25532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25532_cast_fp16 = einsum(equation = var_25532_equation_0, values = (var_25374_cast_fp16, var_25291_cast_fp16))[name = tensor("op_25532_cast_fp16")]; + tensor var_25533_to_fp16 = const()[name = tensor("op_25533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2123_cast_fp16 = mul(x = var_25532_cast_fp16, y = var_25533_to_fp16)[name = tensor("aw_2123_cast_fp16")]; + tensor var_25536_equation_0 = const()[name = tensor("op_25536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25536_cast_fp16 = einsum(equation = var_25536_equation_0, values = (var_25378_cast_fp16, var_25295_cast_fp16))[name = tensor("op_25536_cast_fp16")]; + tensor var_25537_to_fp16 = const()[name = tensor("op_25537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2125_cast_fp16 = mul(x = var_25536_cast_fp16, y = var_25537_to_fp16)[name = tensor("aw_2125_cast_fp16")]; + tensor var_25540_equation_0 = const()[name = tensor("op_25540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25540_cast_fp16 = einsum(equation = var_25540_equation_0, values = (var_25382_cast_fp16, var_25299_cast_fp16))[name = tensor("op_25540_cast_fp16")]; + tensor var_25541_to_fp16 = const()[name = tensor("op_25541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2127_cast_fp16 = mul(x = var_25540_cast_fp16, y = var_25541_to_fp16)[name = tensor("aw_2127_cast_fp16")]; + tensor var_25544_equation_0 = const()[name = tensor("op_25544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25544_cast_fp16 = einsum(equation = var_25544_equation_0, values = (var_25386_cast_fp16, var_25303_cast_fp16))[name = tensor("op_25544_cast_fp16")]; + tensor var_25545_to_fp16 = const()[name = tensor("op_25545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2129_cast_fp16 = mul(x = var_25544_cast_fp16, y = var_25545_to_fp16)[name = tensor("aw_2129_cast_fp16")]; + tensor var_25548_equation_0 = const()[name = tensor("op_25548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25548_cast_fp16 = einsum(equation = var_25548_equation_0, values = (var_25390_cast_fp16, var_25307_cast_fp16))[name = tensor("op_25548_cast_fp16")]; + tensor var_25549_to_fp16 = const()[name = tensor("op_25549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2131_cast_fp16 = mul(x = var_25548_cast_fp16, y = var_25549_to_fp16)[name = tensor("aw_2131_cast_fp16")]; + tensor var_25552_equation_0 = const()[name = tensor("op_25552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25552_cast_fp16 = einsum(equation = var_25552_equation_0, values = (var_25394_cast_fp16, var_25311_cast_fp16))[name = tensor("op_25552_cast_fp16")]; + tensor var_25553_to_fp16 = const()[name = tensor("op_25553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2133_cast_fp16 = mul(x = var_25552_cast_fp16, y = var_25553_to_fp16)[name = tensor("aw_2133_cast_fp16")]; + tensor var_25556_equation_0 = const()[name = tensor("op_25556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25556_cast_fp16 = einsum(equation = var_25556_equation_0, values = (var_25398_cast_fp16, var_25315_cast_fp16))[name = tensor("op_25556_cast_fp16")]; + tensor var_25557_to_fp16 = const()[name = tensor("op_25557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2135_cast_fp16 = mul(x = var_25556_cast_fp16, y = var_25557_to_fp16)[name = tensor("aw_2135_cast_fp16")]; + tensor var_25560_equation_0 = const()[name = tensor("op_25560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25560_cast_fp16 = einsum(equation = var_25560_equation_0, values = (var_25402_cast_fp16, var_25319_cast_fp16))[name = tensor("op_25560_cast_fp16")]; + tensor var_25561_to_fp16 = const()[name = tensor("op_25561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2137_cast_fp16 = mul(x = var_25560_cast_fp16, y = var_25561_to_fp16)[name = tensor("aw_2137_cast_fp16")]; + tensor var_25564_equation_0 = const()[name = tensor("op_25564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25564_cast_fp16 = einsum(equation = var_25564_equation_0, values = (var_25406_cast_fp16, var_25323_cast_fp16))[name = tensor("op_25564_cast_fp16")]; + tensor var_25565_to_fp16 = const()[name = tensor("op_25565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2139_cast_fp16 = mul(x = var_25564_cast_fp16, y = var_25565_to_fp16)[name = tensor("aw_2139_cast_fp16")]; + tensor var_25568_equation_0 = const()[name = tensor("op_25568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25568_cast_fp16 = einsum(equation = var_25568_equation_0, values = (var_25410_cast_fp16, var_25327_cast_fp16))[name = tensor("op_25568_cast_fp16")]; + tensor var_25569_to_fp16 = const()[name = tensor("op_25569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2141_cast_fp16 = mul(x = var_25568_cast_fp16, y = var_25569_to_fp16)[name = tensor("aw_2141_cast_fp16")]; + tensor var_25572_equation_0 = const()[name = tensor("op_25572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25572_cast_fp16 = einsum(equation = var_25572_equation_0, values = (var_25414_cast_fp16, var_25331_cast_fp16))[name = tensor("op_25572_cast_fp16")]; + tensor var_25573_to_fp16 = const()[name = tensor("op_25573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2143_cast_fp16 = mul(x = var_25572_cast_fp16, y = var_25573_to_fp16)[name = tensor("aw_2143_cast_fp16")]; + tensor var_25576_equation_0 = const()[name = tensor("op_25576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25576_cast_fp16 = einsum(equation = var_25576_equation_0, values = (var_25418_cast_fp16, var_25335_cast_fp16))[name = tensor("op_25576_cast_fp16")]; + tensor var_25577_to_fp16 = const()[name = tensor("op_25577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2145_cast_fp16 = mul(x = var_25576_cast_fp16, y = var_25577_to_fp16)[name = tensor("aw_2145_cast_fp16")]; + tensor var_25580_equation_0 = const()[name = tensor("op_25580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25580_cast_fp16 = einsum(equation = var_25580_equation_0, values = (var_25422_cast_fp16, var_25339_cast_fp16))[name = tensor("op_25580_cast_fp16")]; + tensor var_25581_to_fp16 = const()[name = tensor("op_25581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2147_cast_fp16 = mul(x = var_25580_cast_fp16, y = var_25581_to_fp16)[name = tensor("aw_2147_cast_fp16")]; + tensor var_25584_equation_0 = const()[name = tensor("op_25584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25584_cast_fp16 = einsum(equation = var_25584_equation_0, values = (var_25426_cast_fp16, var_25343_cast_fp16))[name = tensor("op_25584_cast_fp16")]; + tensor var_25585_to_fp16 = const()[name = tensor("op_25585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2149_cast_fp16 = mul(x = var_25584_cast_fp16, y = var_25585_to_fp16)[name = tensor("aw_2149_cast_fp16")]; + tensor var_25588_equation_0 = const()[name = tensor("op_25588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25588_cast_fp16 = einsum(equation = var_25588_equation_0, values = (var_25430_cast_fp16, var_25347_cast_fp16))[name = tensor("op_25588_cast_fp16")]; + tensor var_25589_to_fp16 = const()[name = tensor("op_25589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2151_cast_fp16 = mul(x = var_25588_cast_fp16, y = var_25589_to_fp16)[name = tensor("aw_2151_cast_fp16")]; + tensor var_25592_equation_0 = const()[name = tensor("op_25592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25592_cast_fp16 = einsum(equation = var_25592_equation_0, values = (var_25434_cast_fp16, var_25351_cast_fp16))[name = tensor("op_25592_cast_fp16")]; + tensor var_25593_to_fp16 = const()[name = tensor("op_25593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2153_cast_fp16 = mul(x = var_25592_cast_fp16, y = var_25593_to_fp16)[name = tensor("aw_2153_cast_fp16")]; + tensor var_25596_equation_0 = const()[name = tensor("op_25596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25596_cast_fp16 = einsum(equation = var_25596_equation_0, values = (var_25438_cast_fp16, var_25355_cast_fp16))[name = tensor("op_25596_cast_fp16")]; + tensor var_25597_to_fp16 = const()[name = tensor("op_25597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2155_cast_fp16 = mul(x = var_25596_cast_fp16, y = var_25597_to_fp16)[name = tensor("aw_2155_cast_fp16")]; + tensor var_25600_equation_0 = const()[name = tensor("op_25600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25600_cast_fp16 = einsum(equation = var_25600_equation_0, values = (var_25442_cast_fp16, var_25359_cast_fp16))[name = tensor("op_25600_cast_fp16")]; + tensor var_25601_to_fp16 = const()[name = tensor("op_25601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2157_cast_fp16 = mul(x = var_25600_cast_fp16, y = var_25601_to_fp16)[name = tensor("aw_2157_cast_fp16")]; + tensor var_25604_equation_0 = const()[name = tensor("op_25604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25604_cast_fp16 = einsum(equation = var_25604_equation_0, values = (var_25446_cast_fp16, var_25363_cast_fp16))[name = tensor("op_25604_cast_fp16")]; + tensor var_25605_to_fp16 = const()[name = tensor("op_25605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2159_cast_fp16 = mul(x = var_25604_cast_fp16, y = var_25605_to_fp16)[name = tensor("aw_2159_cast_fp16")]; + tensor var_25607_cast_fp16 = softmax(axis = var_21077, x = aw_2121_cast_fp16)[name = tensor("op_25607_cast_fp16")]; + tensor var_25608_cast_fp16 = softmax(axis = var_21077, x = aw_2123_cast_fp16)[name = tensor("op_25608_cast_fp16")]; + tensor var_25609_cast_fp16 = softmax(axis = var_21077, x = aw_2125_cast_fp16)[name = tensor("op_25609_cast_fp16")]; + tensor var_25610_cast_fp16 = softmax(axis = var_21077, x = aw_2127_cast_fp16)[name = tensor("op_25610_cast_fp16")]; + tensor var_25611_cast_fp16 = softmax(axis = var_21077, x = aw_2129_cast_fp16)[name = tensor("op_25611_cast_fp16")]; + tensor var_25612_cast_fp16 = softmax(axis = var_21077, x = aw_2131_cast_fp16)[name = tensor("op_25612_cast_fp16")]; + tensor var_25613_cast_fp16 = softmax(axis = var_21077, x = aw_2133_cast_fp16)[name = tensor("op_25613_cast_fp16")]; + tensor var_25614_cast_fp16 = softmax(axis = var_21077, x = aw_2135_cast_fp16)[name = tensor("op_25614_cast_fp16")]; + tensor var_25615_cast_fp16 = softmax(axis = var_21077, x = aw_2137_cast_fp16)[name = tensor("op_25615_cast_fp16")]; + tensor var_25616_cast_fp16 = softmax(axis = var_21077, x = aw_2139_cast_fp16)[name = tensor("op_25616_cast_fp16")]; + tensor var_25617_cast_fp16 = softmax(axis = var_21077, x = aw_2141_cast_fp16)[name = tensor("op_25617_cast_fp16")]; + tensor var_25618_cast_fp16 = softmax(axis = var_21077, x = aw_2143_cast_fp16)[name = tensor("op_25618_cast_fp16")]; + tensor var_25619_cast_fp16 = softmax(axis = var_21077, x = aw_2145_cast_fp16)[name = tensor("op_25619_cast_fp16")]; + tensor var_25620_cast_fp16 = softmax(axis = var_21077, x = aw_2147_cast_fp16)[name = tensor("op_25620_cast_fp16")]; + tensor var_25621_cast_fp16 = softmax(axis = var_21077, x = aw_2149_cast_fp16)[name = tensor("op_25621_cast_fp16")]; + tensor var_25622_cast_fp16 = softmax(axis = var_21077, x = aw_2151_cast_fp16)[name = tensor("op_25622_cast_fp16")]; + tensor var_25623_cast_fp16 = softmax(axis = var_21077, x = aw_2153_cast_fp16)[name = tensor("op_25623_cast_fp16")]; + tensor var_25624_cast_fp16 = softmax(axis = var_21077, x = aw_2155_cast_fp16)[name = tensor("op_25624_cast_fp16")]; + tensor var_25625_cast_fp16 = softmax(axis = var_21077, x = aw_2157_cast_fp16)[name = tensor("op_25625_cast_fp16")]; + tensor var_25626_cast_fp16 = softmax(axis = var_21077, x = aw_2159_cast_fp16)[name = tensor("op_25626_cast_fp16")]; + tensor var_25628_equation_0 = const()[name = tensor("op_25628_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25628_cast_fp16 = einsum(equation = var_25628_equation_0, values = (var_25448_cast_fp16, var_25607_cast_fp16))[name = tensor("op_25628_cast_fp16")]; + tensor var_25630_equation_0 = const()[name = tensor("op_25630_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25630_cast_fp16 = einsum(equation = var_25630_equation_0, values = (var_25452_cast_fp16, var_25608_cast_fp16))[name = tensor("op_25630_cast_fp16")]; + tensor var_25632_equation_0 = const()[name = tensor("op_25632_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25632_cast_fp16 = einsum(equation = var_25632_equation_0, values = (var_25456_cast_fp16, var_25609_cast_fp16))[name = tensor("op_25632_cast_fp16")]; + tensor var_25634_equation_0 = const()[name = tensor("op_25634_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25634_cast_fp16 = einsum(equation = var_25634_equation_0, values = (var_25460_cast_fp16, var_25610_cast_fp16))[name = tensor("op_25634_cast_fp16")]; + tensor var_25636_equation_0 = const()[name = tensor("op_25636_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25636_cast_fp16 = einsum(equation = var_25636_equation_0, values = (var_25464_cast_fp16, var_25611_cast_fp16))[name = tensor("op_25636_cast_fp16")]; + tensor var_25638_equation_0 = const()[name = tensor("op_25638_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25638_cast_fp16 = einsum(equation = var_25638_equation_0, values = (var_25468_cast_fp16, var_25612_cast_fp16))[name = tensor("op_25638_cast_fp16")]; + tensor var_25640_equation_0 = const()[name = tensor("op_25640_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25640_cast_fp16 = einsum(equation = var_25640_equation_0, values = (var_25472_cast_fp16, var_25613_cast_fp16))[name = tensor("op_25640_cast_fp16")]; + tensor var_25642_equation_0 = const()[name = tensor("op_25642_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25642_cast_fp16 = einsum(equation = var_25642_equation_0, values = (var_25476_cast_fp16, var_25614_cast_fp16))[name = tensor("op_25642_cast_fp16")]; + tensor var_25644_equation_0 = const()[name = tensor("op_25644_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25644_cast_fp16 = einsum(equation = var_25644_equation_0, values = (var_25480_cast_fp16, var_25615_cast_fp16))[name = tensor("op_25644_cast_fp16")]; + tensor var_25646_equation_0 = const()[name = tensor("op_25646_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25646_cast_fp16 = einsum(equation = var_25646_equation_0, values = (var_25484_cast_fp16, var_25616_cast_fp16))[name = tensor("op_25646_cast_fp16")]; + tensor var_25648_equation_0 = const()[name = tensor("op_25648_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25648_cast_fp16 = einsum(equation = var_25648_equation_0, values = (var_25488_cast_fp16, var_25617_cast_fp16))[name = tensor("op_25648_cast_fp16")]; + tensor var_25650_equation_0 = const()[name = tensor("op_25650_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25650_cast_fp16 = einsum(equation = var_25650_equation_0, values = (var_25492_cast_fp16, var_25618_cast_fp16))[name = tensor("op_25650_cast_fp16")]; + tensor var_25652_equation_0 = const()[name = tensor("op_25652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25652_cast_fp16 = einsum(equation = var_25652_equation_0, values = (var_25496_cast_fp16, var_25619_cast_fp16))[name = tensor("op_25652_cast_fp16")]; + tensor var_25654_equation_0 = const()[name = tensor("op_25654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25654_cast_fp16 = einsum(equation = var_25654_equation_0, values = (var_25500_cast_fp16, var_25620_cast_fp16))[name = tensor("op_25654_cast_fp16")]; + tensor var_25656_equation_0 = const()[name = tensor("op_25656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25656_cast_fp16 = einsum(equation = var_25656_equation_0, values = (var_25504_cast_fp16, var_25621_cast_fp16))[name = tensor("op_25656_cast_fp16")]; + tensor var_25658_equation_0 = const()[name = tensor("op_25658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25658_cast_fp16 = einsum(equation = var_25658_equation_0, values = (var_25508_cast_fp16, var_25622_cast_fp16))[name = tensor("op_25658_cast_fp16")]; + tensor var_25660_equation_0 = const()[name = tensor("op_25660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25660_cast_fp16 = einsum(equation = var_25660_equation_0, values = (var_25512_cast_fp16, var_25623_cast_fp16))[name = tensor("op_25660_cast_fp16")]; + tensor var_25662_equation_0 = const()[name = tensor("op_25662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25662_cast_fp16 = einsum(equation = var_25662_equation_0, values = (var_25516_cast_fp16, var_25624_cast_fp16))[name = tensor("op_25662_cast_fp16")]; + tensor var_25664_equation_0 = const()[name = tensor("op_25664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25664_cast_fp16 = einsum(equation = var_25664_equation_0, values = (var_25520_cast_fp16, var_25625_cast_fp16))[name = tensor("op_25664_cast_fp16")]; + tensor var_25666_equation_0 = const()[name = tensor("op_25666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25666_cast_fp16 = einsum(equation = var_25666_equation_0, values = (var_25524_cast_fp16, var_25626_cast_fp16))[name = tensor("op_25666_cast_fp16")]; + tensor input_361_interleave_0 = const()[name = tensor("input_361_interleave_0"), val = tensor(false)]; + tensor input_361_cast_fp16 = concat(axis = var_21077, interleave = input_361_interleave_0, values = (var_25628_cast_fp16, var_25630_cast_fp16, var_25632_cast_fp16, var_25634_cast_fp16, var_25636_cast_fp16, var_25638_cast_fp16, var_25640_cast_fp16, var_25642_cast_fp16, var_25644_cast_fp16, var_25646_cast_fp16, var_25648_cast_fp16, var_25650_cast_fp16, var_25652_cast_fp16, var_25654_cast_fp16, var_25656_cast_fp16, var_25658_cast_fp16, var_25660_cast_fp16, var_25662_cast_fp16, var_25664_cast_fp16, var_25666_cast_fp16))[name = tensor("input_361_cast_fp16")]; + tensor var_25676_pad_type_0 = const()[name = tensor("op_25676_pad_type_0"), val = tensor("valid")]; + tensor var_25676_strides_0 = const()[name = tensor("op_25676_strides_0"), val = tensor([1, 1])]; + tensor var_25676_pad_0 = const()[name = tensor("op_25676_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25676_dilations_0 = const()[name = tensor("op_25676_dilations_0"), val = tensor([1, 1])]; + tensor var_25676_groups_0 = const()[name = tensor("op_25676_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762535616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763764480))), name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763764672)))]; + tensor var_25676_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_25676_dilations_0, groups = var_25676_groups_0, pad = var_25676_pad_0, pad_type = var_25676_pad_type_0, strides = var_25676_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized, x = input_361_cast_fp16)[name = tensor("op_25676_cast_fp16")]; + tensor inputs_173_cast_fp16 = add(x = var_25676_cast_fp16, y = inputs_171_cast_fp16)[name = tensor("inputs_173_cast_fp16")]; + tensor input_363_axes_0 = const()[name = tensor("input_363_axes_0"), val = tensor([1])]; + tensor input_363_gamma_0_to_fp16 = const()[name = tensor("input_363_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763767296)))]; + tensor input_363_beta_0_to_fp16 = const()[name = tensor("input_363_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763769920)))]; + tensor var_25686_to_fp16 = const()[name = tensor("op_25686_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_363_cast_fp16 = layer_norm(axes = input_363_axes_0, beta = input_363_beta_0_to_fp16, epsilon = var_25686_to_fp16, gamma = input_363_gamma_0_to_fp16, x = inputs_173_cast_fp16)[name = tensor("input_363_cast_fp16")]; + tensor var_25706_pad_type_0 = const()[name = tensor("op_25706_pad_type_0"), val = tensor("valid")]; + tensor var_25706_strides_0 = const()[name = tensor("op_25706_strides_0"), val = tensor([1, 1])]; + tensor var_25706_pad_0 = const()[name = tensor("op_25706_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25706_dilations_0 = const()[name = tensor("op_25706_dilations_0"), val = tensor([1, 1])]; + tensor var_25706_groups_0 = const()[name = tensor("op_25706_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763772544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773603008))), name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773603200)))]; + tensor var_25706_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_25706_dilations_0, groups = var_25706_groups_0, pad = var_25706_pad_0, pad_type = var_25706_pad_type_0, strides = var_25706_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized, x = input_363_cast_fp16)[name = tensor("op_25706_cast_fp16")]; + tensor var_25707_split_sizes_0 = const()[name = tensor("op_25707_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_25707_axis_0 = const()[name = tensor("op_25707_axis_0"), val = tensor(1)]; + tensor var_25707_cast_fp16_0, tensor var_25707_cast_fp16_1 = split(axis = var_25707_axis_0, split_sizes = var_25707_split_sizes_0, x = var_25706_cast_fp16)[name = tensor("op_25707_cast_fp16")]; + tensor var_25709_mode_0 = const()[name = tensor("op_25709_mode_0"), val = tensor("EXACT")]; + tensor var_25709_cast_fp16 = gelu(mode = var_25709_mode_0, x = var_25707_cast_fp16_1)[name = tensor("op_25709_cast_fp16")]; + tensor input_365_cast_fp16 = mul(x = var_25707_cast_fp16_0, y = var_25709_cast_fp16)[name = tensor("input_365_cast_fp16")]; + tensor var_25717_pad_type_0 = const()[name = tensor("op_25717_pad_type_0"), val = tensor("valid")]; + tensor var_25717_strides_0 = const()[name = tensor("op_25717_strides_0"), val = tensor([1, 1])]; + tensor var_25717_pad_0 = const()[name = tensor("op_25717_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25717_dilations_0 = const()[name = tensor("op_25717_dilations_0"), val = tensor([1, 1])]; + tensor var_25717_groups_0 = const()[name = tensor("op_25717_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773623744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778539008))), name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778539200)))]; + tensor var_25717_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_25717_dilations_0, groups = var_25717_groups_0, pad = var_25717_pad_0, pad_type = var_25717_pad_type_0, strides = var_25717_strides_0, weight = mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized, x = input_365_cast_fp16)[name = tensor("op_25717_cast_fp16")]; + tensor inputs_175_cast_fp16 = add(x = var_25717_cast_fp16, y = inputs_173_cast_fp16)[name = tensor("inputs_175_cast_fp16")]; + tensor hidden_states_239_axes_0 = const()[name = tensor("hidden_states_239_axes_0"), val = tensor([1])]; + tensor hidden_states_239_gamma_0_to_fp16 = const()[name = tensor("hidden_states_239_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778541824)))]; + tensor hidden_states_239_beta_0_to_fp16 = const()[name = tensor("hidden_states_239_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778544448)))]; + tensor var_25733_to_fp16 = const()[name = tensor("op_25733_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_239_cast_fp16 = layer_norm(axes = hidden_states_239_axes_0, beta = hidden_states_239_beta_0_to_fp16, epsilon = var_25733_to_fp16, gamma = hidden_states_239_gamma_0_to_fp16, x = inputs_175_cast_fp16)[name = tensor("hidden_states_239_cast_fp16")]; + tensor q_117_pad_type_0 = const()[name = tensor("q_117_pad_type_0"), val = tensor("valid")]; + tensor q_117_strides_0 = const()[name = tensor("q_117_strides_0"), val = tensor([1, 1])]; + tensor q_117_pad_0 = const()[name = tensor("q_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_117_dilations_0 = const()[name = tensor("q_117_dilations_0"), val = tensor([1, 1])]; + tensor q_117_groups_0 = const()[name = tensor("q_117_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778547072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(779775936))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_117_cast_fp16 = conv(dilations = q_117_dilations_0, groups = q_117_groups_0, pad = q_117_pad_0, pad_type = q_117_pad_type_0, strides = q_117_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_239_cast_fp16)[name = tensor("q_117_cast_fp16")]; + tensor k_233_pad_type_0 = const()[name = tensor("k_233_pad_type_0"), val = tensor("valid")]; + tensor k_233_strides_0 = const()[name = tensor("k_233_strides_0"), val = tensor([1, 1])]; + tensor k_233_pad_0 = const()[name = tensor("k_233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_233_dilations_0 = const()[name = tensor("k_233_dilations_0"), val = tensor([1, 1])]; + tensor k_233_groups_0 = const()[name = tensor("k_233_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(779776128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(781004992))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_233_cast_fp16 = conv(dilations = k_233_dilations_0, groups = k_233_groups_0, pad = k_233_pad_0, pad_type = k_233_pad_type_0, strides = k_233_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_239_cast_fp16)[name = tensor("k_233_cast_fp16")]; + tensor v_117_pad_type_0 = const()[name = tensor("v_117_pad_type_0"), val = tensor("valid")]; + tensor v_117_strides_0 = const()[name = tensor("v_117_strides_0"), val = tensor([1, 1])]; + tensor v_117_pad_0 = const()[name = tensor("v_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_117_dilations_0 = const()[name = tensor("v_117_dilations_0"), val = tensor([1, 1])]; + tensor v_117_groups_0 = const()[name = tensor("v_117_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(781005184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(782234048))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_117_cast_fp16 = conv(dilations = v_117_dilations_0, groups = v_117_groups_0, pad = v_117_pad_0, pad_type = v_117_pad_type_0, strides = v_117_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_239_cast_fp16)[name = tensor("v_117_cast_fp16")]; + tensor var_25766_begin_0 = const()[name = tensor("op_25766_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25766_end_0 = const()[name = tensor("op_25766_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25766_end_mask_0 = const()[name = tensor("op_25766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25766_cast_fp16 = slice_by_index(begin = var_25766_begin_0, end = var_25766_end_0, end_mask = var_25766_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25766_cast_fp16")]; + tensor var_25770_begin_0 = const()[name = tensor("op_25770_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25770_end_0 = const()[name = tensor("op_25770_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25770_end_mask_0 = const()[name = tensor("op_25770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25770_cast_fp16 = slice_by_index(begin = var_25770_begin_0, end = var_25770_end_0, end_mask = var_25770_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25770_cast_fp16")]; + tensor var_25774_begin_0 = const()[name = tensor("op_25774_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25774_end_0 = const()[name = tensor("op_25774_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25774_end_mask_0 = const()[name = tensor("op_25774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25774_cast_fp16 = slice_by_index(begin = var_25774_begin_0, end = var_25774_end_0, end_mask = var_25774_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25774_cast_fp16")]; + tensor var_25778_begin_0 = const()[name = tensor("op_25778_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25778_end_0 = const()[name = tensor("op_25778_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25778_end_mask_0 = const()[name = tensor("op_25778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25778_cast_fp16 = slice_by_index(begin = var_25778_begin_0, end = var_25778_end_0, end_mask = var_25778_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25778_cast_fp16")]; + tensor var_25782_begin_0 = const()[name = tensor("op_25782_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25782_end_0 = const()[name = tensor("op_25782_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25782_end_mask_0 = const()[name = tensor("op_25782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25782_cast_fp16 = slice_by_index(begin = var_25782_begin_0, end = var_25782_end_0, end_mask = var_25782_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25782_cast_fp16")]; + tensor var_25786_begin_0 = const()[name = tensor("op_25786_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25786_end_0 = const()[name = tensor("op_25786_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25786_end_mask_0 = const()[name = tensor("op_25786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25786_cast_fp16 = slice_by_index(begin = var_25786_begin_0, end = var_25786_end_0, end_mask = var_25786_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25786_cast_fp16")]; + tensor var_25790_begin_0 = const()[name = tensor("op_25790_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25790_end_0 = const()[name = tensor("op_25790_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25790_end_mask_0 = const()[name = tensor("op_25790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25790_cast_fp16 = slice_by_index(begin = var_25790_begin_0, end = var_25790_end_0, end_mask = var_25790_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25790_cast_fp16")]; + tensor var_25794_begin_0 = const()[name = tensor("op_25794_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25794_end_0 = const()[name = tensor("op_25794_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25794_end_mask_0 = const()[name = tensor("op_25794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25794_cast_fp16 = slice_by_index(begin = var_25794_begin_0, end = var_25794_end_0, end_mask = var_25794_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25794_cast_fp16")]; + tensor var_25798_begin_0 = const()[name = tensor("op_25798_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25798_end_0 = const()[name = tensor("op_25798_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25798_end_mask_0 = const()[name = tensor("op_25798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25798_cast_fp16 = slice_by_index(begin = var_25798_begin_0, end = var_25798_end_0, end_mask = var_25798_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25798_cast_fp16")]; + tensor var_25802_begin_0 = const()[name = tensor("op_25802_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25802_end_0 = const()[name = tensor("op_25802_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25802_end_mask_0 = const()[name = tensor("op_25802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25802_cast_fp16 = slice_by_index(begin = var_25802_begin_0, end = var_25802_end_0, end_mask = var_25802_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25802_cast_fp16")]; + tensor var_25806_begin_0 = const()[name = tensor("op_25806_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25806_end_0 = const()[name = tensor("op_25806_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25806_end_mask_0 = const()[name = tensor("op_25806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25806_cast_fp16 = slice_by_index(begin = var_25806_begin_0, end = var_25806_end_0, end_mask = var_25806_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25806_cast_fp16")]; + tensor var_25810_begin_0 = const()[name = tensor("op_25810_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25810_end_0 = const()[name = tensor("op_25810_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25810_end_mask_0 = const()[name = tensor("op_25810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25810_cast_fp16 = slice_by_index(begin = var_25810_begin_0, end = var_25810_end_0, end_mask = var_25810_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25810_cast_fp16")]; + tensor var_25814_begin_0 = const()[name = tensor("op_25814_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25814_end_0 = const()[name = tensor("op_25814_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25814_end_mask_0 = const()[name = tensor("op_25814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25814_cast_fp16 = slice_by_index(begin = var_25814_begin_0, end = var_25814_end_0, end_mask = var_25814_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25814_cast_fp16")]; + tensor var_25818_begin_0 = const()[name = tensor("op_25818_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25818_end_0 = const()[name = tensor("op_25818_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25818_end_mask_0 = const()[name = tensor("op_25818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25818_cast_fp16 = slice_by_index(begin = var_25818_begin_0, end = var_25818_end_0, end_mask = var_25818_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25818_cast_fp16")]; + tensor var_25822_begin_0 = const()[name = tensor("op_25822_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25822_end_0 = const()[name = tensor("op_25822_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25822_end_mask_0 = const()[name = tensor("op_25822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25822_cast_fp16 = slice_by_index(begin = var_25822_begin_0, end = var_25822_end_0, end_mask = var_25822_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25822_cast_fp16")]; + tensor var_25826_begin_0 = const()[name = tensor("op_25826_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25826_end_0 = const()[name = tensor("op_25826_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25826_end_mask_0 = const()[name = tensor("op_25826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25826_cast_fp16 = slice_by_index(begin = var_25826_begin_0, end = var_25826_end_0, end_mask = var_25826_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25826_cast_fp16")]; + tensor var_25830_begin_0 = const()[name = tensor("op_25830_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25830_end_0 = const()[name = tensor("op_25830_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25830_end_mask_0 = const()[name = tensor("op_25830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25830_cast_fp16 = slice_by_index(begin = var_25830_begin_0, end = var_25830_end_0, end_mask = var_25830_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25830_cast_fp16")]; + tensor var_25834_begin_0 = const()[name = tensor("op_25834_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25834_end_0 = const()[name = tensor("op_25834_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25834_end_mask_0 = const()[name = tensor("op_25834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25834_cast_fp16 = slice_by_index(begin = var_25834_begin_0, end = var_25834_end_0, end_mask = var_25834_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25834_cast_fp16")]; + tensor var_25838_begin_0 = const()[name = tensor("op_25838_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25838_end_0 = const()[name = tensor("op_25838_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25838_end_mask_0 = const()[name = tensor("op_25838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25838_cast_fp16 = slice_by_index(begin = var_25838_begin_0, end = var_25838_end_0, end_mask = var_25838_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25838_cast_fp16")]; + tensor var_25842_begin_0 = const()[name = tensor("op_25842_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25842_end_0 = const()[name = tensor("op_25842_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_25842_end_mask_0 = const()[name = tensor("op_25842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25842_cast_fp16 = slice_by_index(begin = var_25842_begin_0, end = var_25842_end_0, end_mask = var_25842_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25842_cast_fp16")]; + tensor k_235_perm_0 = const()[name = tensor("k_235_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_25849_begin_0 = const()[name = tensor("op_25849_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25849_end_0 = const()[name = tensor("op_25849_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_25849_end_mask_0 = const()[name = tensor("op_25849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_235_cast_fp16 = transpose(perm = k_235_perm_0, x = k_233_cast_fp16)[name = tensor("transpose_9")]; + tensor var_25849_cast_fp16 = slice_by_index(begin = var_25849_begin_0, end = var_25849_end_0, end_mask = var_25849_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25849_cast_fp16")]; + tensor var_25853_begin_0 = const()[name = tensor("op_25853_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_25853_end_0 = const()[name = tensor("op_25853_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_25853_end_mask_0 = const()[name = tensor("op_25853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25853_cast_fp16 = slice_by_index(begin = var_25853_begin_0, end = var_25853_end_0, end_mask = var_25853_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25853_cast_fp16")]; + tensor var_25857_begin_0 = const()[name = tensor("op_25857_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_25857_end_0 = const()[name = tensor("op_25857_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_25857_end_mask_0 = const()[name = tensor("op_25857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25857_cast_fp16 = slice_by_index(begin = var_25857_begin_0, end = var_25857_end_0, end_mask = var_25857_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25857_cast_fp16")]; + tensor var_25861_begin_0 = const()[name = tensor("op_25861_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_25861_end_0 = const()[name = tensor("op_25861_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_25861_end_mask_0 = const()[name = tensor("op_25861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25861_cast_fp16 = slice_by_index(begin = var_25861_begin_0, end = var_25861_end_0, end_mask = var_25861_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25861_cast_fp16")]; + tensor var_25865_begin_0 = const()[name = tensor("op_25865_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_25865_end_0 = const()[name = tensor("op_25865_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_25865_end_mask_0 = const()[name = tensor("op_25865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25865_cast_fp16 = slice_by_index(begin = var_25865_begin_0, end = var_25865_end_0, end_mask = var_25865_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25865_cast_fp16")]; + tensor var_25869_begin_0 = const()[name = tensor("op_25869_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_25869_end_0 = const()[name = tensor("op_25869_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_25869_end_mask_0 = const()[name = tensor("op_25869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25869_cast_fp16 = slice_by_index(begin = var_25869_begin_0, end = var_25869_end_0, end_mask = var_25869_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25869_cast_fp16")]; + tensor var_25873_begin_0 = const()[name = tensor("op_25873_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_25873_end_0 = const()[name = tensor("op_25873_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_25873_end_mask_0 = const()[name = tensor("op_25873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25873_cast_fp16 = slice_by_index(begin = var_25873_begin_0, end = var_25873_end_0, end_mask = var_25873_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25873_cast_fp16")]; + tensor var_25877_begin_0 = const()[name = tensor("op_25877_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_25877_end_0 = const()[name = tensor("op_25877_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_25877_end_mask_0 = const()[name = tensor("op_25877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25877_cast_fp16 = slice_by_index(begin = var_25877_begin_0, end = var_25877_end_0, end_mask = var_25877_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25877_cast_fp16")]; + tensor var_25881_begin_0 = const()[name = tensor("op_25881_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_25881_end_0 = const()[name = tensor("op_25881_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_25881_end_mask_0 = const()[name = tensor("op_25881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25881_cast_fp16 = slice_by_index(begin = var_25881_begin_0, end = var_25881_end_0, end_mask = var_25881_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25881_cast_fp16")]; + tensor var_25885_begin_0 = const()[name = tensor("op_25885_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_25885_end_0 = const()[name = tensor("op_25885_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_25885_end_mask_0 = const()[name = tensor("op_25885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25885_cast_fp16 = slice_by_index(begin = var_25885_begin_0, end = var_25885_end_0, end_mask = var_25885_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25885_cast_fp16")]; + tensor var_25889_begin_0 = const()[name = tensor("op_25889_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_25889_end_0 = const()[name = tensor("op_25889_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_25889_end_mask_0 = const()[name = tensor("op_25889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25889_cast_fp16 = slice_by_index(begin = var_25889_begin_0, end = var_25889_end_0, end_mask = var_25889_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25889_cast_fp16")]; + tensor var_25893_begin_0 = const()[name = tensor("op_25893_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_25893_end_0 = const()[name = tensor("op_25893_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_25893_end_mask_0 = const()[name = tensor("op_25893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25893_cast_fp16 = slice_by_index(begin = var_25893_begin_0, end = var_25893_end_0, end_mask = var_25893_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25893_cast_fp16")]; + tensor var_25897_begin_0 = const()[name = tensor("op_25897_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_25897_end_0 = const()[name = tensor("op_25897_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_25897_end_mask_0 = const()[name = tensor("op_25897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25897_cast_fp16 = slice_by_index(begin = var_25897_begin_0, end = var_25897_end_0, end_mask = var_25897_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25897_cast_fp16")]; + tensor var_25901_begin_0 = const()[name = tensor("op_25901_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_25901_end_0 = const()[name = tensor("op_25901_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_25901_end_mask_0 = const()[name = tensor("op_25901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25901_cast_fp16 = slice_by_index(begin = var_25901_begin_0, end = var_25901_end_0, end_mask = var_25901_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25901_cast_fp16")]; + tensor var_25905_begin_0 = const()[name = tensor("op_25905_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_25905_end_0 = const()[name = tensor("op_25905_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_25905_end_mask_0 = const()[name = tensor("op_25905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25905_cast_fp16 = slice_by_index(begin = var_25905_begin_0, end = var_25905_end_0, end_mask = var_25905_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25905_cast_fp16")]; + tensor var_25909_begin_0 = const()[name = tensor("op_25909_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_25909_end_0 = const()[name = tensor("op_25909_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25909_end_mask_0 = const()[name = tensor("op_25909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25909_cast_fp16 = slice_by_index(begin = var_25909_begin_0, end = var_25909_end_0, end_mask = var_25909_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25909_cast_fp16")]; + tensor var_25913_begin_0 = const()[name = tensor("op_25913_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25913_end_0 = const()[name = tensor("op_25913_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_25913_end_mask_0 = const()[name = tensor("op_25913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25913_cast_fp16 = slice_by_index(begin = var_25913_begin_0, end = var_25913_end_0, end_mask = var_25913_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25913_cast_fp16")]; + tensor var_25917_begin_0 = const()[name = tensor("op_25917_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25917_end_0 = const()[name = tensor("op_25917_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_25917_end_mask_0 = const()[name = tensor("op_25917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25917_cast_fp16 = slice_by_index(begin = var_25917_begin_0, end = var_25917_end_0, end_mask = var_25917_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25917_cast_fp16")]; + tensor var_25921_begin_0 = const()[name = tensor("op_25921_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25921_end_0 = const()[name = tensor("op_25921_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_25921_end_mask_0 = const()[name = tensor("op_25921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25921_cast_fp16 = slice_by_index(begin = var_25921_begin_0, end = var_25921_end_0, end_mask = var_25921_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25921_cast_fp16")]; + tensor var_25925_begin_0 = const()[name = tensor("op_25925_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25925_end_0 = const()[name = tensor("op_25925_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_25925_end_mask_0 = const()[name = tensor("op_25925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25925_cast_fp16 = slice_by_index(begin = var_25925_begin_0, end = var_25925_end_0, end_mask = var_25925_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25925_cast_fp16")]; + tensor var_25927_begin_0 = const()[name = tensor("op_25927_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25927_end_0 = const()[name = tensor("op_25927_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25927_end_mask_0 = const()[name = tensor("op_25927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25927_cast_fp16 = slice_by_index(begin = var_25927_begin_0, end = var_25927_end_0, end_mask = var_25927_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25927_cast_fp16")]; + tensor var_25931_begin_0 = const()[name = tensor("op_25931_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25931_end_0 = const()[name = tensor("op_25931_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25931_end_mask_0 = const()[name = tensor("op_25931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25931_cast_fp16 = slice_by_index(begin = var_25931_begin_0, end = var_25931_end_0, end_mask = var_25931_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25931_cast_fp16")]; + tensor var_25935_begin_0 = const()[name = tensor("op_25935_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25935_end_0 = const()[name = tensor("op_25935_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25935_end_mask_0 = const()[name = tensor("op_25935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25935_cast_fp16 = slice_by_index(begin = var_25935_begin_0, end = var_25935_end_0, end_mask = var_25935_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25935_cast_fp16")]; + tensor var_25939_begin_0 = const()[name = tensor("op_25939_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25939_end_0 = const()[name = tensor("op_25939_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25939_end_mask_0 = const()[name = tensor("op_25939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25939_cast_fp16 = slice_by_index(begin = var_25939_begin_0, end = var_25939_end_0, end_mask = var_25939_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25939_cast_fp16")]; + tensor var_25943_begin_0 = const()[name = tensor("op_25943_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25943_end_0 = const()[name = tensor("op_25943_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25943_end_mask_0 = const()[name = tensor("op_25943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25943_cast_fp16 = slice_by_index(begin = var_25943_begin_0, end = var_25943_end_0, end_mask = var_25943_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25943_cast_fp16")]; + tensor var_25947_begin_0 = const()[name = tensor("op_25947_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25947_end_0 = const()[name = tensor("op_25947_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25947_end_mask_0 = const()[name = tensor("op_25947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25947_cast_fp16 = slice_by_index(begin = var_25947_begin_0, end = var_25947_end_0, end_mask = var_25947_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25947_cast_fp16")]; + tensor var_25951_begin_0 = const()[name = tensor("op_25951_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25951_end_0 = const()[name = tensor("op_25951_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25951_end_mask_0 = const()[name = tensor("op_25951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25951_cast_fp16 = slice_by_index(begin = var_25951_begin_0, end = var_25951_end_0, end_mask = var_25951_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25951_cast_fp16")]; + tensor var_25955_begin_0 = const()[name = tensor("op_25955_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25955_end_0 = const()[name = tensor("op_25955_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25955_end_mask_0 = const()[name = tensor("op_25955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25955_cast_fp16 = slice_by_index(begin = var_25955_begin_0, end = var_25955_end_0, end_mask = var_25955_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25955_cast_fp16")]; + tensor var_25959_begin_0 = const()[name = tensor("op_25959_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25959_end_0 = const()[name = tensor("op_25959_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25959_end_mask_0 = const()[name = tensor("op_25959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25959_cast_fp16 = slice_by_index(begin = var_25959_begin_0, end = var_25959_end_0, end_mask = var_25959_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25959_cast_fp16")]; + tensor var_25963_begin_0 = const()[name = tensor("op_25963_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25963_end_0 = const()[name = tensor("op_25963_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25963_end_mask_0 = const()[name = tensor("op_25963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25963_cast_fp16 = slice_by_index(begin = var_25963_begin_0, end = var_25963_end_0, end_mask = var_25963_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25963_cast_fp16")]; + tensor var_25967_begin_0 = const()[name = tensor("op_25967_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25967_end_0 = const()[name = tensor("op_25967_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25967_end_mask_0 = const()[name = tensor("op_25967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25967_cast_fp16 = slice_by_index(begin = var_25967_begin_0, end = var_25967_end_0, end_mask = var_25967_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25967_cast_fp16")]; + tensor var_25971_begin_0 = const()[name = tensor("op_25971_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25971_end_0 = const()[name = tensor("op_25971_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25971_end_mask_0 = const()[name = tensor("op_25971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25971_cast_fp16 = slice_by_index(begin = var_25971_begin_0, end = var_25971_end_0, end_mask = var_25971_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25971_cast_fp16")]; + tensor var_25975_begin_0 = const()[name = tensor("op_25975_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25975_end_0 = const()[name = tensor("op_25975_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25975_end_mask_0 = const()[name = tensor("op_25975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25975_cast_fp16 = slice_by_index(begin = var_25975_begin_0, end = var_25975_end_0, end_mask = var_25975_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25975_cast_fp16")]; + tensor var_25979_begin_0 = const()[name = tensor("op_25979_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25979_end_0 = const()[name = tensor("op_25979_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25979_end_mask_0 = const()[name = tensor("op_25979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25979_cast_fp16 = slice_by_index(begin = var_25979_begin_0, end = var_25979_end_0, end_mask = var_25979_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25979_cast_fp16")]; + tensor var_25983_begin_0 = const()[name = tensor("op_25983_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25983_end_0 = const()[name = tensor("op_25983_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25983_end_mask_0 = const()[name = tensor("op_25983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25983_cast_fp16 = slice_by_index(begin = var_25983_begin_0, end = var_25983_end_0, end_mask = var_25983_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25983_cast_fp16")]; + tensor var_25987_begin_0 = const()[name = tensor("op_25987_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25987_end_0 = const()[name = tensor("op_25987_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25987_end_mask_0 = const()[name = tensor("op_25987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25987_cast_fp16 = slice_by_index(begin = var_25987_begin_0, end = var_25987_end_0, end_mask = var_25987_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25987_cast_fp16")]; + tensor var_25991_begin_0 = const()[name = tensor("op_25991_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25991_end_0 = const()[name = tensor("op_25991_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25991_end_mask_0 = const()[name = tensor("op_25991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25991_cast_fp16 = slice_by_index(begin = var_25991_begin_0, end = var_25991_end_0, end_mask = var_25991_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25991_cast_fp16")]; + tensor var_25995_begin_0 = const()[name = tensor("op_25995_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25995_end_0 = const()[name = tensor("op_25995_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25995_end_mask_0 = const()[name = tensor("op_25995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25995_cast_fp16 = slice_by_index(begin = var_25995_begin_0, end = var_25995_end_0, end_mask = var_25995_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25995_cast_fp16")]; + tensor var_25999_begin_0 = const()[name = tensor("op_25999_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25999_end_0 = const()[name = tensor("op_25999_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25999_end_mask_0 = const()[name = tensor("op_25999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25999_cast_fp16 = slice_by_index(begin = var_25999_begin_0, end = var_25999_end_0, end_mask = var_25999_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25999_cast_fp16")]; + tensor var_26003_begin_0 = const()[name = tensor("op_26003_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26003_end_0 = const()[name = tensor("op_26003_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26003_end_mask_0 = const()[name = tensor("op_26003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26003_cast_fp16 = slice_by_index(begin = var_26003_begin_0, end = var_26003_end_0, end_mask = var_26003_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_26003_cast_fp16")]; + tensor var_26007_equation_0 = const()[name = tensor("op_26007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26007_cast_fp16 = einsum(equation = var_26007_equation_0, values = (var_25849_cast_fp16, var_25766_cast_fp16))[name = tensor("op_26007_cast_fp16")]; + tensor var_26008_to_fp16 = const()[name = tensor("op_26008_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2161_cast_fp16 = mul(x = var_26007_cast_fp16, y = var_26008_to_fp16)[name = tensor("aw_2161_cast_fp16")]; + tensor var_26011_equation_0 = const()[name = tensor("op_26011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26011_cast_fp16 = einsum(equation = var_26011_equation_0, values = (var_25853_cast_fp16, var_25770_cast_fp16))[name = tensor("op_26011_cast_fp16")]; + tensor var_26012_to_fp16 = const()[name = tensor("op_26012_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2163_cast_fp16 = mul(x = var_26011_cast_fp16, y = var_26012_to_fp16)[name = tensor("aw_2163_cast_fp16")]; + tensor var_26015_equation_0 = const()[name = tensor("op_26015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26015_cast_fp16 = einsum(equation = var_26015_equation_0, values = (var_25857_cast_fp16, var_25774_cast_fp16))[name = tensor("op_26015_cast_fp16")]; + tensor var_26016_to_fp16 = const()[name = tensor("op_26016_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2165_cast_fp16 = mul(x = var_26015_cast_fp16, y = var_26016_to_fp16)[name = tensor("aw_2165_cast_fp16")]; + tensor var_26019_equation_0 = const()[name = tensor("op_26019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26019_cast_fp16 = einsum(equation = var_26019_equation_0, values = (var_25861_cast_fp16, var_25778_cast_fp16))[name = tensor("op_26019_cast_fp16")]; + tensor var_26020_to_fp16 = const()[name = tensor("op_26020_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2167_cast_fp16 = mul(x = var_26019_cast_fp16, y = var_26020_to_fp16)[name = tensor("aw_2167_cast_fp16")]; + tensor var_26023_equation_0 = const()[name = tensor("op_26023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26023_cast_fp16 = einsum(equation = var_26023_equation_0, values = (var_25865_cast_fp16, var_25782_cast_fp16))[name = tensor("op_26023_cast_fp16")]; + tensor var_26024_to_fp16 = const()[name = tensor("op_26024_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2169_cast_fp16 = mul(x = var_26023_cast_fp16, y = var_26024_to_fp16)[name = tensor("aw_2169_cast_fp16")]; + tensor var_26027_equation_0 = const()[name = tensor("op_26027_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26027_cast_fp16 = einsum(equation = var_26027_equation_0, values = (var_25869_cast_fp16, var_25786_cast_fp16))[name = tensor("op_26027_cast_fp16")]; + tensor var_26028_to_fp16 = const()[name = tensor("op_26028_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2171_cast_fp16 = mul(x = var_26027_cast_fp16, y = var_26028_to_fp16)[name = tensor("aw_2171_cast_fp16")]; + tensor var_26031_equation_0 = const()[name = tensor("op_26031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26031_cast_fp16 = einsum(equation = var_26031_equation_0, values = (var_25873_cast_fp16, var_25790_cast_fp16))[name = tensor("op_26031_cast_fp16")]; + tensor var_26032_to_fp16 = const()[name = tensor("op_26032_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2173_cast_fp16 = mul(x = var_26031_cast_fp16, y = var_26032_to_fp16)[name = tensor("aw_2173_cast_fp16")]; + tensor var_26035_equation_0 = const()[name = tensor("op_26035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26035_cast_fp16 = einsum(equation = var_26035_equation_0, values = (var_25877_cast_fp16, var_25794_cast_fp16))[name = tensor("op_26035_cast_fp16")]; + tensor var_26036_to_fp16 = const()[name = tensor("op_26036_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2175_cast_fp16 = mul(x = var_26035_cast_fp16, y = var_26036_to_fp16)[name = tensor("aw_2175_cast_fp16")]; + tensor var_26039_equation_0 = const()[name = tensor("op_26039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26039_cast_fp16 = einsum(equation = var_26039_equation_0, values = (var_25881_cast_fp16, var_25798_cast_fp16))[name = tensor("op_26039_cast_fp16")]; + tensor var_26040_to_fp16 = const()[name = tensor("op_26040_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2177_cast_fp16 = mul(x = var_26039_cast_fp16, y = var_26040_to_fp16)[name = tensor("aw_2177_cast_fp16")]; + tensor var_26043_equation_0 = const()[name = tensor("op_26043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26043_cast_fp16 = einsum(equation = var_26043_equation_0, values = (var_25885_cast_fp16, var_25802_cast_fp16))[name = tensor("op_26043_cast_fp16")]; + tensor var_26044_to_fp16 = const()[name = tensor("op_26044_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2179_cast_fp16 = mul(x = var_26043_cast_fp16, y = var_26044_to_fp16)[name = tensor("aw_2179_cast_fp16")]; + tensor var_26047_equation_0 = const()[name = tensor("op_26047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26047_cast_fp16 = einsum(equation = var_26047_equation_0, values = (var_25889_cast_fp16, var_25806_cast_fp16))[name = tensor("op_26047_cast_fp16")]; + tensor var_26048_to_fp16 = const()[name = tensor("op_26048_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2181_cast_fp16 = mul(x = var_26047_cast_fp16, y = var_26048_to_fp16)[name = tensor("aw_2181_cast_fp16")]; + tensor var_26051_equation_0 = const()[name = tensor("op_26051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26051_cast_fp16 = einsum(equation = var_26051_equation_0, values = (var_25893_cast_fp16, var_25810_cast_fp16))[name = tensor("op_26051_cast_fp16")]; + tensor var_26052_to_fp16 = const()[name = tensor("op_26052_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2183_cast_fp16 = mul(x = var_26051_cast_fp16, y = var_26052_to_fp16)[name = tensor("aw_2183_cast_fp16")]; + tensor var_26055_equation_0 = const()[name = tensor("op_26055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26055_cast_fp16 = einsum(equation = var_26055_equation_0, values = (var_25897_cast_fp16, var_25814_cast_fp16))[name = tensor("op_26055_cast_fp16")]; + tensor var_26056_to_fp16 = const()[name = tensor("op_26056_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2185_cast_fp16 = mul(x = var_26055_cast_fp16, y = var_26056_to_fp16)[name = tensor("aw_2185_cast_fp16")]; + tensor var_26059_equation_0 = const()[name = tensor("op_26059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26059_cast_fp16 = einsum(equation = var_26059_equation_0, values = (var_25901_cast_fp16, var_25818_cast_fp16))[name = tensor("op_26059_cast_fp16")]; + tensor var_26060_to_fp16 = const()[name = tensor("op_26060_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2187_cast_fp16 = mul(x = var_26059_cast_fp16, y = var_26060_to_fp16)[name = tensor("aw_2187_cast_fp16")]; + tensor var_26063_equation_0 = const()[name = tensor("op_26063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26063_cast_fp16 = einsum(equation = var_26063_equation_0, values = (var_25905_cast_fp16, var_25822_cast_fp16))[name = tensor("op_26063_cast_fp16")]; + tensor var_26064_to_fp16 = const()[name = tensor("op_26064_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2189_cast_fp16 = mul(x = var_26063_cast_fp16, y = var_26064_to_fp16)[name = tensor("aw_2189_cast_fp16")]; + tensor var_26067_equation_0 = const()[name = tensor("op_26067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26067_cast_fp16 = einsum(equation = var_26067_equation_0, values = (var_25909_cast_fp16, var_25826_cast_fp16))[name = tensor("op_26067_cast_fp16")]; + tensor var_26068_to_fp16 = const()[name = tensor("op_26068_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2191_cast_fp16 = mul(x = var_26067_cast_fp16, y = var_26068_to_fp16)[name = tensor("aw_2191_cast_fp16")]; + tensor var_26071_equation_0 = const()[name = tensor("op_26071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26071_cast_fp16 = einsum(equation = var_26071_equation_0, values = (var_25913_cast_fp16, var_25830_cast_fp16))[name = tensor("op_26071_cast_fp16")]; + tensor var_26072_to_fp16 = const()[name = tensor("op_26072_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2193_cast_fp16 = mul(x = var_26071_cast_fp16, y = var_26072_to_fp16)[name = tensor("aw_2193_cast_fp16")]; + tensor var_26075_equation_0 = const()[name = tensor("op_26075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26075_cast_fp16 = einsum(equation = var_26075_equation_0, values = (var_25917_cast_fp16, var_25834_cast_fp16))[name = tensor("op_26075_cast_fp16")]; + tensor var_26076_to_fp16 = const()[name = tensor("op_26076_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2195_cast_fp16 = mul(x = var_26075_cast_fp16, y = var_26076_to_fp16)[name = tensor("aw_2195_cast_fp16")]; + tensor var_26079_equation_0 = const()[name = tensor("op_26079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26079_cast_fp16 = einsum(equation = var_26079_equation_0, values = (var_25921_cast_fp16, var_25838_cast_fp16))[name = tensor("op_26079_cast_fp16")]; + tensor var_26080_to_fp16 = const()[name = tensor("op_26080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2197_cast_fp16 = mul(x = var_26079_cast_fp16, y = var_26080_to_fp16)[name = tensor("aw_2197_cast_fp16")]; + tensor var_26083_equation_0 = const()[name = tensor("op_26083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26083_cast_fp16 = einsum(equation = var_26083_equation_0, values = (var_25925_cast_fp16, var_25842_cast_fp16))[name = tensor("op_26083_cast_fp16")]; + tensor var_26084_to_fp16 = const()[name = tensor("op_26084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2199_cast_fp16 = mul(x = var_26083_cast_fp16, y = var_26084_to_fp16)[name = tensor("aw_2199_cast_fp16")]; + tensor var_26086_cast_fp16 = softmax(axis = var_21077, x = aw_2161_cast_fp16)[name = tensor("op_26086_cast_fp16")]; + tensor var_26087_cast_fp16 = softmax(axis = var_21077, x = aw_2163_cast_fp16)[name = tensor("op_26087_cast_fp16")]; + tensor var_26088_cast_fp16 = softmax(axis = var_21077, x = aw_2165_cast_fp16)[name = tensor("op_26088_cast_fp16")]; + tensor var_26089_cast_fp16 = softmax(axis = var_21077, x = aw_2167_cast_fp16)[name = tensor("op_26089_cast_fp16")]; + tensor var_26090_cast_fp16 = softmax(axis = var_21077, x = aw_2169_cast_fp16)[name = tensor("op_26090_cast_fp16")]; + tensor var_26091_cast_fp16 = softmax(axis = var_21077, x = aw_2171_cast_fp16)[name = tensor("op_26091_cast_fp16")]; + tensor var_26092_cast_fp16 = softmax(axis = var_21077, x = aw_2173_cast_fp16)[name = tensor("op_26092_cast_fp16")]; + tensor var_26093_cast_fp16 = softmax(axis = var_21077, x = aw_2175_cast_fp16)[name = tensor("op_26093_cast_fp16")]; + tensor var_26094_cast_fp16 = softmax(axis = var_21077, x = aw_2177_cast_fp16)[name = tensor("op_26094_cast_fp16")]; + tensor var_26095_cast_fp16 = softmax(axis = var_21077, x = aw_2179_cast_fp16)[name = tensor("op_26095_cast_fp16")]; + tensor var_26096_cast_fp16 = softmax(axis = var_21077, x = aw_2181_cast_fp16)[name = tensor("op_26096_cast_fp16")]; + tensor var_26097_cast_fp16 = softmax(axis = var_21077, x = aw_2183_cast_fp16)[name = tensor("op_26097_cast_fp16")]; + tensor var_26098_cast_fp16 = softmax(axis = var_21077, x = aw_2185_cast_fp16)[name = tensor("op_26098_cast_fp16")]; + tensor var_26099_cast_fp16 = softmax(axis = var_21077, x = aw_2187_cast_fp16)[name = tensor("op_26099_cast_fp16")]; + tensor var_26100_cast_fp16 = softmax(axis = var_21077, x = aw_2189_cast_fp16)[name = tensor("op_26100_cast_fp16")]; + tensor var_26101_cast_fp16 = softmax(axis = var_21077, x = aw_2191_cast_fp16)[name = tensor("op_26101_cast_fp16")]; + tensor var_26102_cast_fp16 = softmax(axis = var_21077, x = aw_2193_cast_fp16)[name = tensor("op_26102_cast_fp16")]; + tensor var_26103_cast_fp16 = softmax(axis = var_21077, x = aw_2195_cast_fp16)[name = tensor("op_26103_cast_fp16")]; + tensor var_26104_cast_fp16 = softmax(axis = var_21077, x = aw_2197_cast_fp16)[name = tensor("op_26104_cast_fp16")]; + tensor var_26105_cast_fp16 = softmax(axis = var_21077, x = aw_2199_cast_fp16)[name = tensor("op_26105_cast_fp16")]; + tensor var_26107_equation_0 = const()[name = tensor("op_26107_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26107_cast_fp16 = einsum(equation = var_26107_equation_0, values = (var_25927_cast_fp16, var_26086_cast_fp16))[name = tensor("op_26107_cast_fp16")]; + tensor var_26109_equation_0 = const()[name = tensor("op_26109_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26109_cast_fp16 = einsum(equation = var_26109_equation_0, values = (var_25931_cast_fp16, var_26087_cast_fp16))[name = tensor("op_26109_cast_fp16")]; + tensor var_26111_equation_0 = const()[name = tensor("op_26111_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26111_cast_fp16 = einsum(equation = var_26111_equation_0, values = (var_25935_cast_fp16, var_26088_cast_fp16))[name = tensor("op_26111_cast_fp16")]; + tensor var_26113_equation_0 = const()[name = tensor("op_26113_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26113_cast_fp16 = einsum(equation = var_26113_equation_0, values = (var_25939_cast_fp16, var_26089_cast_fp16))[name = tensor("op_26113_cast_fp16")]; + tensor var_26115_equation_0 = const()[name = tensor("op_26115_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26115_cast_fp16 = einsum(equation = var_26115_equation_0, values = (var_25943_cast_fp16, var_26090_cast_fp16))[name = tensor("op_26115_cast_fp16")]; + tensor var_26117_equation_0 = const()[name = tensor("op_26117_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26117_cast_fp16 = einsum(equation = var_26117_equation_0, values = (var_25947_cast_fp16, var_26091_cast_fp16))[name = tensor("op_26117_cast_fp16")]; + tensor var_26119_equation_0 = const()[name = tensor("op_26119_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26119_cast_fp16 = einsum(equation = var_26119_equation_0, values = (var_25951_cast_fp16, var_26092_cast_fp16))[name = tensor("op_26119_cast_fp16")]; + tensor var_26121_equation_0 = const()[name = tensor("op_26121_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26121_cast_fp16 = einsum(equation = var_26121_equation_0, values = (var_25955_cast_fp16, var_26093_cast_fp16))[name = tensor("op_26121_cast_fp16")]; + tensor var_26123_equation_0 = const()[name = tensor("op_26123_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26123_cast_fp16 = einsum(equation = var_26123_equation_0, values = (var_25959_cast_fp16, var_26094_cast_fp16))[name = tensor("op_26123_cast_fp16")]; + tensor var_26125_equation_0 = const()[name = tensor("op_26125_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26125_cast_fp16 = einsum(equation = var_26125_equation_0, values = (var_25963_cast_fp16, var_26095_cast_fp16))[name = tensor("op_26125_cast_fp16")]; + tensor var_26127_equation_0 = const()[name = tensor("op_26127_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26127_cast_fp16 = einsum(equation = var_26127_equation_0, values = (var_25967_cast_fp16, var_26096_cast_fp16))[name = tensor("op_26127_cast_fp16")]; + tensor var_26129_equation_0 = const()[name = tensor("op_26129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26129_cast_fp16 = einsum(equation = var_26129_equation_0, values = (var_25971_cast_fp16, var_26097_cast_fp16))[name = tensor("op_26129_cast_fp16")]; + tensor var_26131_equation_0 = const()[name = tensor("op_26131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26131_cast_fp16 = einsum(equation = var_26131_equation_0, values = (var_25975_cast_fp16, var_26098_cast_fp16))[name = tensor("op_26131_cast_fp16")]; + tensor var_26133_equation_0 = const()[name = tensor("op_26133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26133_cast_fp16 = einsum(equation = var_26133_equation_0, values = (var_25979_cast_fp16, var_26099_cast_fp16))[name = tensor("op_26133_cast_fp16")]; + tensor var_26135_equation_0 = const()[name = tensor("op_26135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26135_cast_fp16 = einsum(equation = var_26135_equation_0, values = (var_25983_cast_fp16, var_26100_cast_fp16))[name = tensor("op_26135_cast_fp16")]; + tensor var_26137_equation_0 = const()[name = tensor("op_26137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26137_cast_fp16 = einsum(equation = var_26137_equation_0, values = (var_25987_cast_fp16, var_26101_cast_fp16))[name = tensor("op_26137_cast_fp16")]; + tensor var_26139_equation_0 = const()[name = tensor("op_26139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26139_cast_fp16 = einsum(equation = var_26139_equation_0, values = (var_25991_cast_fp16, var_26102_cast_fp16))[name = tensor("op_26139_cast_fp16")]; + tensor var_26141_equation_0 = const()[name = tensor("op_26141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26141_cast_fp16 = einsum(equation = var_26141_equation_0, values = (var_25995_cast_fp16, var_26103_cast_fp16))[name = tensor("op_26141_cast_fp16")]; + tensor var_26143_equation_0 = const()[name = tensor("op_26143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26143_cast_fp16 = einsum(equation = var_26143_equation_0, values = (var_25999_cast_fp16, var_26104_cast_fp16))[name = tensor("op_26143_cast_fp16")]; + tensor var_26145_equation_0 = const()[name = tensor("op_26145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26145_cast_fp16 = einsum(equation = var_26145_equation_0, values = (var_26003_cast_fp16, var_26105_cast_fp16))[name = tensor("op_26145_cast_fp16")]; + tensor input_367_interleave_0 = const()[name = tensor("input_367_interleave_0"), val = tensor(false)]; + tensor input_367_cast_fp16 = concat(axis = var_21077, interleave = input_367_interleave_0, values = (var_26107_cast_fp16, var_26109_cast_fp16, var_26111_cast_fp16, var_26113_cast_fp16, var_26115_cast_fp16, var_26117_cast_fp16, var_26119_cast_fp16, var_26121_cast_fp16, var_26123_cast_fp16, var_26125_cast_fp16, var_26127_cast_fp16, var_26129_cast_fp16, var_26131_cast_fp16, var_26133_cast_fp16, var_26135_cast_fp16, var_26137_cast_fp16, var_26139_cast_fp16, var_26141_cast_fp16, var_26143_cast_fp16, var_26145_cast_fp16))[name = tensor("input_367_cast_fp16")]; + tensor var_26155_pad_type_0 = const()[name = tensor("op_26155_pad_type_0"), val = tensor("valid")]; + tensor var_26155_strides_0 = const()[name = tensor("op_26155_strides_0"), val = tensor([1, 1])]; + tensor var_26155_pad_0 = const()[name = tensor("op_26155_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26155_dilations_0 = const()[name = tensor("op_26155_dilations_0"), val = tensor([1, 1])]; + tensor var_26155_groups_0 = const()[name = tensor("op_26155_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(782234240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(783463104))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(783463296)))]; + tensor var_26155_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_26155_dilations_0, groups = var_26155_groups_0, pad = var_26155_pad_0, pad_type = var_26155_pad_type_0, strides = var_26155_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized, x = input_367_cast_fp16)[name = tensor("op_26155_cast_fp16")]; + tensor inputs_177_cast_fp16 = add(x = var_26155_cast_fp16, y = inputs_175_cast_fp16)[name = tensor("inputs_177_cast_fp16")]; + tensor hidden_states_241_axes_0 = const()[name = tensor("hidden_states_241_axes_0"), val = tensor([1])]; + tensor hidden_states_241_gamma_0_to_fp16 = const()[name = tensor("hidden_states_241_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(783465920)))]; + tensor hidden_states_241_beta_0_to_fp16 = const()[name = tensor("hidden_states_241_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(783468544)))]; + tensor var_26165_to_fp16 = const()[name = tensor("op_26165_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_241_cast_fp16 = layer_norm(axes = hidden_states_241_axes_0, beta = hidden_states_241_beta_0_to_fp16, epsilon = var_26165_to_fp16, gamma = hidden_states_241_gamma_0_to_fp16, x = inputs_177_cast_fp16)[name = tensor("hidden_states_241_cast_fp16")]; + tensor q_119_pad_type_0 = const()[name = tensor("q_119_pad_type_0"), val = tensor("valid")]; + tensor q_119_strides_0 = const()[name = tensor("q_119_strides_0"), val = tensor([1, 1])]; + tensor q_119_pad_0 = const()[name = tensor("q_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_119_dilations_0 = const()[name = tensor("q_119_dilations_0"), val = tensor([1, 1])]; + tensor q_119_groups_0 = const()[name = tensor("q_119_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(783471168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(784700032))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_119_cast_fp16 = conv(dilations = q_119_dilations_0, groups = q_119_groups_0, pad = q_119_pad_0, pad_type = q_119_pad_type_0, strides = q_119_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_241_cast_fp16)[name = tensor("q_119_cast_fp16")]; + tensor k_237_pad_type_0 = const()[name = tensor("k_237_pad_type_0"), val = tensor("valid")]; + tensor k_237_strides_0 = const()[name = tensor("k_237_strides_0"), val = tensor([1, 1])]; + tensor k_237_pad_0 = const()[name = tensor("k_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_237_dilations_0 = const()[name = tensor("k_237_dilations_0"), val = tensor([1, 1])]; + tensor k_237_groups_0 = const()[name = tensor("k_237_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(784700224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(786666368))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_237_cast_fp16 = conv(dilations = k_237_dilations_0, groups = k_237_groups_0, pad = k_237_pad_0, pad_type = k_237_pad_type_0, strides = k_237_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_237_cast_fp16")]; + tensor v_119_pad_type_0 = const()[name = tensor("v_119_pad_type_0"), val = tensor("valid")]; + tensor v_119_strides_0 = const()[name = tensor("v_119_strides_0"), val = tensor([1, 1])]; + tensor v_119_pad_0 = const()[name = tensor("v_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_119_dilations_0 = const()[name = tensor("v_119_dilations_0"), val = tensor([1, 1])]; + tensor v_119_groups_0 = const()[name = tensor("v_119_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(786666560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788632704))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_119_cast_fp16 = conv(dilations = v_119_dilations_0, groups = v_119_groups_0, pad = v_119_pad_0, pad_type = v_119_pad_type_0, strides = v_119_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_119_cast_fp16")]; + tensor var_26198_begin_0 = const()[name = tensor("op_26198_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26198_end_0 = const()[name = tensor("op_26198_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_26198_end_mask_0 = const()[name = tensor("op_26198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26198_cast_fp16 = slice_by_index(begin = var_26198_begin_0, end = var_26198_end_0, end_mask = var_26198_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26198_cast_fp16")]; + tensor var_26202_begin_0 = const()[name = tensor("op_26202_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26202_end_0 = const()[name = tensor("op_26202_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_26202_end_mask_0 = const()[name = tensor("op_26202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26202_cast_fp16 = slice_by_index(begin = var_26202_begin_0, end = var_26202_end_0, end_mask = var_26202_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26202_cast_fp16")]; + tensor var_26206_begin_0 = const()[name = tensor("op_26206_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26206_end_0 = const()[name = tensor("op_26206_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_26206_end_mask_0 = const()[name = tensor("op_26206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26206_cast_fp16 = slice_by_index(begin = var_26206_begin_0, end = var_26206_end_0, end_mask = var_26206_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26206_cast_fp16")]; + tensor var_26210_begin_0 = const()[name = tensor("op_26210_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26210_end_0 = const()[name = tensor("op_26210_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_26210_end_mask_0 = const()[name = tensor("op_26210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26210_cast_fp16 = slice_by_index(begin = var_26210_begin_0, end = var_26210_end_0, end_mask = var_26210_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26210_cast_fp16")]; + tensor var_26214_begin_0 = const()[name = tensor("op_26214_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26214_end_0 = const()[name = tensor("op_26214_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_26214_end_mask_0 = const()[name = tensor("op_26214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26214_cast_fp16 = slice_by_index(begin = var_26214_begin_0, end = var_26214_end_0, end_mask = var_26214_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26214_cast_fp16")]; + tensor var_26218_begin_0 = const()[name = tensor("op_26218_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26218_end_0 = const()[name = tensor("op_26218_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_26218_end_mask_0 = const()[name = tensor("op_26218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26218_cast_fp16 = slice_by_index(begin = var_26218_begin_0, end = var_26218_end_0, end_mask = var_26218_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26218_cast_fp16")]; + tensor var_26222_begin_0 = const()[name = tensor("op_26222_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26222_end_0 = const()[name = tensor("op_26222_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_26222_end_mask_0 = const()[name = tensor("op_26222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26222_cast_fp16 = slice_by_index(begin = var_26222_begin_0, end = var_26222_end_0, end_mask = var_26222_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26222_cast_fp16")]; + tensor var_26226_begin_0 = const()[name = tensor("op_26226_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26226_end_0 = const()[name = tensor("op_26226_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_26226_end_mask_0 = const()[name = tensor("op_26226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26226_cast_fp16 = slice_by_index(begin = var_26226_begin_0, end = var_26226_end_0, end_mask = var_26226_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26226_cast_fp16")]; + tensor var_26230_begin_0 = const()[name = tensor("op_26230_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26230_end_0 = const()[name = tensor("op_26230_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_26230_end_mask_0 = const()[name = tensor("op_26230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26230_cast_fp16 = slice_by_index(begin = var_26230_begin_0, end = var_26230_end_0, end_mask = var_26230_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26230_cast_fp16")]; + tensor var_26234_begin_0 = const()[name = tensor("op_26234_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26234_end_0 = const()[name = tensor("op_26234_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_26234_end_mask_0 = const()[name = tensor("op_26234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26234_cast_fp16 = slice_by_index(begin = var_26234_begin_0, end = var_26234_end_0, end_mask = var_26234_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26234_cast_fp16")]; + tensor var_26238_begin_0 = const()[name = tensor("op_26238_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26238_end_0 = const()[name = tensor("op_26238_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_26238_end_mask_0 = const()[name = tensor("op_26238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26238_cast_fp16 = slice_by_index(begin = var_26238_begin_0, end = var_26238_end_0, end_mask = var_26238_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26238_cast_fp16")]; + tensor var_26242_begin_0 = const()[name = tensor("op_26242_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26242_end_0 = const()[name = tensor("op_26242_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_26242_end_mask_0 = const()[name = tensor("op_26242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26242_cast_fp16 = slice_by_index(begin = var_26242_begin_0, end = var_26242_end_0, end_mask = var_26242_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26242_cast_fp16")]; + tensor var_26246_begin_0 = const()[name = tensor("op_26246_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26246_end_0 = const()[name = tensor("op_26246_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_26246_end_mask_0 = const()[name = tensor("op_26246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26246_cast_fp16 = slice_by_index(begin = var_26246_begin_0, end = var_26246_end_0, end_mask = var_26246_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26246_cast_fp16")]; + tensor var_26250_begin_0 = const()[name = tensor("op_26250_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26250_end_0 = const()[name = tensor("op_26250_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_26250_end_mask_0 = const()[name = tensor("op_26250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26250_cast_fp16 = slice_by_index(begin = var_26250_begin_0, end = var_26250_end_0, end_mask = var_26250_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26250_cast_fp16")]; + tensor var_26254_begin_0 = const()[name = tensor("op_26254_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26254_end_0 = const()[name = tensor("op_26254_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_26254_end_mask_0 = const()[name = tensor("op_26254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26254_cast_fp16 = slice_by_index(begin = var_26254_begin_0, end = var_26254_end_0, end_mask = var_26254_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26254_cast_fp16")]; + tensor var_26258_begin_0 = const()[name = tensor("op_26258_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26258_end_0 = const()[name = tensor("op_26258_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26258_end_mask_0 = const()[name = tensor("op_26258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26258_cast_fp16 = slice_by_index(begin = var_26258_begin_0, end = var_26258_end_0, end_mask = var_26258_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26258_cast_fp16")]; + tensor var_26262_begin_0 = const()[name = tensor("op_26262_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26262_end_0 = const()[name = tensor("op_26262_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_26262_end_mask_0 = const()[name = tensor("op_26262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26262_cast_fp16 = slice_by_index(begin = var_26262_begin_0, end = var_26262_end_0, end_mask = var_26262_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26262_cast_fp16")]; + tensor var_26266_begin_0 = const()[name = tensor("op_26266_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26266_end_0 = const()[name = tensor("op_26266_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_26266_end_mask_0 = const()[name = tensor("op_26266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26266_cast_fp16 = slice_by_index(begin = var_26266_begin_0, end = var_26266_end_0, end_mask = var_26266_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26266_cast_fp16")]; + tensor var_26270_begin_0 = const()[name = tensor("op_26270_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26270_end_0 = const()[name = tensor("op_26270_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_26270_end_mask_0 = const()[name = tensor("op_26270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26270_cast_fp16 = slice_by_index(begin = var_26270_begin_0, end = var_26270_end_0, end_mask = var_26270_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26270_cast_fp16")]; + tensor var_26274_begin_0 = const()[name = tensor("op_26274_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26274_end_0 = const()[name = tensor("op_26274_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26274_end_mask_0 = const()[name = tensor("op_26274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26274_cast_fp16 = slice_by_index(begin = var_26274_begin_0, end = var_26274_end_0, end_mask = var_26274_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26274_cast_fp16")]; + tensor k_239_perm_0 = const()[name = tensor("k_239_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_26281_begin_0 = const()[name = tensor("op_26281_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26281_end_0 = const()[name = tensor("op_26281_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_26281_end_mask_0 = const()[name = tensor("op_26281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_239_cast_fp16 = transpose(perm = k_239_perm_0, x = k_237_cast_fp16)[name = tensor("transpose_8")]; + tensor var_26281_cast_fp16 = slice_by_index(begin = var_26281_begin_0, end = var_26281_end_0, end_mask = var_26281_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26281_cast_fp16")]; + tensor var_26285_begin_0 = const()[name = tensor("op_26285_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_26285_end_0 = const()[name = tensor("op_26285_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_26285_end_mask_0 = const()[name = tensor("op_26285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26285_cast_fp16 = slice_by_index(begin = var_26285_begin_0, end = var_26285_end_0, end_mask = var_26285_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26285_cast_fp16")]; + tensor var_26289_begin_0 = const()[name = tensor("op_26289_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_26289_end_0 = const()[name = tensor("op_26289_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_26289_end_mask_0 = const()[name = tensor("op_26289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26289_cast_fp16 = slice_by_index(begin = var_26289_begin_0, end = var_26289_end_0, end_mask = var_26289_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26289_cast_fp16")]; + tensor var_26293_begin_0 = const()[name = tensor("op_26293_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_26293_end_0 = const()[name = tensor("op_26293_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_26293_end_mask_0 = const()[name = tensor("op_26293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26293_cast_fp16 = slice_by_index(begin = var_26293_begin_0, end = var_26293_end_0, end_mask = var_26293_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26293_cast_fp16")]; + tensor var_26297_begin_0 = const()[name = tensor("op_26297_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_26297_end_0 = const()[name = tensor("op_26297_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_26297_end_mask_0 = const()[name = tensor("op_26297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26297_cast_fp16 = slice_by_index(begin = var_26297_begin_0, end = var_26297_end_0, end_mask = var_26297_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26297_cast_fp16")]; + tensor var_26301_begin_0 = const()[name = tensor("op_26301_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_26301_end_0 = const()[name = tensor("op_26301_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_26301_end_mask_0 = const()[name = tensor("op_26301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26301_cast_fp16 = slice_by_index(begin = var_26301_begin_0, end = var_26301_end_0, end_mask = var_26301_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26301_cast_fp16")]; + tensor var_26305_begin_0 = const()[name = tensor("op_26305_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_26305_end_0 = const()[name = tensor("op_26305_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_26305_end_mask_0 = const()[name = tensor("op_26305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26305_cast_fp16 = slice_by_index(begin = var_26305_begin_0, end = var_26305_end_0, end_mask = var_26305_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26305_cast_fp16")]; + tensor var_26309_begin_0 = const()[name = tensor("op_26309_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_26309_end_0 = const()[name = tensor("op_26309_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_26309_end_mask_0 = const()[name = tensor("op_26309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26309_cast_fp16 = slice_by_index(begin = var_26309_begin_0, end = var_26309_end_0, end_mask = var_26309_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26309_cast_fp16")]; + tensor var_26313_begin_0 = const()[name = tensor("op_26313_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_26313_end_0 = const()[name = tensor("op_26313_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_26313_end_mask_0 = const()[name = tensor("op_26313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26313_cast_fp16 = slice_by_index(begin = var_26313_begin_0, end = var_26313_end_0, end_mask = var_26313_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26313_cast_fp16")]; + tensor var_26317_begin_0 = const()[name = tensor("op_26317_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_26317_end_0 = const()[name = tensor("op_26317_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_26317_end_mask_0 = const()[name = tensor("op_26317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26317_cast_fp16 = slice_by_index(begin = var_26317_begin_0, end = var_26317_end_0, end_mask = var_26317_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26317_cast_fp16")]; + tensor var_26321_begin_0 = const()[name = tensor("op_26321_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_26321_end_0 = const()[name = tensor("op_26321_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_26321_end_mask_0 = const()[name = tensor("op_26321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26321_cast_fp16 = slice_by_index(begin = var_26321_begin_0, end = var_26321_end_0, end_mask = var_26321_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26321_cast_fp16")]; + tensor var_26325_begin_0 = const()[name = tensor("op_26325_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_26325_end_0 = const()[name = tensor("op_26325_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_26325_end_mask_0 = const()[name = tensor("op_26325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26325_cast_fp16 = slice_by_index(begin = var_26325_begin_0, end = var_26325_end_0, end_mask = var_26325_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26325_cast_fp16")]; + tensor var_26329_begin_0 = const()[name = tensor("op_26329_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_26329_end_0 = const()[name = tensor("op_26329_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_26329_end_mask_0 = const()[name = tensor("op_26329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26329_cast_fp16 = slice_by_index(begin = var_26329_begin_0, end = var_26329_end_0, end_mask = var_26329_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26329_cast_fp16")]; + tensor var_26333_begin_0 = const()[name = tensor("op_26333_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_26333_end_0 = const()[name = tensor("op_26333_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_26333_end_mask_0 = const()[name = tensor("op_26333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26333_cast_fp16 = slice_by_index(begin = var_26333_begin_0, end = var_26333_end_0, end_mask = var_26333_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26333_cast_fp16")]; + tensor var_26337_begin_0 = const()[name = tensor("op_26337_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_26337_end_0 = const()[name = tensor("op_26337_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_26337_end_mask_0 = const()[name = tensor("op_26337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26337_cast_fp16 = slice_by_index(begin = var_26337_begin_0, end = var_26337_end_0, end_mask = var_26337_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26337_cast_fp16")]; + tensor var_26341_begin_0 = const()[name = tensor("op_26341_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_26341_end_0 = const()[name = tensor("op_26341_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_26341_end_mask_0 = const()[name = tensor("op_26341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26341_cast_fp16 = slice_by_index(begin = var_26341_begin_0, end = var_26341_end_0, end_mask = var_26341_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26341_cast_fp16")]; + tensor var_26345_begin_0 = const()[name = tensor("op_26345_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_26345_end_0 = const()[name = tensor("op_26345_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_26345_end_mask_0 = const()[name = tensor("op_26345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26345_cast_fp16 = slice_by_index(begin = var_26345_begin_0, end = var_26345_end_0, end_mask = var_26345_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26345_cast_fp16")]; + tensor var_26349_begin_0 = const()[name = tensor("op_26349_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_26349_end_0 = const()[name = tensor("op_26349_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_26349_end_mask_0 = const()[name = tensor("op_26349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26349_cast_fp16 = slice_by_index(begin = var_26349_begin_0, end = var_26349_end_0, end_mask = var_26349_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26349_cast_fp16")]; + tensor var_26353_begin_0 = const()[name = tensor("op_26353_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_26353_end_0 = const()[name = tensor("op_26353_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_26353_end_mask_0 = const()[name = tensor("op_26353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26353_cast_fp16 = slice_by_index(begin = var_26353_begin_0, end = var_26353_end_0, end_mask = var_26353_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26353_cast_fp16")]; + tensor var_26357_begin_0 = const()[name = tensor("op_26357_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_26357_end_0 = const()[name = tensor("op_26357_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_26357_end_mask_0 = const()[name = tensor("op_26357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26357_cast_fp16 = slice_by_index(begin = var_26357_begin_0, end = var_26357_end_0, end_mask = var_26357_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26357_cast_fp16")]; + tensor var_26359_begin_0 = const()[name = tensor("op_26359_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26359_end_0 = const()[name = tensor("op_26359_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_26359_end_mask_0 = const()[name = tensor("op_26359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26359_cast_fp16 = slice_by_index(begin = var_26359_begin_0, end = var_26359_end_0, end_mask = var_26359_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26359_cast_fp16")]; + tensor var_26363_begin_0 = const()[name = tensor("op_26363_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26363_end_0 = const()[name = tensor("op_26363_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_26363_end_mask_0 = const()[name = tensor("op_26363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26363_cast_fp16 = slice_by_index(begin = var_26363_begin_0, end = var_26363_end_0, end_mask = var_26363_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26363_cast_fp16")]; + tensor var_26367_begin_0 = const()[name = tensor("op_26367_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26367_end_0 = const()[name = tensor("op_26367_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_26367_end_mask_0 = const()[name = tensor("op_26367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26367_cast_fp16 = slice_by_index(begin = var_26367_begin_0, end = var_26367_end_0, end_mask = var_26367_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26367_cast_fp16")]; + tensor var_26371_begin_0 = const()[name = tensor("op_26371_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26371_end_0 = const()[name = tensor("op_26371_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_26371_end_mask_0 = const()[name = tensor("op_26371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26371_cast_fp16 = slice_by_index(begin = var_26371_begin_0, end = var_26371_end_0, end_mask = var_26371_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26371_cast_fp16")]; + tensor var_26375_begin_0 = const()[name = tensor("op_26375_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26375_end_0 = const()[name = tensor("op_26375_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_26375_end_mask_0 = const()[name = tensor("op_26375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26375_cast_fp16 = slice_by_index(begin = var_26375_begin_0, end = var_26375_end_0, end_mask = var_26375_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26375_cast_fp16")]; + tensor var_26379_begin_0 = const()[name = tensor("op_26379_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26379_end_0 = const()[name = tensor("op_26379_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_26379_end_mask_0 = const()[name = tensor("op_26379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26379_cast_fp16 = slice_by_index(begin = var_26379_begin_0, end = var_26379_end_0, end_mask = var_26379_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26379_cast_fp16")]; + tensor var_26383_begin_0 = const()[name = tensor("op_26383_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26383_end_0 = const()[name = tensor("op_26383_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_26383_end_mask_0 = const()[name = tensor("op_26383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26383_cast_fp16 = slice_by_index(begin = var_26383_begin_0, end = var_26383_end_0, end_mask = var_26383_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26383_cast_fp16")]; + tensor var_26387_begin_0 = const()[name = tensor("op_26387_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26387_end_0 = const()[name = tensor("op_26387_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_26387_end_mask_0 = const()[name = tensor("op_26387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26387_cast_fp16 = slice_by_index(begin = var_26387_begin_0, end = var_26387_end_0, end_mask = var_26387_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26387_cast_fp16")]; + tensor var_26391_begin_0 = const()[name = tensor("op_26391_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26391_end_0 = const()[name = tensor("op_26391_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_26391_end_mask_0 = const()[name = tensor("op_26391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26391_cast_fp16 = slice_by_index(begin = var_26391_begin_0, end = var_26391_end_0, end_mask = var_26391_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26391_cast_fp16")]; + tensor var_26395_begin_0 = const()[name = tensor("op_26395_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26395_end_0 = const()[name = tensor("op_26395_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_26395_end_mask_0 = const()[name = tensor("op_26395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26395_cast_fp16 = slice_by_index(begin = var_26395_begin_0, end = var_26395_end_0, end_mask = var_26395_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26395_cast_fp16")]; + tensor var_26399_begin_0 = const()[name = tensor("op_26399_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26399_end_0 = const()[name = tensor("op_26399_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_26399_end_mask_0 = const()[name = tensor("op_26399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26399_cast_fp16 = slice_by_index(begin = var_26399_begin_0, end = var_26399_end_0, end_mask = var_26399_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26399_cast_fp16")]; + tensor var_26403_begin_0 = const()[name = tensor("op_26403_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26403_end_0 = const()[name = tensor("op_26403_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_26403_end_mask_0 = const()[name = tensor("op_26403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26403_cast_fp16 = slice_by_index(begin = var_26403_begin_0, end = var_26403_end_0, end_mask = var_26403_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26403_cast_fp16")]; + tensor var_26407_begin_0 = const()[name = tensor("op_26407_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26407_end_0 = const()[name = tensor("op_26407_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_26407_end_mask_0 = const()[name = tensor("op_26407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26407_cast_fp16 = slice_by_index(begin = var_26407_begin_0, end = var_26407_end_0, end_mask = var_26407_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26407_cast_fp16")]; + tensor var_26411_begin_0 = const()[name = tensor("op_26411_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26411_end_0 = const()[name = tensor("op_26411_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_26411_end_mask_0 = const()[name = tensor("op_26411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26411_cast_fp16 = slice_by_index(begin = var_26411_begin_0, end = var_26411_end_0, end_mask = var_26411_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26411_cast_fp16")]; + tensor var_26415_begin_0 = const()[name = tensor("op_26415_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26415_end_0 = const()[name = tensor("op_26415_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_26415_end_mask_0 = const()[name = tensor("op_26415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26415_cast_fp16 = slice_by_index(begin = var_26415_begin_0, end = var_26415_end_0, end_mask = var_26415_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26415_cast_fp16")]; + tensor var_26419_begin_0 = const()[name = tensor("op_26419_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26419_end_0 = const()[name = tensor("op_26419_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_26419_end_mask_0 = const()[name = tensor("op_26419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26419_cast_fp16 = slice_by_index(begin = var_26419_begin_0, end = var_26419_end_0, end_mask = var_26419_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26419_cast_fp16")]; + tensor var_26423_begin_0 = const()[name = tensor("op_26423_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26423_end_0 = const()[name = tensor("op_26423_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_26423_end_mask_0 = const()[name = tensor("op_26423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26423_cast_fp16 = slice_by_index(begin = var_26423_begin_0, end = var_26423_end_0, end_mask = var_26423_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26423_cast_fp16")]; + tensor var_26427_begin_0 = const()[name = tensor("op_26427_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26427_end_0 = const()[name = tensor("op_26427_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_26427_end_mask_0 = const()[name = tensor("op_26427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26427_cast_fp16 = slice_by_index(begin = var_26427_begin_0, end = var_26427_end_0, end_mask = var_26427_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26427_cast_fp16")]; + tensor var_26431_begin_0 = const()[name = tensor("op_26431_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26431_end_0 = const()[name = tensor("op_26431_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_26431_end_mask_0 = const()[name = tensor("op_26431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26431_cast_fp16 = slice_by_index(begin = var_26431_begin_0, end = var_26431_end_0, end_mask = var_26431_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26431_cast_fp16")]; + tensor var_26435_begin_0 = const()[name = tensor("op_26435_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26435_end_0 = const()[name = tensor("op_26435_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_26435_end_mask_0 = const()[name = tensor("op_26435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26435_cast_fp16 = slice_by_index(begin = var_26435_begin_0, end = var_26435_end_0, end_mask = var_26435_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26435_cast_fp16")]; + tensor var_26439_equation_0 = const()[name = tensor("op_26439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26439_cast_fp16 = einsum(equation = var_26439_equation_0, values = (var_26281_cast_fp16, var_26198_cast_fp16))[name = tensor("op_26439_cast_fp16")]; + tensor var_26440_to_fp16 = const()[name = tensor("op_26440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2201_cast_fp16 = mul(x = var_26439_cast_fp16, y = var_26440_to_fp16)[name = tensor("aw_2201_cast_fp16")]; + tensor var_26443_equation_0 = const()[name = tensor("op_26443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26443_cast_fp16 = einsum(equation = var_26443_equation_0, values = (var_26285_cast_fp16, var_26202_cast_fp16))[name = tensor("op_26443_cast_fp16")]; + tensor var_26444_to_fp16 = const()[name = tensor("op_26444_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2203_cast_fp16 = mul(x = var_26443_cast_fp16, y = var_26444_to_fp16)[name = tensor("aw_2203_cast_fp16")]; + tensor var_26447_equation_0 = const()[name = tensor("op_26447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26447_cast_fp16 = einsum(equation = var_26447_equation_0, values = (var_26289_cast_fp16, var_26206_cast_fp16))[name = tensor("op_26447_cast_fp16")]; + tensor var_26448_to_fp16 = const()[name = tensor("op_26448_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2205_cast_fp16 = mul(x = var_26447_cast_fp16, y = var_26448_to_fp16)[name = tensor("aw_2205_cast_fp16")]; + tensor var_26451_equation_0 = const()[name = tensor("op_26451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26451_cast_fp16 = einsum(equation = var_26451_equation_0, values = (var_26293_cast_fp16, var_26210_cast_fp16))[name = tensor("op_26451_cast_fp16")]; + tensor var_26452_to_fp16 = const()[name = tensor("op_26452_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2207_cast_fp16 = mul(x = var_26451_cast_fp16, y = var_26452_to_fp16)[name = tensor("aw_2207_cast_fp16")]; + tensor var_26455_equation_0 = const()[name = tensor("op_26455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26455_cast_fp16 = einsum(equation = var_26455_equation_0, values = (var_26297_cast_fp16, var_26214_cast_fp16))[name = tensor("op_26455_cast_fp16")]; + tensor var_26456_to_fp16 = const()[name = tensor("op_26456_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2209_cast_fp16 = mul(x = var_26455_cast_fp16, y = var_26456_to_fp16)[name = tensor("aw_2209_cast_fp16")]; + tensor var_26459_equation_0 = const()[name = tensor("op_26459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26459_cast_fp16 = einsum(equation = var_26459_equation_0, values = (var_26301_cast_fp16, var_26218_cast_fp16))[name = tensor("op_26459_cast_fp16")]; + tensor var_26460_to_fp16 = const()[name = tensor("op_26460_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2211_cast_fp16 = mul(x = var_26459_cast_fp16, y = var_26460_to_fp16)[name = tensor("aw_2211_cast_fp16")]; + tensor var_26463_equation_0 = const()[name = tensor("op_26463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26463_cast_fp16 = einsum(equation = var_26463_equation_0, values = (var_26305_cast_fp16, var_26222_cast_fp16))[name = tensor("op_26463_cast_fp16")]; + tensor var_26464_to_fp16 = const()[name = tensor("op_26464_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2213_cast_fp16 = mul(x = var_26463_cast_fp16, y = var_26464_to_fp16)[name = tensor("aw_2213_cast_fp16")]; + tensor var_26467_equation_0 = const()[name = tensor("op_26467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26467_cast_fp16 = einsum(equation = var_26467_equation_0, values = (var_26309_cast_fp16, var_26226_cast_fp16))[name = tensor("op_26467_cast_fp16")]; + tensor var_26468_to_fp16 = const()[name = tensor("op_26468_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2215_cast_fp16 = mul(x = var_26467_cast_fp16, y = var_26468_to_fp16)[name = tensor("aw_2215_cast_fp16")]; + tensor var_26471_equation_0 = const()[name = tensor("op_26471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26471_cast_fp16 = einsum(equation = var_26471_equation_0, values = (var_26313_cast_fp16, var_26230_cast_fp16))[name = tensor("op_26471_cast_fp16")]; + tensor var_26472_to_fp16 = const()[name = tensor("op_26472_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2217_cast_fp16 = mul(x = var_26471_cast_fp16, y = var_26472_to_fp16)[name = tensor("aw_2217_cast_fp16")]; + tensor var_26475_equation_0 = const()[name = tensor("op_26475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26475_cast_fp16 = einsum(equation = var_26475_equation_0, values = (var_26317_cast_fp16, var_26234_cast_fp16))[name = tensor("op_26475_cast_fp16")]; + tensor var_26476_to_fp16 = const()[name = tensor("op_26476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2219_cast_fp16 = mul(x = var_26475_cast_fp16, y = var_26476_to_fp16)[name = tensor("aw_2219_cast_fp16")]; + tensor var_26479_equation_0 = const()[name = tensor("op_26479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26479_cast_fp16 = einsum(equation = var_26479_equation_0, values = (var_26321_cast_fp16, var_26238_cast_fp16))[name = tensor("op_26479_cast_fp16")]; + tensor var_26480_to_fp16 = const()[name = tensor("op_26480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2221_cast_fp16 = mul(x = var_26479_cast_fp16, y = var_26480_to_fp16)[name = tensor("aw_2221_cast_fp16")]; + tensor var_26483_equation_0 = const()[name = tensor("op_26483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26483_cast_fp16 = einsum(equation = var_26483_equation_0, values = (var_26325_cast_fp16, var_26242_cast_fp16))[name = tensor("op_26483_cast_fp16")]; + tensor var_26484_to_fp16 = const()[name = tensor("op_26484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2223_cast_fp16 = mul(x = var_26483_cast_fp16, y = var_26484_to_fp16)[name = tensor("aw_2223_cast_fp16")]; + tensor var_26487_equation_0 = const()[name = tensor("op_26487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26487_cast_fp16 = einsum(equation = var_26487_equation_0, values = (var_26329_cast_fp16, var_26246_cast_fp16))[name = tensor("op_26487_cast_fp16")]; + tensor var_26488_to_fp16 = const()[name = tensor("op_26488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2225_cast_fp16 = mul(x = var_26487_cast_fp16, y = var_26488_to_fp16)[name = tensor("aw_2225_cast_fp16")]; + tensor var_26491_equation_0 = const()[name = tensor("op_26491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26491_cast_fp16 = einsum(equation = var_26491_equation_0, values = (var_26333_cast_fp16, var_26250_cast_fp16))[name = tensor("op_26491_cast_fp16")]; + tensor var_26492_to_fp16 = const()[name = tensor("op_26492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2227_cast_fp16 = mul(x = var_26491_cast_fp16, y = var_26492_to_fp16)[name = tensor("aw_2227_cast_fp16")]; + tensor var_26495_equation_0 = const()[name = tensor("op_26495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26495_cast_fp16 = einsum(equation = var_26495_equation_0, values = (var_26337_cast_fp16, var_26254_cast_fp16))[name = tensor("op_26495_cast_fp16")]; + tensor var_26496_to_fp16 = const()[name = tensor("op_26496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2229_cast_fp16 = mul(x = var_26495_cast_fp16, y = var_26496_to_fp16)[name = tensor("aw_2229_cast_fp16")]; + tensor var_26499_equation_0 = const()[name = tensor("op_26499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26499_cast_fp16 = einsum(equation = var_26499_equation_0, values = (var_26341_cast_fp16, var_26258_cast_fp16))[name = tensor("op_26499_cast_fp16")]; + tensor var_26500_to_fp16 = const()[name = tensor("op_26500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2231_cast_fp16 = mul(x = var_26499_cast_fp16, y = var_26500_to_fp16)[name = tensor("aw_2231_cast_fp16")]; + tensor var_26503_equation_0 = const()[name = tensor("op_26503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26503_cast_fp16 = einsum(equation = var_26503_equation_0, values = (var_26345_cast_fp16, var_26262_cast_fp16))[name = tensor("op_26503_cast_fp16")]; + tensor var_26504_to_fp16 = const()[name = tensor("op_26504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2233_cast_fp16 = mul(x = var_26503_cast_fp16, y = var_26504_to_fp16)[name = tensor("aw_2233_cast_fp16")]; + tensor var_26507_equation_0 = const()[name = tensor("op_26507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26507_cast_fp16 = einsum(equation = var_26507_equation_0, values = (var_26349_cast_fp16, var_26266_cast_fp16))[name = tensor("op_26507_cast_fp16")]; + tensor var_26508_to_fp16 = const()[name = tensor("op_26508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2235_cast_fp16 = mul(x = var_26507_cast_fp16, y = var_26508_to_fp16)[name = tensor("aw_2235_cast_fp16")]; + tensor var_26511_equation_0 = const()[name = tensor("op_26511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26511_cast_fp16 = einsum(equation = var_26511_equation_0, values = (var_26353_cast_fp16, var_26270_cast_fp16))[name = tensor("op_26511_cast_fp16")]; + tensor var_26512_to_fp16 = const()[name = tensor("op_26512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2237_cast_fp16 = mul(x = var_26511_cast_fp16, y = var_26512_to_fp16)[name = tensor("aw_2237_cast_fp16")]; + tensor var_26515_equation_0 = const()[name = tensor("op_26515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26515_cast_fp16 = einsum(equation = var_26515_equation_0, values = (var_26357_cast_fp16, var_26274_cast_fp16))[name = tensor("op_26515_cast_fp16")]; + tensor var_26516_to_fp16 = const()[name = tensor("op_26516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2239_cast_fp16 = mul(x = var_26515_cast_fp16, y = var_26516_to_fp16)[name = tensor("aw_2239_cast_fp16")]; + tensor var_26518_cast_fp16 = softmax(axis = var_21077, x = aw_2201_cast_fp16)[name = tensor("op_26518_cast_fp16")]; + tensor var_26519_cast_fp16 = softmax(axis = var_21077, x = aw_2203_cast_fp16)[name = tensor("op_26519_cast_fp16")]; + tensor var_26520_cast_fp16 = softmax(axis = var_21077, x = aw_2205_cast_fp16)[name = tensor("op_26520_cast_fp16")]; + tensor var_26521_cast_fp16 = softmax(axis = var_21077, x = aw_2207_cast_fp16)[name = tensor("op_26521_cast_fp16")]; + tensor var_26522_cast_fp16 = softmax(axis = var_21077, x = aw_2209_cast_fp16)[name = tensor("op_26522_cast_fp16")]; + tensor var_26523_cast_fp16 = softmax(axis = var_21077, x = aw_2211_cast_fp16)[name = tensor("op_26523_cast_fp16")]; + tensor var_26524_cast_fp16 = softmax(axis = var_21077, x = aw_2213_cast_fp16)[name = tensor("op_26524_cast_fp16")]; + tensor var_26525_cast_fp16 = softmax(axis = var_21077, x = aw_2215_cast_fp16)[name = tensor("op_26525_cast_fp16")]; + tensor var_26526_cast_fp16 = softmax(axis = var_21077, x = aw_2217_cast_fp16)[name = tensor("op_26526_cast_fp16")]; + tensor var_26527_cast_fp16 = softmax(axis = var_21077, x = aw_2219_cast_fp16)[name = tensor("op_26527_cast_fp16")]; + tensor var_26528_cast_fp16 = softmax(axis = var_21077, x = aw_2221_cast_fp16)[name = tensor("op_26528_cast_fp16")]; + tensor var_26529_cast_fp16 = softmax(axis = var_21077, x = aw_2223_cast_fp16)[name = tensor("op_26529_cast_fp16")]; + tensor var_26530_cast_fp16 = softmax(axis = var_21077, x = aw_2225_cast_fp16)[name = tensor("op_26530_cast_fp16")]; + tensor var_26531_cast_fp16 = softmax(axis = var_21077, x = aw_2227_cast_fp16)[name = tensor("op_26531_cast_fp16")]; + tensor var_26532_cast_fp16 = softmax(axis = var_21077, x = aw_2229_cast_fp16)[name = tensor("op_26532_cast_fp16")]; + tensor var_26533_cast_fp16 = softmax(axis = var_21077, x = aw_2231_cast_fp16)[name = tensor("op_26533_cast_fp16")]; + tensor var_26534_cast_fp16 = softmax(axis = var_21077, x = aw_2233_cast_fp16)[name = tensor("op_26534_cast_fp16")]; + tensor var_26535_cast_fp16 = softmax(axis = var_21077, x = aw_2235_cast_fp16)[name = tensor("op_26535_cast_fp16")]; + tensor var_26536_cast_fp16 = softmax(axis = var_21077, x = aw_2237_cast_fp16)[name = tensor("op_26536_cast_fp16")]; + tensor var_26537_cast_fp16 = softmax(axis = var_21077, x = aw_2239_cast_fp16)[name = tensor("op_26537_cast_fp16")]; + tensor var_26539_equation_0 = const()[name = tensor("op_26539_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26539_cast_fp16 = einsum(equation = var_26539_equation_0, values = (var_26359_cast_fp16, var_26518_cast_fp16))[name = tensor("op_26539_cast_fp16")]; + tensor var_26541_equation_0 = const()[name = tensor("op_26541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26541_cast_fp16 = einsum(equation = var_26541_equation_0, values = (var_26363_cast_fp16, var_26519_cast_fp16))[name = tensor("op_26541_cast_fp16")]; + tensor var_26543_equation_0 = const()[name = tensor("op_26543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26543_cast_fp16 = einsum(equation = var_26543_equation_0, values = (var_26367_cast_fp16, var_26520_cast_fp16))[name = tensor("op_26543_cast_fp16")]; + tensor var_26545_equation_0 = const()[name = tensor("op_26545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26545_cast_fp16 = einsum(equation = var_26545_equation_0, values = (var_26371_cast_fp16, var_26521_cast_fp16))[name = tensor("op_26545_cast_fp16")]; + tensor var_26547_equation_0 = const()[name = tensor("op_26547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26547_cast_fp16 = einsum(equation = var_26547_equation_0, values = (var_26375_cast_fp16, var_26522_cast_fp16))[name = tensor("op_26547_cast_fp16")]; + tensor var_26549_equation_0 = const()[name = tensor("op_26549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26549_cast_fp16 = einsum(equation = var_26549_equation_0, values = (var_26379_cast_fp16, var_26523_cast_fp16))[name = tensor("op_26549_cast_fp16")]; + tensor var_26551_equation_0 = const()[name = tensor("op_26551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26551_cast_fp16 = einsum(equation = var_26551_equation_0, values = (var_26383_cast_fp16, var_26524_cast_fp16))[name = tensor("op_26551_cast_fp16")]; + tensor var_26553_equation_0 = const()[name = tensor("op_26553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26553_cast_fp16 = einsum(equation = var_26553_equation_0, values = (var_26387_cast_fp16, var_26525_cast_fp16))[name = tensor("op_26553_cast_fp16")]; + tensor var_26555_equation_0 = const()[name = tensor("op_26555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26555_cast_fp16 = einsum(equation = var_26555_equation_0, values = (var_26391_cast_fp16, var_26526_cast_fp16))[name = tensor("op_26555_cast_fp16")]; + tensor var_26557_equation_0 = const()[name = tensor("op_26557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26557_cast_fp16 = einsum(equation = var_26557_equation_0, values = (var_26395_cast_fp16, var_26527_cast_fp16))[name = tensor("op_26557_cast_fp16")]; + tensor var_26559_equation_0 = const()[name = tensor("op_26559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26559_cast_fp16 = einsum(equation = var_26559_equation_0, values = (var_26399_cast_fp16, var_26528_cast_fp16))[name = tensor("op_26559_cast_fp16")]; + tensor var_26561_equation_0 = const()[name = tensor("op_26561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26561_cast_fp16 = einsum(equation = var_26561_equation_0, values = (var_26403_cast_fp16, var_26529_cast_fp16))[name = tensor("op_26561_cast_fp16")]; + tensor var_26563_equation_0 = const()[name = tensor("op_26563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26563_cast_fp16 = einsum(equation = var_26563_equation_0, values = (var_26407_cast_fp16, var_26530_cast_fp16))[name = tensor("op_26563_cast_fp16")]; + tensor var_26565_equation_0 = const()[name = tensor("op_26565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26565_cast_fp16 = einsum(equation = var_26565_equation_0, values = (var_26411_cast_fp16, var_26531_cast_fp16))[name = tensor("op_26565_cast_fp16")]; + tensor var_26567_equation_0 = const()[name = tensor("op_26567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26567_cast_fp16 = einsum(equation = var_26567_equation_0, values = (var_26415_cast_fp16, var_26532_cast_fp16))[name = tensor("op_26567_cast_fp16")]; + tensor var_26569_equation_0 = const()[name = tensor("op_26569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26569_cast_fp16 = einsum(equation = var_26569_equation_0, values = (var_26419_cast_fp16, var_26533_cast_fp16))[name = tensor("op_26569_cast_fp16")]; + tensor var_26571_equation_0 = const()[name = tensor("op_26571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26571_cast_fp16 = einsum(equation = var_26571_equation_0, values = (var_26423_cast_fp16, var_26534_cast_fp16))[name = tensor("op_26571_cast_fp16")]; + tensor var_26573_equation_0 = const()[name = tensor("op_26573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26573_cast_fp16 = einsum(equation = var_26573_equation_0, values = (var_26427_cast_fp16, var_26535_cast_fp16))[name = tensor("op_26573_cast_fp16")]; + tensor var_26575_equation_0 = const()[name = tensor("op_26575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26575_cast_fp16 = einsum(equation = var_26575_equation_0, values = (var_26431_cast_fp16, var_26536_cast_fp16))[name = tensor("op_26575_cast_fp16")]; + tensor var_26577_equation_0 = const()[name = tensor("op_26577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26577_cast_fp16 = einsum(equation = var_26577_equation_0, values = (var_26435_cast_fp16, var_26537_cast_fp16))[name = tensor("op_26577_cast_fp16")]; + tensor input_369_interleave_0 = const()[name = tensor("input_369_interleave_0"), val = tensor(false)]; + tensor input_369_cast_fp16 = concat(axis = var_21077, interleave = input_369_interleave_0, values = (var_26539_cast_fp16, var_26541_cast_fp16, var_26543_cast_fp16, var_26545_cast_fp16, var_26547_cast_fp16, var_26549_cast_fp16, var_26551_cast_fp16, var_26553_cast_fp16, var_26555_cast_fp16, var_26557_cast_fp16, var_26559_cast_fp16, var_26561_cast_fp16, var_26563_cast_fp16, var_26565_cast_fp16, var_26567_cast_fp16, var_26569_cast_fp16, var_26571_cast_fp16, var_26573_cast_fp16, var_26575_cast_fp16, var_26577_cast_fp16))[name = tensor("input_369_cast_fp16")]; + tensor var_26587_pad_type_0 = const()[name = tensor("op_26587_pad_type_0"), val = tensor("valid")]; + tensor var_26587_strides_0 = const()[name = tensor("op_26587_strides_0"), val = tensor([1, 1])]; + tensor var_26587_pad_0 = const()[name = tensor("op_26587_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26587_dilations_0 = const()[name = tensor("op_26587_dilations_0"), val = tensor([1, 1])]; + tensor var_26587_groups_0 = const()[name = tensor("op_26587_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788632896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789861760))), name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789861952)))]; + tensor var_26587_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_26587_dilations_0, groups = var_26587_groups_0, pad = var_26587_pad_0, pad_type = var_26587_pad_type_0, strides = var_26587_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized, x = input_369_cast_fp16)[name = tensor("op_26587_cast_fp16")]; + tensor inputs_179_cast_fp16 = add(x = var_26587_cast_fp16, y = inputs_177_cast_fp16)[name = tensor("inputs_179_cast_fp16")]; + tensor input_371_axes_0 = const()[name = tensor("input_371_axes_0"), val = tensor([1])]; + tensor input_371_gamma_0_to_fp16 = const()[name = tensor("input_371_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789864576)))]; + tensor input_371_beta_0_to_fp16 = const()[name = tensor("input_371_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789867200)))]; + tensor var_26597_to_fp16 = const()[name = tensor("op_26597_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = input_371_beta_0_to_fp16, epsilon = var_26597_to_fp16, gamma = input_371_gamma_0_to_fp16, x = inputs_179_cast_fp16)[name = tensor("input_371_cast_fp16")]; + tensor var_26617_pad_type_0 = const()[name = tensor("op_26617_pad_type_0"), val = tensor("valid")]; + tensor var_26617_strides_0 = const()[name = tensor("op_26617_strides_0"), val = tensor([1, 1])]; + tensor var_26617_pad_0 = const()[name = tensor("op_26617_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26617_dilations_0 = const()[name = tensor("op_26617_dilations_0"), val = tensor([1, 1])]; + tensor var_26617_groups_0 = const()[name = tensor("op_26617_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789869824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799700288))), name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799700480)))]; + tensor var_26617_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_26617_dilations_0, groups = var_26617_groups_0, pad = var_26617_pad_0, pad_type = var_26617_pad_type_0, strides = var_26617_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized, x = input_371_cast_fp16)[name = tensor("op_26617_cast_fp16")]; + tensor var_26618_split_sizes_0 = const()[name = tensor("op_26618_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_26618_axis_0 = const()[name = tensor("op_26618_axis_0"), val = tensor(1)]; + tensor var_26618_cast_fp16_0, tensor var_26618_cast_fp16_1 = split(axis = var_26618_axis_0, split_sizes = var_26618_split_sizes_0, x = var_26617_cast_fp16)[name = tensor("op_26618_cast_fp16")]; + tensor var_26620_mode_0 = const()[name = tensor("op_26620_mode_0"), val = tensor("EXACT")]; + tensor var_26620_cast_fp16 = gelu(mode = var_26620_mode_0, x = var_26618_cast_fp16_1)[name = tensor("op_26620_cast_fp16")]; + tensor input_373_cast_fp16 = mul(x = var_26618_cast_fp16_0, y = var_26620_cast_fp16)[name = tensor("input_373_cast_fp16")]; + tensor var_26628_pad_type_0 = const()[name = tensor("op_26628_pad_type_0"), val = tensor("valid")]; + tensor var_26628_strides_0 = const()[name = tensor("op_26628_strides_0"), val = tensor([1, 1])]; + tensor var_26628_pad_0 = const()[name = tensor("op_26628_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26628_dilations_0 = const()[name = tensor("op_26628_dilations_0"), val = tensor([1, 1])]; + tensor var_26628_groups_0 = const()[name = tensor("op_26628_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799721024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804636288))), name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804636480)))]; + tensor var_26628_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_26628_dilations_0, groups = var_26628_groups_0, pad = var_26628_pad_0, pad_type = var_26628_pad_type_0, strides = var_26628_strides_0, weight = mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized, x = input_373_cast_fp16)[name = tensor("op_26628_cast_fp16")]; + tensor inputs_181_cast_fp16 = add(x = var_26628_cast_fp16, y = inputs_179_cast_fp16)[name = tensor("inputs_181_cast_fp16")]; + tensor hidden_states_245_axes_0 = const()[name = tensor("hidden_states_245_axes_0"), val = tensor([1])]; + tensor hidden_states_245_gamma_0_to_fp16 = const()[name = tensor("hidden_states_245_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804639104)))]; + tensor hidden_states_245_beta_0_to_fp16 = const()[name = tensor("hidden_states_245_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804641728)))]; + tensor var_26644_to_fp16 = const()[name = tensor("op_26644_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_245_cast_fp16 = layer_norm(axes = hidden_states_245_axes_0, beta = hidden_states_245_beta_0_to_fp16, epsilon = var_26644_to_fp16, gamma = hidden_states_245_gamma_0_to_fp16, x = inputs_181_cast_fp16)[name = tensor("hidden_states_245_cast_fp16")]; + tensor q_121_pad_type_0 = const()[name = tensor("q_121_pad_type_0"), val = tensor("valid")]; + tensor q_121_strides_0 = const()[name = tensor("q_121_strides_0"), val = tensor([1, 1])]; + tensor q_121_pad_0 = const()[name = tensor("q_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_121_dilations_0 = const()[name = tensor("q_121_dilations_0"), val = tensor([1, 1])]; + tensor q_121_groups_0 = const()[name = tensor("q_121_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804644352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805873216))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_121_cast_fp16 = conv(dilations = q_121_dilations_0, groups = q_121_groups_0, pad = q_121_pad_0, pad_type = q_121_pad_type_0, strides = q_121_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_245_cast_fp16)[name = tensor("q_121_cast_fp16")]; + tensor k_241_pad_type_0 = const()[name = tensor("k_241_pad_type_0"), val = tensor("valid")]; + tensor k_241_strides_0 = const()[name = tensor("k_241_strides_0"), val = tensor([1, 1])]; + tensor k_241_pad_0 = const()[name = tensor("k_241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_241_dilations_0 = const()[name = tensor("k_241_dilations_0"), val = tensor([1, 1])]; + tensor k_241_groups_0 = const()[name = tensor("k_241_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805873408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807102272))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_241_cast_fp16 = conv(dilations = k_241_dilations_0, groups = k_241_groups_0, pad = k_241_pad_0, pad_type = k_241_pad_type_0, strides = k_241_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_245_cast_fp16)[name = tensor("k_241_cast_fp16")]; + tensor v_121_pad_type_0 = const()[name = tensor("v_121_pad_type_0"), val = tensor("valid")]; + tensor v_121_strides_0 = const()[name = tensor("v_121_strides_0"), val = tensor([1, 1])]; + tensor v_121_pad_0 = const()[name = tensor("v_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_121_dilations_0 = const()[name = tensor("v_121_dilations_0"), val = tensor([1, 1])]; + tensor v_121_groups_0 = const()[name = tensor("v_121_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807102464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808331328))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_121_cast_fp16 = conv(dilations = v_121_dilations_0, groups = v_121_groups_0, pad = v_121_pad_0, pad_type = v_121_pad_type_0, strides = v_121_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_245_cast_fp16)[name = tensor("v_121_cast_fp16")]; + tensor var_26677_begin_0 = const()[name = tensor("op_26677_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26677_end_0 = const()[name = tensor("op_26677_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_26677_end_mask_0 = const()[name = tensor("op_26677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26677_cast_fp16 = slice_by_index(begin = var_26677_begin_0, end = var_26677_end_0, end_mask = var_26677_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26677_cast_fp16")]; + tensor var_26681_begin_0 = const()[name = tensor("op_26681_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26681_end_0 = const()[name = tensor("op_26681_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_26681_end_mask_0 = const()[name = tensor("op_26681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26681_cast_fp16 = slice_by_index(begin = var_26681_begin_0, end = var_26681_end_0, end_mask = var_26681_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26681_cast_fp16")]; + tensor var_26685_begin_0 = const()[name = tensor("op_26685_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26685_end_0 = const()[name = tensor("op_26685_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_26685_end_mask_0 = const()[name = tensor("op_26685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26685_cast_fp16 = slice_by_index(begin = var_26685_begin_0, end = var_26685_end_0, end_mask = var_26685_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26685_cast_fp16")]; + tensor var_26689_begin_0 = const()[name = tensor("op_26689_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26689_end_0 = const()[name = tensor("op_26689_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_26689_end_mask_0 = const()[name = tensor("op_26689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26689_cast_fp16 = slice_by_index(begin = var_26689_begin_0, end = var_26689_end_0, end_mask = var_26689_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26689_cast_fp16")]; + tensor var_26693_begin_0 = const()[name = tensor("op_26693_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26693_end_0 = const()[name = tensor("op_26693_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_26693_end_mask_0 = const()[name = tensor("op_26693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26693_cast_fp16 = slice_by_index(begin = var_26693_begin_0, end = var_26693_end_0, end_mask = var_26693_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26693_cast_fp16")]; + tensor var_26697_begin_0 = const()[name = tensor("op_26697_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26697_end_0 = const()[name = tensor("op_26697_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_26697_end_mask_0 = const()[name = tensor("op_26697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26697_cast_fp16 = slice_by_index(begin = var_26697_begin_0, end = var_26697_end_0, end_mask = var_26697_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26697_cast_fp16")]; + tensor var_26701_begin_0 = const()[name = tensor("op_26701_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26701_end_0 = const()[name = tensor("op_26701_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_26701_end_mask_0 = const()[name = tensor("op_26701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26701_cast_fp16 = slice_by_index(begin = var_26701_begin_0, end = var_26701_end_0, end_mask = var_26701_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26701_cast_fp16")]; + tensor var_26705_begin_0 = const()[name = tensor("op_26705_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26705_end_0 = const()[name = tensor("op_26705_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_26705_end_mask_0 = const()[name = tensor("op_26705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26705_cast_fp16 = slice_by_index(begin = var_26705_begin_0, end = var_26705_end_0, end_mask = var_26705_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26705_cast_fp16")]; + tensor var_26709_begin_0 = const()[name = tensor("op_26709_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26709_end_0 = const()[name = tensor("op_26709_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_26709_end_mask_0 = const()[name = tensor("op_26709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26709_cast_fp16 = slice_by_index(begin = var_26709_begin_0, end = var_26709_end_0, end_mask = var_26709_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26709_cast_fp16")]; + tensor var_26713_begin_0 = const()[name = tensor("op_26713_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26713_end_0 = const()[name = tensor("op_26713_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_26713_end_mask_0 = const()[name = tensor("op_26713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26713_cast_fp16 = slice_by_index(begin = var_26713_begin_0, end = var_26713_end_0, end_mask = var_26713_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26713_cast_fp16")]; + tensor var_26717_begin_0 = const()[name = tensor("op_26717_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26717_end_0 = const()[name = tensor("op_26717_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_26717_end_mask_0 = const()[name = tensor("op_26717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26717_cast_fp16 = slice_by_index(begin = var_26717_begin_0, end = var_26717_end_0, end_mask = var_26717_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26717_cast_fp16")]; + tensor var_26721_begin_0 = const()[name = tensor("op_26721_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26721_end_0 = const()[name = tensor("op_26721_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_26721_end_mask_0 = const()[name = tensor("op_26721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26721_cast_fp16 = slice_by_index(begin = var_26721_begin_0, end = var_26721_end_0, end_mask = var_26721_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26721_cast_fp16")]; + tensor var_26725_begin_0 = const()[name = tensor("op_26725_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26725_end_0 = const()[name = tensor("op_26725_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_26725_end_mask_0 = const()[name = tensor("op_26725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26725_cast_fp16 = slice_by_index(begin = var_26725_begin_0, end = var_26725_end_0, end_mask = var_26725_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26725_cast_fp16")]; + tensor var_26729_begin_0 = const()[name = tensor("op_26729_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26729_end_0 = const()[name = tensor("op_26729_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_26729_end_mask_0 = const()[name = tensor("op_26729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26729_cast_fp16 = slice_by_index(begin = var_26729_begin_0, end = var_26729_end_0, end_mask = var_26729_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26729_cast_fp16")]; + tensor var_26733_begin_0 = const()[name = tensor("op_26733_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26733_end_0 = const()[name = tensor("op_26733_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_26733_end_mask_0 = const()[name = tensor("op_26733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26733_cast_fp16 = slice_by_index(begin = var_26733_begin_0, end = var_26733_end_0, end_mask = var_26733_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26733_cast_fp16")]; + tensor var_26737_begin_0 = const()[name = tensor("op_26737_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26737_end_0 = const()[name = tensor("op_26737_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26737_end_mask_0 = const()[name = tensor("op_26737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26737_cast_fp16 = slice_by_index(begin = var_26737_begin_0, end = var_26737_end_0, end_mask = var_26737_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26737_cast_fp16")]; + tensor var_26741_begin_0 = const()[name = tensor("op_26741_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26741_end_0 = const()[name = tensor("op_26741_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_26741_end_mask_0 = const()[name = tensor("op_26741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26741_cast_fp16 = slice_by_index(begin = var_26741_begin_0, end = var_26741_end_0, end_mask = var_26741_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26741_cast_fp16")]; + tensor var_26745_begin_0 = const()[name = tensor("op_26745_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26745_end_0 = const()[name = tensor("op_26745_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_26745_end_mask_0 = const()[name = tensor("op_26745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26745_cast_fp16 = slice_by_index(begin = var_26745_begin_0, end = var_26745_end_0, end_mask = var_26745_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26745_cast_fp16")]; + tensor var_26749_begin_0 = const()[name = tensor("op_26749_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26749_end_0 = const()[name = tensor("op_26749_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_26749_end_mask_0 = const()[name = tensor("op_26749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26749_cast_fp16 = slice_by_index(begin = var_26749_begin_0, end = var_26749_end_0, end_mask = var_26749_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26749_cast_fp16")]; + tensor var_26753_begin_0 = const()[name = tensor("op_26753_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26753_end_0 = const()[name = tensor("op_26753_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26753_end_mask_0 = const()[name = tensor("op_26753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26753_cast_fp16 = slice_by_index(begin = var_26753_begin_0, end = var_26753_end_0, end_mask = var_26753_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26753_cast_fp16")]; + tensor k_243_perm_0 = const()[name = tensor("k_243_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_26760_begin_0 = const()[name = tensor("op_26760_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26760_end_0 = const()[name = tensor("op_26760_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_26760_end_mask_0 = const()[name = tensor("op_26760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_243_cast_fp16 = transpose(perm = k_243_perm_0, x = k_241_cast_fp16)[name = tensor("transpose_7")]; + tensor var_26760_cast_fp16 = slice_by_index(begin = var_26760_begin_0, end = var_26760_end_0, end_mask = var_26760_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26760_cast_fp16")]; + tensor var_26764_begin_0 = const()[name = tensor("op_26764_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_26764_end_0 = const()[name = tensor("op_26764_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_26764_end_mask_0 = const()[name = tensor("op_26764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26764_cast_fp16 = slice_by_index(begin = var_26764_begin_0, end = var_26764_end_0, end_mask = var_26764_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26764_cast_fp16")]; + tensor var_26768_begin_0 = const()[name = tensor("op_26768_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_26768_end_0 = const()[name = tensor("op_26768_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_26768_end_mask_0 = const()[name = tensor("op_26768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26768_cast_fp16 = slice_by_index(begin = var_26768_begin_0, end = var_26768_end_0, end_mask = var_26768_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26768_cast_fp16")]; + tensor var_26772_begin_0 = const()[name = tensor("op_26772_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_26772_end_0 = const()[name = tensor("op_26772_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_26772_end_mask_0 = const()[name = tensor("op_26772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26772_cast_fp16 = slice_by_index(begin = var_26772_begin_0, end = var_26772_end_0, end_mask = var_26772_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26772_cast_fp16")]; + tensor var_26776_begin_0 = const()[name = tensor("op_26776_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_26776_end_0 = const()[name = tensor("op_26776_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_26776_end_mask_0 = const()[name = tensor("op_26776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26776_cast_fp16 = slice_by_index(begin = var_26776_begin_0, end = var_26776_end_0, end_mask = var_26776_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26776_cast_fp16")]; + tensor var_26780_begin_0 = const()[name = tensor("op_26780_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_26780_end_0 = const()[name = tensor("op_26780_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_26780_end_mask_0 = const()[name = tensor("op_26780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26780_cast_fp16 = slice_by_index(begin = var_26780_begin_0, end = var_26780_end_0, end_mask = var_26780_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26780_cast_fp16")]; + tensor var_26784_begin_0 = const()[name = tensor("op_26784_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_26784_end_0 = const()[name = tensor("op_26784_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_26784_end_mask_0 = const()[name = tensor("op_26784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26784_cast_fp16 = slice_by_index(begin = var_26784_begin_0, end = var_26784_end_0, end_mask = var_26784_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26784_cast_fp16")]; + tensor var_26788_begin_0 = const()[name = tensor("op_26788_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_26788_end_0 = const()[name = tensor("op_26788_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_26788_end_mask_0 = const()[name = tensor("op_26788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26788_cast_fp16 = slice_by_index(begin = var_26788_begin_0, end = var_26788_end_0, end_mask = var_26788_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26788_cast_fp16")]; + tensor var_26792_begin_0 = const()[name = tensor("op_26792_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_26792_end_0 = const()[name = tensor("op_26792_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_26792_end_mask_0 = const()[name = tensor("op_26792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26792_cast_fp16 = slice_by_index(begin = var_26792_begin_0, end = var_26792_end_0, end_mask = var_26792_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26792_cast_fp16")]; + tensor var_26796_begin_0 = const()[name = tensor("op_26796_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_26796_end_0 = const()[name = tensor("op_26796_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_26796_end_mask_0 = const()[name = tensor("op_26796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26796_cast_fp16 = slice_by_index(begin = var_26796_begin_0, end = var_26796_end_0, end_mask = var_26796_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26796_cast_fp16")]; + tensor var_26800_begin_0 = const()[name = tensor("op_26800_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_26800_end_0 = const()[name = tensor("op_26800_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_26800_end_mask_0 = const()[name = tensor("op_26800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26800_cast_fp16 = slice_by_index(begin = var_26800_begin_0, end = var_26800_end_0, end_mask = var_26800_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26800_cast_fp16")]; + tensor var_26804_begin_0 = const()[name = tensor("op_26804_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_26804_end_0 = const()[name = tensor("op_26804_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_26804_end_mask_0 = const()[name = tensor("op_26804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26804_cast_fp16 = slice_by_index(begin = var_26804_begin_0, end = var_26804_end_0, end_mask = var_26804_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26804_cast_fp16")]; + tensor var_26808_begin_0 = const()[name = tensor("op_26808_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_26808_end_0 = const()[name = tensor("op_26808_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_26808_end_mask_0 = const()[name = tensor("op_26808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26808_cast_fp16 = slice_by_index(begin = var_26808_begin_0, end = var_26808_end_0, end_mask = var_26808_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26808_cast_fp16")]; + tensor var_26812_begin_0 = const()[name = tensor("op_26812_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_26812_end_0 = const()[name = tensor("op_26812_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_26812_end_mask_0 = const()[name = tensor("op_26812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26812_cast_fp16 = slice_by_index(begin = var_26812_begin_0, end = var_26812_end_0, end_mask = var_26812_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26812_cast_fp16")]; + tensor var_26816_begin_0 = const()[name = tensor("op_26816_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_26816_end_0 = const()[name = tensor("op_26816_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_26816_end_mask_0 = const()[name = tensor("op_26816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26816_cast_fp16 = slice_by_index(begin = var_26816_begin_0, end = var_26816_end_0, end_mask = var_26816_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26816_cast_fp16")]; + tensor var_26820_begin_0 = const()[name = tensor("op_26820_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_26820_end_0 = const()[name = tensor("op_26820_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26820_end_mask_0 = const()[name = tensor("op_26820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26820_cast_fp16 = slice_by_index(begin = var_26820_begin_0, end = var_26820_end_0, end_mask = var_26820_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26820_cast_fp16")]; + tensor var_26824_begin_0 = const()[name = tensor("op_26824_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_26824_end_0 = const()[name = tensor("op_26824_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_26824_end_mask_0 = const()[name = tensor("op_26824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26824_cast_fp16 = slice_by_index(begin = var_26824_begin_0, end = var_26824_end_0, end_mask = var_26824_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26824_cast_fp16")]; + tensor var_26828_begin_0 = const()[name = tensor("op_26828_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_26828_end_0 = const()[name = tensor("op_26828_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_26828_end_mask_0 = const()[name = tensor("op_26828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26828_cast_fp16 = slice_by_index(begin = var_26828_begin_0, end = var_26828_end_0, end_mask = var_26828_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26828_cast_fp16")]; + tensor var_26832_begin_0 = const()[name = tensor("op_26832_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_26832_end_0 = const()[name = tensor("op_26832_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_26832_end_mask_0 = const()[name = tensor("op_26832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26832_cast_fp16 = slice_by_index(begin = var_26832_begin_0, end = var_26832_end_0, end_mask = var_26832_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26832_cast_fp16")]; + tensor var_26836_begin_0 = const()[name = tensor("op_26836_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_26836_end_0 = const()[name = tensor("op_26836_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_26836_end_mask_0 = const()[name = tensor("op_26836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26836_cast_fp16 = slice_by_index(begin = var_26836_begin_0, end = var_26836_end_0, end_mask = var_26836_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26836_cast_fp16")]; + tensor var_26838_begin_0 = const()[name = tensor("op_26838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26838_end_0 = const()[name = tensor("op_26838_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_26838_end_mask_0 = const()[name = tensor("op_26838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26838_cast_fp16 = slice_by_index(begin = var_26838_begin_0, end = var_26838_end_0, end_mask = var_26838_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26838_cast_fp16")]; + tensor var_26842_begin_0 = const()[name = tensor("op_26842_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26842_end_0 = const()[name = tensor("op_26842_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_26842_end_mask_0 = const()[name = tensor("op_26842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26842_cast_fp16 = slice_by_index(begin = var_26842_begin_0, end = var_26842_end_0, end_mask = var_26842_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26842_cast_fp16")]; + tensor var_26846_begin_0 = const()[name = tensor("op_26846_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26846_end_0 = const()[name = tensor("op_26846_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_26846_end_mask_0 = const()[name = tensor("op_26846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26846_cast_fp16 = slice_by_index(begin = var_26846_begin_0, end = var_26846_end_0, end_mask = var_26846_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26846_cast_fp16")]; + tensor var_26850_begin_0 = const()[name = tensor("op_26850_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26850_end_0 = const()[name = tensor("op_26850_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_26850_end_mask_0 = const()[name = tensor("op_26850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26850_cast_fp16 = slice_by_index(begin = var_26850_begin_0, end = var_26850_end_0, end_mask = var_26850_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26850_cast_fp16")]; + tensor var_26854_begin_0 = const()[name = tensor("op_26854_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26854_end_0 = const()[name = tensor("op_26854_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_26854_end_mask_0 = const()[name = tensor("op_26854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26854_cast_fp16 = slice_by_index(begin = var_26854_begin_0, end = var_26854_end_0, end_mask = var_26854_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26854_cast_fp16")]; + tensor var_26858_begin_0 = const()[name = tensor("op_26858_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26858_end_0 = const()[name = tensor("op_26858_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_26858_end_mask_0 = const()[name = tensor("op_26858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26858_cast_fp16 = slice_by_index(begin = var_26858_begin_0, end = var_26858_end_0, end_mask = var_26858_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26858_cast_fp16")]; + tensor var_26862_begin_0 = const()[name = tensor("op_26862_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26862_end_0 = const()[name = tensor("op_26862_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_26862_end_mask_0 = const()[name = tensor("op_26862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26862_cast_fp16 = slice_by_index(begin = var_26862_begin_0, end = var_26862_end_0, end_mask = var_26862_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26862_cast_fp16")]; + tensor var_26866_begin_0 = const()[name = tensor("op_26866_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26866_end_0 = const()[name = tensor("op_26866_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_26866_end_mask_0 = const()[name = tensor("op_26866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26866_cast_fp16 = slice_by_index(begin = var_26866_begin_0, end = var_26866_end_0, end_mask = var_26866_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26866_cast_fp16")]; + tensor var_26870_begin_0 = const()[name = tensor("op_26870_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26870_end_0 = const()[name = tensor("op_26870_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_26870_end_mask_0 = const()[name = tensor("op_26870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26870_cast_fp16 = slice_by_index(begin = var_26870_begin_0, end = var_26870_end_0, end_mask = var_26870_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26870_cast_fp16")]; + tensor var_26874_begin_0 = const()[name = tensor("op_26874_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26874_end_0 = const()[name = tensor("op_26874_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_26874_end_mask_0 = const()[name = tensor("op_26874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26874_cast_fp16 = slice_by_index(begin = var_26874_begin_0, end = var_26874_end_0, end_mask = var_26874_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26874_cast_fp16")]; + tensor var_26878_begin_0 = const()[name = tensor("op_26878_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26878_end_0 = const()[name = tensor("op_26878_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_26878_end_mask_0 = const()[name = tensor("op_26878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26878_cast_fp16 = slice_by_index(begin = var_26878_begin_0, end = var_26878_end_0, end_mask = var_26878_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26878_cast_fp16")]; + tensor var_26882_begin_0 = const()[name = tensor("op_26882_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26882_end_0 = const()[name = tensor("op_26882_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_26882_end_mask_0 = const()[name = tensor("op_26882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26882_cast_fp16 = slice_by_index(begin = var_26882_begin_0, end = var_26882_end_0, end_mask = var_26882_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26882_cast_fp16")]; + tensor var_26886_begin_0 = const()[name = tensor("op_26886_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26886_end_0 = const()[name = tensor("op_26886_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_26886_end_mask_0 = const()[name = tensor("op_26886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26886_cast_fp16 = slice_by_index(begin = var_26886_begin_0, end = var_26886_end_0, end_mask = var_26886_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26886_cast_fp16")]; + tensor var_26890_begin_0 = const()[name = tensor("op_26890_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26890_end_0 = const()[name = tensor("op_26890_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_26890_end_mask_0 = const()[name = tensor("op_26890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26890_cast_fp16 = slice_by_index(begin = var_26890_begin_0, end = var_26890_end_0, end_mask = var_26890_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26890_cast_fp16")]; + tensor var_26894_begin_0 = const()[name = tensor("op_26894_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26894_end_0 = const()[name = tensor("op_26894_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_26894_end_mask_0 = const()[name = tensor("op_26894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26894_cast_fp16 = slice_by_index(begin = var_26894_begin_0, end = var_26894_end_0, end_mask = var_26894_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26894_cast_fp16")]; + tensor var_26898_begin_0 = const()[name = tensor("op_26898_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26898_end_0 = const()[name = tensor("op_26898_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26898_end_mask_0 = const()[name = tensor("op_26898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26898_cast_fp16 = slice_by_index(begin = var_26898_begin_0, end = var_26898_end_0, end_mask = var_26898_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26898_cast_fp16")]; + tensor var_26902_begin_0 = const()[name = tensor("op_26902_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26902_end_0 = const()[name = tensor("op_26902_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_26902_end_mask_0 = const()[name = tensor("op_26902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26902_cast_fp16 = slice_by_index(begin = var_26902_begin_0, end = var_26902_end_0, end_mask = var_26902_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26902_cast_fp16")]; + tensor var_26906_begin_0 = const()[name = tensor("op_26906_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26906_end_0 = const()[name = tensor("op_26906_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_26906_end_mask_0 = const()[name = tensor("op_26906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26906_cast_fp16 = slice_by_index(begin = var_26906_begin_0, end = var_26906_end_0, end_mask = var_26906_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26906_cast_fp16")]; + tensor var_26910_begin_0 = const()[name = tensor("op_26910_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26910_end_0 = const()[name = tensor("op_26910_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_26910_end_mask_0 = const()[name = tensor("op_26910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26910_cast_fp16 = slice_by_index(begin = var_26910_begin_0, end = var_26910_end_0, end_mask = var_26910_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26910_cast_fp16")]; + tensor var_26914_begin_0 = const()[name = tensor("op_26914_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26914_end_0 = const()[name = tensor("op_26914_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26914_end_mask_0 = const()[name = tensor("op_26914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26914_cast_fp16 = slice_by_index(begin = var_26914_begin_0, end = var_26914_end_0, end_mask = var_26914_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26914_cast_fp16")]; + tensor var_26918_equation_0 = const()[name = tensor("op_26918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26918_cast_fp16 = einsum(equation = var_26918_equation_0, values = (var_26760_cast_fp16, var_26677_cast_fp16))[name = tensor("op_26918_cast_fp16")]; + tensor var_26919_to_fp16 = const()[name = tensor("op_26919_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2241_cast_fp16 = mul(x = var_26918_cast_fp16, y = var_26919_to_fp16)[name = tensor("aw_2241_cast_fp16")]; + tensor var_26922_equation_0 = const()[name = tensor("op_26922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26922_cast_fp16 = einsum(equation = var_26922_equation_0, values = (var_26764_cast_fp16, var_26681_cast_fp16))[name = tensor("op_26922_cast_fp16")]; + tensor var_26923_to_fp16 = const()[name = tensor("op_26923_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2243_cast_fp16 = mul(x = var_26922_cast_fp16, y = var_26923_to_fp16)[name = tensor("aw_2243_cast_fp16")]; + tensor var_26926_equation_0 = const()[name = tensor("op_26926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26926_cast_fp16 = einsum(equation = var_26926_equation_0, values = (var_26768_cast_fp16, var_26685_cast_fp16))[name = tensor("op_26926_cast_fp16")]; + tensor var_26927_to_fp16 = const()[name = tensor("op_26927_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2245_cast_fp16 = mul(x = var_26926_cast_fp16, y = var_26927_to_fp16)[name = tensor("aw_2245_cast_fp16")]; + tensor var_26930_equation_0 = const()[name = tensor("op_26930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26930_cast_fp16 = einsum(equation = var_26930_equation_0, values = (var_26772_cast_fp16, var_26689_cast_fp16))[name = tensor("op_26930_cast_fp16")]; + tensor var_26931_to_fp16 = const()[name = tensor("op_26931_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2247_cast_fp16 = mul(x = var_26930_cast_fp16, y = var_26931_to_fp16)[name = tensor("aw_2247_cast_fp16")]; + tensor var_26934_equation_0 = const()[name = tensor("op_26934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26934_cast_fp16 = einsum(equation = var_26934_equation_0, values = (var_26776_cast_fp16, var_26693_cast_fp16))[name = tensor("op_26934_cast_fp16")]; + tensor var_26935_to_fp16 = const()[name = tensor("op_26935_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2249_cast_fp16 = mul(x = var_26934_cast_fp16, y = var_26935_to_fp16)[name = tensor("aw_2249_cast_fp16")]; + tensor var_26938_equation_0 = const()[name = tensor("op_26938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26938_cast_fp16 = einsum(equation = var_26938_equation_0, values = (var_26780_cast_fp16, var_26697_cast_fp16))[name = tensor("op_26938_cast_fp16")]; + tensor var_26939_to_fp16 = const()[name = tensor("op_26939_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2251_cast_fp16 = mul(x = var_26938_cast_fp16, y = var_26939_to_fp16)[name = tensor("aw_2251_cast_fp16")]; + tensor var_26942_equation_0 = const()[name = tensor("op_26942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26942_cast_fp16 = einsum(equation = var_26942_equation_0, values = (var_26784_cast_fp16, var_26701_cast_fp16))[name = tensor("op_26942_cast_fp16")]; + tensor var_26943_to_fp16 = const()[name = tensor("op_26943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2253_cast_fp16 = mul(x = var_26942_cast_fp16, y = var_26943_to_fp16)[name = tensor("aw_2253_cast_fp16")]; + tensor var_26946_equation_0 = const()[name = tensor("op_26946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26946_cast_fp16 = einsum(equation = var_26946_equation_0, values = (var_26788_cast_fp16, var_26705_cast_fp16))[name = tensor("op_26946_cast_fp16")]; + tensor var_26947_to_fp16 = const()[name = tensor("op_26947_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2255_cast_fp16 = mul(x = var_26946_cast_fp16, y = var_26947_to_fp16)[name = tensor("aw_2255_cast_fp16")]; + tensor var_26950_equation_0 = const()[name = tensor("op_26950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26950_cast_fp16 = einsum(equation = var_26950_equation_0, values = (var_26792_cast_fp16, var_26709_cast_fp16))[name = tensor("op_26950_cast_fp16")]; + tensor var_26951_to_fp16 = const()[name = tensor("op_26951_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2257_cast_fp16 = mul(x = var_26950_cast_fp16, y = var_26951_to_fp16)[name = tensor("aw_2257_cast_fp16")]; + tensor var_26954_equation_0 = const()[name = tensor("op_26954_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26954_cast_fp16 = einsum(equation = var_26954_equation_0, values = (var_26796_cast_fp16, var_26713_cast_fp16))[name = tensor("op_26954_cast_fp16")]; + tensor var_26955_to_fp16 = const()[name = tensor("op_26955_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2259_cast_fp16 = mul(x = var_26954_cast_fp16, y = var_26955_to_fp16)[name = tensor("aw_2259_cast_fp16")]; + tensor var_26958_equation_0 = const()[name = tensor("op_26958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26958_cast_fp16 = einsum(equation = var_26958_equation_0, values = (var_26800_cast_fp16, var_26717_cast_fp16))[name = tensor("op_26958_cast_fp16")]; + tensor var_26959_to_fp16 = const()[name = tensor("op_26959_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2261_cast_fp16 = mul(x = var_26958_cast_fp16, y = var_26959_to_fp16)[name = tensor("aw_2261_cast_fp16")]; + tensor var_26962_equation_0 = const()[name = tensor("op_26962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26962_cast_fp16 = einsum(equation = var_26962_equation_0, values = (var_26804_cast_fp16, var_26721_cast_fp16))[name = tensor("op_26962_cast_fp16")]; + tensor var_26963_to_fp16 = const()[name = tensor("op_26963_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2263_cast_fp16 = mul(x = var_26962_cast_fp16, y = var_26963_to_fp16)[name = tensor("aw_2263_cast_fp16")]; + tensor var_26966_equation_0 = const()[name = tensor("op_26966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26966_cast_fp16 = einsum(equation = var_26966_equation_0, values = (var_26808_cast_fp16, var_26725_cast_fp16))[name = tensor("op_26966_cast_fp16")]; + tensor var_26967_to_fp16 = const()[name = tensor("op_26967_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2265_cast_fp16 = mul(x = var_26966_cast_fp16, y = var_26967_to_fp16)[name = tensor("aw_2265_cast_fp16")]; + tensor var_26970_equation_0 = const()[name = tensor("op_26970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26970_cast_fp16 = einsum(equation = var_26970_equation_0, values = (var_26812_cast_fp16, var_26729_cast_fp16))[name = tensor("op_26970_cast_fp16")]; + tensor var_26971_to_fp16 = const()[name = tensor("op_26971_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2267_cast_fp16 = mul(x = var_26970_cast_fp16, y = var_26971_to_fp16)[name = tensor("aw_2267_cast_fp16")]; + tensor var_26974_equation_0 = const()[name = tensor("op_26974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26974_cast_fp16 = einsum(equation = var_26974_equation_0, values = (var_26816_cast_fp16, var_26733_cast_fp16))[name = tensor("op_26974_cast_fp16")]; + tensor var_26975_to_fp16 = const()[name = tensor("op_26975_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2269_cast_fp16 = mul(x = var_26974_cast_fp16, y = var_26975_to_fp16)[name = tensor("aw_2269_cast_fp16")]; + tensor var_26978_equation_0 = const()[name = tensor("op_26978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26978_cast_fp16 = einsum(equation = var_26978_equation_0, values = (var_26820_cast_fp16, var_26737_cast_fp16))[name = tensor("op_26978_cast_fp16")]; + tensor var_26979_to_fp16 = const()[name = tensor("op_26979_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2271_cast_fp16 = mul(x = var_26978_cast_fp16, y = var_26979_to_fp16)[name = tensor("aw_2271_cast_fp16")]; + tensor var_26982_equation_0 = const()[name = tensor("op_26982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26982_cast_fp16 = einsum(equation = var_26982_equation_0, values = (var_26824_cast_fp16, var_26741_cast_fp16))[name = tensor("op_26982_cast_fp16")]; + tensor var_26983_to_fp16 = const()[name = tensor("op_26983_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2273_cast_fp16 = mul(x = var_26982_cast_fp16, y = var_26983_to_fp16)[name = tensor("aw_2273_cast_fp16")]; + tensor var_26986_equation_0 = const()[name = tensor("op_26986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26986_cast_fp16 = einsum(equation = var_26986_equation_0, values = (var_26828_cast_fp16, var_26745_cast_fp16))[name = tensor("op_26986_cast_fp16")]; + tensor var_26987_to_fp16 = const()[name = tensor("op_26987_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2275_cast_fp16 = mul(x = var_26986_cast_fp16, y = var_26987_to_fp16)[name = tensor("aw_2275_cast_fp16")]; + tensor var_26990_equation_0 = const()[name = tensor("op_26990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26990_cast_fp16 = einsum(equation = var_26990_equation_0, values = (var_26832_cast_fp16, var_26749_cast_fp16))[name = tensor("op_26990_cast_fp16")]; + tensor var_26991_to_fp16 = const()[name = tensor("op_26991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2277_cast_fp16 = mul(x = var_26990_cast_fp16, y = var_26991_to_fp16)[name = tensor("aw_2277_cast_fp16")]; + tensor var_26994_equation_0 = const()[name = tensor("op_26994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26994_cast_fp16 = einsum(equation = var_26994_equation_0, values = (var_26836_cast_fp16, var_26753_cast_fp16))[name = tensor("op_26994_cast_fp16")]; + tensor var_26995_to_fp16 = const()[name = tensor("op_26995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2279_cast_fp16 = mul(x = var_26994_cast_fp16, y = var_26995_to_fp16)[name = tensor("aw_2279_cast_fp16")]; + tensor var_26997_cast_fp16 = softmax(axis = var_21077, x = aw_2241_cast_fp16)[name = tensor("op_26997_cast_fp16")]; + tensor var_26998_cast_fp16 = softmax(axis = var_21077, x = aw_2243_cast_fp16)[name = tensor("op_26998_cast_fp16")]; + tensor var_26999_cast_fp16 = softmax(axis = var_21077, x = aw_2245_cast_fp16)[name = tensor("op_26999_cast_fp16")]; + tensor var_27000_cast_fp16 = softmax(axis = var_21077, x = aw_2247_cast_fp16)[name = tensor("op_27000_cast_fp16")]; + tensor var_27001_cast_fp16 = softmax(axis = var_21077, x = aw_2249_cast_fp16)[name = tensor("op_27001_cast_fp16")]; + tensor var_27002_cast_fp16 = softmax(axis = var_21077, x = aw_2251_cast_fp16)[name = tensor("op_27002_cast_fp16")]; + tensor var_27003_cast_fp16 = softmax(axis = var_21077, x = aw_2253_cast_fp16)[name = tensor("op_27003_cast_fp16")]; + tensor var_27004_cast_fp16 = softmax(axis = var_21077, x = aw_2255_cast_fp16)[name = tensor("op_27004_cast_fp16")]; + tensor var_27005_cast_fp16 = softmax(axis = var_21077, x = aw_2257_cast_fp16)[name = tensor("op_27005_cast_fp16")]; + tensor var_27006_cast_fp16 = softmax(axis = var_21077, x = aw_2259_cast_fp16)[name = tensor("op_27006_cast_fp16")]; + tensor var_27007_cast_fp16 = softmax(axis = var_21077, x = aw_2261_cast_fp16)[name = tensor("op_27007_cast_fp16")]; + tensor var_27008_cast_fp16 = softmax(axis = var_21077, x = aw_2263_cast_fp16)[name = tensor("op_27008_cast_fp16")]; + tensor var_27009_cast_fp16 = softmax(axis = var_21077, x = aw_2265_cast_fp16)[name = tensor("op_27009_cast_fp16")]; + tensor var_27010_cast_fp16 = softmax(axis = var_21077, x = aw_2267_cast_fp16)[name = tensor("op_27010_cast_fp16")]; + tensor var_27011_cast_fp16 = softmax(axis = var_21077, x = aw_2269_cast_fp16)[name = tensor("op_27011_cast_fp16")]; + tensor var_27012_cast_fp16 = softmax(axis = var_21077, x = aw_2271_cast_fp16)[name = tensor("op_27012_cast_fp16")]; + tensor var_27013_cast_fp16 = softmax(axis = var_21077, x = aw_2273_cast_fp16)[name = tensor("op_27013_cast_fp16")]; + tensor var_27014_cast_fp16 = softmax(axis = var_21077, x = aw_2275_cast_fp16)[name = tensor("op_27014_cast_fp16")]; + tensor var_27015_cast_fp16 = softmax(axis = var_21077, x = aw_2277_cast_fp16)[name = tensor("op_27015_cast_fp16")]; + tensor var_27016_cast_fp16 = softmax(axis = var_21077, x = aw_2279_cast_fp16)[name = tensor("op_27016_cast_fp16")]; + tensor var_27018_equation_0 = const()[name = tensor("op_27018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27018_cast_fp16 = einsum(equation = var_27018_equation_0, values = (var_26838_cast_fp16, var_26997_cast_fp16))[name = tensor("op_27018_cast_fp16")]; + tensor var_27020_equation_0 = const()[name = tensor("op_27020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27020_cast_fp16 = einsum(equation = var_27020_equation_0, values = (var_26842_cast_fp16, var_26998_cast_fp16))[name = tensor("op_27020_cast_fp16")]; + tensor var_27022_equation_0 = const()[name = tensor("op_27022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27022_cast_fp16 = einsum(equation = var_27022_equation_0, values = (var_26846_cast_fp16, var_26999_cast_fp16))[name = tensor("op_27022_cast_fp16")]; + tensor var_27024_equation_0 = const()[name = tensor("op_27024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27024_cast_fp16 = einsum(equation = var_27024_equation_0, values = (var_26850_cast_fp16, var_27000_cast_fp16))[name = tensor("op_27024_cast_fp16")]; + tensor var_27026_equation_0 = const()[name = tensor("op_27026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27026_cast_fp16 = einsum(equation = var_27026_equation_0, values = (var_26854_cast_fp16, var_27001_cast_fp16))[name = tensor("op_27026_cast_fp16")]; + tensor var_27028_equation_0 = const()[name = tensor("op_27028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27028_cast_fp16 = einsum(equation = var_27028_equation_0, values = (var_26858_cast_fp16, var_27002_cast_fp16))[name = tensor("op_27028_cast_fp16")]; + tensor var_27030_equation_0 = const()[name = tensor("op_27030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27030_cast_fp16 = einsum(equation = var_27030_equation_0, values = (var_26862_cast_fp16, var_27003_cast_fp16))[name = tensor("op_27030_cast_fp16")]; + tensor var_27032_equation_0 = const()[name = tensor("op_27032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27032_cast_fp16 = einsum(equation = var_27032_equation_0, values = (var_26866_cast_fp16, var_27004_cast_fp16))[name = tensor("op_27032_cast_fp16")]; + tensor var_27034_equation_0 = const()[name = tensor("op_27034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27034_cast_fp16 = einsum(equation = var_27034_equation_0, values = (var_26870_cast_fp16, var_27005_cast_fp16))[name = tensor("op_27034_cast_fp16")]; + tensor var_27036_equation_0 = const()[name = tensor("op_27036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27036_cast_fp16 = einsum(equation = var_27036_equation_0, values = (var_26874_cast_fp16, var_27006_cast_fp16))[name = tensor("op_27036_cast_fp16")]; + tensor var_27038_equation_0 = const()[name = tensor("op_27038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27038_cast_fp16 = einsum(equation = var_27038_equation_0, values = (var_26878_cast_fp16, var_27007_cast_fp16))[name = tensor("op_27038_cast_fp16")]; + tensor var_27040_equation_0 = const()[name = tensor("op_27040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27040_cast_fp16 = einsum(equation = var_27040_equation_0, values = (var_26882_cast_fp16, var_27008_cast_fp16))[name = tensor("op_27040_cast_fp16")]; + tensor var_27042_equation_0 = const()[name = tensor("op_27042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27042_cast_fp16 = einsum(equation = var_27042_equation_0, values = (var_26886_cast_fp16, var_27009_cast_fp16))[name = tensor("op_27042_cast_fp16")]; + tensor var_27044_equation_0 = const()[name = tensor("op_27044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27044_cast_fp16 = einsum(equation = var_27044_equation_0, values = (var_26890_cast_fp16, var_27010_cast_fp16))[name = tensor("op_27044_cast_fp16")]; + tensor var_27046_equation_0 = const()[name = tensor("op_27046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27046_cast_fp16 = einsum(equation = var_27046_equation_0, values = (var_26894_cast_fp16, var_27011_cast_fp16))[name = tensor("op_27046_cast_fp16")]; + tensor var_27048_equation_0 = const()[name = tensor("op_27048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27048_cast_fp16 = einsum(equation = var_27048_equation_0, values = (var_26898_cast_fp16, var_27012_cast_fp16))[name = tensor("op_27048_cast_fp16")]; + tensor var_27050_equation_0 = const()[name = tensor("op_27050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27050_cast_fp16 = einsum(equation = var_27050_equation_0, values = (var_26902_cast_fp16, var_27013_cast_fp16))[name = tensor("op_27050_cast_fp16")]; + tensor var_27052_equation_0 = const()[name = tensor("op_27052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27052_cast_fp16 = einsum(equation = var_27052_equation_0, values = (var_26906_cast_fp16, var_27014_cast_fp16))[name = tensor("op_27052_cast_fp16")]; + tensor var_27054_equation_0 = const()[name = tensor("op_27054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27054_cast_fp16 = einsum(equation = var_27054_equation_0, values = (var_26910_cast_fp16, var_27015_cast_fp16))[name = tensor("op_27054_cast_fp16")]; + tensor var_27056_equation_0 = const()[name = tensor("op_27056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27056_cast_fp16 = einsum(equation = var_27056_equation_0, values = (var_26914_cast_fp16, var_27016_cast_fp16))[name = tensor("op_27056_cast_fp16")]; + tensor input_375_interleave_0 = const()[name = tensor("input_375_interleave_0"), val = tensor(false)]; + tensor input_375_cast_fp16 = concat(axis = var_21077, interleave = input_375_interleave_0, values = (var_27018_cast_fp16, var_27020_cast_fp16, var_27022_cast_fp16, var_27024_cast_fp16, var_27026_cast_fp16, var_27028_cast_fp16, var_27030_cast_fp16, var_27032_cast_fp16, var_27034_cast_fp16, var_27036_cast_fp16, var_27038_cast_fp16, var_27040_cast_fp16, var_27042_cast_fp16, var_27044_cast_fp16, var_27046_cast_fp16, var_27048_cast_fp16, var_27050_cast_fp16, var_27052_cast_fp16, var_27054_cast_fp16, var_27056_cast_fp16))[name = tensor("input_375_cast_fp16")]; + tensor var_27066_pad_type_0 = const()[name = tensor("op_27066_pad_type_0"), val = tensor("valid")]; + tensor var_27066_strides_0 = const()[name = tensor("op_27066_strides_0"), val = tensor([1, 1])]; + tensor var_27066_pad_0 = const()[name = tensor("op_27066_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27066_dilations_0 = const()[name = tensor("op_27066_dilations_0"), val = tensor([1, 1])]; + tensor var_27066_groups_0 = const()[name = tensor("op_27066_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808331520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809560384))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809560576)))]; + tensor var_27066_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_27066_dilations_0, groups = var_27066_groups_0, pad = var_27066_pad_0, pad_type = var_27066_pad_type_0, strides = var_27066_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized, x = input_375_cast_fp16)[name = tensor("op_27066_cast_fp16")]; + tensor inputs_183_cast_fp16 = add(x = var_27066_cast_fp16, y = inputs_181_cast_fp16)[name = tensor("inputs_183_cast_fp16")]; + tensor hidden_states_247_axes_0 = const()[name = tensor("hidden_states_247_axes_0"), val = tensor([1])]; + tensor hidden_states_247_gamma_0_to_fp16 = const()[name = tensor("hidden_states_247_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809563200)))]; + tensor hidden_states_247_beta_0_to_fp16 = const()[name = tensor("hidden_states_247_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809565824)))]; + tensor var_27076_to_fp16 = const()[name = tensor("op_27076_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_247_cast_fp16 = layer_norm(axes = hidden_states_247_axes_0, beta = hidden_states_247_beta_0_to_fp16, epsilon = var_27076_to_fp16, gamma = hidden_states_247_gamma_0_to_fp16, x = inputs_183_cast_fp16)[name = tensor("hidden_states_247_cast_fp16")]; + tensor q_123_pad_type_0 = const()[name = tensor("q_123_pad_type_0"), val = tensor("valid")]; + tensor q_123_strides_0 = const()[name = tensor("q_123_strides_0"), val = tensor([1, 1])]; + tensor q_123_pad_0 = const()[name = tensor("q_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_123_dilations_0 = const()[name = tensor("q_123_dilations_0"), val = tensor([1, 1])]; + tensor q_123_groups_0 = const()[name = tensor("q_123_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809568448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(810797312))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_123_cast_fp16 = conv(dilations = q_123_dilations_0, groups = q_123_groups_0, pad = q_123_pad_0, pad_type = q_123_pad_type_0, strides = q_123_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_247_cast_fp16)[name = tensor("q_123_cast_fp16")]; + tensor k_245_pad_type_0 = const()[name = tensor("k_245_pad_type_0"), val = tensor("valid")]; + tensor k_245_strides_0 = const()[name = tensor("k_245_strides_0"), val = tensor([1, 1])]; + tensor k_245_pad_0 = const()[name = tensor("k_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_245_dilations_0 = const()[name = tensor("k_245_dilations_0"), val = tensor([1, 1])]; + tensor k_245_groups_0 = const()[name = tensor("k_245_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(810797504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(812763648))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_245_cast_fp16 = conv(dilations = k_245_dilations_0, groups = k_245_groups_0, pad = k_245_pad_0, pad_type = k_245_pad_type_0, strides = k_245_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_245_cast_fp16")]; + tensor v_123_pad_type_0 = const()[name = tensor("v_123_pad_type_0"), val = tensor("valid")]; + tensor v_123_strides_0 = const()[name = tensor("v_123_strides_0"), val = tensor([1, 1])]; + tensor v_123_pad_0 = const()[name = tensor("v_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_123_dilations_0 = const()[name = tensor("v_123_dilations_0"), val = tensor([1, 1])]; + tensor v_123_groups_0 = const()[name = tensor("v_123_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(812763840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814729984))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_123_cast_fp16 = conv(dilations = v_123_dilations_0, groups = v_123_groups_0, pad = v_123_pad_0, pad_type = v_123_pad_type_0, strides = v_123_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_123_cast_fp16")]; + tensor var_27109_begin_0 = const()[name = tensor("op_27109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27109_end_0 = const()[name = tensor("op_27109_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_27109_end_mask_0 = const()[name = tensor("op_27109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27109_cast_fp16 = slice_by_index(begin = var_27109_begin_0, end = var_27109_end_0, end_mask = var_27109_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27109_cast_fp16")]; + tensor var_27113_begin_0 = const()[name = tensor("op_27113_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27113_end_0 = const()[name = tensor("op_27113_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_27113_end_mask_0 = const()[name = tensor("op_27113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27113_cast_fp16 = slice_by_index(begin = var_27113_begin_0, end = var_27113_end_0, end_mask = var_27113_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27113_cast_fp16")]; + tensor var_27117_begin_0 = const()[name = tensor("op_27117_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27117_end_0 = const()[name = tensor("op_27117_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_27117_end_mask_0 = const()[name = tensor("op_27117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27117_cast_fp16 = slice_by_index(begin = var_27117_begin_0, end = var_27117_end_0, end_mask = var_27117_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27117_cast_fp16")]; + tensor var_27121_begin_0 = const()[name = tensor("op_27121_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27121_end_0 = const()[name = tensor("op_27121_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_27121_end_mask_0 = const()[name = tensor("op_27121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27121_cast_fp16 = slice_by_index(begin = var_27121_begin_0, end = var_27121_end_0, end_mask = var_27121_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27121_cast_fp16")]; + tensor var_27125_begin_0 = const()[name = tensor("op_27125_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27125_end_0 = const()[name = tensor("op_27125_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_27125_end_mask_0 = const()[name = tensor("op_27125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27125_cast_fp16 = slice_by_index(begin = var_27125_begin_0, end = var_27125_end_0, end_mask = var_27125_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27125_cast_fp16")]; + tensor var_27129_begin_0 = const()[name = tensor("op_27129_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27129_end_0 = const()[name = tensor("op_27129_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_27129_end_mask_0 = const()[name = tensor("op_27129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27129_cast_fp16 = slice_by_index(begin = var_27129_begin_0, end = var_27129_end_0, end_mask = var_27129_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27129_cast_fp16")]; + tensor var_27133_begin_0 = const()[name = tensor("op_27133_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27133_end_0 = const()[name = tensor("op_27133_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_27133_end_mask_0 = const()[name = tensor("op_27133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27133_cast_fp16 = slice_by_index(begin = var_27133_begin_0, end = var_27133_end_0, end_mask = var_27133_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27133_cast_fp16")]; + tensor var_27137_begin_0 = const()[name = tensor("op_27137_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27137_end_0 = const()[name = tensor("op_27137_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_27137_end_mask_0 = const()[name = tensor("op_27137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27137_cast_fp16 = slice_by_index(begin = var_27137_begin_0, end = var_27137_end_0, end_mask = var_27137_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27137_cast_fp16")]; + tensor var_27141_begin_0 = const()[name = tensor("op_27141_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27141_end_0 = const()[name = tensor("op_27141_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_27141_end_mask_0 = const()[name = tensor("op_27141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27141_cast_fp16 = slice_by_index(begin = var_27141_begin_0, end = var_27141_end_0, end_mask = var_27141_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27141_cast_fp16")]; + tensor var_27145_begin_0 = const()[name = tensor("op_27145_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27145_end_0 = const()[name = tensor("op_27145_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_27145_end_mask_0 = const()[name = tensor("op_27145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27145_cast_fp16 = slice_by_index(begin = var_27145_begin_0, end = var_27145_end_0, end_mask = var_27145_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27145_cast_fp16")]; + tensor var_27149_begin_0 = const()[name = tensor("op_27149_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27149_end_0 = const()[name = tensor("op_27149_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_27149_end_mask_0 = const()[name = tensor("op_27149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27149_cast_fp16 = slice_by_index(begin = var_27149_begin_0, end = var_27149_end_0, end_mask = var_27149_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27149_cast_fp16")]; + tensor var_27153_begin_0 = const()[name = tensor("op_27153_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27153_end_0 = const()[name = tensor("op_27153_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_27153_end_mask_0 = const()[name = tensor("op_27153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27153_cast_fp16 = slice_by_index(begin = var_27153_begin_0, end = var_27153_end_0, end_mask = var_27153_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27153_cast_fp16")]; + tensor var_27157_begin_0 = const()[name = tensor("op_27157_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27157_end_0 = const()[name = tensor("op_27157_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_27157_end_mask_0 = const()[name = tensor("op_27157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27157_cast_fp16 = slice_by_index(begin = var_27157_begin_0, end = var_27157_end_0, end_mask = var_27157_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27157_cast_fp16")]; + tensor var_27161_begin_0 = const()[name = tensor("op_27161_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27161_end_0 = const()[name = tensor("op_27161_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_27161_end_mask_0 = const()[name = tensor("op_27161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27161_cast_fp16 = slice_by_index(begin = var_27161_begin_0, end = var_27161_end_0, end_mask = var_27161_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27161_cast_fp16")]; + tensor var_27165_begin_0 = const()[name = tensor("op_27165_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27165_end_0 = const()[name = tensor("op_27165_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_27165_end_mask_0 = const()[name = tensor("op_27165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27165_cast_fp16 = slice_by_index(begin = var_27165_begin_0, end = var_27165_end_0, end_mask = var_27165_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27165_cast_fp16")]; + tensor var_27169_begin_0 = const()[name = tensor("op_27169_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27169_end_0 = const()[name = tensor("op_27169_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27169_end_mask_0 = const()[name = tensor("op_27169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27169_cast_fp16 = slice_by_index(begin = var_27169_begin_0, end = var_27169_end_0, end_mask = var_27169_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27169_cast_fp16")]; + tensor var_27173_begin_0 = const()[name = tensor("op_27173_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27173_end_0 = const()[name = tensor("op_27173_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_27173_end_mask_0 = const()[name = tensor("op_27173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27173_cast_fp16 = slice_by_index(begin = var_27173_begin_0, end = var_27173_end_0, end_mask = var_27173_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27173_cast_fp16")]; + tensor var_27177_begin_0 = const()[name = tensor("op_27177_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27177_end_0 = const()[name = tensor("op_27177_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_27177_end_mask_0 = const()[name = tensor("op_27177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27177_cast_fp16 = slice_by_index(begin = var_27177_begin_0, end = var_27177_end_0, end_mask = var_27177_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27177_cast_fp16")]; + tensor var_27181_begin_0 = const()[name = tensor("op_27181_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27181_end_0 = const()[name = tensor("op_27181_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_27181_end_mask_0 = const()[name = tensor("op_27181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27181_cast_fp16 = slice_by_index(begin = var_27181_begin_0, end = var_27181_end_0, end_mask = var_27181_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27181_cast_fp16")]; + tensor var_27185_begin_0 = const()[name = tensor("op_27185_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27185_end_0 = const()[name = tensor("op_27185_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_27185_end_mask_0 = const()[name = tensor("op_27185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27185_cast_fp16 = slice_by_index(begin = var_27185_begin_0, end = var_27185_end_0, end_mask = var_27185_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27185_cast_fp16")]; + tensor k_247_perm_0 = const()[name = tensor("k_247_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_27192_begin_0 = const()[name = tensor("op_27192_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27192_end_0 = const()[name = tensor("op_27192_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_27192_end_mask_0 = const()[name = tensor("op_27192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_247_cast_fp16 = transpose(perm = k_247_perm_0, x = k_245_cast_fp16)[name = tensor("transpose_6")]; + tensor var_27192_cast_fp16 = slice_by_index(begin = var_27192_begin_0, end = var_27192_end_0, end_mask = var_27192_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27192_cast_fp16")]; + tensor var_27196_begin_0 = const()[name = tensor("op_27196_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_27196_end_0 = const()[name = tensor("op_27196_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_27196_end_mask_0 = const()[name = tensor("op_27196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27196_cast_fp16 = slice_by_index(begin = var_27196_begin_0, end = var_27196_end_0, end_mask = var_27196_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27196_cast_fp16")]; + tensor var_27200_begin_0 = const()[name = tensor("op_27200_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_27200_end_0 = const()[name = tensor("op_27200_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_27200_end_mask_0 = const()[name = tensor("op_27200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27200_cast_fp16 = slice_by_index(begin = var_27200_begin_0, end = var_27200_end_0, end_mask = var_27200_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27200_cast_fp16")]; + tensor var_27204_begin_0 = const()[name = tensor("op_27204_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_27204_end_0 = const()[name = tensor("op_27204_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_27204_end_mask_0 = const()[name = tensor("op_27204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27204_cast_fp16 = slice_by_index(begin = var_27204_begin_0, end = var_27204_end_0, end_mask = var_27204_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27204_cast_fp16")]; + tensor var_27208_begin_0 = const()[name = tensor("op_27208_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_27208_end_0 = const()[name = tensor("op_27208_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_27208_end_mask_0 = const()[name = tensor("op_27208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27208_cast_fp16 = slice_by_index(begin = var_27208_begin_0, end = var_27208_end_0, end_mask = var_27208_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27208_cast_fp16")]; + tensor var_27212_begin_0 = const()[name = tensor("op_27212_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_27212_end_0 = const()[name = tensor("op_27212_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_27212_end_mask_0 = const()[name = tensor("op_27212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27212_cast_fp16 = slice_by_index(begin = var_27212_begin_0, end = var_27212_end_0, end_mask = var_27212_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27212_cast_fp16")]; + tensor var_27216_begin_0 = const()[name = tensor("op_27216_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_27216_end_0 = const()[name = tensor("op_27216_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_27216_end_mask_0 = const()[name = tensor("op_27216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27216_cast_fp16 = slice_by_index(begin = var_27216_begin_0, end = var_27216_end_0, end_mask = var_27216_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27216_cast_fp16")]; + tensor var_27220_begin_0 = const()[name = tensor("op_27220_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_27220_end_0 = const()[name = tensor("op_27220_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_27220_end_mask_0 = const()[name = tensor("op_27220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27220_cast_fp16 = slice_by_index(begin = var_27220_begin_0, end = var_27220_end_0, end_mask = var_27220_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27220_cast_fp16")]; + tensor var_27224_begin_0 = const()[name = tensor("op_27224_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_27224_end_0 = const()[name = tensor("op_27224_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_27224_end_mask_0 = const()[name = tensor("op_27224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27224_cast_fp16 = slice_by_index(begin = var_27224_begin_0, end = var_27224_end_0, end_mask = var_27224_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27224_cast_fp16")]; + tensor var_27228_begin_0 = const()[name = tensor("op_27228_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_27228_end_0 = const()[name = tensor("op_27228_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_27228_end_mask_0 = const()[name = tensor("op_27228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27228_cast_fp16 = slice_by_index(begin = var_27228_begin_0, end = var_27228_end_0, end_mask = var_27228_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27228_cast_fp16")]; + tensor var_27232_begin_0 = const()[name = tensor("op_27232_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_27232_end_0 = const()[name = tensor("op_27232_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_27232_end_mask_0 = const()[name = tensor("op_27232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27232_cast_fp16 = slice_by_index(begin = var_27232_begin_0, end = var_27232_end_0, end_mask = var_27232_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27232_cast_fp16")]; + tensor var_27236_begin_0 = const()[name = tensor("op_27236_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_27236_end_0 = const()[name = tensor("op_27236_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_27236_end_mask_0 = const()[name = tensor("op_27236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27236_cast_fp16 = slice_by_index(begin = var_27236_begin_0, end = var_27236_end_0, end_mask = var_27236_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27236_cast_fp16")]; + tensor var_27240_begin_0 = const()[name = tensor("op_27240_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_27240_end_0 = const()[name = tensor("op_27240_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_27240_end_mask_0 = const()[name = tensor("op_27240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27240_cast_fp16 = slice_by_index(begin = var_27240_begin_0, end = var_27240_end_0, end_mask = var_27240_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27240_cast_fp16")]; + tensor var_27244_begin_0 = const()[name = tensor("op_27244_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_27244_end_0 = const()[name = tensor("op_27244_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_27244_end_mask_0 = const()[name = tensor("op_27244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27244_cast_fp16 = slice_by_index(begin = var_27244_begin_0, end = var_27244_end_0, end_mask = var_27244_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27244_cast_fp16")]; + tensor var_27248_begin_0 = const()[name = tensor("op_27248_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_27248_end_0 = const()[name = tensor("op_27248_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_27248_end_mask_0 = const()[name = tensor("op_27248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27248_cast_fp16 = slice_by_index(begin = var_27248_begin_0, end = var_27248_end_0, end_mask = var_27248_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27248_cast_fp16")]; + tensor var_27252_begin_0 = const()[name = tensor("op_27252_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_27252_end_0 = const()[name = tensor("op_27252_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_27252_end_mask_0 = const()[name = tensor("op_27252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27252_cast_fp16 = slice_by_index(begin = var_27252_begin_0, end = var_27252_end_0, end_mask = var_27252_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27252_cast_fp16")]; + tensor var_27256_begin_0 = const()[name = tensor("op_27256_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_27256_end_0 = const()[name = tensor("op_27256_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_27256_end_mask_0 = const()[name = tensor("op_27256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27256_cast_fp16 = slice_by_index(begin = var_27256_begin_0, end = var_27256_end_0, end_mask = var_27256_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27256_cast_fp16")]; + tensor var_27260_begin_0 = const()[name = tensor("op_27260_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_27260_end_0 = const()[name = tensor("op_27260_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_27260_end_mask_0 = const()[name = tensor("op_27260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27260_cast_fp16 = slice_by_index(begin = var_27260_begin_0, end = var_27260_end_0, end_mask = var_27260_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27260_cast_fp16")]; + tensor var_27264_begin_0 = const()[name = tensor("op_27264_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_27264_end_0 = const()[name = tensor("op_27264_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_27264_end_mask_0 = const()[name = tensor("op_27264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27264_cast_fp16 = slice_by_index(begin = var_27264_begin_0, end = var_27264_end_0, end_mask = var_27264_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27264_cast_fp16")]; + tensor var_27268_begin_0 = const()[name = tensor("op_27268_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_27268_end_0 = const()[name = tensor("op_27268_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_27268_end_mask_0 = const()[name = tensor("op_27268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27268_cast_fp16 = slice_by_index(begin = var_27268_begin_0, end = var_27268_end_0, end_mask = var_27268_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27268_cast_fp16")]; + tensor var_27270_begin_0 = const()[name = tensor("op_27270_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27270_end_0 = const()[name = tensor("op_27270_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_27270_end_mask_0 = const()[name = tensor("op_27270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27270_cast_fp16 = slice_by_index(begin = var_27270_begin_0, end = var_27270_end_0, end_mask = var_27270_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27270_cast_fp16")]; + tensor var_27274_begin_0 = const()[name = tensor("op_27274_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27274_end_0 = const()[name = tensor("op_27274_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_27274_end_mask_0 = const()[name = tensor("op_27274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27274_cast_fp16 = slice_by_index(begin = var_27274_begin_0, end = var_27274_end_0, end_mask = var_27274_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27274_cast_fp16")]; + tensor var_27278_begin_0 = const()[name = tensor("op_27278_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27278_end_0 = const()[name = tensor("op_27278_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_27278_end_mask_0 = const()[name = tensor("op_27278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27278_cast_fp16 = slice_by_index(begin = var_27278_begin_0, end = var_27278_end_0, end_mask = var_27278_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27278_cast_fp16")]; + tensor var_27282_begin_0 = const()[name = tensor("op_27282_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27282_end_0 = const()[name = tensor("op_27282_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_27282_end_mask_0 = const()[name = tensor("op_27282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27282_cast_fp16 = slice_by_index(begin = var_27282_begin_0, end = var_27282_end_0, end_mask = var_27282_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27282_cast_fp16")]; + tensor var_27286_begin_0 = const()[name = tensor("op_27286_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27286_end_0 = const()[name = tensor("op_27286_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_27286_end_mask_0 = const()[name = tensor("op_27286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27286_cast_fp16 = slice_by_index(begin = var_27286_begin_0, end = var_27286_end_0, end_mask = var_27286_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27286_cast_fp16")]; + tensor var_27290_begin_0 = const()[name = tensor("op_27290_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27290_end_0 = const()[name = tensor("op_27290_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_27290_end_mask_0 = const()[name = tensor("op_27290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27290_cast_fp16 = slice_by_index(begin = var_27290_begin_0, end = var_27290_end_0, end_mask = var_27290_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27290_cast_fp16")]; + tensor var_27294_begin_0 = const()[name = tensor("op_27294_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27294_end_0 = const()[name = tensor("op_27294_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_27294_end_mask_0 = const()[name = tensor("op_27294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27294_cast_fp16 = slice_by_index(begin = var_27294_begin_0, end = var_27294_end_0, end_mask = var_27294_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27294_cast_fp16")]; + tensor var_27298_begin_0 = const()[name = tensor("op_27298_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27298_end_0 = const()[name = tensor("op_27298_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_27298_end_mask_0 = const()[name = tensor("op_27298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27298_cast_fp16 = slice_by_index(begin = var_27298_begin_0, end = var_27298_end_0, end_mask = var_27298_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27298_cast_fp16")]; + tensor var_27302_begin_0 = const()[name = tensor("op_27302_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27302_end_0 = const()[name = tensor("op_27302_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_27302_end_mask_0 = const()[name = tensor("op_27302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27302_cast_fp16 = slice_by_index(begin = var_27302_begin_0, end = var_27302_end_0, end_mask = var_27302_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27302_cast_fp16")]; + tensor var_27306_begin_0 = const()[name = tensor("op_27306_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27306_end_0 = const()[name = tensor("op_27306_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_27306_end_mask_0 = const()[name = tensor("op_27306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27306_cast_fp16 = slice_by_index(begin = var_27306_begin_0, end = var_27306_end_0, end_mask = var_27306_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27306_cast_fp16")]; + tensor var_27310_begin_0 = const()[name = tensor("op_27310_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27310_end_0 = const()[name = tensor("op_27310_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_27310_end_mask_0 = const()[name = tensor("op_27310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27310_cast_fp16 = slice_by_index(begin = var_27310_begin_0, end = var_27310_end_0, end_mask = var_27310_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27310_cast_fp16")]; + tensor var_27314_begin_0 = const()[name = tensor("op_27314_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27314_end_0 = const()[name = tensor("op_27314_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_27314_end_mask_0 = const()[name = tensor("op_27314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27314_cast_fp16 = slice_by_index(begin = var_27314_begin_0, end = var_27314_end_0, end_mask = var_27314_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27314_cast_fp16")]; + tensor var_27318_begin_0 = const()[name = tensor("op_27318_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27318_end_0 = const()[name = tensor("op_27318_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_27318_end_mask_0 = const()[name = tensor("op_27318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27318_cast_fp16 = slice_by_index(begin = var_27318_begin_0, end = var_27318_end_0, end_mask = var_27318_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27318_cast_fp16")]; + tensor var_27322_begin_0 = const()[name = tensor("op_27322_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27322_end_0 = const()[name = tensor("op_27322_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_27322_end_mask_0 = const()[name = tensor("op_27322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27322_cast_fp16 = slice_by_index(begin = var_27322_begin_0, end = var_27322_end_0, end_mask = var_27322_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27322_cast_fp16")]; + tensor var_27326_begin_0 = const()[name = tensor("op_27326_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27326_end_0 = const()[name = tensor("op_27326_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_27326_end_mask_0 = const()[name = tensor("op_27326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27326_cast_fp16 = slice_by_index(begin = var_27326_begin_0, end = var_27326_end_0, end_mask = var_27326_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27326_cast_fp16")]; + tensor var_27330_begin_0 = const()[name = tensor("op_27330_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27330_end_0 = const()[name = tensor("op_27330_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_27330_end_mask_0 = const()[name = tensor("op_27330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27330_cast_fp16 = slice_by_index(begin = var_27330_begin_0, end = var_27330_end_0, end_mask = var_27330_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27330_cast_fp16")]; + tensor var_27334_begin_0 = const()[name = tensor("op_27334_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27334_end_0 = const()[name = tensor("op_27334_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_27334_end_mask_0 = const()[name = tensor("op_27334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27334_cast_fp16 = slice_by_index(begin = var_27334_begin_0, end = var_27334_end_0, end_mask = var_27334_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27334_cast_fp16")]; + tensor var_27338_begin_0 = const()[name = tensor("op_27338_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27338_end_0 = const()[name = tensor("op_27338_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_27338_end_mask_0 = const()[name = tensor("op_27338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27338_cast_fp16 = slice_by_index(begin = var_27338_begin_0, end = var_27338_end_0, end_mask = var_27338_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27338_cast_fp16")]; + tensor var_27342_begin_0 = const()[name = tensor("op_27342_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27342_end_0 = const()[name = tensor("op_27342_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_27342_end_mask_0 = const()[name = tensor("op_27342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27342_cast_fp16 = slice_by_index(begin = var_27342_begin_0, end = var_27342_end_0, end_mask = var_27342_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27342_cast_fp16")]; + tensor var_27346_begin_0 = const()[name = tensor("op_27346_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27346_end_0 = const()[name = tensor("op_27346_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_27346_end_mask_0 = const()[name = tensor("op_27346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27346_cast_fp16 = slice_by_index(begin = var_27346_begin_0, end = var_27346_end_0, end_mask = var_27346_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27346_cast_fp16")]; + tensor var_27350_equation_0 = const()[name = tensor("op_27350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27350_cast_fp16 = einsum(equation = var_27350_equation_0, values = (var_27192_cast_fp16, var_27109_cast_fp16))[name = tensor("op_27350_cast_fp16")]; + tensor var_27351_to_fp16 = const()[name = tensor("op_27351_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2281_cast_fp16 = mul(x = var_27350_cast_fp16, y = var_27351_to_fp16)[name = tensor("aw_2281_cast_fp16")]; + tensor var_27354_equation_0 = const()[name = tensor("op_27354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27354_cast_fp16 = einsum(equation = var_27354_equation_0, values = (var_27196_cast_fp16, var_27113_cast_fp16))[name = tensor("op_27354_cast_fp16")]; + tensor var_27355_to_fp16 = const()[name = tensor("op_27355_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2283_cast_fp16 = mul(x = var_27354_cast_fp16, y = var_27355_to_fp16)[name = tensor("aw_2283_cast_fp16")]; + tensor var_27358_equation_0 = const()[name = tensor("op_27358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27358_cast_fp16 = einsum(equation = var_27358_equation_0, values = (var_27200_cast_fp16, var_27117_cast_fp16))[name = tensor("op_27358_cast_fp16")]; + tensor var_27359_to_fp16 = const()[name = tensor("op_27359_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2285_cast_fp16 = mul(x = var_27358_cast_fp16, y = var_27359_to_fp16)[name = tensor("aw_2285_cast_fp16")]; + tensor var_27362_equation_0 = const()[name = tensor("op_27362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27362_cast_fp16 = einsum(equation = var_27362_equation_0, values = (var_27204_cast_fp16, var_27121_cast_fp16))[name = tensor("op_27362_cast_fp16")]; + tensor var_27363_to_fp16 = const()[name = tensor("op_27363_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2287_cast_fp16 = mul(x = var_27362_cast_fp16, y = var_27363_to_fp16)[name = tensor("aw_2287_cast_fp16")]; + tensor var_27366_equation_0 = const()[name = tensor("op_27366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27366_cast_fp16 = einsum(equation = var_27366_equation_0, values = (var_27208_cast_fp16, var_27125_cast_fp16))[name = tensor("op_27366_cast_fp16")]; + tensor var_27367_to_fp16 = const()[name = tensor("op_27367_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2289_cast_fp16 = mul(x = var_27366_cast_fp16, y = var_27367_to_fp16)[name = tensor("aw_2289_cast_fp16")]; + tensor var_27370_equation_0 = const()[name = tensor("op_27370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27370_cast_fp16 = einsum(equation = var_27370_equation_0, values = (var_27212_cast_fp16, var_27129_cast_fp16))[name = tensor("op_27370_cast_fp16")]; + tensor var_27371_to_fp16 = const()[name = tensor("op_27371_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2291_cast_fp16 = mul(x = var_27370_cast_fp16, y = var_27371_to_fp16)[name = tensor("aw_2291_cast_fp16")]; + tensor var_27374_equation_0 = const()[name = tensor("op_27374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27374_cast_fp16 = einsum(equation = var_27374_equation_0, values = (var_27216_cast_fp16, var_27133_cast_fp16))[name = tensor("op_27374_cast_fp16")]; + tensor var_27375_to_fp16 = const()[name = tensor("op_27375_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2293_cast_fp16 = mul(x = var_27374_cast_fp16, y = var_27375_to_fp16)[name = tensor("aw_2293_cast_fp16")]; + tensor var_27378_equation_0 = const()[name = tensor("op_27378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27378_cast_fp16 = einsum(equation = var_27378_equation_0, values = (var_27220_cast_fp16, var_27137_cast_fp16))[name = tensor("op_27378_cast_fp16")]; + tensor var_27379_to_fp16 = const()[name = tensor("op_27379_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2295_cast_fp16 = mul(x = var_27378_cast_fp16, y = var_27379_to_fp16)[name = tensor("aw_2295_cast_fp16")]; + tensor var_27382_equation_0 = const()[name = tensor("op_27382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27382_cast_fp16 = einsum(equation = var_27382_equation_0, values = (var_27224_cast_fp16, var_27141_cast_fp16))[name = tensor("op_27382_cast_fp16")]; + tensor var_27383_to_fp16 = const()[name = tensor("op_27383_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2297_cast_fp16 = mul(x = var_27382_cast_fp16, y = var_27383_to_fp16)[name = tensor("aw_2297_cast_fp16")]; + tensor var_27386_equation_0 = const()[name = tensor("op_27386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27386_cast_fp16 = einsum(equation = var_27386_equation_0, values = (var_27228_cast_fp16, var_27145_cast_fp16))[name = tensor("op_27386_cast_fp16")]; + tensor var_27387_to_fp16 = const()[name = tensor("op_27387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2299_cast_fp16 = mul(x = var_27386_cast_fp16, y = var_27387_to_fp16)[name = tensor("aw_2299_cast_fp16")]; + tensor var_27390_equation_0 = const()[name = tensor("op_27390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27390_cast_fp16 = einsum(equation = var_27390_equation_0, values = (var_27232_cast_fp16, var_27149_cast_fp16))[name = tensor("op_27390_cast_fp16")]; + tensor var_27391_to_fp16 = const()[name = tensor("op_27391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2301_cast_fp16 = mul(x = var_27390_cast_fp16, y = var_27391_to_fp16)[name = tensor("aw_2301_cast_fp16")]; + tensor var_27394_equation_0 = const()[name = tensor("op_27394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27394_cast_fp16 = einsum(equation = var_27394_equation_0, values = (var_27236_cast_fp16, var_27153_cast_fp16))[name = tensor("op_27394_cast_fp16")]; + tensor var_27395_to_fp16 = const()[name = tensor("op_27395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2303_cast_fp16 = mul(x = var_27394_cast_fp16, y = var_27395_to_fp16)[name = tensor("aw_2303_cast_fp16")]; + tensor var_27398_equation_0 = const()[name = tensor("op_27398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27398_cast_fp16 = einsum(equation = var_27398_equation_0, values = (var_27240_cast_fp16, var_27157_cast_fp16))[name = tensor("op_27398_cast_fp16")]; + tensor var_27399_to_fp16 = const()[name = tensor("op_27399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2305_cast_fp16 = mul(x = var_27398_cast_fp16, y = var_27399_to_fp16)[name = tensor("aw_2305_cast_fp16")]; + tensor var_27402_equation_0 = const()[name = tensor("op_27402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27402_cast_fp16 = einsum(equation = var_27402_equation_0, values = (var_27244_cast_fp16, var_27161_cast_fp16))[name = tensor("op_27402_cast_fp16")]; + tensor var_27403_to_fp16 = const()[name = tensor("op_27403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2307_cast_fp16 = mul(x = var_27402_cast_fp16, y = var_27403_to_fp16)[name = tensor("aw_2307_cast_fp16")]; + tensor var_27406_equation_0 = const()[name = tensor("op_27406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27406_cast_fp16 = einsum(equation = var_27406_equation_0, values = (var_27248_cast_fp16, var_27165_cast_fp16))[name = tensor("op_27406_cast_fp16")]; + tensor var_27407_to_fp16 = const()[name = tensor("op_27407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2309_cast_fp16 = mul(x = var_27406_cast_fp16, y = var_27407_to_fp16)[name = tensor("aw_2309_cast_fp16")]; + tensor var_27410_equation_0 = const()[name = tensor("op_27410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27410_cast_fp16 = einsum(equation = var_27410_equation_0, values = (var_27252_cast_fp16, var_27169_cast_fp16))[name = tensor("op_27410_cast_fp16")]; + tensor var_27411_to_fp16 = const()[name = tensor("op_27411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2311_cast_fp16 = mul(x = var_27410_cast_fp16, y = var_27411_to_fp16)[name = tensor("aw_2311_cast_fp16")]; + tensor var_27414_equation_0 = const()[name = tensor("op_27414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27414_cast_fp16 = einsum(equation = var_27414_equation_0, values = (var_27256_cast_fp16, var_27173_cast_fp16))[name = tensor("op_27414_cast_fp16")]; + tensor var_27415_to_fp16 = const()[name = tensor("op_27415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2313_cast_fp16 = mul(x = var_27414_cast_fp16, y = var_27415_to_fp16)[name = tensor("aw_2313_cast_fp16")]; + tensor var_27418_equation_0 = const()[name = tensor("op_27418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27418_cast_fp16 = einsum(equation = var_27418_equation_0, values = (var_27260_cast_fp16, var_27177_cast_fp16))[name = tensor("op_27418_cast_fp16")]; + tensor var_27419_to_fp16 = const()[name = tensor("op_27419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2315_cast_fp16 = mul(x = var_27418_cast_fp16, y = var_27419_to_fp16)[name = tensor("aw_2315_cast_fp16")]; + tensor var_27422_equation_0 = const()[name = tensor("op_27422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27422_cast_fp16 = einsum(equation = var_27422_equation_0, values = (var_27264_cast_fp16, var_27181_cast_fp16))[name = tensor("op_27422_cast_fp16")]; + tensor var_27423_to_fp16 = const()[name = tensor("op_27423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2317_cast_fp16 = mul(x = var_27422_cast_fp16, y = var_27423_to_fp16)[name = tensor("aw_2317_cast_fp16")]; + tensor var_27426_equation_0 = const()[name = tensor("op_27426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27426_cast_fp16 = einsum(equation = var_27426_equation_0, values = (var_27268_cast_fp16, var_27185_cast_fp16))[name = tensor("op_27426_cast_fp16")]; + tensor var_27427_to_fp16 = const()[name = tensor("op_27427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2319_cast_fp16 = mul(x = var_27426_cast_fp16, y = var_27427_to_fp16)[name = tensor("aw_2319_cast_fp16")]; + tensor var_27429_cast_fp16 = softmax(axis = var_21077, x = aw_2281_cast_fp16)[name = tensor("op_27429_cast_fp16")]; + tensor var_27430_cast_fp16 = softmax(axis = var_21077, x = aw_2283_cast_fp16)[name = tensor("op_27430_cast_fp16")]; + tensor var_27431_cast_fp16 = softmax(axis = var_21077, x = aw_2285_cast_fp16)[name = tensor("op_27431_cast_fp16")]; + tensor var_27432_cast_fp16 = softmax(axis = var_21077, x = aw_2287_cast_fp16)[name = tensor("op_27432_cast_fp16")]; + tensor var_27433_cast_fp16 = softmax(axis = var_21077, x = aw_2289_cast_fp16)[name = tensor("op_27433_cast_fp16")]; + tensor var_27434_cast_fp16 = softmax(axis = var_21077, x = aw_2291_cast_fp16)[name = tensor("op_27434_cast_fp16")]; + tensor var_27435_cast_fp16 = softmax(axis = var_21077, x = aw_2293_cast_fp16)[name = tensor("op_27435_cast_fp16")]; + tensor var_27436_cast_fp16 = softmax(axis = var_21077, x = aw_2295_cast_fp16)[name = tensor("op_27436_cast_fp16")]; + tensor var_27437_cast_fp16 = softmax(axis = var_21077, x = aw_2297_cast_fp16)[name = tensor("op_27437_cast_fp16")]; + tensor var_27438_cast_fp16 = softmax(axis = var_21077, x = aw_2299_cast_fp16)[name = tensor("op_27438_cast_fp16")]; + tensor var_27439_cast_fp16 = softmax(axis = var_21077, x = aw_2301_cast_fp16)[name = tensor("op_27439_cast_fp16")]; + tensor var_27440_cast_fp16 = softmax(axis = var_21077, x = aw_2303_cast_fp16)[name = tensor("op_27440_cast_fp16")]; + tensor var_27441_cast_fp16 = softmax(axis = var_21077, x = aw_2305_cast_fp16)[name = tensor("op_27441_cast_fp16")]; + tensor var_27442_cast_fp16 = softmax(axis = var_21077, x = aw_2307_cast_fp16)[name = tensor("op_27442_cast_fp16")]; + tensor var_27443_cast_fp16 = softmax(axis = var_21077, x = aw_2309_cast_fp16)[name = tensor("op_27443_cast_fp16")]; + tensor var_27444_cast_fp16 = softmax(axis = var_21077, x = aw_2311_cast_fp16)[name = tensor("op_27444_cast_fp16")]; + tensor var_27445_cast_fp16 = softmax(axis = var_21077, x = aw_2313_cast_fp16)[name = tensor("op_27445_cast_fp16")]; + tensor var_27446_cast_fp16 = softmax(axis = var_21077, x = aw_2315_cast_fp16)[name = tensor("op_27446_cast_fp16")]; + tensor var_27447_cast_fp16 = softmax(axis = var_21077, x = aw_2317_cast_fp16)[name = tensor("op_27447_cast_fp16")]; + tensor var_27448_cast_fp16 = softmax(axis = var_21077, x = aw_2319_cast_fp16)[name = tensor("op_27448_cast_fp16")]; + tensor var_27450_equation_0 = const()[name = tensor("op_27450_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27450_cast_fp16 = einsum(equation = var_27450_equation_0, values = (var_27270_cast_fp16, var_27429_cast_fp16))[name = tensor("op_27450_cast_fp16")]; + tensor var_27452_equation_0 = const()[name = tensor("op_27452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27452_cast_fp16 = einsum(equation = var_27452_equation_0, values = (var_27274_cast_fp16, var_27430_cast_fp16))[name = tensor("op_27452_cast_fp16")]; + tensor var_27454_equation_0 = const()[name = tensor("op_27454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27454_cast_fp16 = einsum(equation = var_27454_equation_0, values = (var_27278_cast_fp16, var_27431_cast_fp16))[name = tensor("op_27454_cast_fp16")]; + tensor var_27456_equation_0 = const()[name = tensor("op_27456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27456_cast_fp16 = einsum(equation = var_27456_equation_0, values = (var_27282_cast_fp16, var_27432_cast_fp16))[name = tensor("op_27456_cast_fp16")]; + tensor var_27458_equation_0 = const()[name = tensor("op_27458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27458_cast_fp16 = einsum(equation = var_27458_equation_0, values = (var_27286_cast_fp16, var_27433_cast_fp16))[name = tensor("op_27458_cast_fp16")]; + tensor var_27460_equation_0 = const()[name = tensor("op_27460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27460_cast_fp16 = einsum(equation = var_27460_equation_0, values = (var_27290_cast_fp16, var_27434_cast_fp16))[name = tensor("op_27460_cast_fp16")]; + tensor var_27462_equation_0 = const()[name = tensor("op_27462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27462_cast_fp16 = einsum(equation = var_27462_equation_0, values = (var_27294_cast_fp16, var_27435_cast_fp16))[name = tensor("op_27462_cast_fp16")]; + tensor var_27464_equation_0 = const()[name = tensor("op_27464_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27464_cast_fp16 = einsum(equation = var_27464_equation_0, values = (var_27298_cast_fp16, var_27436_cast_fp16))[name = tensor("op_27464_cast_fp16")]; + tensor var_27466_equation_0 = const()[name = tensor("op_27466_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27466_cast_fp16 = einsum(equation = var_27466_equation_0, values = (var_27302_cast_fp16, var_27437_cast_fp16))[name = tensor("op_27466_cast_fp16")]; + tensor var_27468_equation_0 = const()[name = tensor("op_27468_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27468_cast_fp16 = einsum(equation = var_27468_equation_0, values = (var_27306_cast_fp16, var_27438_cast_fp16))[name = tensor("op_27468_cast_fp16")]; + tensor var_27470_equation_0 = const()[name = tensor("op_27470_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27470_cast_fp16 = einsum(equation = var_27470_equation_0, values = (var_27310_cast_fp16, var_27439_cast_fp16))[name = tensor("op_27470_cast_fp16")]; + tensor var_27472_equation_0 = const()[name = tensor("op_27472_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27472_cast_fp16 = einsum(equation = var_27472_equation_0, values = (var_27314_cast_fp16, var_27440_cast_fp16))[name = tensor("op_27472_cast_fp16")]; + tensor var_27474_equation_0 = const()[name = tensor("op_27474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27474_cast_fp16 = einsum(equation = var_27474_equation_0, values = (var_27318_cast_fp16, var_27441_cast_fp16))[name = tensor("op_27474_cast_fp16")]; + tensor var_27476_equation_0 = const()[name = tensor("op_27476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27476_cast_fp16 = einsum(equation = var_27476_equation_0, values = (var_27322_cast_fp16, var_27442_cast_fp16))[name = tensor("op_27476_cast_fp16")]; + tensor var_27478_equation_0 = const()[name = tensor("op_27478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27478_cast_fp16 = einsum(equation = var_27478_equation_0, values = (var_27326_cast_fp16, var_27443_cast_fp16))[name = tensor("op_27478_cast_fp16")]; + tensor var_27480_equation_0 = const()[name = tensor("op_27480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27480_cast_fp16 = einsum(equation = var_27480_equation_0, values = (var_27330_cast_fp16, var_27444_cast_fp16))[name = tensor("op_27480_cast_fp16")]; + tensor var_27482_equation_0 = const()[name = tensor("op_27482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27482_cast_fp16 = einsum(equation = var_27482_equation_0, values = (var_27334_cast_fp16, var_27445_cast_fp16))[name = tensor("op_27482_cast_fp16")]; + tensor var_27484_equation_0 = const()[name = tensor("op_27484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27484_cast_fp16 = einsum(equation = var_27484_equation_0, values = (var_27338_cast_fp16, var_27446_cast_fp16))[name = tensor("op_27484_cast_fp16")]; + tensor var_27486_equation_0 = const()[name = tensor("op_27486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27486_cast_fp16 = einsum(equation = var_27486_equation_0, values = (var_27342_cast_fp16, var_27447_cast_fp16))[name = tensor("op_27486_cast_fp16")]; + tensor var_27488_equation_0 = const()[name = tensor("op_27488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27488_cast_fp16 = einsum(equation = var_27488_equation_0, values = (var_27346_cast_fp16, var_27448_cast_fp16))[name = tensor("op_27488_cast_fp16")]; + tensor input_377_interleave_0 = const()[name = tensor("input_377_interleave_0"), val = tensor(false)]; + tensor input_377_cast_fp16 = concat(axis = var_21077, interleave = input_377_interleave_0, values = (var_27450_cast_fp16, var_27452_cast_fp16, var_27454_cast_fp16, var_27456_cast_fp16, var_27458_cast_fp16, var_27460_cast_fp16, var_27462_cast_fp16, var_27464_cast_fp16, var_27466_cast_fp16, var_27468_cast_fp16, var_27470_cast_fp16, var_27472_cast_fp16, var_27474_cast_fp16, var_27476_cast_fp16, var_27478_cast_fp16, var_27480_cast_fp16, var_27482_cast_fp16, var_27484_cast_fp16, var_27486_cast_fp16, var_27488_cast_fp16))[name = tensor("input_377_cast_fp16")]; + tensor var_27498_pad_type_0 = const()[name = tensor("op_27498_pad_type_0"), val = tensor("valid")]; + tensor var_27498_strides_0 = const()[name = tensor("op_27498_strides_0"), val = tensor([1, 1])]; + tensor var_27498_pad_0 = const()[name = tensor("op_27498_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27498_dilations_0 = const()[name = tensor("op_27498_dilations_0"), val = tensor([1, 1])]; + tensor var_27498_groups_0 = const()[name = tensor("op_27498_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814730176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815959040))), name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815959232)))]; + tensor var_27498_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_27498_dilations_0, groups = var_27498_groups_0, pad = var_27498_pad_0, pad_type = var_27498_pad_type_0, strides = var_27498_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized, x = input_377_cast_fp16)[name = tensor("op_27498_cast_fp16")]; + tensor inputs_185_cast_fp16 = add(x = var_27498_cast_fp16, y = inputs_183_cast_fp16)[name = tensor("inputs_185_cast_fp16")]; + tensor input_379_axes_0 = const()[name = tensor("input_379_axes_0"), val = tensor([1])]; + tensor input_379_gamma_0_to_fp16 = const()[name = tensor("input_379_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815961856)))]; + tensor input_379_beta_0_to_fp16 = const()[name = tensor("input_379_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815964480)))]; + tensor var_27508_to_fp16 = const()[name = tensor("op_27508_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_379_cast_fp16 = layer_norm(axes = input_379_axes_0, beta = input_379_beta_0_to_fp16, epsilon = var_27508_to_fp16, gamma = input_379_gamma_0_to_fp16, x = inputs_185_cast_fp16)[name = tensor("input_379_cast_fp16")]; + tensor var_27528_pad_type_0 = const()[name = tensor("op_27528_pad_type_0"), val = tensor("valid")]; + tensor var_27528_strides_0 = const()[name = tensor("op_27528_strides_0"), val = tensor([1, 1])]; + tensor var_27528_pad_0 = const()[name = tensor("op_27528_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27528_dilations_0 = const()[name = tensor("op_27528_dilations_0"), val = tensor([1, 1])]; + tensor var_27528_groups_0 = const()[name = tensor("op_27528_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(815967104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(825797568))), name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(825797760)))]; + tensor var_27528_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_27528_dilations_0, groups = var_27528_groups_0, pad = var_27528_pad_0, pad_type = var_27528_pad_type_0, strides = var_27528_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized, x = input_379_cast_fp16)[name = tensor("op_27528_cast_fp16")]; + tensor var_27529_split_sizes_0 = const()[name = tensor("op_27529_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_27529_axis_0 = const()[name = tensor("op_27529_axis_0"), val = tensor(1)]; + tensor var_27529_cast_fp16_0, tensor var_27529_cast_fp16_1 = split(axis = var_27529_axis_0, split_sizes = var_27529_split_sizes_0, x = var_27528_cast_fp16)[name = tensor("op_27529_cast_fp16")]; + tensor var_27531_mode_0 = const()[name = tensor("op_27531_mode_0"), val = tensor("EXACT")]; + tensor var_27531_cast_fp16 = gelu(mode = var_27531_mode_0, x = var_27529_cast_fp16_1)[name = tensor("op_27531_cast_fp16")]; + tensor input_381_cast_fp16 = mul(x = var_27529_cast_fp16_0, y = var_27531_cast_fp16)[name = tensor("input_381_cast_fp16")]; + tensor var_27539_pad_type_0 = const()[name = tensor("op_27539_pad_type_0"), val = tensor("valid")]; + tensor var_27539_strides_0 = const()[name = tensor("op_27539_strides_0"), val = tensor([1, 1])]; + tensor var_27539_pad_0 = const()[name = tensor("op_27539_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27539_dilations_0 = const()[name = tensor("op_27539_dilations_0"), val = tensor([1, 1])]; + tensor var_27539_groups_0 = const()[name = tensor("op_27539_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(825818304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830733568))), name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830733760)))]; + tensor var_27539_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_27539_dilations_0, groups = var_27539_groups_0, pad = var_27539_pad_0, pad_type = var_27539_pad_type_0, strides = var_27539_strides_0, weight = mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized, x = input_381_cast_fp16)[name = tensor("op_27539_cast_fp16")]; + tensor inputs_187_cast_fp16 = add(x = var_27539_cast_fp16, y = inputs_185_cast_fp16)[name = tensor("inputs_187_cast_fp16")]; + tensor hidden_states_251_axes_0 = const()[name = tensor("hidden_states_251_axes_0"), val = tensor([1])]; + tensor hidden_states_251_gamma_0_to_fp16 = const()[name = tensor("hidden_states_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830736384)))]; + tensor hidden_states_251_beta_0_to_fp16 = const()[name = tensor("hidden_states_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830739008)))]; + tensor var_27555_to_fp16 = const()[name = tensor("op_27555_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_251_cast_fp16 = layer_norm(axes = hidden_states_251_axes_0, beta = hidden_states_251_beta_0_to_fp16, epsilon = var_27555_to_fp16, gamma = hidden_states_251_gamma_0_to_fp16, x = inputs_187_cast_fp16)[name = tensor("hidden_states_251_cast_fp16")]; + tensor q_125_pad_type_0 = const()[name = tensor("q_125_pad_type_0"), val = tensor("valid")]; + tensor q_125_strides_0 = const()[name = tensor("q_125_strides_0"), val = tensor([1, 1])]; + tensor q_125_pad_0 = const()[name = tensor("q_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_125_dilations_0 = const()[name = tensor("q_125_dilations_0"), val = tensor([1, 1])]; + tensor q_125_groups_0 = const()[name = tensor("q_125_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830741632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831970496))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_125_cast_fp16 = conv(dilations = q_125_dilations_0, groups = q_125_groups_0, pad = q_125_pad_0, pad_type = q_125_pad_type_0, strides = q_125_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_251_cast_fp16)[name = tensor("q_125_cast_fp16")]; + tensor k_249_pad_type_0 = const()[name = tensor("k_249_pad_type_0"), val = tensor("valid")]; + tensor k_249_strides_0 = const()[name = tensor("k_249_strides_0"), val = tensor([1, 1])]; + tensor k_249_pad_0 = const()[name = tensor("k_249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_249_dilations_0 = const()[name = tensor("k_249_dilations_0"), val = tensor([1, 1])]; + tensor k_249_groups_0 = const()[name = tensor("k_249_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(831970688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833199552))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_249_cast_fp16 = conv(dilations = k_249_dilations_0, groups = k_249_groups_0, pad = k_249_pad_0, pad_type = k_249_pad_type_0, strides = k_249_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_251_cast_fp16)[name = tensor("k_249_cast_fp16")]; + tensor v_125_pad_type_0 = const()[name = tensor("v_125_pad_type_0"), val = tensor("valid")]; + tensor v_125_strides_0 = const()[name = tensor("v_125_strides_0"), val = tensor([1, 1])]; + tensor v_125_pad_0 = const()[name = tensor("v_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_125_dilations_0 = const()[name = tensor("v_125_dilations_0"), val = tensor([1, 1])]; + tensor v_125_groups_0 = const()[name = tensor("v_125_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833199744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834428608))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_125_cast_fp16 = conv(dilations = v_125_dilations_0, groups = v_125_groups_0, pad = v_125_pad_0, pad_type = v_125_pad_type_0, strides = v_125_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_251_cast_fp16)[name = tensor("v_125_cast_fp16")]; + tensor var_27588_begin_0 = const()[name = tensor("op_27588_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27588_end_0 = const()[name = tensor("op_27588_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_27588_end_mask_0 = const()[name = tensor("op_27588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27588_cast_fp16 = slice_by_index(begin = var_27588_begin_0, end = var_27588_end_0, end_mask = var_27588_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27588_cast_fp16")]; + tensor var_27592_begin_0 = const()[name = tensor("op_27592_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27592_end_0 = const()[name = tensor("op_27592_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_27592_end_mask_0 = const()[name = tensor("op_27592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27592_cast_fp16 = slice_by_index(begin = var_27592_begin_0, end = var_27592_end_0, end_mask = var_27592_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27592_cast_fp16")]; + tensor var_27596_begin_0 = const()[name = tensor("op_27596_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27596_end_0 = const()[name = tensor("op_27596_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_27596_end_mask_0 = const()[name = tensor("op_27596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27596_cast_fp16 = slice_by_index(begin = var_27596_begin_0, end = var_27596_end_0, end_mask = var_27596_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27596_cast_fp16")]; + tensor var_27600_begin_0 = const()[name = tensor("op_27600_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27600_end_0 = const()[name = tensor("op_27600_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_27600_end_mask_0 = const()[name = tensor("op_27600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27600_cast_fp16 = slice_by_index(begin = var_27600_begin_0, end = var_27600_end_0, end_mask = var_27600_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27600_cast_fp16")]; + tensor var_27604_begin_0 = const()[name = tensor("op_27604_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27604_end_0 = const()[name = tensor("op_27604_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_27604_end_mask_0 = const()[name = tensor("op_27604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27604_cast_fp16 = slice_by_index(begin = var_27604_begin_0, end = var_27604_end_0, end_mask = var_27604_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27604_cast_fp16")]; + tensor var_27608_begin_0 = const()[name = tensor("op_27608_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27608_end_0 = const()[name = tensor("op_27608_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_27608_end_mask_0 = const()[name = tensor("op_27608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27608_cast_fp16 = slice_by_index(begin = var_27608_begin_0, end = var_27608_end_0, end_mask = var_27608_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27608_cast_fp16")]; + tensor var_27612_begin_0 = const()[name = tensor("op_27612_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27612_end_0 = const()[name = tensor("op_27612_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_27612_end_mask_0 = const()[name = tensor("op_27612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27612_cast_fp16 = slice_by_index(begin = var_27612_begin_0, end = var_27612_end_0, end_mask = var_27612_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27612_cast_fp16")]; + tensor var_27616_begin_0 = const()[name = tensor("op_27616_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27616_end_0 = const()[name = tensor("op_27616_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_27616_end_mask_0 = const()[name = tensor("op_27616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27616_cast_fp16 = slice_by_index(begin = var_27616_begin_0, end = var_27616_end_0, end_mask = var_27616_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27616_cast_fp16")]; + tensor var_27620_begin_0 = const()[name = tensor("op_27620_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27620_end_0 = const()[name = tensor("op_27620_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_27620_end_mask_0 = const()[name = tensor("op_27620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27620_cast_fp16 = slice_by_index(begin = var_27620_begin_0, end = var_27620_end_0, end_mask = var_27620_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27620_cast_fp16")]; + tensor var_27624_begin_0 = const()[name = tensor("op_27624_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27624_end_0 = const()[name = tensor("op_27624_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_27624_end_mask_0 = const()[name = tensor("op_27624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27624_cast_fp16 = slice_by_index(begin = var_27624_begin_0, end = var_27624_end_0, end_mask = var_27624_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27624_cast_fp16")]; + tensor var_27628_begin_0 = const()[name = tensor("op_27628_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27628_end_0 = const()[name = tensor("op_27628_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_27628_end_mask_0 = const()[name = tensor("op_27628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27628_cast_fp16 = slice_by_index(begin = var_27628_begin_0, end = var_27628_end_0, end_mask = var_27628_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27628_cast_fp16")]; + tensor var_27632_begin_0 = const()[name = tensor("op_27632_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27632_end_0 = const()[name = tensor("op_27632_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_27632_end_mask_0 = const()[name = tensor("op_27632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27632_cast_fp16 = slice_by_index(begin = var_27632_begin_0, end = var_27632_end_0, end_mask = var_27632_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27632_cast_fp16")]; + tensor var_27636_begin_0 = const()[name = tensor("op_27636_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27636_end_0 = const()[name = tensor("op_27636_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_27636_end_mask_0 = const()[name = tensor("op_27636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27636_cast_fp16 = slice_by_index(begin = var_27636_begin_0, end = var_27636_end_0, end_mask = var_27636_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27636_cast_fp16")]; + tensor var_27640_begin_0 = const()[name = tensor("op_27640_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27640_end_0 = const()[name = tensor("op_27640_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_27640_end_mask_0 = const()[name = tensor("op_27640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27640_cast_fp16 = slice_by_index(begin = var_27640_begin_0, end = var_27640_end_0, end_mask = var_27640_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27640_cast_fp16")]; + tensor var_27644_begin_0 = const()[name = tensor("op_27644_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27644_end_0 = const()[name = tensor("op_27644_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_27644_end_mask_0 = const()[name = tensor("op_27644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27644_cast_fp16 = slice_by_index(begin = var_27644_begin_0, end = var_27644_end_0, end_mask = var_27644_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27644_cast_fp16")]; + tensor var_27648_begin_0 = const()[name = tensor("op_27648_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27648_end_0 = const()[name = tensor("op_27648_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27648_end_mask_0 = const()[name = tensor("op_27648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27648_cast_fp16 = slice_by_index(begin = var_27648_begin_0, end = var_27648_end_0, end_mask = var_27648_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27648_cast_fp16")]; + tensor var_27652_begin_0 = const()[name = tensor("op_27652_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27652_end_0 = const()[name = tensor("op_27652_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_27652_end_mask_0 = const()[name = tensor("op_27652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27652_cast_fp16 = slice_by_index(begin = var_27652_begin_0, end = var_27652_end_0, end_mask = var_27652_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27652_cast_fp16")]; + tensor var_27656_begin_0 = const()[name = tensor("op_27656_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27656_end_0 = const()[name = tensor("op_27656_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_27656_end_mask_0 = const()[name = tensor("op_27656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27656_cast_fp16 = slice_by_index(begin = var_27656_begin_0, end = var_27656_end_0, end_mask = var_27656_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27656_cast_fp16")]; + tensor var_27660_begin_0 = const()[name = tensor("op_27660_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27660_end_0 = const()[name = tensor("op_27660_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_27660_end_mask_0 = const()[name = tensor("op_27660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27660_cast_fp16 = slice_by_index(begin = var_27660_begin_0, end = var_27660_end_0, end_mask = var_27660_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27660_cast_fp16")]; + tensor var_27664_begin_0 = const()[name = tensor("op_27664_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27664_end_0 = const()[name = tensor("op_27664_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_27664_end_mask_0 = const()[name = tensor("op_27664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27664_cast_fp16 = slice_by_index(begin = var_27664_begin_0, end = var_27664_end_0, end_mask = var_27664_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27664_cast_fp16")]; + tensor k_251_perm_0 = const()[name = tensor("k_251_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_27671_begin_0 = const()[name = tensor("op_27671_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27671_end_0 = const()[name = tensor("op_27671_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_27671_end_mask_0 = const()[name = tensor("op_27671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_251_cast_fp16 = transpose(perm = k_251_perm_0, x = k_249_cast_fp16)[name = tensor("transpose_5")]; + tensor var_27671_cast_fp16 = slice_by_index(begin = var_27671_begin_0, end = var_27671_end_0, end_mask = var_27671_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27671_cast_fp16")]; + tensor var_27675_begin_0 = const()[name = tensor("op_27675_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_27675_end_0 = const()[name = tensor("op_27675_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_27675_end_mask_0 = const()[name = tensor("op_27675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27675_cast_fp16 = slice_by_index(begin = var_27675_begin_0, end = var_27675_end_0, end_mask = var_27675_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27675_cast_fp16")]; + tensor var_27679_begin_0 = const()[name = tensor("op_27679_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_27679_end_0 = const()[name = tensor("op_27679_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_27679_end_mask_0 = const()[name = tensor("op_27679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27679_cast_fp16 = slice_by_index(begin = var_27679_begin_0, end = var_27679_end_0, end_mask = var_27679_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27679_cast_fp16")]; + tensor var_27683_begin_0 = const()[name = tensor("op_27683_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_27683_end_0 = const()[name = tensor("op_27683_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_27683_end_mask_0 = const()[name = tensor("op_27683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27683_cast_fp16 = slice_by_index(begin = var_27683_begin_0, end = var_27683_end_0, end_mask = var_27683_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27683_cast_fp16")]; + tensor var_27687_begin_0 = const()[name = tensor("op_27687_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_27687_end_0 = const()[name = tensor("op_27687_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_27687_end_mask_0 = const()[name = tensor("op_27687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27687_cast_fp16 = slice_by_index(begin = var_27687_begin_0, end = var_27687_end_0, end_mask = var_27687_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27687_cast_fp16")]; + tensor var_27691_begin_0 = const()[name = tensor("op_27691_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_27691_end_0 = const()[name = tensor("op_27691_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_27691_end_mask_0 = const()[name = tensor("op_27691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27691_cast_fp16 = slice_by_index(begin = var_27691_begin_0, end = var_27691_end_0, end_mask = var_27691_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27691_cast_fp16")]; + tensor var_27695_begin_0 = const()[name = tensor("op_27695_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_27695_end_0 = const()[name = tensor("op_27695_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_27695_end_mask_0 = const()[name = tensor("op_27695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27695_cast_fp16 = slice_by_index(begin = var_27695_begin_0, end = var_27695_end_0, end_mask = var_27695_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27695_cast_fp16")]; + tensor var_27699_begin_0 = const()[name = tensor("op_27699_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_27699_end_0 = const()[name = tensor("op_27699_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_27699_end_mask_0 = const()[name = tensor("op_27699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27699_cast_fp16 = slice_by_index(begin = var_27699_begin_0, end = var_27699_end_0, end_mask = var_27699_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27699_cast_fp16")]; + tensor var_27703_begin_0 = const()[name = tensor("op_27703_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_27703_end_0 = const()[name = tensor("op_27703_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_27703_end_mask_0 = const()[name = tensor("op_27703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27703_cast_fp16 = slice_by_index(begin = var_27703_begin_0, end = var_27703_end_0, end_mask = var_27703_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27703_cast_fp16")]; + tensor var_27707_begin_0 = const()[name = tensor("op_27707_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_27707_end_0 = const()[name = tensor("op_27707_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_27707_end_mask_0 = const()[name = tensor("op_27707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27707_cast_fp16 = slice_by_index(begin = var_27707_begin_0, end = var_27707_end_0, end_mask = var_27707_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27707_cast_fp16")]; + tensor var_27711_begin_0 = const()[name = tensor("op_27711_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_27711_end_0 = const()[name = tensor("op_27711_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_27711_end_mask_0 = const()[name = tensor("op_27711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27711_cast_fp16 = slice_by_index(begin = var_27711_begin_0, end = var_27711_end_0, end_mask = var_27711_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27711_cast_fp16")]; + tensor var_27715_begin_0 = const()[name = tensor("op_27715_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_27715_end_0 = const()[name = tensor("op_27715_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_27715_end_mask_0 = const()[name = tensor("op_27715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27715_cast_fp16 = slice_by_index(begin = var_27715_begin_0, end = var_27715_end_0, end_mask = var_27715_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27715_cast_fp16")]; + tensor var_27719_begin_0 = const()[name = tensor("op_27719_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_27719_end_0 = const()[name = tensor("op_27719_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_27719_end_mask_0 = const()[name = tensor("op_27719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27719_cast_fp16 = slice_by_index(begin = var_27719_begin_0, end = var_27719_end_0, end_mask = var_27719_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27719_cast_fp16")]; + tensor var_27723_begin_0 = const()[name = tensor("op_27723_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_27723_end_0 = const()[name = tensor("op_27723_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_27723_end_mask_0 = const()[name = tensor("op_27723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27723_cast_fp16 = slice_by_index(begin = var_27723_begin_0, end = var_27723_end_0, end_mask = var_27723_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27723_cast_fp16")]; + tensor var_27727_begin_0 = const()[name = tensor("op_27727_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_27727_end_0 = const()[name = tensor("op_27727_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_27727_end_mask_0 = const()[name = tensor("op_27727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27727_cast_fp16 = slice_by_index(begin = var_27727_begin_0, end = var_27727_end_0, end_mask = var_27727_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27727_cast_fp16")]; + tensor var_27731_begin_0 = const()[name = tensor("op_27731_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_27731_end_0 = const()[name = tensor("op_27731_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27731_end_mask_0 = const()[name = tensor("op_27731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27731_cast_fp16 = slice_by_index(begin = var_27731_begin_0, end = var_27731_end_0, end_mask = var_27731_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27731_cast_fp16")]; + tensor var_27735_begin_0 = const()[name = tensor("op_27735_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_27735_end_0 = const()[name = tensor("op_27735_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_27735_end_mask_0 = const()[name = tensor("op_27735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27735_cast_fp16 = slice_by_index(begin = var_27735_begin_0, end = var_27735_end_0, end_mask = var_27735_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27735_cast_fp16")]; + tensor var_27739_begin_0 = const()[name = tensor("op_27739_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_27739_end_0 = const()[name = tensor("op_27739_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_27739_end_mask_0 = const()[name = tensor("op_27739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27739_cast_fp16 = slice_by_index(begin = var_27739_begin_0, end = var_27739_end_0, end_mask = var_27739_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27739_cast_fp16")]; + tensor var_27743_begin_0 = const()[name = tensor("op_27743_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_27743_end_0 = const()[name = tensor("op_27743_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_27743_end_mask_0 = const()[name = tensor("op_27743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27743_cast_fp16 = slice_by_index(begin = var_27743_begin_0, end = var_27743_end_0, end_mask = var_27743_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27743_cast_fp16")]; + tensor var_27747_begin_0 = const()[name = tensor("op_27747_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_27747_end_0 = const()[name = tensor("op_27747_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_27747_end_mask_0 = const()[name = tensor("op_27747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27747_cast_fp16 = slice_by_index(begin = var_27747_begin_0, end = var_27747_end_0, end_mask = var_27747_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27747_cast_fp16")]; + tensor var_27749_begin_0 = const()[name = tensor("op_27749_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27749_end_0 = const()[name = tensor("op_27749_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_27749_end_mask_0 = const()[name = tensor("op_27749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27749_cast_fp16 = slice_by_index(begin = var_27749_begin_0, end = var_27749_end_0, end_mask = var_27749_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27749_cast_fp16")]; + tensor var_27753_begin_0 = const()[name = tensor("op_27753_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27753_end_0 = const()[name = tensor("op_27753_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_27753_end_mask_0 = const()[name = tensor("op_27753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27753_cast_fp16 = slice_by_index(begin = var_27753_begin_0, end = var_27753_end_0, end_mask = var_27753_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27753_cast_fp16")]; + tensor var_27757_begin_0 = const()[name = tensor("op_27757_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27757_end_0 = const()[name = tensor("op_27757_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_27757_end_mask_0 = const()[name = tensor("op_27757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27757_cast_fp16 = slice_by_index(begin = var_27757_begin_0, end = var_27757_end_0, end_mask = var_27757_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27757_cast_fp16")]; + tensor var_27761_begin_0 = const()[name = tensor("op_27761_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27761_end_0 = const()[name = tensor("op_27761_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_27761_end_mask_0 = const()[name = tensor("op_27761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27761_cast_fp16 = slice_by_index(begin = var_27761_begin_0, end = var_27761_end_0, end_mask = var_27761_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27761_cast_fp16")]; + tensor var_27765_begin_0 = const()[name = tensor("op_27765_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27765_end_0 = const()[name = tensor("op_27765_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_27765_end_mask_0 = const()[name = tensor("op_27765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27765_cast_fp16 = slice_by_index(begin = var_27765_begin_0, end = var_27765_end_0, end_mask = var_27765_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27765_cast_fp16")]; + tensor var_27769_begin_0 = const()[name = tensor("op_27769_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27769_end_0 = const()[name = tensor("op_27769_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_27769_end_mask_0 = const()[name = tensor("op_27769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27769_cast_fp16 = slice_by_index(begin = var_27769_begin_0, end = var_27769_end_0, end_mask = var_27769_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27769_cast_fp16")]; + tensor var_27773_begin_0 = const()[name = tensor("op_27773_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27773_end_0 = const()[name = tensor("op_27773_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_27773_end_mask_0 = const()[name = tensor("op_27773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27773_cast_fp16 = slice_by_index(begin = var_27773_begin_0, end = var_27773_end_0, end_mask = var_27773_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27773_cast_fp16")]; + tensor var_27777_begin_0 = const()[name = tensor("op_27777_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27777_end_0 = const()[name = tensor("op_27777_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_27777_end_mask_0 = const()[name = tensor("op_27777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27777_cast_fp16 = slice_by_index(begin = var_27777_begin_0, end = var_27777_end_0, end_mask = var_27777_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27777_cast_fp16")]; + tensor var_27781_begin_0 = const()[name = tensor("op_27781_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27781_end_0 = const()[name = tensor("op_27781_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_27781_end_mask_0 = const()[name = tensor("op_27781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27781_cast_fp16 = slice_by_index(begin = var_27781_begin_0, end = var_27781_end_0, end_mask = var_27781_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27781_cast_fp16")]; + tensor var_27785_begin_0 = const()[name = tensor("op_27785_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27785_end_0 = const()[name = tensor("op_27785_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_27785_end_mask_0 = const()[name = tensor("op_27785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27785_cast_fp16 = slice_by_index(begin = var_27785_begin_0, end = var_27785_end_0, end_mask = var_27785_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27785_cast_fp16")]; + tensor var_27789_begin_0 = const()[name = tensor("op_27789_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27789_end_0 = const()[name = tensor("op_27789_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_27789_end_mask_0 = const()[name = tensor("op_27789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27789_cast_fp16 = slice_by_index(begin = var_27789_begin_0, end = var_27789_end_0, end_mask = var_27789_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27789_cast_fp16")]; + tensor var_27793_begin_0 = const()[name = tensor("op_27793_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27793_end_0 = const()[name = tensor("op_27793_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_27793_end_mask_0 = const()[name = tensor("op_27793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27793_cast_fp16 = slice_by_index(begin = var_27793_begin_0, end = var_27793_end_0, end_mask = var_27793_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27793_cast_fp16")]; + tensor var_27797_begin_0 = const()[name = tensor("op_27797_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27797_end_0 = const()[name = tensor("op_27797_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_27797_end_mask_0 = const()[name = tensor("op_27797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27797_cast_fp16 = slice_by_index(begin = var_27797_begin_0, end = var_27797_end_0, end_mask = var_27797_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27797_cast_fp16")]; + tensor var_27801_begin_0 = const()[name = tensor("op_27801_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27801_end_0 = const()[name = tensor("op_27801_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_27801_end_mask_0 = const()[name = tensor("op_27801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27801_cast_fp16 = slice_by_index(begin = var_27801_begin_0, end = var_27801_end_0, end_mask = var_27801_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27801_cast_fp16")]; + tensor var_27805_begin_0 = const()[name = tensor("op_27805_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27805_end_0 = const()[name = tensor("op_27805_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_27805_end_mask_0 = const()[name = tensor("op_27805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27805_cast_fp16 = slice_by_index(begin = var_27805_begin_0, end = var_27805_end_0, end_mask = var_27805_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27805_cast_fp16")]; + tensor var_27809_begin_0 = const()[name = tensor("op_27809_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27809_end_0 = const()[name = tensor("op_27809_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27809_end_mask_0 = const()[name = tensor("op_27809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27809_cast_fp16 = slice_by_index(begin = var_27809_begin_0, end = var_27809_end_0, end_mask = var_27809_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27809_cast_fp16")]; + tensor var_27813_begin_0 = const()[name = tensor("op_27813_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27813_end_0 = const()[name = tensor("op_27813_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_27813_end_mask_0 = const()[name = tensor("op_27813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27813_cast_fp16 = slice_by_index(begin = var_27813_begin_0, end = var_27813_end_0, end_mask = var_27813_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27813_cast_fp16")]; + tensor var_27817_begin_0 = const()[name = tensor("op_27817_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27817_end_0 = const()[name = tensor("op_27817_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_27817_end_mask_0 = const()[name = tensor("op_27817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27817_cast_fp16 = slice_by_index(begin = var_27817_begin_0, end = var_27817_end_0, end_mask = var_27817_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27817_cast_fp16")]; + tensor var_27821_begin_0 = const()[name = tensor("op_27821_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27821_end_0 = const()[name = tensor("op_27821_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_27821_end_mask_0 = const()[name = tensor("op_27821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27821_cast_fp16 = slice_by_index(begin = var_27821_begin_0, end = var_27821_end_0, end_mask = var_27821_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27821_cast_fp16")]; + tensor var_27825_begin_0 = const()[name = tensor("op_27825_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27825_end_0 = const()[name = tensor("op_27825_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_27825_end_mask_0 = const()[name = tensor("op_27825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27825_cast_fp16 = slice_by_index(begin = var_27825_begin_0, end = var_27825_end_0, end_mask = var_27825_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27825_cast_fp16")]; + tensor var_27829_equation_0 = const()[name = tensor("op_27829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27829_cast_fp16 = einsum(equation = var_27829_equation_0, values = (var_27671_cast_fp16, var_27588_cast_fp16))[name = tensor("op_27829_cast_fp16")]; + tensor var_27830_to_fp16 = const()[name = tensor("op_27830_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2321_cast_fp16 = mul(x = var_27829_cast_fp16, y = var_27830_to_fp16)[name = tensor("aw_2321_cast_fp16")]; + tensor var_27833_equation_0 = const()[name = tensor("op_27833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27833_cast_fp16 = einsum(equation = var_27833_equation_0, values = (var_27675_cast_fp16, var_27592_cast_fp16))[name = tensor("op_27833_cast_fp16")]; + tensor var_27834_to_fp16 = const()[name = tensor("op_27834_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2323_cast_fp16 = mul(x = var_27833_cast_fp16, y = var_27834_to_fp16)[name = tensor("aw_2323_cast_fp16")]; + tensor var_27837_equation_0 = const()[name = tensor("op_27837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27837_cast_fp16 = einsum(equation = var_27837_equation_0, values = (var_27679_cast_fp16, var_27596_cast_fp16))[name = tensor("op_27837_cast_fp16")]; + tensor var_27838_to_fp16 = const()[name = tensor("op_27838_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2325_cast_fp16 = mul(x = var_27837_cast_fp16, y = var_27838_to_fp16)[name = tensor("aw_2325_cast_fp16")]; + tensor var_27841_equation_0 = const()[name = tensor("op_27841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27841_cast_fp16 = einsum(equation = var_27841_equation_0, values = (var_27683_cast_fp16, var_27600_cast_fp16))[name = tensor("op_27841_cast_fp16")]; + tensor var_27842_to_fp16 = const()[name = tensor("op_27842_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2327_cast_fp16 = mul(x = var_27841_cast_fp16, y = var_27842_to_fp16)[name = tensor("aw_2327_cast_fp16")]; + tensor var_27845_equation_0 = const()[name = tensor("op_27845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27845_cast_fp16 = einsum(equation = var_27845_equation_0, values = (var_27687_cast_fp16, var_27604_cast_fp16))[name = tensor("op_27845_cast_fp16")]; + tensor var_27846_to_fp16 = const()[name = tensor("op_27846_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2329_cast_fp16 = mul(x = var_27845_cast_fp16, y = var_27846_to_fp16)[name = tensor("aw_2329_cast_fp16")]; + tensor var_27849_equation_0 = const()[name = tensor("op_27849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27849_cast_fp16 = einsum(equation = var_27849_equation_0, values = (var_27691_cast_fp16, var_27608_cast_fp16))[name = tensor("op_27849_cast_fp16")]; + tensor var_27850_to_fp16 = const()[name = tensor("op_27850_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2331_cast_fp16 = mul(x = var_27849_cast_fp16, y = var_27850_to_fp16)[name = tensor("aw_2331_cast_fp16")]; + tensor var_27853_equation_0 = const()[name = tensor("op_27853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27853_cast_fp16 = einsum(equation = var_27853_equation_0, values = (var_27695_cast_fp16, var_27612_cast_fp16))[name = tensor("op_27853_cast_fp16")]; + tensor var_27854_to_fp16 = const()[name = tensor("op_27854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2333_cast_fp16 = mul(x = var_27853_cast_fp16, y = var_27854_to_fp16)[name = tensor("aw_2333_cast_fp16")]; + tensor var_27857_equation_0 = const()[name = tensor("op_27857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27857_cast_fp16 = einsum(equation = var_27857_equation_0, values = (var_27699_cast_fp16, var_27616_cast_fp16))[name = tensor("op_27857_cast_fp16")]; + tensor var_27858_to_fp16 = const()[name = tensor("op_27858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2335_cast_fp16 = mul(x = var_27857_cast_fp16, y = var_27858_to_fp16)[name = tensor("aw_2335_cast_fp16")]; + tensor var_27861_equation_0 = const()[name = tensor("op_27861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27861_cast_fp16 = einsum(equation = var_27861_equation_0, values = (var_27703_cast_fp16, var_27620_cast_fp16))[name = tensor("op_27861_cast_fp16")]; + tensor var_27862_to_fp16 = const()[name = tensor("op_27862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2337_cast_fp16 = mul(x = var_27861_cast_fp16, y = var_27862_to_fp16)[name = tensor("aw_2337_cast_fp16")]; + tensor var_27865_equation_0 = const()[name = tensor("op_27865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27865_cast_fp16 = einsum(equation = var_27865_equation_0, values = (var_27707_cast_fp16, var_27624_cast_fp16))[name = tensor("op_27865_cast_fp16")]; + tensor var_27866_to_fp16 = const()[name = tensor("op_27866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2339_cast_fp16 = mul(x = var_27865_cast_fp16, y = var_27866_to_fp16)[name = tensor("aw_2339_cast_fp16")]; + tensor var_27869_equation_0 = const()[name = tensor("op_27869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27869_cast_fp16 = einsum(equation = var_27869_equation_0, values = (var_27711_cast_fp16, var_27628_cast_fp16))[name = tensor("op_27869_cast_fp16")]; + tensor var_27870_to_fp16 = const()[name = tensor("op_27870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2341_cast_fp16 = mul(x = var_27869_cast_fp16, y = var_27870_to_fp16)[name = tensor("aw_2341_cast_fp16")]; + tensor var_27873_equation_0 = const()[name = tensor("op_27873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27873_cast_fp16 = einsum(equation = var_27873_equation_0, values = (var_27715_cast_fp16, var_27632_cast_fp16))[name = tensor("op_27873_cast_fp16")]; + tensor var_27874_to_fp16 = const()[name = tensor("op_27874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2343_cast_fp16 = mul(x = var_27873_cast_fp16, y = var_27874_to_fp16)[name = tensor("aw_2343_cast_fp16")]; + tensor var_27877_equation_0 = const()[name = tensor("op_27877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27877_cast_fp16 = einsum(equation = var_27877_equation_0, values = (var_27719_cast_fp16, var_27636_cast_fp16))[name = tensor("op_27877_cast_fp16")]; + tensor var_27878_to_fp16 = const()[name = tensor("op_27878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2345_cast_fp16 = mul(x = var_27877_cast_fp16, y = var_27878_to_fp16)[name = tensor("aw_2345_cast_fp16")]; + tensor var_27881_equation_0 = const()[name = tensor("op_27881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27881_cast_fp16 = einsum(equation = var_27881_equation_0, values = (var_27723_cast_fp16, var_27640_cast_fp16))[name = tensor("op_27881_cast_fp16")]; + tensor var_27882_to_fp16 = const()[name = tensor("op_27882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2347_cast_fp16 = mul(x = var_27881_cast_fp16, y = var_27882_to_fp16)[name = tensor("aw_2347_cast_fp16")]; + tensor var_27885_equation_0 = const()[name = tensor("op_27885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27885_cast_fp16 = einsum(equation = var_27885_equation_0, values = (var_27727_cast_fp16, var_27644_cast_fp16))[name = tensor("op_27885_cast_fp16")]; + tensor var_27886_to_fp16 = const()[name = tensor("op_27886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2349_cast_fp16 = mul(x = var_27885_cast_fp16, y = var_27886_to_fp16)[name = tensor("aw_2349_cast_fp16")]; + tensor var_27889_equation_0 = const()[name = tensor("op_27889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27889_cast_fp16 = einsum(equation = var_27889_equation_0, values = (var_27731_cast_fp16, var_27648_cast_fp16))[name = tensor("op_27889_cast_fp16")]; + tensor var_27890_to_fp16 = const()[name = tensor("op_27890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2351_cast_fp16 = mul(x = var_27889_cast_fp16, y = var_27890_to_fp16)[name = tensor("aw_2351_cast_fp16")]; + tensor var_27893_equation_0 = const()[name = tensor("op_27893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27893_cast_fp16 = einsum(equation = var_27893_equation_0, values = (var_27735_cast_fp16, var_27652_cast_fp16))[name = tensor("op_27893_cast_fp16")]; + tensor var_27894_to_fp16 = const()[name = tensor("op_27894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2353_cast_fp16 = mul(x = var_27893_cast_fp16, y = var_27894_to_fp16)[name = tensor("aw_2353_cast_fp16")]; + tensor var_27897_equation_0 = const()[name = tensor("op_27897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27897_cast_fp16 = einsum(equation = var_27897_equation_0, values = (var_27739_cast_fp16, var_27656_cast_fp16))[name = tensor("op_27897_cast_fp16")]; + tensor var_27898_to_fp16 = const()[name = tensor("op_27898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2355_cast_fp16 = mul(x = var_27897_cast_fp16, y = var_27898_to_fp16)[name = tensor("aw_2355_cast_fp16")]; + tensor var_27901_equation_0 = const()[name = tensor("op_27901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27901_cast_fp16 = einsum(equation = var_27901_equation_0, values = (var_27743_cast_fp16, var_27660_cast_fp16))[name = tensor("op_27901_cast_fp16")]; + tensor var_27902_to_fp16 = const()[name = tensor("op_27902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2357_cast_fp16 = mul(x = var_27901_cast_fp16, y = var_27902_to_fp16)[name = tensor("aw_2357_cast_fp16")]; + tensor var_27905_equation_0 = const()[name = tensor("op_27905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27905_cast_fp16 = einsum(equation = var_27905_equation_0, values = (var_27747_cast_fp16, var_27664_cast_fp16))[name = tensor("op_27905_cast_fp16")]; + tensor var_27906_to_fp16 = const()[name = tensor("op_27906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2359_cast_fp16 = mul(x = var_27905_cast_fp16, y = var_27906_to_fp16)[name = tensor("aw_2359_cast_fp16")]; + tensor var_27908_cast_fp16 = softmax(axis = var_21077, x = aw_2321_cast_fp16)[name = tensor("op_27908_cast_fp16")]; + tensor var_27909_cast_fp16 = softmax(axis = var_21077, x = aw_2323_cast_fp16)[name = tensor("op_27909_cast_fp16")]; + tensor var_27910_cast_fp16 = softmax(axis = var_21077, x = aw_2325_cast_fp16)[name = tensor("op_27910_cast_fp16")]; + tensor var_27911_cast_fp16 = softmax(axis = var_21077, x = aw_2327_cast_fp16)[name = tensor("op_27911_cast_fp16")]; + tensor var_27912_cast_fp16 = softmax(axis = var_21077, x = aw_2329_cast_fp16)[name = tensor("op_27912_cast_fp16")]; + tensor var_27913_cast_fp16 = softmax(axis = var_21077, x = aw_2331_cast_fp16)[name = tensor("op_27913_cast_fp16")]; + tensor var_27914_cast_fp16 = softmax(axis = var_21077, x = aw_2333_cast_fp16)[name = tensor("op_27914_cast_fp16")]; + tensor var_27915_cast_fp16 = softmax(axis = var_21077, x = aw_2335_cast_fp16)[name = tensor("op_27915_cast_fp16")]; + tensor var_27916_cast_fp16 = softmax(axis = var_21077, x = aw_2337_cast_fp16)[name = tensor("op_27916_cast_fp16")]; + tensor var_27917_cast_fp16 = softmax(axis = var_21077, x = aw_2339_cast_fp16)[name = tensor("op_27917_cast_fp16")]; + tensor var_27918_cast_fp16 = softmax(axis = var_21077, x = aw_2341_cast_fp16)[name = tensor("op_27918_cast_fp16")]; + tensor var_27919_cast_fp16 = softmax(axis = var_21077, x = aw_2343_cast_fp16)[name = tensor("op_27919_cast_fp16")]; + tensor var_27920_cast_fp16 = softmax(axis = var_21077, x = aw_2345_cast_fp16)[name = tensor("op_27920_cast_fp16")]; + tensor var_27921_cast_fp16 = softmax(axis = var_21077, x = aw_2347_cast_fp16)[name = tensor("op_27921_cast_fp16")]; + tensor var_27922_cast_fp16 = softmax(axis = var_21077, x = aw_2349_cast_fp16)[name = tensor("op_27922_cast_fp16")]; + tensor var_27923_cast_fp16 = softmax(axis = var_21077, x = aw_2351_cast_fp16)[name = tensor("op_27923_cast_fp16")]; + tensor var_27924_cast_fp16 = softmax(axis = var_21077, x = aw_2353_cast_fp16)[name = tensor("op_27924_cast_fp16")]; + tensor var_27925_cast_fp16 = softmax(axis = var_21077, x = aw_2355_cast_fp16)[name = tensor("op_27925_cast_fp16")]; + tensor var_27926_cast_fp16 = softmax(axis = var_21077, x = aw_2357_cast_fp16)[name = tensor("op_27926_cast_fp16")]; + tensor var_27927_cast_fp16 = softmax(axis = var_21077, x = aw_2359_cast_fp16)[name = tensor("op_27927_cast_fp16")]; + tensor var_27929_equation_0 = const()[name = tensor("op_27929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27929_cast_fp16 = einsum(equation = var_27929_equation_0, values = (var_27749_cast_fp16, var_27908_cast_fp16))[name = tensor("op_27929_cast_fp16")]; + tensor var_27931_equation_0 = const()[name = tensor("op_27931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27931_cast_fp16 = einsum(equation = var_27931_equation_0, values = (var_27753_cast_fp16, var_27909_cast_fp16))[name = tensor("op_27931_cast_fp16")]; + tensor var_27933_equation_0 = const()[name = tensor("op_27933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27933_cast_fp16 = einsum(equation = var_27933_equation_0, values = (var_27757_cast_fp16, var_27910_cast_fp16))[name = tensor("op_27933_cast_fp16")]; + tensor var_27935_equation_0 = const()[name = tensor("op_27935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27935_cast_fp16 = einsum(equation = var_27935_equation_0, values = (var_27761_cast_fp16, var_27911_cast_fp16))[name = tensor("op_27935_cast_fp16")]; + tensor var_27937_equation_0 = const()[name = tensor("op_27937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27937_cast_fp16 = einsum(equation = var_27937_equation_0, values = (var_27765_cast_fp16, var_27912_cast_fp16))[name = tensor("op_27937_cast_fp16")]; + tensor var_27939_equation_0 = const()[name = tensor("op_27939_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27939_cast_fp16 = einsum(equation = var_27939_equation_0, values = (var_27769_cast_fp16, var_27913_cast_fp16))[name = tensor("op_27939_cast_fp16")]; + tensor var_27941_equation_0 = const()[name = tensor("op_27941_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27941_cast_fp16 = einsum(equation = var_27941_equation_0, values = (var_27773_cast_fp16, var_27914_cast_fp16))[name = tensor("op_27941_cast_fp16")]; + tensor var_27943_equation_0 = const()[name = tensor("op_27943_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27943_cast_fp16 = einsum(equation = var_27943_equation_0, values = (var_27777_cast_fp16, var_27915_cast_fp16))[name = tensor("op_27943_cast_fp16")]; + tensor var_27945_equation_0 = const()[name = tensor("op_27945_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27945_cast_fp16 = einsum(equation = var_27945_equation_0, values = (var_27781_cast_fp16, var_27916_cast_fp16))[name = tensor("op_27945_cast_fp16")]; + tensor var_27947_equation_0 = const()[name = tensor("op_27947_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27947_cast_fp16 = einsum(equation = var_27947_equation_0, values = (var_27785_cast_fp16, var_27917_cast_fp16))[name = tensor("op_27947_cast_fp16")]; + tensor var_27949_equation_0 = const()[name = tensor("op_27949_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27949_cast_fp16 = einsum(equation = var_27949_equation_0, values = (var_27789_cast_fp16, var_27918_cast_fp16))[name = tensor("op_27949_cast_fp16")]; + tensor var_27951_equation_0 = const()[name = tensor("op_27951_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27951_cast_fp16 = einsum(equation = var_27951_equation_0, values = (var_27793_cast_fp16, var_27919_cast_fp16))[name = tensor("op_27951_cast_fp16")]; + tensor var_27953_equation_0 = const()[name = tensor("op_27953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27953_cast_fp16 = einsum(equation = var_27953_equation_0, values = (var_27797_cast_fp16, var_27920_cast_fp16))[name = tensor("op_27953_cast_fp16")]; + tensor var_27955_equation_0 = const()[name = tensor("op_27955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27955_cast_fp16 = einsum(equation = var_27955_equation_0, values = (var_27801_cast_fp16, var_27921_cast_fp16))[name = tensor("op_27955_cast_fp16")]; + tensor var_27957_equation_0 = const()[name = tensor("op_27957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27957_cast_fp16 = einsum(equation = var_27957_equation_0, values = (var_27805_cast_fp16, var_27922_cast_fp16))[name = tensor("op_27957_cast_fp16")]; + tensor var_27959_equation_0 = const()[name = tensor("op_27959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27959_cast_fp16 = einsum(equation = var_27959_equation_0, values = (var_27809_cast_fp16, var_27923_cast_fp16))[name = tensor("op_27959_cast_fp16")]; + tensor var_27961_equation_0 = const()[name = tensor("op_27961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27961_cast_fp16 = einsum(equation = var_27961_equation_0, values = (var_27813_cast_fp16, var_27924_cast_fp16))[name = tensor("op_27961_cast_fp16")]; + tensor var_27963_equation_0 = const()[name = tensor("op_27963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27963_cast_fp16 = einsum(equation = var_27963_equation_0, values = (var_27817_cast_fp16, var_27925_cast_fp16))[name = tensor("op_27963_cast_fp16")]; + tensor var_27965_equation_0 = const()[name = tensor("op_27965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27965_cast_fp16 = einsum(equation = var_27965_equation_0, values = (var_27821_cast_fp16, var_27926_cast_fp16))[name = tensor("op_27965_cast_fp16")]; + tensor var_27967_equation_0 = const()[name = tensor("op_27967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27967_cast_fp16 = einsum(equation = var_27967_equation_0, values = (var_27825_cast_fp16, var_27927_cast_fp16))[name = tensor("op_27967_cast_fp16")]; + tensor input_383_interleave_0 = const()[name = tensor("input_383_interleave_0"), val = tensor(false)]; + tensor input_383_cast_fp16 = concat(axis = var_21077, interleave = input_383_interleave_0, values = (var_27929_cast_fp16, var_27931_cast_fp16, var_27933_cast_fp16, var_27935_cast_fp16, var_27937_cast_fp16, var_27939_cast_fp16, var_27941_cast_fp16, var_27943_cast_fp16, var_27945_cast_fp16, var_27947_cast_fp16, var_27949_cast_fp16, var_27951_cast_fp16, var_27953_cast_fp16, var_27955_cast_fp16, var_27957_cast_fp16, var_27959_cast_fp16, var_27961_cast_fp16, var_27963_cast_fp16, var_27965_cast_fp16, var_27967_cast_fp16))[name = tensor("input_383_cast_fp16")]; + tensor var_27977_pad_type_0 = const()[name = tensor("op_27977_pad_type_0"), val = tensor("valid")]; + tensor var_27977_strides_0 = const()[name = tensor("op_27977_strides_0"), val = tensor([1, 1])]; + tensor var_27977_pad_0 = const()[name = tensor("op_27977_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27977_dilations_0 = const()[name = tensor("op_27977_dilations_0"), val = tensor([1, 1])]; + tensor var_27977_groups_0 = const()[name = tensor("op_27977_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834428800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835657664))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835657856)))]; + tensor var_27977_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_27977_dilations_0, groups = var_27977_groups_0, pad = var_27977_pad_0, pad_type = var_27977_pad_type_0, strides = var_27977_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized, x = input_383_cast_fp16)[name = tensor("op_27977_cast_fp16")]; + tensor inputs_189_cast_fp16 = add(x = var_27977_cast_fp16, y = inputs_187_cast_fp16)[name = tensor("inputs_189_cast_fp16")]; + tensor hidden_states_253_axes_0 = const()[name = tensor("hidden_states_253_axes_0"), val = tensor([1])]; + tensor hidden_states_253_gamma_0_to_fp16 = const()[name = tensor("hidden_states_253_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835660480)))]; + tensor hidden_states_253_beta_0_to_fp16 = const()[name = tensor("hidden_states_253_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835663104)))]; + tensor var_27987_to_fp16 = const()[name = tensor("op_27987_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_253_cast_fp16 = layer_norm(axes = hidden_states_253_axes_0, beta = hidden_states_253_beta_0_to_fp16, epsilon = var_27987_to_fp16, gamma = hidden_states_253_gamma_0_to_fp16, x = inputs_189_cast_fp16)[name = tensor("hidden_states_253_cast_fp16")]; + tensor q_127_pad_type_0 = const()[name = tensor("q_127_pad_type_0"), val = tensor("valid")]; + tensor q_127_strides_0 = const()[name = tensor("q_127_strides_0"), val = tensor([1, 1])]; + tensor q_127_pad_0 = const()[name = tensor("q_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_127_dilations_0 = const()[name = tensor("q_127_dilations_0"), val = tensor([1, 1])]; + tensor q_127_groups_0 = const()[name = tensor("q_127_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835665728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(836894592))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_127_cast_fp16 = conv(dilations = q_127_dilations_0, groups = q_127_groups_0, pad = q_127_pad_0, pad_type = q_127_pad_type_0, strides = q_127_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_253_cast_fp16)[name = tensor("q_127_cast_fp16")]; + tensor k_253_pad_type_0 = const()[name = tensor("k_253_pad_type_0"), val = tensor("valid")]; + tensor k_253_strides_0 = const()[name = tensor("k_253_strides_0"), val = tensor([1, 1])]; + tensor k_253_pad_0 = const()[name = tensor("k_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_253_dilations_0 = const()[name = tensor("k_253_dilations_0"), val = tensor([1, 1])]; + tensor k_253_groups_0 = const()[name = tensor("k_253_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(836894784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838860928))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_253_cast_fp16 = conv(dilations = k_253_dilations_0, groups = k_253_groups_0, pad = k_253_pad_0, pad_type = k_253_pad_type_0, strides = k_253_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_253_cast_fp16")]; + tensor v_127_pad_type_0 = const()[name = tensor("v_127_pad_type_0"), val = tensor("valid")]; + tensor v_127_strides_0 = const()[name = tensor("v_127_strides_0"), val = tensor([1, 1])]; + tensor v_127_pad_0 = const()[name = tensor("v_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_127_dilations_0 = const()[name = tensor("v_127_dilations_0"), val = tensor([1, 1])]; + tensor v_127_groups_0 = const()[name = tensor("v_127_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838861120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840827264))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_127_cast_fp16 = conv(dilations = v_127_dilations_0, groups = v_127_groups_0, pad = v_127_pad_0, pad_type = v_127_pad_type_0, strides = v_127_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_127_cast_fp16")]; + tensor var_28020_begin_0 = const()[name = tensor("op_28020_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28020_end_0 = const()[name = tensor("op_28020_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28020_end_mask_0 = const()[name = tensor("op_28020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28020_cast_fp16 = slice_by_index(begin = var_28020_begin_0, end = var_28020_end_0, end_mask = var_28020_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28020_cast_fp16")]; + tensor var_28024_begin_0 = const()[name = tensor("op_28024_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28024_end_0 = const()[name = tensor("op_28024_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28024_end_mask_0 = const()[name = tensor("op_28024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28024_cast_fp16 = slice_by_index(begin = var_28024_begin_0, end = var_28024_end_0, end_mask = var_28024_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28024_cast_fp16")]; + tensor var_28028_begin_0 = const()[name = tensor("op_28028_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28028_end_0 = const()[name = tensor("op_28028_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28028_end_mask_0 = const()[name = tensor("op_28028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28028_cast_fp16 = slice_by_index(begin = var_28028_begin_0, end = var_28028_end_0, end_mask = var_28028_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28028_cast_fp16")]; + tensor var_28032_begin_0 = const()[name = tensor("op_28032_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28032_end_0 = const()[name = tensor("op_28032_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28032_end_mask_0 = const()[name = tensor("op_28032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28032_cast_fp16 = slice_by_index(begin = var_28032_begin_0, end = var_28032_end_0, end_mask = var_28032_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28032_cast_fp16")]; + tensor var_28036_begin_0 = const()[name = tensor("op_28036_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28036_end_0 = const()[name = tensor("op_28036_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28036_end_mask_0 = const()[name = tensor("op_28036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28036_cast_fp16 = slice_by_index(begin = var_28036_begin_0, end = var_28036_end_0, end_mask = var_28036_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28036_cast_fp16")]; + tensor var_28040_begin_0 = const()[name = tensor("op_28040_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28040_end_0 = const()[name = tensor("op_28040_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28040_end_mask_0 = const()[name = tensor("op_28040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28040_cast_fp16 = slice_by_index(begin = var_28040_begin_0, end = var_28040_end_0, end_mask = var_28040_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28040_cast_fp16")]; + tensor var_28044_begin_0 = const()[name = tensor("op_28044_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28044_end_0 = const()[name = tensor("op_28044_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28044_end_mask_0 = const()[name = tensor("op_28044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28044_cast_fp16 = slice_by_index(begin = var_28044_begin_0, end = var_28044_end_0, end_mask = var_28044_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28044_cast_fp16")]; + tensor var_28048_begin_0 = const()[name = tensor("op_28048_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28048_end_0 = const()[name = tensor("op_28048_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28048_end_mask_0 = const()[name = tensor("op_28048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28048_cast_fp16 = slice_by_index(begin = var_28048_begin_0, end = var_28048_end_0, end_mask = var_28048_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28048_cast_fp16")]; + tensor var_28052_begin_0 = const()[name = tensor("op_28052_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28052_end_0 = const()[name = tensor("op_28052_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28052_end_mask_0 = const()[name = tensor("op_28052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28052_cast_fp16 = slice_by_index(begin = var_28052_begin_0, end = var_28052_end_0, end_mask = var_28052_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28052_cast_fp16")]; + tensor var_28056_begin_0 = const()[name = tensor("op_28056_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28056_end_0 = const()[name = tensor("op_28056_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28056_end_mask_0 = const()[name = tensor("op_28056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28056_cast_fp16 = slice_by_index(begin = var_28056_begin_0, end = var_28056_end_0, end_mask = var_28056_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28056_cast_fp16")]; + tensor var_28060_begin_0 = const()[name = tensor("op_28060_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28060_end_0 = const()[name = tensor("op_28060_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28060_end_mask_0 = const()[name = tensor("op_28060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28060_cast_fp16 = slice_by_index(begin = var_28060_begin_0, end = var_28060_end_0, end_mask = var_28060_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28060_cast_fp16")]; + tensor var_28064_begin_0 = const()[name = tensor("op_28064_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28064_end_0 = const()[name = tensor("op_28064_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28064_end_mask_0 = const()[name = tensor("op_28064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28064_cast_fp16 = slice_by_index(begin = var_28064_begin_0, end = var_28064_end_0, end_mask = var_28064_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28064_cast_fp16")]; + tensor var_28068_begin_0 = const()[name = tensor("op_28068_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28068_end_0 = const()[name = tensor("op_28068_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28068_end_mask_0 = const()[name = tensor("op_28068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28068_cast_fp16 = slice_by_index(begin = var_28068_begin_0, end = var_28068_end_0, end_mask = var_28068_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28068_cast_fp16")]; + tensor var_28072_begin_0 = const()[name = tensor("op_28072_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28072_end_0 = const()[name = tensor("op_28072_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28072_end_mask_0 = const()[name = tensor("op_28072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28072_cast_fp16 = slice_by_index(begin = var_28072_begin_0, end = var_28072_end_0, end_mask = var_28072_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28072_cast_fp16")]; + tensor var_28076_begin_0 = const()[name = tensor("op_28076_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28076_end_0 = const()[name = tensor("op_28076_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28076_end_mask_0 = const()[name = tensor("op_28076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28076_cast_fp16 = slice_by_index(begin = var_28076_begin_0, end = var_28076_end_0, end_mask = var_28076_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28076_cast_fp16")]; + tensor var_28080_begin_0 = const()[name = tensor("op_28080_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28080_end_0 = const()[name = tensor("op_28080_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28080_end_mask_0 = const()[name = tensor("op_28080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28080_cast_fp16 = slice_by_index(begin = var_28080_begin_0, end = var_28080_end_0, end_mask = var_28080_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28080_cast_fp16")]; + tensor var_28084_begin_0 = const()[name = tensor("op_28084_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28084_end_0 = const()[name = tensor("op_28084_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28084_end_mask_0 = const()[name = tensor("op_28084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28084_cast_fp16 = slice_by_index(begin = var_28084_begin_0, end = var_28084_end_0, end_mask = var_28084_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28084_cast_fp16")]; + tensor var_28088_begin_0 = const()[name = tensor("op_28088_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28088_end_0 = const()[name = tensor("op_28088_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28088_end_mask_0 = const()[name = tensor("op_28088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28088_cast_fp16 = slice_by_index(begin = var_28088_begin_0, end = var_28088_end_0, end_mask = var_28088_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28088_cast_fp16")]; + tensor var_28092_begin_0 = const()[name = tensor("op_28092_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28092_end_0 = const()[name = tensor("op_28092_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_28092_end_mask_0 = const()[name = tensor("op_28092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28092_cast_fp16 = slice_by_index(begin = var_28092_begin_0, end = var_28092_end_0, end_mask = var_28092_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28092_cast_fp16")]; + tensor var_28096_begin_0 = const()[name = tensor("op_28096_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28096_end_0 = const()[name = tensor("op_28096_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_28096_end_mask_0 = const()[name = tensor("op_28096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28096_cast_fp16 = slice_by_index(begin = var_28096_begin_0, end = var_28096_end_0, end_mask = var_28096_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28096_cast_fp16")]; + tensor k_255_perm_0 = const()[name = tensor("k_255_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_28103_begin_0 = const()[name = tensor("op_28103_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28103_end_0 = const()[name = tensor("op_28103_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_28103_end_mask_0 = const()[name = tensor("op_28103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_255_cast_fp16 = transpose(perm = k_255_perm_0, x = k_253_cast_fp16)[name = tensor("transpose_4")]; + tensor var_28103_cast_fp16 = slice_by_index(begin = var_28103_begin_0, end = var_28103_end_0, end_mask = var_28103_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28103_cast_fp16")]; + tensor var_28107_begin_0 = const()[name = tensor("op_28107_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_28107_end_0 = const()[name = tensor("op_28107_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_28107_end_mask_0 = const()[name = tensor("op_28107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28107_cast_fp16 = slice_by_index(begin = var_28107_begin_0, end = var_28107_end_0, end_mask = var_28107_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28107_cast_fp16")]; + tensor var_28111_begin_0 = const()[name = tensor("op_28111_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_28111_end_0 = const()[name = tensor("op_28111_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_28111_end_mask_0 = const()[name = tensor("op_28111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28111_cast_fp16 = slice_by_index(begin = var_28111_begin_0, end = var_28111_end_0, end_mask = var_28111_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28111_cast_fp16")]; + tensor var_28115_begin_0 = const()[name = tensor("op_28115_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_28115_end_0 = const()[name = tensor("op_28115_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_28115_end_mask_0 = const()[name = tensor("op_28115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28115_cast_fp16 = slice_by_index(begin = var_28115_begin_0, end = var_28115_end_0, end_mask = var_28115_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28115_cast_fp16")]; + tensor var_28119_begin_0 = const()[name = tensor("op_28119_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_28119_end_0 = const()[name = tensor("op_28119_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_28119_end_mask_0 = const()[name = tensor("op_28119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28119_cast_fp16 = slice_by_index(begin = var_28119_begin_0, end = var_28119_end_0, end_mask = var_28119_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28119_cast_fp16")]; + tensor var_28123_begin_0 = const()[name = tensor("op_28123_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_28123_end_0 = const()[name = tensor("op_28123_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_28123_end_mask_0 = const()[name = tensor("op_28123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28123_cast_fp16 = slice_by_index(begin = var_28123_begin_0, end = var_28123_end_0, end_mask = var_28123_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28123_cast_fp16")]; + tensor var_28127_begin_0 = const()[name = tensor("op_28127_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_28127_end_0 = const()[name = tensor("op_28127_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_28127_end_mask_0 = const()[name = tensor("op_28127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28127_cast_fp16 = slice_by_index(begin = var_28127_begin_0, end = var_28127_end_0, end_mask = var_28127_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28127_cast_fp16")]; + tensor var_28131_begin_0 = const()[name = tensor("op_28131_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_28131_end_0 = const()[name = tensor("op_28131_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_28131_end_mask_0 = const()[name = tensor("op_28131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28131_cast_fp16 = slice_by_index(begin = var_28131_begin_0, end = var_28131_end_0, end_mask = var_28131_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28131_cast_fp16")]; + tensor var_28135_begin_0 = const()[name = tensor("op_28135_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_28135_end_0 = const()[name = tensor("op_28135_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_28135_end_mask_0 = const()[name = tensor("op_28135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28135_cast_fp16 = slice_by_index(begin = var_28135_begin_0, end = var_28135_end_0, end_mask = var_28135_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28135_cast_fp16")]; + tensor var_28139_begin_0 = const()[name = tensor("op_28139_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_28139_end_0 = const()[name = tensor("op_28139_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_28139_end_mask_0 = const()[name = tensor("op_28139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28139_cast_fp16 = slice_by_index(begin = var_28139_begin_0, end = var_28139_end_0, end_mask = var_28139_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28139_cast_fp16")]; + tensor var_28143_begin_0 = const()[name = tensor("op_28143_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_28143_end_0 = const()[name = tensor("op_28143_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_28143_end_mask_0 = const()[name = tensor("op_28143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28143_cast_fp16 = slice_by_index(begin = var_28143_begin_0, end = var_28143_end_0, end_mask = var_28143_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28143_cast_fp16")]; + tensor var_28147_begin_0 = const()[name = tensor("op_28147_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_28147_end_0 = const()[name = tensor("op_28147_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_28147_end_mask_0 = const()[name = tensor("op_28147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28147_cast_fp16 = slice_by_index(begin = var_28147_begin_0, end = var_28147_end_0, end_mask = var_28147_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28147_cast_fp16")]; + tensor var_28151_begin_0 = const()[name = tensor("op_28151_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_28151_end_0 = const()[name = tensor("op_28151_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_28151_end_mask_0 = const()[name = tensor("op_28151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28151_cast_fp16 = slice_by_index(begin = var_28151_begin_0, end = var_28151_end_0, end_mask = var_28151_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28151_cast_fp16")]; + tensor var_28155_begin_0 = const()[name = tensor("op_28155_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_28155_end_0 = const()[name = tensor("op_28155_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_28155_end_mask_0 = const()[name = tensor("op_28155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28155_cast_fp16 = slice_by_index(begin = var_28155_begin_0, end = var_28155_end_0, end_mask = var_28155_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28155_cast_fp16")]; + tensor var_28159_begin_0 = const()[name = tensor("op_28159_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_28159_end_0 = const()[name = tensor("op_28159_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_28159_end_mask_0 = const()[name = tensor("op_28159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28159_cast_fp16 = slice_by_index(begin = var_28159_begin_0, end = var_28159_end_0, end_mask = var_28159_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28159_cast_fp16")]; + tensor var_28163_begin_0 = const()[name = tensor("op_28163_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_28163_end_0 = const()[name = tensor("op_28163_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_28163_end_mask_0 = const()[name = tensor("op_28163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28163_cast_fp16 = slice_by_index(begin = var_28163_begin_0, end = var_28163_end_0, end_mask = var_28163_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28163_cast_fp16")]; + tensor var_28167_begin_0 = const()[name = tensor("op_28167_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_28167_end_0 = const()[name = tensor("op_28167_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_28167_end_mask_0 = const()[name = tensor("op_28167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28167_cast_fp16 = slice_by_index(begin = var_28167_begin_0, end = var_28167_end_0, end_mask = var_28167_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28167_cast_fp16")]; + tensor var_28171_begin_0 = const()[name = tensor("op_28171_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_28171_end_0 = const()[name = tensor("op_28171_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_28171_end_mask_0 = const()[name = tensor("op_28171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28171_cast_fp16 = slice_by_index(begin = var_28171_begin_0, end = var_28171_end_0, end_mask = var_28171_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28171_cast_fp16")]; + tensor var_28175_begin_0 = const()[name = tensor("op_28175_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_28175_end_0 = const()[name = tensor("op_28175_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_28175_end_mask_0 = const()[name = tensor("op_28175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28175_cast_fp16 = slice_by_index(begin = var_28175_begin_0, end = var_28175_end_0, end_mask = var_28175_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28175_cast_fp16")]; + tensor var_28179_begin_0 = const()[name = tensor("op_28179_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_28179_end_0 = const()[name = tensor("op_28179_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_28179_end_mask_0 = const()[name = tensor("op_28179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28179_cast_fp16 = slice_by_index(begin = var_28179_begin_0, end = var_28179_end_0, end_mask = var_28179_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28179_cast_fp16")]; + tensor var_28181_begin_0 = const()[name = tensor("op_28181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28181_end_0 = const()[name = tensor("op_28181_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_28181_end_mask_0 = const()[name = tensor("op_28181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28181_cast_fp16 = slice_by_index(begin = var_28181_begin_0, end = var_28181_end_0, end_mask = var_28181_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28181_cast_fp16")]; + tensor var_28185_begin_0 = const()[name = tensor("op_28185_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28185_end_0 = const()[name = tensor("op_28185_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_28185_end_mask_0 = const()[name = tensor("op_28185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28185_cast_fp16 = slice_by_index(begin = var_28185_begin_0, end = var_28185_end_0, end_mask = var_28185_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28185_cast_fp16")]; + tensor var_28189_begin_0 = const()[name = tensor("op_28189_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28189_end_0 = const()[name = tensor("op_28189_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_28189_end_mask_0 = const()[name = tensor("op_28189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28189_cast_fp16 = slice_by_index(begin = var_28189_begin_0, end = var_28189_end_0, end_mask = var_28189_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28189_cast_fp16")]; + tensor var_28193_begin_0 = const()[name = tensor("op_28193_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28193_end_0 = const()[name = tensor("op_28193_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_28193_end_mask_0 = const()[name = tensor("op_28193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28193_cast_fp16 = slice_by_index(begin = var_28193_begin_0, end = var_28193_end_0, end_mask = var_28193_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28193_cast_fp16")]; + tensor var_28197_begin_0 = const()[name = tensor("op_28197_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28197_end_0 = const()[name = tensor("op_28197_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_28197_end_mask_0 = const()[name = tensor("op_28197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28197_cast_fp16 = slice_by_index(begin = var_28197_begin_0, end = var_28197_end_0, end_mask = var_28197_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28197_cast_fp16")]; + tensor var_28201_begin_0 = const()[name = tensor("op_28201_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28201_end_0 = const()[name = tensor("op_28201_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_28201_end_mask_0 = const()[name = tensor("op_28201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28201_cast_fp16 = slice_by_index(begin = var_28201_begin_0, end = var_28201_end_0, end_mask = var_28201_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28201_cast_fp16")]; + tensor var_28205_begin_0 = const()[name = tensor("op_28205_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28205_end_0 = const()[name = tensor("op_28205_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_28205_end_mask_0 = const()[name = tensor("op_28205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28205_cast_fp16 = slice_by_index(begin = var_28205_begin_0, end = var_28205_end_0, end_mask = var_28205_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28205_cast_fp16")]; + tensor var_28209_begin_0 = const()[name = tensor("op_28209_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28209_end_0 = const()[name = tensor("op_28209_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_28209_end_mask_0 = const()[name = tensor("op_28209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28209_cast_fp16 = slice_by_index(begin = var_28209_begin_0, end = var_28209_end_0, end_mask = var_28209_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28209_cast_fp16")]; + tensor var_28213_begin_0 = const()[name = tensor("op_28213_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28213_end_0 = const()[name = tensor("op_28213_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_28213_end_mask_0 = const()[name = tensor("op_28213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28213_cast_fp16 = slice_by_index(begin = var_28213_begin_0, end = var_28213_end_0, end_mask = var_28213_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28213_cast_fp16")]; + tensor var_28217_begin_0 = const()[name = tensor("op_28217_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28217_end_0 = const()[name = tensor("op_28217_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_28217_end_mask_0 = const()[name = tensor("op_28217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28217_cast_fp16 = slice_by_index(begin = var_28217_begin_0, end = var_28217_end_0, end_mask = var_28217_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28217_cast_fp16")]; + tensor var_28221_begin_0 = const()[name = tensor("op_28221_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28221_end_0 = const()[name = tensor("op_28221_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_28221_end_mask_0 = const()[name = tensor("op_28221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28221_cast_fp16 = slice_by_index(begin = var_28221_begin_0, end = var_28221_end_0, end_mask = var_28221_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28221_cast_fp16")]; + tensor var_28225_begin_0 = const()[name = tensor("op_28225_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28225_end_0 = const()[name = tensor("op_28225_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_28225_end_mask_0 = const()[name = tensor("op_28225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28225_cast_fp16 = slice_by_index(begin = var_28225_begin_0, end = var_28225_end_0, end_mask = var_28225_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28225_cast_fp16")]; + tensor var_28229_begin_0 = const()[name = tensor("op_28229_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28229_end_0 = const()[name = tensor("op_28229_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_28229_end_mask_0 = const()[name = tensor("op_28229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28229_cast_fp16 = slice_by_index(begin = var_28229_begin_0, end = var_28229_end_0, end_mask = var_28229_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28229_cast_fp16")]; + tensor var_28233_begin_0 = const()[name = tensor("op_28233_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28233_end_0 = const()[name = tensor("op_28233_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_28233_end_mask_0 = const()[name = tensor("op_28233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28233_cast_fp16 = slice_by_index(begin = var_28233_begin_0, end = var_28233_end_0, end_mask = var_28233_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28233_cast_fp16")]; + tensor var_28237_begin_0 = const()[name = tensor("op_28237_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28237_end_0 = const()[name = tensor("op_28237_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_28237_end_mask_0 = const()[name = tensor("op_28237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28237_cast_fp16 = slice_by_index(begin = var_28237_begin_0, end = var_28237_end_0, end_mask = var_28237_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28237_cast_fp16")]; + tensor var_28241_begin_0 = const()[name = tensor("op_28241_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28241_end_0 = const()[name = tensor("op_28241_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_28241_end_mask_0 = const()[name = tensor("op_28241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28241_cast_fp16 = slice_by_index(begin = var_28241_begin_0, end = var_28241_end_0, end_mask = var_28241_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28241_cast_fp16")]; + tensor var_28245_begin_0 = const()[name = tensor("op_28245_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28245_end_0 = const()[name = tensor("op_28245_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_28245_end_mask_0 = const()[name = tensor("op_28245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28245_cast_fp16 = slice_by_index(begin = var_28245_begin_0, end = var_28245_end_0, end_mask = var_28245_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28245_cast_fp16")]; + tensor var_28249_begin_0 = const()[name = tensor("op_28249_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28249_end_0 = const()[name = tensor("op_28249_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_28249_end_mask_0 = const()[name = tensor("op_28249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28249_cast_fp16 = slice_by_index(begin = var_28249_begin_0, end = var_28249_end_0, end_mask = var_28249_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28249_cast_fp16")]; + tensor var_28253_begin_0 = const()[name = tensor("op_28253_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28253_end_0 = const()[name = tensor("op_28253_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_28253_end_mask_0 = const()[name = tensor("op_28253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28253_cast_fp16 = slice_by_index(begin = var_28253_begin_0, end = var_28253_end_0, end_mask = var_28253_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28253_cast_fp16")]; + tensor var_28257_begin_0 = const()[name = tensor("op_28257_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28257_end_0 = const()[name = tensor("op_28257_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_28257_end_mask_0 = const()[name = tensor("op_28257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28257_cast_fp16 = slice_by_index(begin = var_28257_begin_0, end = var_28257_end_0, end_mask = var_28257_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28257_cast_fp16")]; + tensor var_28261_equation_0 = const()[name = tensor("op_28261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28261_cast_fp16 = einsum(equation = var_28261_equation_0, values = (var_28103_cast_fp16, var_28020_cast_fp16))[name = tensor("op_28261_cast_fp16")]; + tensor var_28262_to_fp16 = const()[name = tensor("op_28262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2361_cast_fp16 = mul(x = var_28261_cast_fp16, y = var_28262_to_fp16)[name = tensor("aw_2361_cast_fp16")]; + tensor var_28265_equation_0 = const()[name = tensor("op_28265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28265_cast_fp16 = einsum(equation = var_28265_equation_0, values = (var_28107_cast_fp16, var_28024_cast_fp16))[name = tensor("op_28265_cast_fp16")]; + tensor var_28266_to_fp16 = const()[name = tensor("op_28266_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2363_cast_fp16 = mul(x = var_28265_cast_fp16, y = var_28266_to_fp16)[name = tensor("aw_2363_cast_fp16")]; + tensor var_28269_equation_0 = const()[name = tensor("op_28269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28269_cast_fp16 = einsum(equation = var_28269_equation_0, values = (var_28111_cast_fp16, var_28028_cast_fp16))[name = tensor("op_28269_cast_fp16")]; + tensor var_28270_to_fp16 = const()[name = tensor("op_28270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2365_cast_fp16 = mul(x = var_28269_cast_fp16, y = var_28270_to_fp16)[name = tensor("aw_2365_cast_fp16")]; + tensor var_28273_equation_0 = const()[name = tensor("op_28273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28273_cast_fp16 = einsum(equation = var_28273_equation_0, values = (var_28115_cast_fp16, var_28032_cast_fp16))[name = tensor("op_28273_cast_fp16")]; + tensor var_28274_to_fp16 = const()[name = tensor("op_28274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2367_cast_fp16 = mul(x = var_28273_cast_fp16, y = var_28274_to_fp16)[name = tensor("aw_2367_cast_fp16")]; + tensor var_28277_equation_0 = const()[name = tensor("op_28277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28277_cast_fp16 = einsum(equation = var_28277_equation_0, values = (var_28119_cast_fp16, var_28036_cast_fp16))[name = tensor("op_28277_cast_fp16")]; + tensor var_28278_to_fp16 = const()[name = tensor("op_28278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2369_cast_fp16 = mul(x = var_28277_cast_fp16, y = var_28278_to_fp16)[name = tensor("aw_2369_cast_fp16")]; + tensor var_28281_equation_0 = const()[name = tensor("op_28281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28281_cast_fp16 = einsum(equation = var_28281_equation_0, values = (var_28123_cast_fp16, var_28040_cast_fp16))[name = tensor("op_28281_cast_fp16")]; + tensor var_28282_to_fp16 = const()[name = tensor("op_28282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2371_cast_fp16 = mul(x = var_28281_cast_fp16, y = var_28282_to_fp16)[name = tensor("aw_2371_cast_fp16")]; + tensor var_28285_equation_0 = const()[name = tensor("op_28285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28285_cast_fp16 = einsum(equation = var_28285_equation_0, values = (var_28127_cast_fp16, var_28044_cast_fp16))[name = tensor("op_28285_cast_fp16")]; + tensor var_28286_to_fp16 = const()[name = tensor("op_28286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2373_cast_fp16 = mul(x = var_28285_cast_fp16, y = var_28286_to_fp16)[name = tensor("aw_2373_cast_fp16")]; + tensor var_28289_equation_0 = const()[name = tensor("op_28289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28289_cast_fp16 = einsum(equation = var_28289_equation_0, values = (var_28131_cast_fp16, var_28048_cast_fp16))[name = tensor("op_28289_cast_fp16")]; + tensor var_28290_to_fp16 = const()[name = tensor("op_28290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2375_cast_fp16 = mul(x = var_28289_cast_fp16, y = var_28290_to_fp16)[name = tensor("aw_2375_cast_fp16")]; + tensor var_28293_equation_0 = const()[name = tensor("op_28293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28293_cast_fp16 = einsum(equation = var_28293_equation_0, values = (var_28135_cast_fp16, var_28052_cast_fp16))[name = tensor("op_28293_cast_fp16")]; + tensor var_28294_to_fp16 = const()[name = tensor("op_28294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2377_cast_fp16 = mul(x = var_28293_cast_fp16, y = var_28294_to_fp16)[name = tensor("aw_2377_cast_fp16")]; + tensor var_28297_equation_0 = const()[name = tensor("op_28297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28297_cast_fp16 = einsum(equation = var_28297_equation_0, values = (var_28139_cast_fp16, var_28056_cast_fp16))[name = tensor("op_28297_cast_fp16")]; + tensor var_28298_to_fp16 = const()[name = tensor("op_28298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2379_cast_fp16 = mul(x = var_28297_cast_fp16, y = var_28298_to_fp16)[name = tensor("aw_2379_cast_fp16")]; + tensor var_28301_equation_0 = const()[name = tensor("op_28301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28301_cast_fp16 = einsum(equation = var_28301_equation_0, values = (var_28143_cast_fp16, var_28060_cast_fp16))[name = tensor("op_28301_cast_fp16")]; + tensor var_28302_to_fp16 = const()[name = tensor("op_28302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2381_cast_fp16 = mul(x = var_28301_cast_fp16, y = var_28302_to_fp16)[name = tensor("aw_2381_cast_fp16")]; + tensor var_28305_equation_0 = const()[name = tensor("op_28305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28305_cast_fp16 = einsum(equation = var_28305_equation_0, values = (var_28147_cast_fp16, var_28064_cast_fp16))[name = tensor("op_28305_cast_fp16")]; + tensor var_28306_to_fp16 = const()[name = tensor("op_28306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2383_cast_fp16 = mul(x = var_28305_cast_fp16, y = var_28306_to_fp16)[name = tensor("aw_2383_cast_fp16")]; + tensor var_28309_equation_0 = const()[name = tensor("op_28309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28309_cast_fp16 = einsum(equation = var_28309_equation_0, values = (var_28151_cast_fp16, var_28068_cast_fp16))[name = tensor("op_28309_cast_fp16")]; + tensor var_28310_to_fp16 = const()[name = tensor("op_28310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2385_cast_fp16 = mul(x = var_28309_cast_fp16, y = var_28310_to_fp16)[name = tensor("aw_2385_cast_fp16")]; + tensor var_28313_equation_0 = const()[name = tensor("op_28313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28313_cast_fp16 = einsum(equation = var_28313_equation_0, values = (var_28155_cast_fp16, var_28072_cast_fp16))[name = tensor("op_28313_cast_fp16")]; + tensor var_28314_to_fp16 = const()[name = tensor("op_28314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2387_cast_fp16 = mul(x = var_28313_cast_fp16, y = var_28314_to_fp16)[name = tensor("aw_2387_cast_fp16")]; + tensor var_28317_equation_0 = const()[name = tensor("op_28317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28317_cast_fp16 = einsum(equation = var_28317_equation_0, values = (var_28159_cast_fp16, var_28076_cast_fp16))[name = tensor("op_28317_cast_fp16")]; + tensor var_28318_to_fp16 = const()[name = tensor("op_28318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2389_cast_fp16 = mul(x = var_28317_cast_fp16, y = var_28318_to_fp16)[name = tensor("aw_2389_cast_fp16")]; + tensor var_28321_equation_0 = const()[name = tensor("op_28321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28321_cast_fp16 = einsum(equation = var_28321_equation_0, values = (var_28163_cast_fp16, var_28080_cast_fp16))[name = tensor("op_28321_cast_fp16")]; + tensor var_28322_to_fp16 = const()[name = tensor("op_28322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2391_cast_fp16 = mul(x = var_28321_cast_fp16, y = var_28322_to_fp16)[name = tensor("aw_2391_cast_fp16")]; + tensor var_28325_equation_0 = const()[name = tensor("op_28325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28325_cast_fp16 = einsum(equation = var_28325_equation_0, values = (var_28167_cast_fp16, var_28084_cast_fp16))[name = tensor("op_28325_cast_fp16")]; + tensor var_28326_to_fp16 = const()[name = tensor("op_28326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2393_cast_fp16 = mul(x = var_28325_cast_fp16, y = var_28326_to_fp16)[name = tensor("aw_2393_cast_fp16")]; + tensor var_28329_equation_0 = const()[name = tensor("op_28329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28329_cast_fp16 = einsum(equation = var_28329_equation_0, values = (var_28171_cast_fp16, var_28088_cast_fp16))[name = tensor("op_28329_cast_fp16")]; + tensor var_28330_to_fp16 = const()[name = tensor("op_28330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2395_cast_fp16 = mul(x = var_28329_cast_fp16, y = var_28330_to_fp16)[name = tensor("aw_2395_cast_fp16")]; + tensor var_28333_equation_0 = const()[name = tensor("op_28333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28333_cast_fp16 = einsum(equation = var_28333_equation_0, values = (var_28175_cast_fp16, var_28092_cast_fp16))[name = tensor("op_28333_cast_fp16")]; + tensor var_28334_to_fp16 = const()[name = tensor("op_28334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2397_cast_fp16 = mul(x = var_28333_cast_fp16, y = var_28334_to_fp16)[name = tensor("aw_2397_cast_fp16")]; + tensor var_28337_equation_0 = const()[name = tensor("op_28337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28337_cast_fp16 = einsum(equation = var_28337_equation_0, values = (var_28179_cast_fp16, var_28096_cast_fp16))[name = tensor("op_28337_cast_fp16")]; + tensor var_28338_to_fp16 = const()[name = tensor("op_28338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2399_cast_fp16 = mul(x = var_28337_cast_fp16, y = var_28338_to_fp16)[name = tensor("aw_2399_cast_fp16")]; + tensor var_28340_cast_fp16 = softmax(axis = var_21077, x = aw_2361_cast_fp16)[name = tensor("op_28340_cast_fp16")]; + tensor var_28341_cast_fp16 = softmax(axis = var_21077, x = aw_2363_cast_fp16)[name = tensor("op_28341_cast_fp16")]; + tensor var_28342_cast_fp16 = softmax(axis = var_21077, x = aw_2365_cast_fp16)[name = tensor("op_28342_cast_fp16")]; + tensor var_28343_cast_fp16 = softmax(axis = var_21077, x = aw_2367_cast_fp16)[name = tensor("op_28343_cast_fp16")]; + tensor var_28344_cast_fp16 = softmax(axis = var_21077, x = aw_2369_cast_fp16)[name = tensor("op_28344_cast_fp16")]; + tensor var_28345_cast_fp16 = softmax(axis = var_21077, x = aw_2371_cast_fp16)[name = tensor("op_28345_cast_fp16")]; + tensor var_28346_cast_fp16 = softmax(axis = var_21077, x = aw_2373_cast_fp16)[name = tensor("op_28346_cast_fp16")]; + tensor var_28347_cast_fp16 = softmax(axis = var_21077, x = aw_2375_cast_fp16)[name = tensor("op_28347_cast_fp16")]; + tensor var_28348_cast_fp16 = softmax(axis = var_21077, x = aw_2377_cast_fp16)[name = tensor("op_28348_cast_fp16")]; + tensor var_28349_cast_fp16 = softmax(axis = var_21077, x = aw_2379_cast_fp16)[name = tensor("op_28349_cast_fp16")]; + tensor var_28350_cast_fp16 = softmax(axis = var_21077, x = aw_2381_cast_fp16)[name = tensor("op_28350_cast_fp16")]; + tensor var_28351_cast_fp16 = softmax(axis = var_21077, x = aw_2383_cast_fp16)[name = tensor("op_28351_cast_fp16")]; + tensor var_28352_cast_fp16 = softmax(axis = var_21077, x = aw_2385_cast_fp16)[name = tensor("op_28352_cast_fp16")]; + tensor var_28353_cast_fp16 = softmax(axis = var_21077, x = aw_2387_cast_fp16)[name = tensor("op_28353_cast_fp16")]; + tensor var_28354_cast_fp16 = softmax(axis = var_21077, x = aw_2389_cast_fp16)[name = tensor("op_28354_cast_fp16")]; + tensor var_28355_cast_fp16 = softmax(axis = var_21077, x = aw_2391_cast_fp16)[name = tensor("op_28355_cast_fp16")]; + tensor var_28356_cast_fp16 = softmax(axis = var_21077, x = aw_2393_cast_fp16)[name = tensor("op_28356_cast_fp16")]; + tensor var_28357_cast_fp16 = softmax(axis = var_21077, x = aw_2395_cast_fp16)[name = tensor("op_28357_cast_fp16")]; + tensor var_28358_cast_fp16 = softmax(axis = var_21077, x = aw_2397_cast_fp16)[name = tensor("op_28358_cast_fp16")]; + tensor var_28359_cast_fp16 = softmax(axis = var_21077, x = aw_2399_cast_fp16)[name = tensor("op_28359_cast_fp16")]; + tensor var_28361_equation_0 = const()[name = tensor("op_28361_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28361_cast_fp16 = einsum(equation = var_28361_equation_0, values = (var_28181_cast_fp16, var_28340_cast_fp16))[name = tensor("op_28361_cast_fp16")]; + tensor var_28363_equation_0 = const()[name = tensor("op_28363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28363_cast_fp16 = einsum(equation = var_28363_equation_0, values = (var_28185_cast_fp16, var_28341_cast_fp16))[name = tensor("op_28363_cast_fp16")]; + tensor var_28365_equation_0 = const()[name = tensor("op_28365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28365_cast_fp16 = einsum(equation = var_28365_equation_0, values = (var_28189_cast_fp16, var_28342_cast_fp16))[name = tensor("op_28365_cast_fp16")]; + tensor var_28367_equation_0 = const()[name = tensor("op_28367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28367_cast_fp16 = einsum(equation = var_28367_equation_0, values = (var_28193_cast_fp16, var_28343_cast_fp16))[name = tensor("op_28367_cast_fp16")]; + tensor var_28369_equation_0 = const()[name = tensor("op_28369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28369_cast_fp16 = einsum(equation = var_28369_equation_0, values = (var_28197_cast_fp16, var_28344_cast_fp16))[name = tensor("op_28369_cast_fp16")]; + tensor var_28371_equation_0 = const()[name = tensor("op_28371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28371_cast_fp16 = einsum(equation = var_28371_equation_0, values = (var_28201_cast_fp16, var_28345_cast_fp16))[name = tensor("op_28371_cast_fp16")]; + tensor var_28373_equation_0 = const()[name = tensor("op_28373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28373_cast_fp16 = einsum(equation = var_28373_equation_0, values = (var_28205_cast_fp16, var_28346_cast_fp16))[name = tensor("op_28373_cast_fp16")]; + tensor var_28375_equation_0 = const()[name = tensor("op_28375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28375_cast_fp16 = einsum(equation = var_28375_equation_0, values = (var_28209_cast_fp16, var_28347_cast_fp16))[name = tensor("op_28375_cast_fp16")]; + tensor var_28377_equation_0 = const()[name = tensor("op_28377_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28377_cast_fp16 = einsum(equation = var_28377_equation_0, values = (var_28213_cast_fp16, var_28348_cast_fp16))[name = tensor("op_28377_cast_fp16")]; + tensor var_28379_equation_0 = const()[name = tensor("op_28379_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28379_cast_fp16 = einsum(equation = var_28379_equation_0, values = (var_28217_cast_fp16, var_28349_cast_fp16))[name = tensor("op_28379_cast_fp16")]; + tensor var_28381_equation_0 = const()[name = tensor("op_28381_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28381_cast_fp16 = einsum(equation = var_28381_equation_0, values = (var_28221_cast_fp16, var_28350_cast_fp16))[name = tensor("op_28381_cast_fp16")]; + tensor var_28383_equation_0 = const()[name = tensor("op_28383_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28383_cast_fp16 = einsum(equation = var_28383_equation_0, values = (var_28225_cast_fp16, var_28351_cast_fp16))[name = tensor("op_28383_cast_fp16")]; + tensor var_28385_equation_0 = const()[name = tensor("op_28385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28385_cast_fp16 = einsum(equation = var_28385_equation_0, values = (var_28229_cast_fp16, var_28352_cast_fp16))[name = tensor("op_28385_cast_fp16")]; + tensor var_28387_equation_0 = const()[name = tensor("op_28387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28387_cast_fp16 = einsum(equation = var_28387_equation_0, values = (var_28233_cast_fp16, var_28353_cast_fp16))[name = tensor("op_28387_cast_fp16")]; + tensor var_28389_equation_0 = const()[name = tensor("op_28389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28389_cast_fp16 = einsum(equation = var_28389_equation_0, values = (var_28237_cast_fp16, var_28354_cast_fp16))[name = tensor("op_28389_cast_fp16")]; + tensor var_28391_equation_0 = const()[name = tensor("op_28391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28391_cast_fp16 = einsum(equation = var_28391_equation_0, values = (var_28241_cast_fp16, var_28355_cast_fp16))[name = tensor("op_28391_cast_fp16")]; + tensor var_28393_equation_0 = const()[name = tensor("op_28393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28393_cast_fp16 = einsum(equation = var_28393_equation_0, values = (var_28245_cast_fp16, var_28356_cast_fp16))[name = tensor("op_28393_cast_fp16")]; + tensor var_28395_equation_0 = const()[name = tensor("op_28395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28395_cast_fp16 = einsum(equation = var_28395_equation_0, values = (var_28249_cast_fp16, var_28357_cast_fp16))[name = tensor("op_28395_cast_fp16")]; + tensor var_28397_equation_0 = const()[name = tensor("op_28397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28397_cast_fp16 = einsum(equation = var_28397_equation_0, values = (var_28253_cast_fp16, var_28358_cast_fp16))[name = tensor("op_28397_cast_fp16")]; + tensor var_28399_equation_0 = const()[name = tensor("op_28399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28399_cast_fp16 = einsum(equation = var_28399_equation_0, values = (var_28257_cast_fp16, var_28359_cast_fp16))[name = tensor("op_28399_cast_fp16")]; + tensor input_385_interleave_0 = const()[name = tensor("input_385_interleave_0"), val = tensor(false)]; + tensor input_385_cast_fp16 = concat(axis = var_21077, interleave = input_385_interleave_0, values = (var_28361_cast_fp16, var_28363_cast_fp16, var_28365_cast_fp16, var_28367_cast_fp16, var_28369_cast_fp16, var_28371_cast_fp16, var_28373_cast_fp16, var_28375_cast_fp16, var_28377_cast_fp16, var_28379_cast_fp16, var_28381_cast_fp16, var_28383_cast_fp16, var_28385_cast_fp16, var_28387_cast_fp16, var_28389_cast_fp16, var_28391_cast_fp16, var_28393_cast_fp16, var_28395_cast_fp16, var_28397_cast_fp16, var_28399_cast_fp16))[name = tensor("input_385_cast_fp16")]; + tensor var_28409_pad_type_0 = const()[name = tensor("op_28409_pad_type_0"), val = tensor("valid")]; + tensor var_28409_strides_0 = const()[name = tensor("op_28409_strides_0"), val = tensor([1, 1])]; + tensor var_28409_pad_0 = const()[name = tensor("op_28409_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28409_dilations_0 = const()[name = tensor("op_28409_dilations_0"), val = tensor([1, 1])]; + tensor var_28409_groups_0 = const()[name = tensor("op_28409_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840827456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842056320))), name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842056512)))]; + tensor var_28409_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_28409_dilations_0, groups = var_28409_groups_0, pad = var_28409_pad_0, pad_type = var_28409_pad_type_0, strides = var_28409_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized, x = input_385_cast_fp16)[name = tensor("op_28409_cast_fp16")]; + tensor inputs_191_cast_fp16 = add(x = var_28409_cast_fp16, y = inputs_189_cast_fp16)[name = tensor("inputs_191_cast_fp16")]; + tensor input_387_axes_0 = const()[name = tensor("input_387_axes_0"), val = tensor([1])]; + tensor input_387_gamma_0_to_fp16 = const()[name = tensor("input_387_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842059136)))]; + tensor input_387_beta_0_to_fp16 = const()[name = tensor("input_387_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842061760)))]; + tensor var_28419_to_fp16 = const()[name = tensor("op_28419_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_387_cast_fp16 = layer_norm(axes = input_387_axes_0, beta = input_387_beta_0_to_fp16, epsilon = var_28419_to_fp16, gamma = input_387_gamma_0_to_fp16, x = inputs_191_cast_fp16)[name = tensor("input_387_cast_fp16")]; + tensor var_28439_pad_type_0 = const()[name = tensor("op_28439_pad_type_0"), val = tensor("valid")]; + tensor var_28439_strides_0 = const()[name = tensor("op_28439_strides_0"), val = tensor([1, 1])]; + tensor var_28439_pad_0 = const()[name = tensor("op_28439_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28439_dilations_0 = const()[name = tensor("op_28439_dilations_0"), val = tensor([1, 1])]; + tensor var_28439_groups_0 = const()[name = tensor("op_28439_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842064384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(851894848))), name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(851895040)))]; + tensor var_28439_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_28439_dilations_0, groups = var_28439_groups_0, pad = var_28439_pad_0, pad_type = var_28439_pad_type_0, strides = var_28439_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized, x = input_387_cast_fp16)[name = tensor("op_28439_cast_fp16")]; + tensor var_28440_split_sizes_0 = const()[name = tensor("op_28440_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_28440_axis_0 = const()[name = tensor("op_28440_axis_0"), val = tensor(1)]; + tensor var_28440_cast_fp16_0, tensor var_28440_cast_fp16_1 = split(axis = var_28440_axis_0, split_sizes = var_28440_split_sizes_0, x = var_28439_cast_fp16)[name = tensor("op_28440_cast_fp16")]; + tensor var_28442_mode_0 = const()[name = tensor("op_28442_mode_0"), val = tensor("EXACT")]; + tensor var_28442_cast_fp16 = gelu(mode = var_28442_mode_0, x = var_28440_cast_fp16_1)[name = tensor("op_28442_cast_fp16")]; + tensor input_389_cast_fp16 = mul(x = var_28440_cast_fp16_0, y = var_28442_cast_fp16)[name = tensor("input_389_cast_fp16")]; + tensor var_28450_pad_type_0 = const()[name = tensor("op_28450_pad_type_0"), val = tensor("valid")]; + tensor var_28450_strides_0 = const()[name = tensor("op_28450_strides_0"), val = tensor([1, 1])]; + tensor var_28450_pad_0 = const()[name = tensor("op_28450_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28450_dilations_0 = const()[name = tensor("op_28450_dilations_0"), val = tensor([1, 1])]; + tensor var_28450_groups_0 = const()[name = tensor("op_28450_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(851915584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(856830848))), name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(856831040)))]; + tensor var_28450_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_28450_dilations_0, groups = var_28450_groups_0, pad = var_28450_pad_0, pad_type = var_28450_pad_type_0, strides = var_28450_strides_0, weight = mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized, x = input_389_cast_fp16)[name = tensor("op_28450_cast_fp16")]; + tensor inputs_193_cast_fp16 = add(x = var_28450_cast_fp16, y = inputs_191_cast_fp16)[name = tensor("inputs_193_cast_fp16")]; + tensor hidden_states_257_axes_0 = const()[name = tensor("hidden_states_257_axes_0"), val = tensor([1])]; + tensor hidden_states_257_gamma_0_to_fp16 = const()[name = tensor("hidden_states_257_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(856833664)))]; + tensor hidden_states_257_beta_0_to_fp16 = const()[name = tensor("hidden_states_257_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(856836288)))]; + tensor var_28466_to_fp16 = const()[name = tensor("op_28466_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_257_cast_fp16 = layer_norm(axes = hidden_states_257_axes_0, beta = hidden_states_257_beta_0_to_fp16, epsilon = var_28466_to_fp16, gamma = hidden_states_257_gamma_0_to_fp16, x = inputs_193_cast_fp16)[name = tensor("hidden_states_257_cast_fp16")]; + tensor q_129_pad_type_0 = const()[name = tensor("q_129_pad_type_0"), val = tensor("valid")]; + tensor q_129_strides_0 = const()[name = tensor("q_129_strides_0"), val = tensor([1, 1])]; + tensor q_129_pad_0 = const()[name = tensor("q_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_129_dilations_0 = const()[name = tensor("q_129_dilations_0"), val = tensor([1, 1])]; + tensor q_129_groups_0 = const()[name = tensor("q_129_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(856838912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(858067776))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_129_cast_fp16 = conv(dilations = q_129_dilations_0, groups = q_129_groups_0, pad = q_129_pad_0, pad_type = q_129_pad_type_0, strides = q_129_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_257_cast_fp16)[name = tensor("q_129_cast_fp16")]; + tensor k_257_pad_type_0 = const()[name = tensor("k_257_pad_type_0"), val = tensor("valid")]; + tensor k_257_strides_0 = const()[name = tensor("k_257_strides_0"), val = tensor([1, 1])]; + tensor k_257_pad_0 = const()[name = tensor("k_257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_257_dilations_0 = const()[name = tensor("k_257_dilations_0"), val = tensor([1, 1])]; + tensor k_257_groups_0 = const()[name = tensor("k_257_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(858067968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(859296832))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_257_cast_fp16 = conv(dilations = k_257_dilations_0, groups = k_257_groups_0, pad = k_257_pad_0, pad_type = k_257_pad_type_0, strides = k_257_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_257_cast_fp16)[name = tensor("k_257_cast_fp16")]; + tensor v_129_pad_type_0 = const()[name = tensor("v_129_pad_type_0"), val = tensor("valid")]; + tensor v_129_strides_0 = const()[name = tensor("v_129_strides_0"), val = tensor([1, 1])]; + tensor v_129_pad_0 = const()[name = tensor("v_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_129_dilations_0 = const()[name = tensor("v_129_dilations_0"), val = tensor([1, 1])]; + tensor v_129_groups_0 = const()[name = tensor("v_129_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(859297024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(860525888))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_129_cast_fp16 = conv(dilations = v_129_dilations_0, groups = v_129_groups_0, pad = v_129_pad_0, pad_type = v_129_pad_type_0, strides = v_129_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_257_cast_fp16)[name = tensor("v_129_cast_fp16")]; + tensor var_28499_begin_0 = const()[name = tensor("op_28499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28499_end_0 = const()[name = tensor("op_28499_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28499_end_mask_0 = const()[name = tensor("op_28499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28499_cast_fp16 = slice_by_index(begin = var_28499_begin_0, end = var_28499_end_0, end_mask = var_28499_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28499_cast_fp16")]; + tensor var_28503_begin_0 = const()[name = tensor("op_28503_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28503_end_0 = const()[name = tensor("op_28503_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28503_end_mask_0 = const()[name = tensor("op_28503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28503_cast_fp16 = slice_by_index(begin = var_28503_begin_0, end = var_28503_end_0, end_mask = var_28503_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28503_cast_fp16")]; + tensor var_28507_begin_0 = const()[name = tensor("op_28507_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28507_end_0 = const()[name = tensor("op_28507_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28507_end_mask_0 = const()[name = tensor("op_28507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28507_cast_fp16 = slice_by_index(begin = var_28507_begin_0, end = var_28507_end_0, end_mask = var_28507_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28507_cast_fp16")]; + tensor var_28511_begin_0 = const()[name = tensor("op_28511_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28511_end_0 = const()[name = tensor("op_28511_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28511_end_mask_0 = const()[name = tensor("op_28511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28511_cast_fp16 = slice_by_index(begin = var_28511_begin_0, end = var_28511_end_0, end_mask = var_28511_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28511_cast_fp16")]; + tensor var_28515_begin_0 = const()[name = tensor("op_28515_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28515_end_0 = const()[name = tensor("op_28515_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28515_end_mask_0 = const()[name = tensor("op_28515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28515_cast_fp16 = slice_by_index(begin = var_28515_begin_0, end = var_28515_end_0, end_mask = var_28515_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28515_cast_fp16")]; + tensor var_28519_begin_0 = const()[name = tensor("op_28519_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28519_end_0 = const()[name = tensor("op_28519_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28519_end_mask_0 = const()[name = tensor("op_28519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28519_cast_fp16 = slice_by_index(begin = var_28519_begin_0, end = var_28519_end_0, end_mask = var_28519_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28519_cast_fp16")]; + tensor var_28523_begin_0 = const()[name = tensor("op_28523_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28523_end_0 = const()[name = tensor("op_28523_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28523_end_mask_0 = const()[name = tensor("op_28523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28523_cast_fp16 = slice_by_index(begin = var_28523_begin_0, end = var_28523_end_0, end_mask = var_28523_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28523_cast_fp16")]; + tensor var_28527_begin_0 = const()[name = tensor("op_28527_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28527_end_0 = const()[name = tensor("op_28527_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28527_end_mask_0 = const()[name = tensor("op_28527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28527_cast_fp16 = slice_by_index(begin = var_28527_begin_0, end = var_28527_end_0, end_mask = var_28527_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28527_cast_fp16")]; + tensor var_28531_begin_0 = const()[name = tensor("op_28531_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28531_end_0 = const()[name = tensor("op_28531_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28531_end_mask_0 = const()[name = tensor("op_28531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28531_cast_fp16 = slice_by_index(begin = var_28531_begin_0, end = var_28531_end_0, end_mask = var_28531_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28531_cast_fp16")]; + tensor var_28535_begin_0 = const()[name = tensor("op_28535_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28535_end_0 = const()[name = tensor("op_28535_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28535_end_mask_0 = const()[name = tensor("op_28535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28535_cast_fp16 = slice_by_index(begin = var_28535_begin_0, end = var_28535_end_0, end_mask = var_28535_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28535_cast_fp16")]; + tensor var_28539_begin_0 = const()[name = tensor("op_28539_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28539_end_0 = const()[name = tensor("op_28539_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28539_end_mask_0 = const()[name = tensor("op_28539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28539_cast_fp16 = slice_by_index(begin = var_28539_begin_0, end = var_28539_end_0, end_mask = var_28539_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28539_cast_fp16")]; + tensor var_28543_begin_0 = const()[name = tensor("op_28543_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28543_end_0 = const()[name = tensor("op_28543_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28543_end_mask_0 = const()[name = tensor("op_28543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28543_cast_fp16 = slice_by_index(begin = var_28543_begin_0, end = var_28543_end_0, end_mask = var_28543_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28543_cast_fp16")]; + tensor var_28547_begin_0 = const()[name = tensor("op_28547_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28547_end_0 = const()[name = tensor("op_28547_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28547_end_mask_0 = const()[name = tensor("op_28547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28547_cast_fp16 = slice_by_index(begin = var_28547_begin_0, end = var_28547_end_0, end_mask = var_28547_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28547_cast_fp16")]; + tensor var_28551_begin_0 = const()[name = tensor("op_28551_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28551_end_0 = const()[name = tensor("op_28551_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28551_end_mask_0 = const()[name = tensor("op_28551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28551_cast_fp16 = slice_by_index(begin = var_28551_begin_0, end = var_28551_end_0, end_mask = var_28551_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28551_cast_fp16")]; + tensor var_28555_begin_0 = const()[name = tensor("op_28555_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28555_end_0 = const()[name = tensor("op_28555_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28555_end_mask_0 = const()[name = tensor("op_28555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28555_cast_fp16 = slice_by_index(begin = var_28555_begin_0, end = var_28555_end_0, end_mask = var_28555_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28555_cast_fp16")]; + tensor var_28559_begin_0 = const()[name = tensor("op_28559_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28559_end_0 = const()[name = tensor("op_28559_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28559_end_mask_0 = const()[name = tensor("op_28559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28559_cast_fp16 = slice_by_index(begin = var_28559_begin_0, end = var_28559_end_0, end_mask = var_28559_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28559_cast_fp16")]; + tensor var_28563_begin_0 = const()[name = tensor("op_28563_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28563_end_0 = const()[name = tensor("op_28563_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28563_end_mask_0 = const()[name = tensor("op_28563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28563_cast_fp16 = slice_by_index(begin = var_28563_begin_0, end = var_28563_end_0, end_mask = var_28563_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28563_cast_fp16")]; + tensor var_28567_begin_0 = const()[name = tensor("op_28567_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28567_end_0 = const()[name = tensor("op_28567_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28567_end_mask_0 = const()[name = tensor("op_28567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28567_cast_fp16 = slice_by_index(begin = var_28567_begin_0, end = var_28567_end_0, end_mask = var_28567_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28567_cast_fp16")]; + tensor var_28571_begin_0 = const()[name = tensor("op_28571_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28571_end_0 = const()[name = tensor("op_28571_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_28571_end_mask_0 = const()[name = tensor("op_28571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28571_cast_fp16 = slice_by_index(begin = var_28571_begin_0, end = var_28571_end_0, end_mask = var_28571_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28571_cast_fp16")]; + tensor var_28575_begin_0 = const()[name = tensor("op_28575_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28575_end_0 = const()[name = tensor("op_28575_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_28575_end_mask_0 = const()[name = tensor("op_28575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28575_cast_fp16 = slice_by_index(begin = var_28575_begin_0, end = var_28575_end_0, end_mask = var_28575_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28575_cast_fp16")]; + tensor k_259_perm_0 = const()[name = tensor("k_259_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_28582_begin_0 = const()[name = tensor("op_28582_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28582_end_0 = const()[name = tensor("op_28582_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_28582_end_mask_0 = const()[name = tensor("op_28582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_259_cast_fp16 = transpose(perm = k_259_perm_0, x = k_257_cast_fp16)[name = tensor("transpose_3")]; + tensor var_28582_cast_fp16 = slice_by_index(begin = var_28582_begin_0, end = var_28582_end_0, end_mask = var_28582_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28582_cast_fp16")]; + tensor var_28586_begin_0 = const()[name = tensor("op_28586_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_28586_end_0 = const()[name = tensor("op_28586_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_28586_end_mask_0 = const()[name = tensor("op_28586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28586_cast_fp16 = slice_by_index(begin = var_28586_begin_0, end = var_28586_end_0, end_mask = var_28586_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28586_cast_fp16")]; + tensor var_28590_begin_0 = const()[name = tensor("op_28590_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_28590_end_0 = const()[name = tensor("op_28590_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_28590_end_mask_0 = const()[name = tensor("op_28590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28590_cast_fp16 = slice_by_index(begin = var_28590_begin_0, end = var_28590_end_0, end_mask = var_28590_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28590_cast_fp16")]; + tensor var_28594_begin_0 = const()[name = tensor("op_28594_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_28594_end_0 = const()[name = tensor("op_28594_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_28594_end_mask_0 = const()[name = tensor("op_28594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28594_cast_fp16 = slice_by_index(begin = var_28594_begin_0, end = var_28594_end_0, end_mask = var_28594_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28594_cast_fp16")]; + tensor var_28598_begin_0 = const()[name = tensor("op_28598_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_28598_end_0 = const()[name = tensor("op_28598_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_28598_end_mask_0 = const()[name = tensor("op_28598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28598_cast_fp16 = slice_by_index(begin = var_28598_begin_0, end = var_28598_end_0, end_mask = var_28598_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28598_cast_fp16")]; + tensor var_28602_begin_0 = const()[name = tensor("op_28602_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_28602_end_0 = const()[name = tensor("op_28602_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_28602_end_mask_0 = const()[name = tensor("op_28602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28602_cast_fp16 = slice_by_index(begin = var_28602_begin_0, end = var_28602_end_0, end_mask = var_28602_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28602_cast_fp16")]; + tensor var_28606_begin_0 = const()[name = tensor("op_28606_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_28606_end_0 = const()[name = tensor("op_28606_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_28606_end_mask_0 = const()[name = tensor("op_28606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28606_cast_fp16 = slice_by_index(begin = var_28606_begin_0, end = var_28606_end_0, end_mask = var_28606_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28606_cast_fp16")]; + tensor var_28610_begin_0 = const()[name = tensor("op_28610_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_28610_end_0 = const()[name = tensor("op_28610_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_28610_end_mask_0 = const()[name = tensor("op_28610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28610_cast_fp16 = slice_by_index(begin = var_28610_begin_0, end = var_28610_end_0, end_mask = var_28610_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28610_cast_fp16")]; + tensor var_28614_begin_0 = const()[name = tensor("op_28614_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_28614_end_0 = const()[name = tensor("op_28614_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_28614_end_mask_0 = const()[name = tensor("op_28614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28614_cast_fp16 = slice_by_index(begin = var_28614_begin_0, end = var_28614_end_0, end_mask = var_28614_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28614_cast_fp16")]; + tensor var_28618_begin_0 = const()[name = tensor("op_28618_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_28618_end_0 = const()[name = tensor("op_28618_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_28618_end_mask_0 = const()[name = tensor("op_28618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28618_cast_fp16 = slice_by_index(begin = var_28618_begin_0, end = var_28618_end_0, end_mask = var_28618_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28618_cast_fp16")]; + tensor var_28622_begin_0 = const()[name = tensor("op_28622_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_28622_end_0 = const()[name = tensor("op_28622_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_28622_end_mask_0 = const()[name = tensor("op_28622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28622_cast_fp16 = slice_by_index(begin = var_28622_begin_0, end = var_28622_end_0, end_mask = var_28622_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28622_cast_fp16")]; + tensor var_28626_begin_0 = const()[name = tensor("op_28626_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_28626_end_0 = const()[name = tensor("op_28626_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_28626_end_mask_0 = const()[name = tensor("op_28626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28626_cast_fp16 = slice_by_index(begin = var_28626_begin_0, end = var_28626_end_0, end_mask = var_28626_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28626_cast_fp16")]; + tensor var_28630_begin_0 = const()[name = tensor("op_28630_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_28630_end_0 = const()[name = tensor("op_28630_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_28630_end_mask_0 = const()[name = tensor("op_28630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28630_cast_fp16 = slice_by_index(begin = var_28630_begin_0, end = var_28630_end_0, end_mask = var_28630_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28630_cast_fp16")]; + tensor var_28634_begin_0 = const()[name = tensor("op_28634_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_28634_end_0 = const()[name = tensor("op_28634_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_28634_end_mask_0 = const()[name = tensor("op_28634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28634_cast_fp16 = slice_by_index(begin = var_28634_begin_0, end = var_28634_end_0, end_mask = var_28634_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28634_cast_fp16")]; + tensor var_28638_begin_0 = const()[name = tensor("op_28638_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_28638_end_0 = const()[name = tensor("op_28638_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_28638_end_mask_0 = const()[name = tensor("op_28638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28638_cast_fp16 = slice_by_index(begin = var_28638_begin_0, end = var_28638_end_0, end_mask = var_28638_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28638_cast_fp16")]; + tensor var_28642_begin_0 = const()[name = tensor("op_28642_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_28642_end_0 = const()[name = tensor("op_28642_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28642_end_mask_0 = const()[name = tensor("op_28642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28642_cast_fp16 = slice_by_index(begin = var_28642_begin_0, end = var_28642_end_0, end_mask = var_28642_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28642_cast_fp16")]; + tensor var_28646_begin_0 = const()[name = tensor("op_28646_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_28646_end_0 = const()[name = tensor("op_28646_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_28646_end_mask_0 = const()[name = tensor("op_28646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28646_cast_fp16 = slice_by_index(begin = var_28646_begin_0, end = var_28646_end_0, end_mask = var_28646_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28646_cast_fp16")]; + tensor var_28650_begin_0 = const()[name = tensor("op_28650_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_28650_end_0 = const()[name = tensor("op_28650_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_28650_end_mask_0 = const()[name = tensor("op_28650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28650_cast_fp16 = slice_by_index(begin = var_28650_begin_0, end = var_28650_end_0, end_mask = var_28650_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28650_cast_fp16")]; + tensor var_28654_begin_0 = const()[name = tensor("op_28654_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_28654_end_0 = const()[name = tensor("op_28654_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_28654_end_mask_0 = const()[name = tensor("op_28654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28654_cast_fp16 = slice_by_index(begin = var_28654_begin_0, end = var_28654_end_0, end_mask = var_28654_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28654_cast_fp16")]; + tensor var_28658_begin_0 = const()[name = tensor("op_28658_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_28658_end_0 = const()[name = tensor("op_28658_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_28658_end_mask_0 = const()[name = tensor("op_28658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28658_cast_fp16 = slice_by_index(begin = var_28658_begin_0, end = var_28658_end_0, end_mask = var_28658_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28658_cast_fp16")]; + tensor var_28660_begin_0 = const()[name = tensor("op_28660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28660_end_0 = const()[name = tensor("op_28660_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28660_end_mask_0 = const()[name = tensor("op_28660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28660_cast_fp16 = slice_by_index(begin = var_28660_begin_0, end = var_28660_end_0, end_mask = var_28660_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28660_cast_fp16")]; + tensor var_28664_begin_0 = const()[name = tensor("op_28664_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28664_end_0 = const()[name = tensor("op_28664_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28664_end_mask_0 = const()[name = tensor("op_28664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28664_cast_fp16 = slice_by_index(begin = var_28664_begin_0, end = var_28664_end_0, end_mask = var_28664_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28664_cast_fp16")]; + tensor var_28668_begin_0 = const()[name = tensor("op_28668_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28668_end_0 = const()[name = tensor("op_28668_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28668_end_mask_0 = const()[name = tensor("op_28668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28668_cast_fp16 = slice_by_index(begin = var_28668_begin_0, end = var_28668_end_0, end_mask = var_28668_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28668_cast_fp16")]; + tensor var_28672_begin_0 = const()[name = tensor("op_28672_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28672_end_0 = const()[name = tensor("op_28672_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28672_end_mask_0 = const()[name = tensor("op_28672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28672_cast_fp16 = slice_by_index(begin = var_28672_begin_0, end = var_28672_end_0, end_mask = var_28672_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28672_cast_fp16")]; + tensor var_28676_begin_0 = const()[name = tensor("op_28676_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28676_end_0 = const()[name = tensor("op_28676_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28676_end_mask_0 = const()[name = tensor("op_28676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28676_cast_fp16 = slice_by_index(begin = var_28676_begin_0, end = var_28676_end_0, end_mask = var_28676_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28676_cast_fp16")]; + tensor var_28680_begin_0 = const()[name = tensor("op_28680_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28680_end_0 = const()[name = tensor("op_28680_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28680_end_mask_0 = const()[name = tensor("op_28680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28680_cast_fp16 = slice_by_index(begin = var_28680_begin_0, end = var_28680_end_0, end_mask = var_28680_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28680_cast_fp16")]; + tensor var_28684_begin_0 = const()[name = tensor("op_28684_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28684_end_0 = const()[name = tensor("op_28684_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28684_end_mask_0 = const()[name = tensor("op_28684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28684_cast_fp16 = slice_by_index(begin = var_28684_begin_0, end = var_28684_end_0, end_mask = var_28684_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28684_cast_fp16")]; + tensor var_28688_begin_0 = const()[name = tensor("op_28688_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28688_end_0 = const()[name = tensor("op_28688_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28688_end_mask_0 = const()[name = tensor("op_28688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28688_cast_fp16 = slice_by_index(begin = var_28688_begin_0, end = var_28688_end_0, end_mask = var_28688_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28688_cast_fp16")]; + tensor var_28692_begin_0 = const()[name = tensor("op_28692_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28692_end_0 = const()[name = tensor("op_28692_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28692_end_mask_0 = const()[name = tensor("op_28692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28692_cast_fp16 = slice_by_index(begin = var_28692_begin_0, end = var_28692_end_0, end_mask = var_28692_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28692_cast_fp16")]; + tensor var_28696_begin_0 = const()[name = tensor("op_28696_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28696_end_0 = const()[name = tensor("op_28696_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28696_end_mask_0 = const()[name = tensor("op_28696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28696_cast_fp16 = slice_by_index(begin = var_28696_begin_0, end = var_28696_end_0, end_mask = var_28696_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28696_cast_fp16")]; + tensor var_28700_begin_0 = const()[name = tensor("op_28700_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28700_end_0 = const()[name = tensor("op_28700_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28700_end_mask_0 = const()[name = tensor("op_28700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28700_cast_fp16 = slice_by_index(begin = var_28700_begin_0, end = var_28700_end_0, end_mask = var_28700_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28700_cast_fp16")]; + tensor var_28704_begin_0 = const()[name = tensor("op_28704_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28704_end_0 = const()[name = tensor("op_28704_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28704_end_mask_0 = const()[name = tensor("op_28704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28704_cast_fp16 = slice_by_index(begin = var_28704_begin_0, end = var_28704_end_0, end_mask = var_28704_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28704_cast_fp16")]; + tensor var_28708_begin_0 = const()[name = tensor("op_28708_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28708_end_0 = const()[name = tensor("op_28708_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28708_end_mask_0 = const()[name = tensor("op_28708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28708_cast_fp16 = slice_by_index(begin = var_28708_begin_0, end = var_28708_end_0, end_mask = var_28708_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28708_cast_fp16")]; + tensor var_28712_begin_0 = const()[name = tensor("op_28712_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28712_end_0 = const()[name = tensor("op_28712_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28712_end_mask_0 = const()[name = tensor("op_28712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28712_cast_fp16 = slice_by_index(begin = var_28712_begin_0, end = var_28712_end_0, end_mask = var_28712_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28712_cast_fp16")]; + tensor var_28716_begin_0 = const()[name = tensor("op_28716_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28716_end_0 = const()[name = tensor("op_28716_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28716_end_mask_0 = const()[name = tensor("op_28716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28716_cast_fp16 = slice_by_index(begin = var_28716_begin_0, end = var_28716_end_0, end_mask = var_28716_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28716_cast_fp16")]; + tensor var_28720_begin_0 = const()[name = tensor("op_28720_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28720_end_0 = const()[name = tensor("op_28720_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28720_end_mask_0 = const()[name = tensor("op_28720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28720_cast_fp16 = slice_by_index(begin = var_28720_begin_0, end = var_28720_end_0, end_mask = var_28720_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28720_cast_fp16")]; + tensor var_28724_begin_0 = const()[name = tensor("op_28724_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28724_end_0 = const()[name = tensor("op_28724_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28724_end_mask_0 = const()[name = tensor("op_28724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28724_cast_fp16 = slice_by_index(begin = var_28724_begin_0, end = var_28724_end_0, end_mask = var_28724_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28724_cast_fp16")]; + tensor var_28728_begin_0 = const()[name = tensor("op_28728_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28728_end_0 = const()[name = tensor("op_28728_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28728_end_mask_0 = const()[name = tensor("op_28728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28728_cast_fp16 = slice_by_index(begin = var_28728_begin_0, end = var_28728_end_0, end_mask = var_28728_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28728_cast_fp16")]; + tensor var_28732_begin_0 = const()[name = tensor("op_28732_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28732_end_0 = const()[name = tensor("op_28732_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_28732_end_mask_0 = const()[name = tensor("op_28732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28732_cast_fp16 = slice_by_index(begin = var_28732_begin_0, end = var_28732_end_0, end_mask = var_28732_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28732_cast_fp16")]; + tensor var_28736_begin_0 = const()[name = tensor("op_28736_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28736_end_0 = const()[name = tensor("op_28736_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_28736_end_mask_0 = const()[name = tensor("op_28736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28736_cast_fp16 = slice_by_index(begin = var_28736_begin_0, end = var_28736_end_0, end_mask = var_28736_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28736_cast_fp16")]; + tensor var_28740_equation_0 = const()[name = tensor("op_28740_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28740_cast_fp16 = einsum(equation = var_28740_equation_0, values = (var_28582_cast_fp16, var_28499_cast_fp16))[name = tensor("op_28740_cast_fp16")]; + tensor var_28741_to_fp16 = const()[name = tensor("op_28741_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2401_cast_fp16 = mul(x = var_28740_cast_fp16, y = var_28741_to_fp16)[name = tensor("aw_2401_cast_fp16")]; + tensor var_28744_equation_0 = const()[name = tensor("op_28744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28744_cast_fp16 = einsum(equation = var_28744_equation_0, values = (var_28586_cast_fp16, var_28503_cast_fp16))[name = tensor("op_28744_cast_fp16")]; + tensor var_28745_to_fp16 = const()[name = tensor("op_28745_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2403_cast_fp16 = mul(x = var_28744_cast_fp16, y = var_28745_to_fp16)[name = tensor("aw_2403_cast_fp16")]; + tensor var_28748_equation_0 = const()[name = tensor("op_28748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28748_cast_fp16 = einsum(equation = var_28748_equation_0, values = (var_28590_cast_fp16, var_28507_cast_fp16))[name = tensor("op_28748_cast_fp16")]; + tensor var_28749_to_fp16 = const()[name = tensor("op_28749_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2405_cast_fp16 = mul(x = var_28748_cast_fp16, y = var_28749_to_fp16)[name = tensor("aw_2405_cast_fp16")]; + tensor var_28752_equation_0 = const()[name = tensor("op_28752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28752_cast_fp16 = einsum(equation = var_28752_equation_0, values = (var_28594_cast_fp16, var_28511_cast_fp16))[name = tensor("op_28752_cast_fp16")]; + tensor var_28753_to_fp16 = const()[name = tensor("op_28753_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2407_cast_fp16 = mul(x = var_28752_cast_fp16, y = var_28753_to_fp16)[name = tensor("aw_2407_cast_fp16")]; + tensor var_28756_equation_0 = const()[name = tensor("op_28756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28756_cast_fp16 = einsum(equation = var_28756_equation_0, values = (var_28598_cast_fp16, var_28515_cast_fp16))[name = tensor("op_28756_cast_fp16")]; + tensor var_28757_to_fp16 = const()[name = tensor("op_28757_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2409_cast_fp16 = mul(x = var_28756_cast_fp16, y = var_28757_to_fp16)[name = tensor("aw_2409_cast_fp16")]; + tensor var_28760_equation_0 = const()[name = tensor("op_28760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28760_cast_fp16 = einsum(equation = var_28760_equation_0, values = (var_28602_cast_fp16, var_28519_cast_fp16))[name = tensor("op_28760_cast_fp16")]; + tensor var_28761_to_fp16 = const()[name = tensor("op_28761_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2411_cast_fp16 = mul(x = var_28760_cast_fp16, y = var_28761_to_fp16)[name = tensor("aw_2411_cast_fp16")]; + tensor var_28764_equation_0 = const()[name = tensor("op_28764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28764_cast_fp16 = einsum(equation = var_28764_equation_0, values = (var_28606_cast_fp16, var_28523_cast_fp16))[name = tensor("op_28764_cast_fp16")]; + tensor var_28765_to_fp16 = const()[name = tensor("op_28765_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2413_cast_fp16 = mul(x = var_28764_cast_fp16, y = var_28765_to_fp16)[name = tensor("aw_2413_cast_fp16")]; + tensor var_28768_equation_0 = const()[name = tensor("op_28768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28768_cast_fp16 = einsum(equation = var_28768_equation_0, values = (var_28610_cast_fp16, var_28527_cast_fp16))[name = tensor("op_28768_cast_fp16")]; + tensor var_28769_to_fp16 = const()[name = tensor("op_28769_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2415_cast_fp16 = mul(x = var_28768_cast_fp16, y = var_28769_to_fp16)[name = tensor("aw_2415_cast_fp16")]; + tensor var_28772_equation_0 = const()[name = tensor("op_28772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28772_cast_fp16 = einsum(equation = var_28772_equation_0, values = (var_28614_cast_fp16, var_28531_cast_fp16))[name = tensor("op_28772_cast_fp16")]; + tensor var_28773_to_fp16 = const()[name = tensor("op_28773_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2417_cast_fp16 = mul(x = var_28772_cast_fp16, y = var_28773_to_fp16)[name = tensor("aw_2417_cast_fp16")]; + tensor var_28776_equation_0 = const()[name = tensor("op_28776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28776_cast_fp16 = einsum(equation = var_28776_equation_0, values = (var_28618_cast_fp16, var_28535_cast_fp16))[name = tensor("op_28776_cast_fp16")]; + tensor var_28777_to_fp16 = const()[name = tensor("op_28777_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2419_cast_fp16 = mul(x = var_28776_cast_fp16, y = var_28777_to_fp16)[name = tensor("aw_2419_cast_fp16")]; + tensor var_28780_equation_0 = const()[name = tensor("op_28780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28780_cast_fp16 = einsum(equation = var_28780_equation_0, values = (var_28622_cast_fp16, var_28539_cast_fp16))[name = tensor("op_28780_cast_fp16")]; + tensor var_28781_to_fp16 = const()[name = tensor("op_28781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2421_cast_fp16 = mul(x = var_28780_cast_fp16, y = var_28781_to_fp16)[name = tensor("aw_2421_cast_fp16")]; + tensor var_28784_equation_0 = const()[name = tensor("op_28784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28784_cast_fp16 = einsum(equation = var_28784_equation_0, values = (var_28626_cast_fp16, var_28543_cast_fp16))[name = tensor("op_28784_cast_fp16")]; + tensor var_28785_to_fp16 = const()[name = tensor("op_28785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2423_cast_fp16 = mul(x = var_28784_cast_fp16, y = var_28785_to_fp16)[name = tensor("aw_2423_cast_fp16")]; + tensor var_28788_equation_0 = const()[name = tensor("op_28788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28788_cast_fp16 = einsum(equation = var_28788_equation_0, values = (var_28630_cast_fp16, var_28547_cast_fp16))[name = tensor("op_28788_cast_fp16")]; + tensor var_28789_to_fp16 = const()[name = tensor("op_28789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2425_cast_fp16 = mul(x = var_28788_cast_fp16, y = var_28789_to_fp16)[name = tensor("aw_2425_cast_fp16")]; + tensor var_28792_equation_0 = const()[name = tensor("op_28792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28792_cast_fp16 = einsum(equation = var_28792_equation_0, values = (var_28634_cast_fp16, var_28551_cast_fp16))[name = tensor("op_28792_cast_fp16")]; + tensor var_28793_to_fp16 = const()[name = tensor("op_28793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2427_cast_fp16 = mul(x = var_28792_cast_fp16, y = var_28793_to_fp16)[name = tensor("aw_2427_cast_fp16")]; + tensor var_28796_equation_0 = const()[name = tensor("op_28796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28796_cast_fp16 = einsum(equation = var_28796_equation_0, values = (var_28638_cast_fp16, var_28555_cast_fp16))[name = tensor("op_28796_cast_fp16")]; + tensor var_28797_to_fp16 = const()[name = tensor("op_28797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2429_cast_fp16 = mul(x = var_28796_cast_fp16, y = var_28797_to_fp16)[name = tensor("aw_2429_cast_fp16")]; + tensor var_28800_equation_0 = const()[name = tensor("op_28800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28800_cast_fp16 = einsum(equation = var_28800_equation_0, values = (var_28642_cast_fp16, var_28559_cast_fp16))[name = tensor("op_28800_cast_fp16")]; + tensor var_28801_to_fp16 = const()[name = tensor("op_28801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2431_cast_fp16 = mul(x = var_28800_cast_fp16, y = var_28801_to_fp16)[name = tensor("aw_2431_cast_fp16")]; + tensor var_28804_equation_0 = const()[name = tensor("op_28804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28804_cast_fp16 = einsum(equation = var_28804_equation_0, values = (var_28646_cast_fp16, var_28563_cast_fp16))[name = tensor("op_28804_cast_fp16")]; + tensor var_28805_to_fp16 = const()[name = tensor("op_28805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2433_cast_fp16 = mul(x = var_28804_cast_fp16, y = var_28805_to_fp16)[name = tensor("aw_2433_cast_fp16")]; + tensor var_28808_equation_0 = const()[name = tensor("op_28808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28808_cast_fp16 = einsum(equation = var_28808_equation_0, values = (var_28650_cast_fp16, var_28567_cast_fp16))[name = tensor("op_28808_cast_fp16")]; + tensor var_28809_to_fp16 = const()[name = tensor("op_28809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2435_cast_fp16 = mul(x = var_28808_cast_fp16, y = var_28809_to_fp16)[name = tensor("aw_2435_cast_fp16")]; + tensor var_28812_equation_0 = const()[name = tensor("op_28812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28812_cast_fp16 = einsum(equation = var_28812_equation_0, values = (var_28654_cast_fp16, var_28571_cast_fp16))[name = tensor("op_28812_cast_fp16")]; + tensor var_28813_to_fp16 = const()[name = tensor("op_28813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2437_cast_fp16 = mul(x = var_28812_cast_fp16, y = var_28813_to_fp16)[name = tensor("aw_2437_cast_fp16")]; + tensor var_28816_equation_0 = const()[name = tensor("op_28816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28816_cast_fp16 = einsum(equation = var_28816_equation_0, values = (var_28658_cast_fp16, var_28575_cast_fp16))[name = tensor("op_28816_cast_fp16")]; + tensor var_28817_to_fp16 = const()[name = tensor("op_28817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2439_cast_fp16 = mul(x = var_28816_cast_fp16, y = var_28817_to_fp16)[name = tensor("aw_2439_cast_fp16")]; + tensor var_28819_cast_fp16 = softmax(axis = var_21077, x = aw_2401_cast_fp16)[name = tensor("op_28819_cast_fp16")]; + tensor var_28820_cast_fp16 = softmax(axis = var_21077, x = aw_2403_cast_fp16)[name = tensor("op_28820_cast_fp16")]; + tensor var_28821_cast_fp16 = softmax(axis = var_21077, x = aw_2405_cast_fp16)[name = tensor("op_28821_cast_fp16")]; + tensor var_28822_cast_fp16 = softmax(axis = var_21077, x = aw_2407_cast_fp16)[name = tensor("op_28822_cast_fp16")]; + tensor var_28823_cast_fp16 = softmax(axis = var_21077, x = aw_2409_cast_fp16)[name = tensor("op_28823_cast_fp16")]; + tensor var_28824_cast_fp16 = softmax(axis = var_21077, x = aw_2411_cast_fp16)[name = tensor("op_28824_cast_fp16")]; + tensor var_28825_cast_fp16 = softmax(axis = var_21077, x = aw_2413_cast_fp16)[name = tensor("op_28825_cast_fp16")]; + tensor var_28826_cast_fp16 = softmax(axis = var_21077, x = aw_2415_cast_fp16)[name = tensor("op_28826_cast_fp16")]; + tensor var_28827_cast_fp16 = softmax(axis = var_21077, x = aw_2417_cast_fp16)[name = tensor("op_28827_cast_fp16")]; + tensor var_28828_cast_fp16 = softmax(axis = var_21077, x = aw_2419_cast_fp16)[name = tensor("op_28828_cast_fp16")]; + tensor var_28829_cast_fp16 = softmax(axis = var_21077, x = aw_2421_cast_fp16)[name = tensor("op_28829_cast_fp16")]; + tensor var_28830_cast_fp16 = softmax(axis = var_21077, x = aw_2423_cast_fp16)[name = tensor("op_28830_cast_fp16")]; + tensor var_28831_cast_fp16 = softmax(axis = var_21077, x = aw_2425_cast_fp16)[name = tensor("op_28831_cast_fp16")]; + tensor var_28832_cast_fp16 = softmax(axis = var_21077, x = aw_2427_cast_fp16)[name = tensor("op_28832_cast_fp16")]; + tensor var_28833_cast_fp16 = softmax(axis = var_21077, x = aw_2429_cast_fp16)[name = tensor("op_28833_cast_fp16")]; + tensor var_28834_cast_fp16 = softmax(axis = var_21077, x = aw_2431_cast_fp16)[name = tensor("op_28834_cast_fp16")]; + tensor var_28835_cast_fp16 = softmax(axis = var_21077, x = aw_2433_cast_fp16)[name = tensor("op_28835_cast_fp16")]; + tensor var_28836_cast_fp16 = softmax(axis = var_21077, x = aw_2435_cast_fp16)[name = tensor("op_28836_cast_fp16")]; + tensor var_28837_cast_fp16 = softmax(axis = var_21077, x = aw_2437_cast_fp16)[name = tensor("op_28837_cast_fp16")]; + tensor var_28838_cast_fp16 = softmax(axis = var_21077, x = aw_2439_cast_fp16)[name = tensor("op_28838_cast_fp16")]; + tensor var_28840_equation_0 = const()[name = tensor("op_28840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28840_cast_fp16 = einsum(equation = var_28840_equation_0, values = (var_28660_cast_fp16, var_28819_cast_fp16))[name = tensor("op_28840_cast_fp16")]; + tensor var_28842_equation_0 = const()[name = tensor("op_28842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28842_cast_fp16 = einsum(equation = var_28842_equation_0, values = (var_28664_cast_fp16, var_28820_cast_fp16))[name = tensor("op_28842_cast_fp16")]; + tensor var_28844_equation_0 = const()[name = tensor("op_28844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28844_cast_fp16 = einsum(equation = var_28844_equation_0, values = (var_28668_cast_fp16, var_28821_cast_fp16))[name = tensor("op_28844_cast_fp16")]; + tensor var_28846_equation_0 = const()[name = tensor("op_28846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28846_cast_fp16 = einsum(equation = var_28846_equation_0, values = (var_28672_cast_fp16, var_28822_cast_fp16))[name = tensor("op_28846_cast_fp16")]; + tensor var_28848_equation_0 = const()[name = tensor("op_28848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28848_cast_fp16 = einsum(equation = var_28848_equation_0, values = (var_28676_cast_fp16, var_28823_cast_fp16))[name = tensor("op_28848_cast_fp16")]; + tensor var_28850_equation_0 = const()[name = tensor("op_28850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28850_cast_fp16 = einsum(equation = var_28850_equation_0, values = (var_28680_cast_fp16, var_28824_cast_fp16))[name = tensor("op_28850_cast_fp16")]; + tensor var_28852_equation_0 = const()[name = tensor("op_28852_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28852_cast_fp16 = einsum(equation = var_28852_equation_0, values = (var_28684_cast_fp16, var_28825_cast_fp16))[name = tensor("op_28852_cast_fp16")]; + tensor var_28854_equation_0 = const()[name = tensor("op_28854_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28854_cast_fp16 = einsum(equation = var_28854_equation_0, values = (var_28688_cast_fp16, var_28826_cast_fp16))[name = tensor("op_28854_cast_fp16")]; + tensor var_28856_equation_0 = const()[name = tensor("op_28856_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28856_cast_fp16 = einsum(equation = var_28856_equation_0, values = (var_28692_cast_fp16, var_28827_cast_fp16))[name = tensor("op_28856_cast_fp16")]; + tensor var_28858_equation_0 = const()[name = tensor("op_28858_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28858_cast_fp16 = einsum(equation = var_28858_equation_0, values = (var_28696_cast_fp16, var_28828_cast_fp16))[name = tensor("op_28858_cast_fp16")]; + tensor var_28860_equation_0 = const()[name = tensor("op_28860_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28860_cast_fp16 = einsum(equation = var_28860_equation_0, values = (var_28700_cast_fp16, var_28829_cast_fp16))[name = tensor("op_28860_cast_fp16")]; + tensor var_28862_equation_0 = const()[name = tensor("op_28862_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28862_cast_fp16 = einsum(equation = var_28862_equation_0, values = (var_28704_cast_fp16, var_28830_cast_fp16))[name = tensor("op_28862_cast_fp16")]; + tensor var_28864_equation_0 = const()[name = tensor("op_28864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28864_cast_fp16 = einsum(equation = var_28864_equation_0, values = (var_28708_cast_fp16, var_28831_cast_fp16))[name = tensor("op_28864_cast_fp16")]; + tensor var_28866_equation_0 = const()[name = tensor("op_28866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28866_cast_fp16 = einsum(equation = var_28866_equation_0, values = (var_28712_cast_fp16, var_28832_cast_fp16))[name = tensor("op_28866_cast_fp16")]; + tensor var_28868_equation_0 = const()[name = tensor("op_28868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28868_cast_fp16 = einsum(equation = var_28868_equation_0, values = (var_28716_cast_fp16, var_28833_cast_fp16))[name = tensor("op_28868_cast_fp16")]; + tensor var_28870_equation_0 = const()[name = tensor("op_28870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28870_cast_fp16 = einsum(equation = var_28870_equation_0, values = (var_28720_cast_fp16, var_28834_cast_fp16))[name = tensor("op_28870_cast_fp16")]; + tensor var_28872_equation_0 = const()[name = tensor("op_28872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28872_cast_fp16 = einsum(equation = var_28872_equation_0, values = (var_28724_cast_fp16, var_28835_cast_fp16))[name = tensor("op_28872_cast_fp16")]; + tensor var_28874_equation_0 = const()[name = tensor("op_28874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28874_cast_fp16 = einsum(equation = var_28874_equation_0, values = (var_28728_cast_fp16, var_28836_cast_fp16))[name = tensor("op_28874_cast_fp16")]; + tensor var_28876_equation_0 = const()[name = tensor("op_28876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28876_cast_fp16 = einsum(equation = var_28876_equation_0, values = (var_28732_cast_fp16, var_28837_cast_fp16))[name = tensor("op_28876_cast_fp16")]; + tensor var_28878_equation_0 = const()[name = tensor("op_28878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28878_cast_fp16 = einsum(equation = var_28878_equation_0, values = (var_28736_cast_fp16, var_28838_cast_fp16))[name = tensor("op_28878_cast_fp16")]; + tensor input_391_interleave_0 = const()[name = tensor("input_391_interleave_0"), val = tensor(false)]; + tensor input_391_cast_fp16 = concat(axis = var_21077, interleave = input_391_interleave_0, values = (var_28840_cast_fp16, var_28842_cast_fp16, var_28844_cast_fp16, var_28846_cast_fp16, var_28848_cast_fp16, var_28850_cast_fp16, var_28852_cast_fp16, var_28854_cast_fp16, var_28856_cast_fp16, var_28858_cast_fp16, var_28860_cast_fp16, var_28862_cast_fp16, var_28864_cast_fp16, var_28866_cast_fp16, var_28868_cast_fp16, var_28870_cast_fp16, var_28872_cast_fp16, var_28874_cast_fp16, var_28876_cast_fp16, var_28878_cast_fp16))[name = tensor("input_391_cast_fp16")]; + tensor var_28888_pad_type_0 = const()[name = tensor("op_28888_pad_type_0"), val = tensor("valid")]; + tensor var_28888_strides_0 = const()[name = tensor("op_28888_strides_0"), val = tensor([1, 1])]; + tensor var_28888_pad_0 = const()[name = tensor("op_28888_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28888_dilations_0 = const()[name = tensor("op_28888_dilations_0"), val = tensor([1, 1])]; + tensor var_28888_groups_0 = const()[name = tensor("op_28888_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(860526080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(861754944))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(861755136)))]; + tensor var_28888_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_28888_dilations_0, groups = var_28888_groups_0, pad = var_28888_pad_0, pad_type = var_28888_pad_type_0, strides = var_28888_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized, x = input_391_cast_fp16)[name = tensor("op_28888_cast_fp16")]; + tensor inputs_195_cast_fp16 = add(x = var_28888_cast_fp16, y = inputs_193_cast_fp16)[name = tensor("inputs_195_cast_fp16")]; + tensor hidden_states_259_axes_0 = const()[name = tensor("hidden_states_259_axes_0"), val = tensor([1])]; + tensor hidden_states_259_gamma_0_to_fp16 = const()[name = tensor("hidden_states_259_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(861757760)))]; + tensor hidden_states_259_beta_0_to_fp16 = const()[name = tensor("hidden_states_259_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(861760384)))]; + tensor var_28898_to_fp16 = const()[name = tensor("op_28898_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_259_cast_fp16 = layer_norm(axes = hidden_states_259_axes_0, beta = hidden_states_259_beta_0_to_fp16, epsilon = var_28898_to_fp16, gamma = hidden_states_259_gamma_0_to_fp16, x = inputs_195_cast_fp16)[name = tensor("hidden_states_259_cast_fp16")]; + tensor q_131_pad_type_0 = const()[name = tensor("q_131_pad_type_0"), val = tensor("valid")]; + tensor q_131_strides_0 = const()[name = tensor("q_131_strides_0"), val = tensor([1, 1])]; + tensor q_131_pad_0 = const()[name = tensor("q_131_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_131_dilations_0 = const()[name = tensor("q_131_dilations_0"), val = tensor([1, 1])]; + tensor q_131_groups_0 = const()[name = tensor("q_131_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(861763008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(862991872))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_131_cast_fp16 = conv(dilations = q_131_dilations_0, groups = q_131_groups_0, pad = q_131_pad_0, pad_type = q_131_pad_type_0, strides = q_131_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_259_cast_fp16)[name = tensor("q_131_cast_fp16")]; + tensor k_261_pad_type_0 = const()[name = tensor("k_261_pad_type_0"), val = tensor("valid")]; + tensor k_261_strides_0 = const()[name = tensor("k_261_strides_0"), val = tensor([1, 1])]; + tensor k_261_pad_0 = const()[name = tensor("k_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_261_dilations_0 = const()[name = tensor("k_261_dilations_0"), val = tensor([1, 1])]; + tensor k_261_groups_0 = const()[name = tensor("k_261_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(862992064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(864958208))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_261_cast_fp16 = conv(dilations = k_261_dilations_0, groups = k_261_groups_0, pad = k_261_pad_0, pad_type = k_261_pad_type_0, strides = k_261_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_261_cast_fp16")]; + tensor v_131_pad_type_0 = const()[name = tensor("v_131_pad_type_0"), val = tensor("valid")]; + tensor v_131_strides_0 = const()[name = tensor("v_131_strides_0"), val = tensor([1, 1])]; + tensor v_131_pad_0 = const()[name = tensor("v_131_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_131_dilations_0 = const()[name = tensor("v_131_dilations_0"), val = tensor([1, 1])]; + tensor v_131_groups_0 = const()[name = tensor("v_131_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(864958400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(866924544))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_131_cast_fp16 = conv(dilations = v_131_dilations_0, groups = v_131_groups_0, pad = v_131_pad_0, pad_type = v_131_pad_type_0, strides = v_131_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_131_cast_fp16")]; + tensor var_28931_begin_0 = const()[name = tensor("op_28931_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28931_end_0 = const()[name = tensor("op_28931_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28931_end_mask_0 = const()[name = tensor("op_28931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28931_cast_fp16 = slice_by_index(begin = var_28931_begin_0, end = var_28931_end_0, end_mask = var_28931_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28931_cast_fp16")]; + tensor var_28935_begin_0 = const()[name = tensor("op_28935_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28935_end_0 = const()[name = tensor("op_28935_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28935_end_mask_0 = const()[name = tensor("op_28935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28935_cast_fp16 = slice_by_index(begin = var_28935_begin_0, end = var_28935_end_0, end_mask = var_28935_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28935_cast_fp16")]; + tensor var_28939_begin_0 = const()[name = tensor("op_28939_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28939_end_0 = const()[name = tensor("op_28939_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28939_end_mask_0 = const()[name = tensor("op_28939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28939_cast_fp16 = slice_by_index(begin = var_28939_begin_0, end = var_28939_end_0, end_mask = var_28939_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28939_cast_fp16")]; + tensor var_28943_begin_0 = const()[name = tensor("op_28943_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28943_end_0 = const()[name = tensor("op_28943_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28943_end_mask_0 = const()[name = tensor("op_28943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28943_cast_fp16 = slice_by_index(begin = var_28943_begin_0, end = var_28943_end_0, end_mask = var_28943_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28943_cast_fp16")]; + tensor var_28947_begin_0 = const()[name = tensor("op_28947_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28947_end_0 = const()[name = tensor("op_28947_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28947_end_mask_0 = const()[name = tensor("op_28947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28947_cast_fp16 = slice_by_index(begin = var_28947_begin_0, end = var_28947_end_0, end_mask = var_28947_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28947_cast_fp16")]; + tensor var_28951_begin_0 = const()[name = tensor("op_28951_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28951_end_0 = const()[name = tensor("op_28951_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28951_end_mask_0 = const()[name = tensor("op_28951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28951_cast_fp16 = slice_by_index(begin = var_28951_begin_0, end = var_28951_end_0, end_mask = var_28951_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28951_cast_fp16")]; + tensor var_28955_begin_0 = const()[name = tensor("op_28955_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28955_end_0 = const()[name = tensor("op_28955_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28955_end_mask_0 = const()[name = tensor("op_28955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28955_cast_fp16 = slice_by_index(begin = var_28955_begin_0, end = var_28955_end_0, end_mask = var_28955_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28955_cast_fp16")]; + tensor var_28959_begin_0 = const()[name = tensor("op_28959_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28959_end_0 = const()[name = tensor("op_28959_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28959_end_mask_0 = const()[name = tensor("op_28959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28959_cast_fp16 = slice_by_index(begin = var_28959_begin_0, end = var_28959_end_0, end_mask = var_28959_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28959_cast_fp16")]; + tensor var_28963_begin_0 = const()[name = tensor("op_28963_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28963_end_0 = const()[name = tensor("op_28963_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28963_end_mask_0 = const()[name = tensor("op_28963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28963_cast_fp16 = slice_by_index(begin = var_28963_begin_0, end = var_28963_end_0, end_mask = var_28963_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28963_cast_fp16")]; + tensor var_28967_begin_0 = const()[name = tensor("op_28967_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28967_end_0 = const()[name = tensor("op_28967_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28967_end_mask_0 = const()[name = tensor("op_28967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28967_cast_fp16 = slice_by_index(begin = var_28967_begin_0, end = var_28967_end_0, end_mask = var_28967_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28967_cast_fp16")]; + tensor var_28971_begin_0 = const()[name = tensor("op_28971_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28971_end_0 = const()[name = tensor("op_28971_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28971_end_mask_0 = const()[name = tensor("op_28971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28971_cast_fp16 = slice_by_index(begin = var_28971_begin_0, end = var_28971_end_0, end_mask = var_28971_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28971_cast_fp16")]; + tensor var_28975_begin_0 = const()[name = tensor("op_28975_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28975_end_0 = const()[name = tensor("op_28975_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28975_end_mask_0 = const()[name = tensor("op_28975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28975_cast_fp16 = slice_by_index(begin = var_28975_begin_0, end = var_28975_end_0, end_mask = var_28975_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28975_cast_fp16")]; + tensor var_28979_begin_0 = const()[name = tensor("op_28979_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28979_end_0 = const()[name = tensor("op_28979_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28979_end_mask_0 = const()[name = tensor("op_28979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28979_cast_fp16 = slice_by_index(begin = var_28979_begin_0, end = var_28979_end_0, end_mask = var_28979_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28979_cast_fp16")]; + tensor var_28983_begin_0 = const()[name = tensor("op_28983_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28983_end_0 = const()[name = tensor("op_28983_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28983_end_mask_0 = const()[name = tensor("op_28983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28983_cast_fp16 = slice_by_index(begin = var_28983_begin_0, end = var_28983_end_0, end_mask = var_28983_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28983_cast_fp16")]; + tensor var_28987_begin_0 = const()[name = tensor("op_28987_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28987_end_0 = const()[name = tensor("op_28987_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28987_end_mask_0 = const()[name = tensor("op_28987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28987_cast_fp16 = slice_by_index(begin = var_28987_begin_0, end = var_28987_end_0, end_mask = var_28987_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28987_cast_fp16")]; + tensor var_28991_begin_0 = const()[name = tensor("op_28991_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28991_end_0 = const()[name = tensor("op_28991_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28991_end_mask_0 = const()[name = tensor("op_28991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28991_cast_fp16 = slice_by_index(begin = var_28991_begin_0, end = var_28991_end_0, end_mask = var_28991_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28991_cast_fp16")]; + tensor var_28995_begin_0 = const()[name = tensor("op_28995_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28995_end_0 = const()[name = tensor("op_28995_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28995_end_mask_0 = const()[name = tensor("op_28995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28995_cast_fp16 = slice_by_index(begin = var_28995_begin_0, end = var_28995_end_0, end_mask = var_28995_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28995_cast_fp16")]; + tensor var_28999_begin_0 = const()[name = tensor("op_28999_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28999_end_0 = const()[name = tensor("op_28999_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28999_end_mask_0 = const()[name = tensor("op_28999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28999_cast_fp16 = slice_by_index(begin = var_28999_begin_0, end = var_28999_end_0, end_mask = var_28999_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28999_cast_fp16")]; + tensor var_29003_begin_0 = const()[name = tensor("op_29003_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29003_end_0 = const()[name = tensor("op_29003_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29003_end_mask_0 = const()[name = tensor("op_29003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29003_cast_fp16 = slice_by_index(begin = var_29003_begin_0, end = var_29003_end_0, end_mask = var_29003_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_29003_cast_fp16")]; + tensor var_29007_begin_0 = const()[name = tensor("op_29007_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29007_end_0 = const()[name = tensor("op_29007_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29007_end_mask_0 = const()[name = tensor("op_29007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29007_cast_fp16 = slice_by_index(begin = var_29007_begin_0, end = var_29007_end_0, end_mask = var_29007_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_29007_cast_fp16")]; + tensor k_263_perm_0 = const()[name = tensor("k_263_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29014_begin_0 = const()[name = tensor("op_29014_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29014_end_0 = const()[name = tensor("op_29014_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_29014_end_mask_0 = const()[name = tensor("op_29014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_263_cast_fp16 = transpose(perm = k_263_perm_0, x = k_261_cast_fp16)[name = tensor("transpose_2")]; + tensor var_29014_cast_fp16 = slice_by_index(begin = var_29014_begin_0, end = var_29014_end_0, end_mask = var_29014_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29014_cast_fp16")]; + tensor var_29018_begin_0 = const()[name = tensor("op_29018_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29018_end_0 = const()[name = tensor("op_29018_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_29018_end_mask_0 = const()[name = tensor("op_29018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29018_cast_fp16 = slice_by_index(begin = var_29018_begin_0, end = var_29018_end_0, end_mask = var_29018_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29018_cast_fp16")]; + tensor var_29022_begin_0 = const()[name = tensor("op_29022_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29022_end_0 = const()[name = tensor("op_29022_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_29022_end_mask_0 = const()[name = tensor("op_29022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29022_cast_fp16 = slice_by_index(begin = var_29022_begin_0, end = var_29022_end_0, end_mask = var_29022_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29022_cast_fp16")]; + tensor var_29026_begin_0 = const()[name = tensor("op_29026_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29026_end_0 = const()[name = tensor("op_29026_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_29026_end_mask_0 = const()[name = tensor("op_29026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29026_cast_fp16 = slice_by_index(begin = var_29026_begin_0, end = var_29026_end_0, end_mask = var_29026_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29026_cast_fp16")]; + tensor var_29030_begin_0 = const()[name = tensor("op_29030_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29030_end_0 = const()[name = tensor("op_29030_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_29030_end_mask_0 = const()[name = tensor("op_29030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29030_cast_fp16 = slice_by_index(begin = var_29030_begin_0, end = var_29030_end_0, end_mask = var_29030_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29030_cast_fp16")]; + tensor var_29034_begin_0 = const()[name = tensor("op_29034_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29034_end_0 = const()[name = tensor("op_29034_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_29034_end_mask_0 = const()[name = tensor("op_29034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29034_cast_fp16 = slice_by_index(begin = var_29034_begin_0, end = var_29034_end_0, end_mask = var_29034_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29034_cast_fp16")]; + tensor var_29038_begin_0 = const()[name = tensor("op_29038_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29038_end_0 = const()[name = tensor("op_29038_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_29038_end_mask_0 = const()[name = tensor("op_29038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29038_cast_fp16 = slice_by_index(begin = var_29038_begin_0, end = var_29038_end_0, end_mask = var_29038_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29038_cast_fp16")]; + tensor var_29042_begin_0 = const()[name = tensor("op_29042_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29042_end_0 = const()[name = tensor("op_29042_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_29042_end_mask_0 = const()[name = tensor("op_29042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29042_cast_fp16 = slice_by_index(begin = var_29042_begin_0, end = var_29042_end_0, end_mask = var_29042_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29042_cast_fp16")]; + tensor var_29046_begin_0 = const()[name = tensor("op_29046_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29046_end_0 = const()[name = tensor("op_29046_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_29046_end_mask_0 = const()[name = tensor("op_29046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29046_cast_fp16 = slice_by_index(begin = var_29046_begin_0, end = var_29046_end_0, end_mask = var_29046_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29046_cast_fp16")]; + tensor var_29050_begin_0 = const()[name = tensor("op_29050_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29050_end_0 = const()[name = tensor("op_29050_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_29050_end_mask_0 = const()[name = tensor("op_29050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29050_cast_fp16 = slice_by_index(begin = var_29050_begin_0, end = var_29050_end_0, end_mask = var_29050_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29050_cast_fp16")]; + tensor var_29054_begin_0 = const()[name = tensor("op_29054_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29054_end_0 = const()[name = tensor("op_29054_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_29054_end_mask_0 = const()[name = tensor("op_29054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29054_cast_fp16 = slice_by_index(begin = var_29054_begin_0, end = var_29054_end_0, end_mask = var_29054_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29054_cast_fp16")]; + tensor var_29058_begin_0 = const()[name = tensor("op_29058_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29058_end_0 = const()[name = tensor("op_29058_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_29058_end_mask_0 = const()[name = tensor("op_29058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29058_cast_fp16 = slice_by_index(begin = var_29058_begin_0, end = var_29058_end_0, end_mask = var_29058_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29058_cast_fp16")]; + tensor var_29062_begin_0 = const()[name = tensor("op_29062_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29062_end_0 = const()[name = tensor("op_29062_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_29062_end_mask_0 = const()[name = tensor("op_29062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29062_cast_fp16 = slice_by_index(begin = var_29062_begin_0, end = var_29062_end_0, end_mask = var_29062_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29062_cast_fp16")]; + tensor var_29066_begin_0 = const()[name = tensor("op_29066_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29066_end_0 = const()[name = tensor("op_29066_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_29066_end_mask_0 = const()[name = tensor("op_29066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29066_cast_fp16 = slice_by_index(begin = var_29066_begin_0, end = var_29066_end_0, end_mask = var_29066_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29066_cast_fp16")]; + tensor var_29070_begin_0 = const()[name = tensor("op_29070_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29070_end_0 = const()[name = tensor("op_29070_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_29070_end_mask_0 = const()[name = tensor("op_29070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29070_cast_fp16 = slice_by_index(begin = var_29070_begin_0, end = var_29070_end_0, end_mask = var_29070_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29070_cast_fp16")]; + tensor var_29074_begin_0 = const()[name = tensor("op_29074_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29074_end_0 = const()[name = tensor("op_29074_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_29074_end_mask_0 = const()[name = tensor("op_29074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29074_cast_fp16 = slice_by_index(begin = var_29074_begin_0, end = var_29074_end_0, end_mask = var_29074_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29074_cast_fp16")]; + tensor var_29078_begin_0 = const()[name = tensor("op_29078_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29078_end_0 = const()[name = tensor("op_29078_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_29078_end_mask_0 = const()[name = tensor("op_29078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29078_cast_fp16 = slice_by_index(begin = var_29078_begin_0, end = var_29078_end_0, end_mask = var_29078_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29078_cast_fp16")]; + tensor var_29082_begin_0 = const()[name = tensor("op_29082_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29082_end_0 = const()[name = tensor("op_29082_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_29082_end_mask_0 = const()[name = tensor("op_29082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29082_cast_fp16 = slice_by_index(begin = var_29082_begin_0, end = var_29082_end_0, end_mask = var_29082_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29082_cast_fp16")]; + tensor var_29086_begin_0 = const()[name = tensor("op_29086_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29086_end_0 = const()[name = tensor("op_29086_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_29086_end_mask_0 = const()[name = tensor("op_29086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29086_cast_fp16 = slice_by_index(begin = var_29086_begin_0, end = var_29086_end_0, end_mask = var_29086_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29086_cast_fp16")]; + tensor var_29090_begin_0 = const()[name = tensor("op_29090_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_29090_end_0 = const()[name = tensor("op_29090_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_29090_end_mask_0 = const()[name = tensor("op_29090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29090_cast_fp16 = slice_by_index(begin = var_29090_begin_0, end = var_29090_end_0, end_mask = var_29090_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29090_cast_fp16")]; + tensor var_29092_begin_0 = const()[name = tensor("op_29092_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29092_end_0 = const()[name = tensor("op_29092_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_29092_end_mask_0 = const()[name = tensor("op_29092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29092_cast_fp16 = slice_by_index(begin = var_29092_begin_0, end = var_29092_end_0, end_mask = var_29092_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29092_cast_fp16")]; + tensor var_29096_begin_0 = const()[name = tensor("op_29096_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29096_end_0 = const()[name = tensor("op_29096_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_29096_end_mask_0 = const()[name = tensor("op_29096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29096_cast_fp16 = slice_by_index(begin = var_29096_begin_0, end = var_29096_end_0, end_mask = var_29096_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29096_cast_fp16")]; + tensor var_29100_begin_0 = const()[name = tensor("op_29100_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29100_end_0 = const()[name = tensor("op_29100_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_29100_end_mask_0 = const()[name = tensor("op_29100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29100_cast_fp16 = slice_by_index(begin = var_29100_begin_0, end = var_29100_end_0, end_mask = var_29100_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29100_cast_fp16")]; + tensor var_29104_begin_0 = const()[name = tensor("op_29104_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29104_end_0 = const()[name = tensor("op_29104_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_29104_end_mask_0 = const()[name = tensor("op_29104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29104_cast_fp16 = slice_by_index(begin = var_29104_begin_0, end = var_29104_end_0, end_mask = var_29104_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29104_cast_fp16")]; + tensor var_29108_begin_0 = const()[name = tensor("op_29108_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29108_end_0 = const()[name = tensor("op_29108_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_29108_end_mask_0 = const()[name = tensor("op_29108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29108_cast_fp16 = slice_by_index(begin = var_29108_begin_0, end = var_29108_end_0, end_mask = var_29108_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29108_cast_fp16")]; + tensor var_29112_begin_0 = const()[name = tensor("op_29112_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29112_end_0 = const()[name = tensor("op_29112_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_29112_end_mask_0 = const()[name = tensor("op_29112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29112_cast_fp16 = slice_by_index(begin = var_29112_begin_0, end = var_29112_end_0, end_mask = var_29112_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29112_cast_fp16")]; + tensor var_29116_begin_0 = const()[name = tensor("op_29116_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29116_end_0 = const()[name = tensor("op_29116_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_29116_end_mask_0 = const()[name = tensor("op_29116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29116_cast_fp16 = slice_by_index(begin = var_29116_begin_0, end = var_29116_end_0, end_mask = var_29116_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29116_cast_fp16")]; + tensor var_29120_begin_0 = const()[name = tensor("op_29120_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29120_end_0 = const()[name = tensor("op_29120_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_29120_end_mask_0 = const()[name = tensor("op_29120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29120_cast_fp16 = slice_by_index(begin = var_29120_begin_0, end = var_29120_end_0, end_mask = var_29120_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29120_cast_fp16")]; + tensor var_29124_begin_0 = const()[name = tensor("op_29124_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29124_end_0 = const()[name = tensor("op_29124_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_29124_end_mask_0 = const()[name = tensor("op_29124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29124_cast_fp16 = slice_by_index(begin = var_29124_begin_0, end = var_29124_end_0, end_mask = var_29124_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29124_cast_fp16")]; + tensor var_29128_begin_0 = const()[name = tensor("op_29128_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29128_end_0 = const()[name = tensor("op_29128_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_29128_end_mask_0 = const()[name = tensor("op_29128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29128_cast_fp16 = slice_by_index(begin = var_29128_begin_0, end = var_29128_end_0, end_mask = var_29128_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29128_cast_fp16")]; + tensor var_29132_begin_0 = const()[name = tensor("op_29132_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29132_end_0 = const()[name = tensor("op_29132_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_29132_end_mask_0 = const()[name = tensor("op_29132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29132_cast_fp16 = slice_by_index(begin = var_29132_begin_0, end = var_29132_end_0, end_mask = var_29132_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29132_cast_fp16")]; + tensor var_29136_begin_0 = const()[name = tensor("op_29136_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29136_end_0 = const()[name = tensor("op_29136_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_29136_end_mask_0 = const()[name = tensor("op_29136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29136_cast_fp16 = slice_by_index(begin = var_29136_begin_0, end = var_29136_end_0, end_mask = var_29136_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29136_cast_fp16")]; + tensor var_29140_begin_0 = const()[name = tensor("op_29140_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29140_end_0 = const()[name = tensor("op_29140_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_29140_end_mask_0 = const()[name = tensor("op_29140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29140_cast_fp16 = slice_by_index(begin = var_29140_begin_0, end = var_29140_end_0, end_mask = var_29140_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29140_cast_fp16")]; + tensor var_29144_begin_0 = const()[name = tensor("op_29144_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29144_end_0 = const()[name = tensor("op_29144_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_29144_end_mask_0 = const()[name = tensor("op_29144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29144_cast_fp16 = slice_by_index(begin = var_29144_begin_0, end = var_29144_end_0, end_mask = var_29144_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29144_cast_fp16")]; + tensor var_29148_begin_0 = const()[name = tensor("op_29148_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29148_end_0 = const()[name = tensor("op_29148_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_29148_end_mask_0 = const()[name = tensor("op_29148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29148_cast_fp16 = slice_by_index(begin = var_29148_begin_0, end = var_29148_end_0, end_mask = var_29148_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29148_cast_fp16")]; + tensor var_29152_begin_0 = const()[name = tensor("op_29152_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29152_end_0 = const()[name = tensor("op_29152_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_29152_end_mask_0 = const()[name = tensor("op_29152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29152_cast_fp16 = slice_by_index(begin = var_29152_begin_0, end = var_29152_end_0, end_mask = var_29152_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29152_cast_fp16")]; + tensor var_29156_begin_0 = const()[name = tensor("op_29156_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29156_end_0 = const()[name = tensor("op_29156_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_29156_end_mask_0 = const()[name = tensor("op_29156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29156_cast_fp16 = slice_by_index(begin = var_29156_begin_0, end = var_29156_end_0, end_mask = var_29156_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29156_cast_fp16")]; + tensor var_29160_begin_0 = const()[name = tensor("op_29160_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29160_end_0 = const()[name = tensor("op_29160_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_29160_end_mask_0 = const()[name = tensor("op_29160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29160_cast_fp16 = slice_by_index(begin = var_29160_begin_0, end = var_29160_end_0, end_mask = var_29160_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29160_cast_fp16")]; + tensor var_29164_begin_0 = const()[name = tensor("op_29164_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29164_end_0 = const()[name = tensor("op_29164_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_29164_end_mask_0 = const()[name = tensor("op_29164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29164_cast_fp16 = slice_by_index(begin = var_29164_begin_0, end = var_29164_end_0, end_mask = var_29164_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29164_cast_fp16")]; + tensor var_29168_begin_0 = const()[name = tensor("op_29168_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29168_end_0 = const()[name = tensor("op_29168_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_29168_end_mask_0 = const()[name = tensor("op_29168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29168_cast_fp16 = slice_by_index(begin = var_29168_begin_0, end = var_29168_end_0, end_mask = var_29168_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29168_cast_fp16")]; + tensor var_29172_equation_0 = const()[name = tensor("op_29172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29172_cast_fp16 = einsum(equation = var_29172_equation_0, values = (var_29014_cast_fp16, var_28931_cast_fp16))[name = tensor("op_29172_cast_fp16")]; + tensor var_29173_to_fp16 = const()[name = tensor("op_29173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2441_cast_fp16 = mul(x = var_29172_cast_fp16, y = var_29173_to_fp16)[name = tensor("aw_2441_cast_fp16")]; + tensor var_29176_equation_0 = const()[name = tensor("op_29176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29176_cast_fp16 = einsum(equation = var_29176_equation_0, values = (var_29018_cast_fp16, var_28935_cast_fp16))[name = tensor("op_29176_cast_fp16")]; + tensor var_29177_to_fp16 = const()[name = tensor("op_29177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2443_cast_fp16 = mul(x = var_29176_cast_fp16, y = var_29177_to_fp16)[name = tensor("aw_2443_cast_fp16")]; + tensor var_29180_equation_0 = const()[name = tensor("op_29180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29180_cast_fp16 = einsum(equation = var_29180_equation_0, values = (var_29022_cast_fp16, var_28939_cast_fp16))[name = tensor("op_29180_cast_fp16")]; + tensor var_29181_to_fp16 = const()[name = tensor("op_29181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2445_cast_fp16 = mul(x = var_29180_cast_fp16, y = var_29181_to_fp16)[name = tensor("aw_2445_cast_fp16")]; + tensor var_29184_equation_0 = const()[name = tensor("op_29184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29184_cast_fp16 = einsum(equation = var_29184_equation_0, values = (var_29026_cast_fp16, var_28943_cast_fp16))[name = tensor("op_29184_cast_fp16")]; + tensor var_29185_to_fp16 = const()[name = tensor("op_29185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2447_cast_fp16 = mul(x = var_29184_cast_fp16, y = var_29185_to_fp16)[name = tensor("aw_2447_cast_fp16")]; + tensor var_29188_equation_0 = const()[name = tensor("op_29188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29188_cast_fp16 = einsum(equation = var_29188_equation_0, values = (var_29030_cast_fp16, var_28947_cast_fp16))[name = tensor("op_29188_cast_fp16")]; + tensor var_29189_to_fp16 = const()[name = tensor("op_29189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2449_cast_fp16 = mul(x = var_29188_cast_fp16, y = var_29189_to_fp16)[name = tensor("aw_2449_cast_fp16")]; + tensor var_29192_equation_0 = const()[name = tensor("op_29192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29192_cast_fp16 = einsum(equation = var_29192_equation_0, values = (var_29034_cast_fp16, var_28951_cast_fp16))[name = tensor("op_29192_cast_fp16")]; + tensor var_29193_to_fp16 = const()[name = tensor("op_29193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2451_cast_fp16 = mul(x = var_29192_cast_fp16, y = var_29193_to_fp16)[name = tensor("aw_2451_cast_fp16")]; + tensor var_29196_equation_0 = const()[name = tensor("op_29196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29196_cast_fp16 = einsum(equation = var_29196_equation_0, values = (var_29038_cast_fp16, var_28955_cast_fp16))[name = tensor("op_29196_cast_fp16")]; + tensor var_29197_to_fp16 = const()[name = tensor("op_29197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2453_cast_fp16 = mul(x = var_29196_cast_fp16, y = var_29197_to_fp16)[name = tensor("aw_2453_cast_fp16")]; + tensor var_29200_equation_0 = const()[name = tensor("op_29200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29200_cast_fp16 = einsum(equation = var_29200_equation_0, values = (var_29042_cast_fp16, var_28959_cast_fp16))[name = tensor("op_29200_cast_fp16")]; + tensor var_29201_to_fp16 = const()[name = tensor("op_29201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2455_cast_fp16 = mul(x = var_29200_cast_fp16, y = var_29201_to_fp16)[name = tensor("aw_2455_cast_fp16")]; + tensor var_29204_equation_0 = const()[name = tensor("op_29204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29204_cast_fp16 = einsum(equation = var_29204_equation_0, values = (var_29046_cast_fp16, var_28963_cast_fp16))[name = tensor("op_29204_cast_fp16")]; + tensor var_29205_to_fp16 = const()[name = tensor("op_29205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2457_cast_fp16 = mul(x = var_29204_cast_fp16, y = var_29205_to_fp16)[name = tensor("aw_2457_cast_fp16")]; + tensor var_29208_equation_0 = const()[name = tensor("op_29208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29208_cast_fp16 = einsum(equation = var_29208_equation_0, values = (var_29050_cast_fp16, var_28967_cast_fp16))[name = tensor("op_29208_cast_fp16")]; + tensor var_29209_to_fp16 = const()[name = tensor("op_29209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2459_cast_fp16 = mul(x = var_29208_cast_fp16, y = var_29209_to_fp16)[name = tensor("aw_2459_cast_fp16")]; + tensor var_29212_equation_0 = const()[name = tensor("op_29212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29212_cast_fp16 = einsum(equation = var_29212_equation_0, values = (var_29054_cast_fp16, var_28971_cast_fp16))[name = tensor("op_29212_cast_fp16")]; + tensor var_29213_to_fp16 = const()[name = tensor("op_29213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2461_cast_fp16 = mul(x = var_29212_cast_fp16, y = var_29213_to_fp16)[name = tensor("aw_2461_cast_fp16")]; + tensor var_29216_equation_0 = const()[name = tensor("op_29216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29216_cast_fp16 = einsum(equation = var_29216_equation_0, values = (var_29058_cast_fp16, var_28975_cast_fp16))[name = tensor("op_29216_cast_fp16")]; + tensor var_29217_to_fp16 = const()[name = tensor("op_29217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2463_cast_fp16 = mul(x = var_29216_cast_fp16, y = var_29217_to_fp16)[name = tensor("aw_2463_cast_fp16")]; + tensor var_29220_equation_0 = const()[name = tensor("op_29220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29220_cast_fp16 = einsum(equation = var_29220_equation_0, values = (var_29062_cast_fp16, var_28979_cast_fp16))[name = tensor("op_29220_cast_fp16")]; + tensor var_29221_to_fp16 = const()[name = tensor("op_29221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2465_cast_fp16 = mul(x = var_29220_cast_fp16, y = var_29221_to_fp16)[name = tensor("aw_2465_cast_fp16")]; + tensor var_29224_equation_0 = const()[name = tensor("op_29224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29224_cast_fp16 = einsum(equation = var_29224_equation_0, values = (var_29066_cast_fp16, var_28983_cast_fp16))[name = tensor("op_29224_cast_fp16")]; + tensor var_29225_to_fp16 = const()[name = tensor("op_29225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2467_cast_fp16 = mul(x = var_29224_cast_fp16, y = var_29225_to_fp16)[name = tensor("aw_2467_cast_fp16")]; + tensor var_29228_equation_0 = const()[name = tensor("op_29228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29228_cast_fp16 = einsum(equation = var_29228_equation_0, values = (var_29070_cast_fp16, var_28987_cast_fp16))[name = tensor("op_29228_cast_fp16")]; + tensor var_29229_to_fp16 = const()[name = tensor("op_29229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2469_cast_fp16 = mul(x = var_29228_cast_fp16, y = var_29229_to_fp16)[name = tensor("aw_2469_cast_fp16")]; + tensor var_29232_equation_0 = const()[name = tensor("op_29232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29232_cast_fp16 = einsum(equation = var_29232_equation_0, values = (var_29074_cast_fp16, var_28991_cast_fp16))[name = tensor("op_29232_cast_fp16")]; + tensor var_29233_to_fp16 = const()[name = tensor("op_29233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2471_cast_fp16 = mul(x = var_29232_cast_fp16, y = var_29233_to_fp16)[name = tensor("aw_2471_cast_fp16")]; + tensor var_29236_equation_0 = const()[name = tensor("op_29236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29236_cast_fp16 = einsum(equation = var_29236_equation_0, values = (var_29078_cast_fp16, var_28995_cast_fp16))[name = tensor("op_29236_cast_fp16")]; + tensor var_29237_to_fp16 = const()[name = tensor("op_29237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2473_cast_fp16 = mul(x = var_29236_cast_fp16, y = var_29237_to_fp16)[name = tensor("aw_2473_cast_fp16")]; + tensor var_29240_equation_0 = const()[name = tensor("op_29240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29240_cast_fp16 = einsum(equation = var_29240_equation_0, values = (var_29082_cast_fp16, var_28999_cast_fp16))[name = tensor("op_29240_cast_fp16")]; + tensor var_29241_to_fp16 = const()[name = tensor("op_29241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2475_cast_fp16 = mul(x = var_29240_cast_fp16, y = var_29241_to_fp16)[name = tensor("aw_2475_cast_fp16")]; + tensor var_29244_equation_0 = const()[name = tensor("op_29244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29244_cast_fp16 = einsum(equation = var_29244_equation_0, values = (var_29086_cast_fp16, var_29003_cast_fp16))[name = tensor("op_29244_cast_fp16")]; + tensor var_29245_to_fp16 = const()[name = tensor("op_29245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2477_cast_fp16 = mul(x = var_29244_cast_fp16, y = var_29245_to_fp16)[name = tensor("aw_2477_cast_fp16")]; + tensor var_29248_equation_0 = const()[name = tensor("op_29248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29248_cast_fp16 = einsum(equation = var_29248_equation_0, values = (var_29090_cast_fp16, var_29007_cast_fp16))[name = tensor("op_29248_cast_fp16")]; + tensor var_29249_to_fp16 = const()[name = tensor("op_29249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2479_cast_fp16 = mul(x = var_29248_cast_fp16, y = var_29249_to_fp16)[name = tensor("aw_2479_cast_fp16")]; + tensor var_29251_cast_fp16 = softmax(axis = var_21077, x = aw_2441_cast_fp16)[name = tensor("op_29251_cast_fp16")]; + tensor var_29252_cast_fp16 = softmax(axis = var_21077, x = aw_2443_cast_fp16)[name = tensor("op_29252_cast_fp16")]; + tensor var_29253_cast_fp16 = softmax(axis = var_21077, x = aw_2445_cast_fp16)[name = tensor("op_29253_cast_fp16")]; + tensor var_29254_cast_fp16 = softmax(axis = var_21077, x = aw_2447_cast_fp16)[name = tensor("op_29254_cast_fp16")]; + tensor var_29255_cast_fp16 = softmax(axis = var_21077, x = aw_2449_cast_fp16)[name = tensor("op_29255_cast_fp16")]; + tensor var_29256_cast_fp16 = softmax(axis = var_21077, x = aw_2451_cast_fp16)[name = tensor("op_29256_cast_fp16")]; + tensor var_29257_cast_fp16 = softmax(axis = var_21077, x = aw_2453_cast_fp16)[name = tensor("op_29257_cast_fp16")]; + tensor var_29258_cast_fp16 = softmax(axis = var_21077, x = aw_2455_cast_fp16)[name = tensor("op_29258_cast_fp16")]; + tensor var_29259_cast_fp16 = softmax(axis = var_21077, x = aw_2457_cast_fp16)[name = tensor("op_29259_cast_fp16")]; + tensor var_29260_cast_fp16 = softmax(axis = var_21077, x = aw_2459_cast_fp16)[name = tensor("op_29260_cast_fp16")]; + tensor var_29261_cast_fp16 = softmax(axis = var_21077, x = aw_2461_cast_fp16)[name = tensor("op_29261_cast_fp16")]; + tensor var_29262_cast_fp16 = softmax(axis = var_21077, x = aw_2463_cast_fp16)[name = tensor("op_29262_cast_fp16")]; + tensor var_29263_cast_fp16 = softmax(axis = var_21077, x = aw_2465_cast_fp16)[name = tensor("op_29263_cast_fp16")]; + tensor var_29264_cast_fp16 = softmax(axis = var_21077, x = aw_2467_cast_fp16)[name = tensor("op_29264_cast_fp16")]; + tensor var_29265_cast_fp16 = softmax(axis = var_21077, x = aw_2469_cast_fp16)[name = tensor("op_29265_cast_fp16")]; + tensor var_29266_cast_fp16 = softmax(axis = var_21077, x = aw_2471_cast_fp16)[name = tensor("op_29266_cast_fp16")]; + tensor var_29267_cast_fp16 = softmax(axis = var_21077, x = aw_2473_cast_fp16)[name = tensor("op_29267_cast_fp16")]; + tensor var_29268_cast_fp16 = softmax(axis = var_21077, x = aw_2475_cast_fp16)[name = tensor("op_29268_cast_fp16")]; + tensor var_29269_cast_fp16 = softmax(axis = var_21077, x = aw_2477_cast_fp16)[name = tensor("op_29269_cast_fp16")]; + tensor var_29270_cast_fp16 = softmax(axis = var_21077, x = aw_2479_cast_fp16)[name = tensor("op_29270_cast_fp16")]; + tensor var_29272_equation_0 = const()[name = tensor("op_29272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29272_cast_fp16 = einsum(equation = var_29272_equation_0, values = (var_29092_cast_fp16, var_29251_cast_fp16))[name = tensor("op_29272_cast_fp16")]; + tensor var_29274_equation_0 = const()[name = tensor("op_29274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29274_cast_fp16 = einsum(equation = var_29274_equation_0, values = (var_29096_cast_fp16, var_29252_cast_fp16))[name = tensor("op_29274_cast_fp16")]; + tensor var_29276_equation_0 = const()[name = tensor("op_29276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29276_cast_fp16 = einsum(equation = var_29276_equation_0, values = (var_29100_cast_fp16, var_29253_cast_fp16))[name = tensor("op_29276_cast_fp16")]; + tensor var_29278_equation_0 = const()[name = tensor("op_29278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29278_cast_fp16 = einsum(equation = var_29278_equation_0, values = (var_29104_cast_fp16, var_29254_cast_fp16))[name = tensor("op_29278_cast_fp16")]; + tensor var_29280_equation_0 = const()[name = tensor("op_29280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29280_cast_fp16 = einsum(equation = var_29280_equation_0, values = (var_29108_cast_fp16, var_29255_cast_fp16))[name = tensor("op_29280_cast_fp16")]; + tensor var_29282_equation_0 = const()[name = tensor("op_29282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29282_cast_fp16 = einsum(equation = var_29282_equation_0, values = (var_29112_cast_fp16, var_29256_cast_fp16))[name = tensor("op_29282_cast_fp16")]; + tensor var_29284_equation_0 = const()[name = tensor("op_29284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29284_cast_fp16 = einsum(equation = var_29284_equation_0, values = (var_29116_cast_fp16, var_29257_cast_fp16))[name = tensor("op_29284_cast_fp16")]; + tensor var_29286_equation_0 = const()[name = tensor("op_29286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29286_cast_fp16 = einsum(equation = var_29286_equation_0, values = (var_29120_cast_fp16, var_29258_cast_fp16))[name = tensor("op_29286_cast_fp16")]; + tensor var_29288_equation_0 = const()[name = tensor("op_29288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29288_cast_fp16 = einsum(equation = var_29288_equation_0, values = (var_29124_cast_fp16, var_29259_cast_fp16))[name = tensor("op_29288_cast_fp16")]; + tensor var_29290_equation_0 = const()[name = tensor("op_29290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29290_cast_fp16 = einsum(equation = var_29290_equation_0, values = (var_29128_cast_fp16, var_29260_cast_fp16))[name = tensor("op_29290_cast_fp16")]; + tensor var_29292_equation_0 = const()[name = tensor("op_29292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29292_cast_fp16 = einsum(equation = var_29292_equation_0, values = (var_29132_cast_fp16, var_29261_cast_fp16))[name = tensor("op_29292_cast_fp16")]; + tensor var_29294_equation_0 = const()[name = tensor("op_29294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29294_cast_fp16 = einsum(equation = var_29294_equation_0, values = (var_29136_cast_fp16, var_29262_cast_fp16))[name = tensor("op_29294_cast_fp16")]; + tensor var_29296_equation_0 = const()[name = tensor("op_29296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29296_cast_fp16 = einsum(equation = var_29296_equation_0, values = (var_29140_cast_fp16, var_29263_cast_fp16))[name = tensor("op_29296_cast_fp16")]; + tensor var_29298_equation_0 = const()[name = tensor("op_29298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29298_cast_fp16 = einsum(equation = var_29298_equation_0, values = (var_29144_cast_fp16, var_29264_cast_fp16))[name = tensor("op_29298_cast_fp16")]; + tensor var_29300_equation_0 = const()[name = tensor("op_29300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29300_cast_fp16 = einsum(equation = var_29300_equation_0, values = (var_29148_cast_fp16, var_29265_cast_fp16))[name = tensor("op_29300_cast_fp16")]; + tensor var_29302_equation_0 = const()[name = tensor("op_29302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29302_cast_fp16 = einsum(equation = var_29302_equation_0, values = (var_29152_cast_fp16, var_29266_cast_fp16))[name = tensor("op_29302_cast_fp16")]; + tensor var_29304_equation_0 = const()[name = tensor("op_29304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29304_cast_fp16 = einsum(equation = var_29304_equation_0, values = (var_29156_cast_fp16, var_29267_cast_fp16))[name = tensor("op_29304_cast_fp16")]; + tensor var_29306_equation_0 = const()[name = tensor("op_29306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29306_cast_fp16 = einsum(equation = var_29306_equation_0, values = (var_29160_cast_fp16, var_29268_cast_fp16))[name = tensor("op_29306_cast_fp16")]; + tensor var_29308_equation_0 = const()[name = tensor("op_29308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29308_cast_fp16 = einsum(equation = var_29308_equation_0, values = (var_29164_cast_fp16, var_29269_cast_fp16))[name = tensor("op_29308_cast_fp16")]; + tensor var_29310_equation_0 = const()[name = tensor("op_29310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29310_cast_fp16 = einsum(equation = var_29310_equation_0, values = (var_29168_cast_fp16, var_29270_cast_fp16))[name = tensor("op_29310_cast_fp16")]; + tensor input_393_interleave_0 = const()[name = tensor("input_393_interleave_0"), val = tensor(false)]; + tensor input_393_cast_fp16 = concat(axis = var_21077, interleave = input_393_interleave_0, values = (var_29272_cast_fp16, var_29274_cast_fp16, var_29276_cast_fp16, var_29278_cast_fp16, var_29280_cast_fp16, var_29282_cast_fp16, var_29284_cast_fp16, var_29286_cast_fp16, var_29288_cast_fp16, var_29290_cast_fp16, var_29292_cast_fp16, var_29294_cast_fp16, var_29296_cast_fp16, var_29298_cast_fp16, var_29300_cast_fp16, var_29302_cast_fp16, var_29304_cast_fp16, var_29306_cast_fp16, var_29308_cast_fp16, var_29310_cast_fp16))[name = tensor("input_393_cast_fp16")]; + tensor var_29320_pad_type_0 = const()[name = tensor("op_29320_pad_type_0"), val = tensor("valid")]; + tensor var_29320_strides_0 = const()[name = tensor("op_29320_strides_0"), val = tensor([1, 1])]; + tensor var_29320_pad_0 = const()[name = tensor("op_29320_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29320_dilations_0 = const()[name = tensor("op_29320_dilations_0"), val = tensor([1, 1])]; + tensor var_29320_groups_0 = const()[name = tensor("op_29320_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(866924736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868153600))), name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868153792)))]; + tensor var_29320_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_29320_dilations_0, groups = var_29320_groups_0, pad = var_29320_pad_0, pad_type = var_29320_pad_type_0, strides = var_29320_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized, x = input_393_cast_fp16)[name = tensor("op_29320_cast_fp16")]; + tensor inputs_197_cast_fp16 = add(x = var_29320_cast_fp16, y = inputs_195_cast_fp16)[name = tensor("inputs_197_cast_fp16")]; + tensor input_395_axes_0 = const()[name = tensor("input_395_axes_0"), val = tensor([1])]; + tensor input_395_gamma_0_to_fp16 = const()[name = tensor("input_395_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868156416)))]; + tensor input_395_beta_0_to_fp16 = const()[name = tensor("input_395_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868159040)))]; + tensor var_29330_to_fp16 = const()[name = tensor("op_29330_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_395_cast_fp16 = layer_norm(axes = input_395_axes_0, beta = input_395_beta_0_to_fp16, epsilon = var_29330_to_fp16, gamma = input_395_gamma_0_to_fp16, x = inputs_197_cast_fp16)[name = tensor("input_395_cast_fp16")]; + tensor var_29350_pad_type_0 = const()[name = tensor("op_29350_pad_type_0"), val = tensor("valid")]; + tensor var_29350_strides_0 = const()[name = tensor("op_29350_strides_0"), val = tensor([1, 1])]; + tensor var_29350_pad_0 = const()[name = tensor("op_29350_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29350_dilations_0 = const()[name = tensor("op_29350_dilations_0"), val = tensor([1, 1])]; + tensor var_29350_groups_0 = const()[name = tensor("op_29350_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868161664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(877992128))), name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(877992320)))]; + tensor var_29350_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_29350_dilations_0, groups = var_29350_groups_0, pad = var_29350_pad_0, pad_type = var_29350_pad_type_0, strides = var_29350_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized, x = input_395_cast_fp16)[name = tensor("op_29350_cast_fp16")]; + tensor var_29351_split_sizes_0 = const()[name = tensor("op_29351_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_29351_axis_0 = const()[name = tensor("op_29351_axis_0"), val = tensor(1)]; + tensor var_29351_cast_fp16_0, tensor var_29351_cast_fp16_1 = split(axis = var_29351_axis_0, split_sizes = var_29351_split_sizes_0, x = var_29350_cast_fp16)[name = tensor("op_29351_cast_fp16")]; + tensor var_29353_mode_0 = const()[name = tensor("op_29353_mode_0"), val = tensor("EXACT")]; + tensor var_29353_cast_fp16 = gelu(mode = var_29353_mode_0, x = var_29351_cast_fp16_1)[name = tensor("op_29353_cast_fp16")]; + tensor input_397_cast_fp16 = mul(x = var_29351_cast_fp16_0, y = var_29353_cast_fp16)[name = tensor("input_397_cast_fp16")]; + tensor var_29361_pad_type_0 = const()[name = tensor("op_29361_pad_type_0"), val = tensor("valid")]; + tensor var_29361_strides_0 = const()[name = tensor("op_29361_strides_0"), val = tensor([1, 1])]; + tensor var_29361_pad_0 = const()[name = tensor("op_29361_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29361_dilations_0 = const()[name = tensor("op_29361_dilations_0"), val = tensor([1, 1])]; + tensor var_29361_groups_0 = const()[name = tensor("op_29361_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(878012864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(882928128))), name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(882928320)))]; + tensor var_29361_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_29361_dilations_0, groups = var_29361_groups_0, pad = var_29361_pad_0, pad_type = var_29361_pad_type_0, strides = var_29361_strides_0, weight = mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized, x = input_397_cast_fp16)[name = tensor("op_29361_cast_fp16")]; + tensor inputs_199_cast_fp16 = add(x = var_29361_cast_fp16, y = inputs_197_cast_fp16)[name = tensor("inputs_199_cast_fp16")]; + tensor hidden_states_263_axes_0 = const()[name = tensor("hidden_states_263_axes_0"), val = tensor([1])]; + tensor hidden_states_263_gamma_0_to_fp16 = const()[name = tensor("hidden_states_263_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(882930944)))]; + tensor hidden_states_263_beta_0_to_fp16 = const()[name = tensor("hidden_states_263_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(882933568)))]; + tensor var_29377_to_fp16 = const()[name = tensor("op_29377_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_263_cast_fp16 = layer_norm(axes = hidden_states_263_axes_0, beta = hidden_states_263_beta_0_to_fp16, epsilon = var_29377_to_fp16, gamma = hidden_states_263_gamma_0_to_fp16, x = inputs_199_cast_fp16)[name = tensor("hidden_states_263_cast_fp16")]; + tensor q_133_pad_type_0 = const()[name = tensor("q_133_pad_type_0"), val = tensor("valid")]; + tensor q_133_strides_0 = const()[name = tensor("q_133_strides_0"), val = tensor([1, 1])]; + tensor q_133_pad_0 = const()[name = tensor("q_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_133_dilations_0 = const()[name = tensor("q_133_dilations_0"), val = tensor([1, 1])]; + tensor q_133_groups_0 = const()[name = tensor("q_133_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(882936192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(884165056))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_133_cast_fp16 = conv(dilations = q_133_dilations_0, groups = q_133_groups_0, pad = q_133_pad_0, pad_type = q_133_pad_type_0, strides = q_133_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_263_cast_fp16)[name = tensor("q_133_cast_fp16")]; + tensor k_265_pad_type_0 = const()[name = tensor("k_265_pad_type_0"), val = tensor("valid")]; + tensor k_265_strides_0 = const()[name = tensor("k_265_strides_0"), val = tensor([1, 1])]; + tensor k_265_pad_0 = const()[name = tensor("k_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_265_dilations_0 = const()[name = tensor("k_265_dilations_0"), val = tensor([1, 1])]; + tensor k_265_groups_0 = const()[name = tensor("k_265_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(884165248))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(885394112))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_265_cast_fp16 = conv(dilations = k_265_dilations_0, groups = k_265_groups_0, pad = k_265_pad_0, pad_type = k_265_pad_type_0, strides = k_265_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_263_cast_fp16)[name = tensor("k_265_cast_fp16")]; + tensor v_133_pad_type_0 = const()[name = tensor("v_133_pad_type_0"), val = tensor("valid")]; + tensor v_133_strides_0 = const()[name = tensor("v_133_strides_0"), val = tensor([1, 1])]; + tensor v_133_pad_0 = const()[name = tensor("v_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_133_dilations_0 = const()[name = tensor("v_133_dilations_0"), val = tensor([1, 1])]; + tensor v_133_groups_0 = const()[name = tensor("v_133_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(885394304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886623168))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_133_cast_fp16 = conv(dilations = v_133_dilations_0, groups = v_133_groups_0, pad = v_133_pad_0, pad_type = v_133_pad_type_0, strides = v_133_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_263_cast_fp16)[name = tensor("v_133_cast_fp16")]; + tensor var_29410_begin_0 = const()[name = tensor("op_29410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29410_end_0 = const()[name = tensor("op_29410_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_29410_end_mask_0 = const()[name = tensor("op_29410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29410_cast_fp16 = slice_by_index(begin = var_29410_begin_0, end = var_29410_end_0, end_mask = var_29410_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29410_cast_fp16")]; + tensor var_29414_begin_0 = const()[name = tensor("op_29414_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29414_end_0 = const()[name = tensor("op_29414_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_29414_end_mask_0 = const()[name = tensor("op_29414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29414_cast_fp16 = slice_by_index(begin = var_29414_begin_0, end = var_29414_end_0, end_mask = var_29414_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29414_cast_fp16")]; + tensor var_29418_begin_0 = const()[name = tensor("op_29418_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29418_end_0 = const()[name = tensor("op_29418_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_29418_end_mask_0 = const()[name = tensor("op_29418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29418_cast_fp16 = slice_by_index(begin = var_29418_begin_0, end = var_29418_end_0, end_mask = var_29418_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29418_cast_fp16")]; + tensor var_29422_begin_0 = const()[name = tensor("op_29422_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29422_end_0 = const()[name = tensor("op_29422_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_29422_end_mask_0 = const()[name = tensor("op_29422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29422_cast_fp16 = slice_by_index(begin = var_29422_begin_0, end = var_29422_end_0, end_mask = var_29422_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29422_cast_fp16")]; + tensor var_29426_begin_0 = const()[name = tensor("op_29426_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29426_end_0 = const()[name = tensor("op_29426_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_29426_end_mask_0 = const()[name = tensor("op_29426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29426_cast_fp16 = slice_by_index(begin = var_29426_begin_0, end = var_29426_end_0, end_mask = var_29426_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29426_cast_fp16")]; + tensor var_29430_begin_0 = const()[name = tensor("op_29430_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29430_end_0 = const()[name = tensor("op_29430_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_29430_end_mask_0 = const()[name = tensor("op_29430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29430_cast_fp16 = slice_by_index(begin = var_29430_begin_0, end = var_29430_end_0, end_mask = var_29430_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29430_cast_fp16")]; + tensor var_29434_begin_0 = const()[name = tensor("op_29434_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29434_end_0 = const()[name = tensor("op_29434_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_29434_end_mask_0 = const()[name = tensor("op_29434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29434_cast_fp16 = slice_by_index(begin = var_29434_begin_0, end = var_29434_end_0, end_mask = var_29434_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29434_cast_fp16")]; + tensor var_29438_begin_0 = const()[name = tensor("op_29438_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29438_end_0 = const()[name = tensor("op_29438_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_29438_end_mask_0 = const()[name = tensor("op_29438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29438_cast_fp16 = slice_by_index(begin = var_29438_begin_0, end = var_29438_end_0, end_mask = var_29438_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29438_cast_fp16")]; + tensor var_29442_begin_0 = const()[name = tensor("op_29442_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29442_end_0 = const()[name = tensor("op_29442_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_29442_end_mask_0 = const()[name = tensor("op_29442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29442_cast_fp16 = slice_by_index(begin = var_29442_begin_0, end = var_29442_end_0, end_mask = var_29442_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29442_cast_fp16")]; + tensor var_29446_begin_0 = const()[name = tensor("op_29446_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29446_end_0 = const()[name = tensor("op_29446_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_29446_end_mask_0 = const()[name = tensor("op_29446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29446_cast_fp16 = slice_by_index(begin = var_29446_begin_0, end = var_29446_end_0, end_mask = var_29446_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29446_cast_fp16")]; + tensor var_29450_begin_0 = const()[name = tensor("op_29450_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29450_end_0 = const()[name = tensor("op_29450_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_29450_end_mask_0 = const()[name = tensor("op_29450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29450_cast_fp16 = slice_by_index(begin = var_29450_begin_0, end = var_29450_end_0, end_mask = var_29450_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29450_cast_fp16")]; + tensor var_29454_begin_0 = const()[name = tensor("op_29454_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29454_end_0 = const()[name = tensor("op_29454_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_29454_end_mask_0 = const()[name = tensor("op_29454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29454_cast_fp16 = slice_by_index(begin = var_29454_begin_0, end = var_29454_end_0, end_mask = var_29454_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29454_cast_fp16")]; + tensor var_29458_begin_0 = const()[name = tensor("op_29458_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29458_end_0 = const()[name = tensor("op_29458_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_29458_end_mask_0 = const()[name = tensor("op_29458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29458_cast_fp16 = slice_by_index(begin = var_29458_begin_0, end = var_29458_end_0, end_mask = var_29458_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29458_cast_fp16")]; + tensor var_29462_begin_0 = const()[name = tensor("op_29462_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29462_end_0 = const()[name = tensor("op_29462_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_29462_end_mask_0 = const()[name = tensor("op_29462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29462_cast_fp16 = slice_by_index(begin = var_29462_begin_0, end = var_29462_end_0, end_mask = var_29462_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29462_cast_fp16")]; + tensor var_29466_begin_0 = const()[name = tensor("op_29466_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29466_end_0 = const()[name = tensor("op_29466_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_29466_end_mask_0 = const()[name = tensor("op_29466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29466_cast_fp16 = slice_by_index(begin = var_29466_begin_0, end = var_29466_end_0, end_mask = var_29466_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29466_cast_fp16")]; + tensor var_29470_begin_0 = const()[name = tensor("op_29470_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29470_end_0 = const()[name = tensor("op_29470_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29470_end_mask_0 = const()[name = tensor("op_29470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29470_cast_fp16 = slice_by_index(begin = var_29470_begin_0, end = var_29470_end_0, end_mask = var_29470_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29470_cast_fp16")]; + tensor var_29474_begin_0 = const()[name = tensor("op_29474_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29474_end_0 = const()[name = tensor("op_29474_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_29474_end_mask_0 = const()[name = tensor("op_29474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29474_cast_fp16 = slice_by_index(begin = var_29474_begin_0, end = var_29474_end_0, end_mask = var_29474_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29474_cast_fp16")]; + tensor var_29478_begin_0 = const()[name = tensor("op_29478_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29478_end_0 = const()[name = tensor("op_29478_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_29478_end_mask_0 = const()[name = tensor("op_29478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29478_cast_fp16 = slice_by_index(begin = var_29478_begin_0, end = var_29478_end_0, end_mask = var_29478_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29478_cast_fp16")]; + tensor var_29482_begin_0 = const()[name = tensor("op_29482_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29482_end_0 = const()[name = tensor("op_29482_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29482_end_mask_0 = const()[name = tensor("op_29482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29482_cast_fp16 = slice_by_index(begin = var_29482_begin_0, end = var_29482_end_0, end_mask = var_29482_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29482_cast_fp16")]; + tensor var_29486_begin_0 = const()[name = tensor("op_29486_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29486_end_0 = const()[name = tensor("op_29486_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29486_end_mask_0 = const()[name = tensor("op_29486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29486_cast_fp16 = slice_by_index(begin = var_29486_begin_0, end = var_29486_end_0, end_mask = var_29486_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29486_cast_fp16")]; + tensor k_267_perm_0 = const()[name = tensor("k_267_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29493_begin_0 = const()[name = tensor("op_29493_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29493_end_0 = const()[name = tensor("op_29493_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_29493_end_mask_0 = const()[name = tensor("op_29493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_267_cast_fp16 = transpose(perm = k_267_perm_0, x = k_265_cast_fp16)[name = tensor("transpose_1")]; + tensor var_29493_cast_fp16 = slice_by_index(begin = var_29493_begin_0, end = var_29493_end_0, end_mask = var_29493_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29493_cast_fp16")]; + tensor var_29497_begin_0 = const()[name = tensor("op_29497_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29497_end_0 = const()[name = tensor("op_29497_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_29497_end_mask_0 = const()[name = tensor("op_29497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29497_cast_fp16 = slice_by_index(begin = var_29497_begin_0, end = var_29497_end_0, end_mask = var_29497_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29497_cast_fp16")]; + tensor var_29501_begin_0 = const()[name = tensor("op_29501_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29501_end_0 = const()[name = tensor("op_29501_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_29501_end_mask_0 = const()[name = tensor("op_29501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29501_cast_fp16 = slice_by_index(begin = var_29501_begin_0, end = var_29501_end_0, end_mask = var_29501_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29501_cast_fp16")]; + tensor var_29505_begin_0 = const()[name = tensor("op_29505_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29505_end_0 = const()[name = tensor("op_29505_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_29505_end_mask_0 = const()[name = tensor("op_29505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29505_cast_fp16 = slice_by_index(begin = var_29505_begin_0, end = var_29505_end_0, end_mask = var_29505_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29505_cast_fp16")]; + tensor var_29509_begin_0 = const()[name = tensor("op_29509_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29509_end_0 = const()[name = tensor("op_29509_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_29509_end_mask_0 = const()[name = tensor("op_29509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29509_cast_fp16 = slice_by_index(begin = var_29509_begin_0, end = var_29509_end_0, end_mask = var_29509_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29509_cast_fp16")]; + tensor var_29513_begin_0 = const()[name = tensor("op_29513_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29513_end_0 = const()[name = tensor("op_29513_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_29513_end_mask_0 = const()[name = tensor("op_29513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29513_cast_fp16 = slice_by_index(begin = var_29513_begin_0, end = var_29513_end_0, end_mask = var_29513_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29513_cast_fp16")]; + tensor var_29517_begin_0 = const()[name = tensor("op_29517_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29517_end_0 = const()[name = tensor("op_29517_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_29517_end_mask_0 = const()[name = tensor("op_29517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29517_cast_fp16 = slice_by_index(begin = var_29517_begin_0, end = var_29517_end_0, end_mask = var_29517_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29517_cast_fp16")]; + tensor var_29521_begin_0 = const()[name = tensor("op_29521_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29521_end_0 = const()[name = tensor("op_29521_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_29521_end_mask_0 = const()[name = tensor("op_29521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29521_cast_fp16 = slice_by_index(begin = var_29521_begin_0, end = var_29521_end_0, end_mask = var_29521_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29521_cast_fp16")]; + tensor var_29525_begin_0 = const()[name = tensor("op_29525_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29525_end_0 = const()[name = tensor("op_29525_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_29525_end_mask_0 = const()[name = tensor("op_29525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29525_cast_fp16 = slice_by_index(begin = var_29525_begin_0, end = var_29525_end_0, end_mask = var_29525_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29525_cast_fp16")]; + tensor var_29529_begin_0 = const()[name = tensor("op_29529_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29529_end_0 = const()[name = tensor("op_29529_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_29529_end_mask_0 = const()[name = tensor("op_29529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29529_cast_fp16 = slice_by_index(begin = var_29529_begin_0, end = var_29529_end_0, end_mask = var_29529_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29529_cast_fp16")]; + tensor var_29533_begin_0 = const()[name = tensor("op_29533_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29533_end_0 = const()[name = tensor("op_29533_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_29533_end_mask_0 = const()[name = tensor("op_29533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29533_cast_fp16 = slice_by_index(begin = var_29533_begin_0, end = var_29533_end_0, end_mask = var_29533_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29533_cast_fp16")]; + tensor var_29537_begin_0 = const()[name = tensor("op_29537_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29537_end_0 = const()[name = tensor("op_29537_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_29537_end_mask_0 = const()[name = tensor("op_29537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29537_cast_fp16 = slice_by_index(begin = var_29537_begin_0, end = var_29537_end_0, end_mask = var_29537_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29537_cast_fp16")]; + tensor var_29541_begin_0 = const()[name = tensor("op_29541_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29541_end_0 = const()[name = tensor("op_29541_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_29541_end_mask_0 = const()[name = tensor("op_29541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29541_cast_fp16 = slice_by_index(begin = var_29541_begin_0, end = var_29541_end_0, end_mask = var_29541_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29541_cast_fp16")]; + tensor var_29545_begin_0 = const()[name = tensor("op_29545_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29545_end_0 = const()[name = tensor("op_29545_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_29545_end_mask_0 = const()[name = tensor("op_29545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29545_cast_fp16 = slice_by_index(begin = var_29545_begin_0, end = var_29545_end_0, end_mask = var_29545_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29545_cast_fp16")]; + tensor var_29549_begin_0 = const()[name = tensor("op_29549_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29549_end_0 = const()[name = tensor("op_29549_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_29549_end_mask_0 = const()[name = tensor("op_29549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29549_cast_fp16 = slice_by_index(begin = var_29549_begin_0, end = var_29549_end_0, end_mask = var_29549_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29549_cast_fp16")]; + tensor var_29553_begin_0 = const()[name = tensor("op_29553_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29553_end_0 = const()[name = tensor("op_29553_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29553_end_mask_0 = const()[name = tensor("op_29553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29553_cast_fp16 = slice_by_index(begin = var_29553_begin_0, end = var_29553_end_0, end_mask = var_29553_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29553_cast_fp16")]; + tensor var_29557_begin_0 = const()[name = tensor("op_29557_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29557_end_0 = const()[name = tensor("op_29557_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_29557_end_mask_0 = const()[name = tensor("op_29557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29557_cast_fp16 = slice_by_index(begin = var_29557_begin_0, end = var_29557_end_0, end_mask = var_29557_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29557_cast_fp16")]; + tensor var_29561_begin_0 = const()[name = tensor("op_29561_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29561_end_0 = const()[name = tensor("op_29561_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_29561_end_mask_0 = const()[name = tensor("op_29561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29561_cast_fp16 = slice_by_index(begin = var_29561_begin_0, end = var_29561_end_0, end_mask = var_29561_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29561_cast_fp16")]; + tensor var_29565_begin_0 = const()[name = tensor("op_29565_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29565_end_0 = const()[name = tensor("op_29565_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_29565_end_mask_0 = const()[name = tensor("op_29565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29565_cast_fp16 = slice_by_index(begin = var_29565_begin_0, end = var_29565_end_0, end_mask = var_29565_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29565_cast_fp16")]; + tensor var_29569_begin_0 = const()[name = tensor("op_29569_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_29569_end_0 = const()[name = tensor("op_29569_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_29569_end_mask_0 = const()[name = tensor("op_29569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29569_cast_fp16 = slice_by_index(begin = var_29569_begin_0, end = var_29569_end_0, end_mask = var_29569_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29569_cast_fp16")]; + tensor var_29571_begin_0 = const()[name = tensor("op_29571_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29571_end_0 = const()[name = tensor("op_29571_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_29571_end_mask_0 = const()[name = tensor("op_29571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29571_cast_fp16 = slice_by_index(begin = var_29571_begin_0, end = var_29571_end_0, end_mask = var_29571_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29571_cast_fp16")]; + tensor var_29575_begin_0 = const()[name = tensor("op_29575_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29575_end_0 = const()[name = tensor("op_29575_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_29575_end_mask_0 = const()[name = tensor("op_29575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29575_cast_fp16 = slice_by_index(begin = var_29575_begin_0, end = var_29575_end_0, end_mask = var_29575_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29575_cast_fp16")]; + tensor var_29579_begin_0 = const()[name = tensor("op_29579_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29579_end_0 = const()[name = tensor("op_29579_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_29579_end_mask_0 = const()[name = tensor("op_29579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29579_cast_fp16 = slice_by_index(begin = var_29579_begin_0, end = var_29579_end_0, end_mask = var_29579_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29579_cast_fp16")]; + tensor var_29583_begin_0 = const()[name = tensor("op_29583_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29583_end_0 = const()[name = tensor("op_29583_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_29583_end_mask_0 = const()[name = tensor("op_29583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29583_cast_fp16 = slice_by_index(begin = var_29583_begin_0, end = var_29583_end_0, end_mask = var_29583_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29583_cast_fp16")]; + tensor var_29587_begin_0 = const()[name = tensor("op_29587_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29587_end_0 = const()[name = tensor("op_29587_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_29587_end_mask_0 = const()[name = tensor("op_29587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29587_cast_fp16 = slice_by_index(begin = var_29587_begin_0, end = var_29587_end_0, end_mask = var_29587_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29587_cast_fp16")]; + tensor var_29591_begin_0 = const()[name = tensor("op_29591_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29591_end_0 = const()[name = tensor("op_29591_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_29591_end_mask_0 = const()[name = tensor("op_29591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29591_cast_fp16 = slice_by_index(begin = var_29591_begin_0, end = var_29591_end_0, end_mask = var_29591_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29591_cast_fp16")]; + tensor var_29595_begin_0 = const()[name = tensor("op_29595_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29595_end_0 = const()[name = tensor("op_29595_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_29595_end_mask_0 = const()[name = tensor("op_29595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29595_cast_fp16 = slice_by_index(begin = var_29595_begin_0, end = var_29595_end_0, end_mask = var_29595_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29595_cast_fp16")]; + tensor var_29599_begin_0 = const()[name = tensor("op_29599_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29599_end_0 = const()[name = tensor("op_29599_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_29599_end_mask_0 = const()[name = tensor("op_29599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29599_cast_fp16 = slice_by_index(begin = var_29599_begin_0, end = var_29599_end_0, end_mask = var_29599_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29599_cast_fp16")]; + tensor var_29603_begin_0 = const()[name = tensor("op_29603_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29603_end_0 = const()[name = tensor("op_29603_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_29603_end_mask_0 = const()[name = tensor("op_29603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29603_cast_fp16 = slice_by_index(begin = var_29603_begin_0, end = var_29603_end_0, end_mask = var_29603_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29603_cast_fp16")]; + tensor var_29607_begin_0 = const()[name = tensor("op_29607_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29607_end_0 = const()[name = tensor("op_29607_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_29607_end_mask_0 = const()[name = tensor("op_29607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29607_cast_fp16 = slice_by_index(begin = var_29607_begin_0, end = var_29607_end_0, end_mask = var_29607_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29607_cast_fp16")]; + tensor var_29611_begin_0 = const()[name = tensor("op_29611_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29611_end_0 = const()[name = tensor("op_29611_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_29611_end_mask_0 = const()[name = tensor("op_29611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29611_cast_fp16 = slice_by_index(begin = var_29611_begin_0, end = var_29611_end_0, end_mask = var_29611_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29611_cast_fp16")]; + tensor var_29615_begin_0 = const()[name = tensor("op_29615_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29615_end_0 = const()[name = tensor("op_29615_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_29615_end_mask_0 = const()[name = tensor("op_29615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29615_cast_fp16 = slice_by_index(begin = var_29615_begin_0, end = var_29615_end_0, end_mask = var_29615_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29615_cast_fp16")]; + tensor var_29619_begin_0 = const()[name = tensor("op_29619_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29619_end_0 = const()[name = tensor("op_29619_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_29619_end_mask_0 = const()[name = tensor("op_29619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29619_cast_fp16 = slice_by_index(begin = var_29619_begin_0, end = var_29619_end_0, end_mask = var_29619_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29619_cast_fp16")]; + tensor var_29623_begin_0 = const()[name = tensor("op_29623_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29623_end_0 = const()[name = tensor("op_29623_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_29623_end_mask_0 = const()[name = tensor("op_29623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29623_cast_fp16 = slice_by_index(begin = var_29623_begin_0, end = var_29623_end_0, end_mask = var_29623_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29623_cast_fp16")]; + tensor var_29627_begin_0 = const()[name = tensor("op_29627_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29627_end_0 = const()[name = tensor("op_29627_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_29627_end_mask_0 = const()[name = tensor("op_29627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29627_cast_fp16 = slice_by_index(begin = var_29627_begin_0, end = var_29627_end_0, end_mask = var_29627_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29627_cast_fp16")]; + tensor var_29631_begin_0 = const()[name = tensor("op_29631_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29631_end_0 = const()[name = tensor("op_29631_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29631_end_mask_0 = const()[name = tensor("op_29631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29631_cast_fp16 = slice_by_index(begin = var_29631_begin_0, end = var_29631_end_0, end_mask = var_29631_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29631_cast_fp16")]; + tensor var_29635_begin_0 = const()[name = tensor("op_29635_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29635_end_0 = const()[name = tensor("op_29635_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_29635_end_mask_0 = const()[name = tensor("op_29635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29635_cast_fp16 = slice_by_index(begin = var_29635_begin_0, end = var_29635_end_0, end_mask = var_29635_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29635_cast_fp16")]; + tensor var_29639_begin_0 = const()[name = tensor("op_29639_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29639_end_0 = const()[name = tensor("op_29639_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_29639_end_mask_0 = const()[name = tensor("op_29639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29639_cast_fp16 = slice_by_index(begin = var_29639_begin_0, end = var_29639_end_0, end_mask = var_29639_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29639_cast_fp16")]; + tensor var_29643_begin_0 = const()[name = tensor("op_29643_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29643_end_0 = const()[name = tensor("op_29643_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29643_end_mask_0 = const()[name = tensor("op_29643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29643_cast_fp16 = slice_by_index(begin = var_29643_begin_0, end = var_29643_end_0, end_mask = var_29643_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29643_cast_fp16")]; + tensor var_29647_begin_0 = const()[name = tensor("op_29647_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29647_end_0 = const()[name = tensor("op_29647_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29647_end_mask_0 = const()[name = tensor("op_29647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29647_cast_fp16 = slice_by_index(begin = var_29647_begin_0, end = var_29647_end_0, end_mask = var_29647_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29647_cast_fp16")]; + tensor var_29651_equation_0 = const()[name = tensor("op_29651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29651_cast_fp16 = einsum(equation = var_29651_equation_0, values = (var_29493_cast_fp16, var_29410_cast_fp16))[name = tensor("op_29651_cast_fp16")]; + tensor var_29652_to_fp16 = const()[name = tensor("op_29652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2481_cast_fp16 = mul(x = var_29651_cast_fp16, y = var_29652_to_fp16)[name = tensor("aw_2481_cast_fp16")]; + tensor var_29655_equation_0 = const()[name = tensor("op_29655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29655_cast_fp16 = einsum(equation = var_29655_equation_0, values = (var_29497_cast_fp16, var_29414_cast_fp16))[name = tensor("op_29655_cast_fp16")]; + tensor var_29656_to_fp16 = const()[name = tensor("op_29656_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2483_cast_fp16 = mul(x = var_29655_cast_fp16, y = var_29656_to_fp16)[name = tensor("aw_2483_cast_fp16")]; + tensor var_29659_equation_0 = const()[name = tensor("op_29659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29659_cast_fp16 = einsum(equation = var_29659_equation_0, values = (var_29501_cast_fp16, var_29418_cast_fp16))[name = tensor("op_29659_cast_fp16")]; + tensor var_29660_to_fp16 = const()[name = tensor("op_29660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2485_cast_fp16 = mul(x = var_29659_cast_fp16, y = var_29660_to_fp16)[name = tensor("aw_2485_cast_fp16")]; + tensor var_29663_equation_0 = const()[name = tensor("op_29663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29663_cast_fp16 = einsum(equation = var_29663_equation_0, values = (var_29505_cast_fp16, var_29422_cast_fp16))[name = tensor("op_29663_cast_fp16")]; + tensor var_29664_to_fp16 = const()[name = tensor("op_29664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2487_cast_fp16 = mul(x = var_29663_cast_fp16, y = var_29664_to_fp16)[name = tensor("aw_2487_cast_fp16")]; + tensor var_29667_equation_0 = const()[name = tensor("op_29667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29667_cast_fp16 = einsum(equation = var_29667_equation_0, values = (var_29509_cast_fp16, var_29426_cast_fp16))[name = tensor("op_29667_cast_fp16")]; + tensor var_29668_to_fp16 = const()[name = tensor("op_29668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2489_cast_fp16 = mul(x = var_29667_cast_fp16, y = var_29668_to_fp16)[name = tensor("aw_2489_cast_fp16")]; + tensor var_29671_equation_0 = const()[name = tensor("op_29671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29671_cast_fp16 = einsum(equation = var_29671_equation_0, values = (var_29513_cast_fp16, var_29430_cast_fp16))[name = tensor("op_29671_cast_fp16")]; + tensor var_29672_to_fp16 = const()[name = tensor("op_29672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2491_cast_fp16 = mul(x = var_29671_cast_fp16, y = var_29672_to_fp16)[name = tensor("aw_2491_cast_fp16")]; + tensor var_29675_equation_0 = const()[name = tensor("op_29675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29675_cast_fp16 = einsum(equation = var_29675_equation_0, values = (var_29517_cast_fp16, var_29434_cast_fp16))[name = tensor("op_29675_cast_fp16")]; + tensor var_29676_to_fp16 = const()[name = tensor("op_29676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2493_cast_fp16 = mul(x = var_29675_cast_fp16, y = var_29676_to_fp16)[name = tensor("aw_2493_cast_fp16")]; + tensor var_29679_equation_0 = const()[name = tensor("op_29679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29679_cast_fp16 = einsum(equation = var_29679_equation_0, values = (var_29521_cast_fp16, var_29438_cast_fp16))[name = tensor("op_29679_cast_fp16")]; + tensor var_29680_to_fp16 = const()[name = tensor("op_29680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2495_cast_fp16 = mul(x = var_29679_cast_fp16, y = var_29680_to_fp16)[name = tensor("aw_2495_cast_fp16")]; + tensor var_29683_equation_0 = const()[name = tensor("op_29683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29683_cast_fp16 = einsum(equation = var_29683_equation_0, values = (var_29525_cast_fp16, var_29442_cast_fp16))[name = tensor("op_29683_cast_fp16")]; + tensor var_29684_to_fp16 = const()[name = tensor("op_29684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2497_cast_fp16 = mul(x = var_29683_cast_fp16, y = var_29684_to_fp16)[name = tensor("aw_2497_cast_fp16")]; + tensor var_29687_equation_0 = const()[name = tensor("op_29687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29687_cast_fp16 = einsum(equation = var_29687_equation_0, values = (var_29529_cast_fp16, var_29446_cast_fp16))[name = tensor("op_29687_cast_fp16")]; + tensor var_29688_to_fp16 = const()[name = tensor("op_29688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2499_cast_fp16 = mul(x = var_29687_cast_fp16, y = var_29688_to_fp16)[name = tensor("aw_2499_cast_fp16")]; + tensor var_29691_equation_0 = const()[name = tensor("op_29691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29691_cast_fp16 = einsum(equation = var_29691_equation_0, values = (var_29533_cast_fp16, var_29450_cast_fp16))[name = tensor("op_29691_cast_fp16")]; + tensor var_29692_to_fp16 = const()[name = tensor("op_29692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2501_cast_fp16 = mul(x = var_29691_cast_fp16, y = var_29692_to_fp16)[name = tensor("aw_2501_cast_fp16")]; + tensor var_29695_equation_0 = const()[name = tensor("op_29695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29695_cast_fp16 = einsum(equation = var_29695_equation_0, values = (var_29537_cast_fp16, var_29454_cast_fp16))[name = tensor("op_29695_cast_fp16")]; + tensor var_29696_to_fp16 = const()[name = tensor("op_29696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2503_cast_fp16 = mul(x = var_29695_cast_fp16, y = var_29696_to_fp16)[name = tensor("aw_2503_cast_fp16")]; + tensor var_29699_equation_0 = const()[name = tensor("op_29699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29699_cast_fp16 = einsum(equation = var_29699_equation_0, values = (var_29541_cast_fp16, var_29458_cast_fp16))[name = tensor("op_29699_cast_fp16")]; + tensor var_29700_to_fp16 = const()[name = tensor("op_29700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2505_cast_fp16 = mul(x = var_29699_cast_fp16, y = var_29700_to_fp16)[name = tensor("aw_2505_cast_fp16")]; + tensor var_29703_equation_0 = const()[name = tensor("op_29703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29703_cast_fp16 = einsum(equation = var_29703_equation_0, values = (var_29545_cast_fp16, var_29462_cast_fp16))[name = tensor("op_29703_cast_fp16")]; + tensor var_29704_to_fp16 = const()[name = tensor("op_29704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2507_cast_fp16 = mul(x = var_29703_cast_fp16, y = var_29704_to_fp16)[name = tensor("aw_2507_cast_fp16")]; + tensor var_29707_equation_0 = const()[name = tensor("op_29707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29707_cast_fp16 = einsum(equation = var_29707_equation_0, values = (var_29549_cast_fp16, var_29466_cast_fp16))[name = tensor("op_29707_cast_fp16")]; + tensor var_29708_to_fp16 = const()[name = tensor("op_29708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2509_cast_fp16 = mul(x = var_29707_cast_fp16, y = var_29708_to_fp16)[name = tensor("aw_2509_cast_fp16")]; + tensor var_29711_equation_0 = const()[name = tensor("op_29711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29711_cast_fp16 = einsum(equation = var_29711_equation_0, values = (var_29553_cast_fp16, var_29470_cast_fp16))[name = tensor("op_29711_cast_fp16")]; + tensor var_29712_to_fp16 = const()[name = tensor("op_29712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2511_cast_fp16 = mul(x = var_29711_cast_fp16, y = var_29712_to_fp16)[name = tensor("aw_2511_cast_fp16")]; + tensor var_29715_equation_0 = const()[name = tensor("op_29715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29715_cast_fp16 = einsum(equation = var_29715_equation_0, values = (var_29557_cast_fp16, var_29474_cast_fp16))[name = tensor("op_29715_cast_fp16")]; + tensor var_29716_to_fp16 = const()[name = tensor("op_29716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2513_cast_fp16 = mul(x = var_29715_cast_fp16, y = var_29716_to_fp16)[name = tensor("aw_2513_cast_fp16")]; + tensor var_29719_equation_0 = const()[name = tensor("op_29719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29719_cast_fp16 = einsum(equation = var_29719_equation_0, values = (var_29561_cast_fp16, var_29478_cast_fp16))[name = tensor("op_29719_cast_fp16")]; + tensor var_29720_to_fp16 = const()[name = tensor("op_29720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2515_cast_fp16 = mul(x = var_29719_cast_fp16, y = var_29720_to_fp16)[name = tensor("aw_2515_cast_fp16")]; + tensor var_29723_equation_0 = const()[name = tensor("op_29723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29723_cast_fp16 = einsum(equation = var_29723_equation_0, values = (var_29565_cast_fp16, var_29482_cast_fp16))[name = tensor("op_29723_cast_fp16")]; + tensor var_29724_to_fp16 = const()[name = tensor("op_29724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2517_cast_fp16 = mul(x = var_29723_cast_fp16, y = var_29724_to_fp16)[name = tensor("aw_2517_cast_fp16")]; + tensor var_29727_equation_0 = const()[name = tensor("op_29727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29727_cast_fp16 = einsum(equation = var_29727_equation_0, values = (var_29569_cast_fp16, var_29486_cast_fp16))[name = tensor("op_29727_cast_fp16")]; + tensor var_29728_to_fp16 = const()[name = tensor("op_29728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2519_cast_fp16 = mul(x = var_29727_cast_fp16, y = var_29728_to_fp16)[name = tensor("aw_2519_cast_fp16")]; + tensor var_29730_cast_fp16 = softmax(axis = var_21077, x = aw_2481_cast_fp16)[name = tensor("op_29730_cast_fp16")]; + tensor var_29731_cast_fp16 = softmax(axis = var_21077, x = aw_2483_cast_fp16)[name = tensor("op_29731_cast_fp16")]; + tensor var_29732_cast_fp16 = softmax(axis = var_21077, x = aw_2485_cast_fp16)[name = tensor("op_29732_cast_fp16")]; + tensor var_29733_cast_fp16 = softmax(axis = var_21077, x = aw_2487_cast_fp16)[name = tensor("op_29733_cast_fp16")]; + tensor var_29734_cast_fp16 = softmax(axis = var_21077, x = aw_2489_cast_fp16)[name = tensor("op_29734_cast_fp16")]; + tensor var_29735_cast_fp16 = softmax(axis = var_21077, x = aw_2491_cast_fp16)[name = tensor("op_29735_cast_fp16")]; + tensor var_29736_cast_fp16 = softmax(axis = var_21077, x = aw_2493_cast_fp16)[name = tensor("op_29736_cast_fp16")]; + tensor var_29737_cast_fp16 = softmax(axis = var_21077, x = aw_2495_cast_fp16)[name = tensor("op_29737_cast_fp16")]; + tensor var_29738_cast_fp16 = softmax(axis = var_21077, x = aw_2497_cast_fp16)[name = tensor("op_29738_cast_fp16")]; + tensor var_29739_cast_fp16 = softmax(axis = var_21077, x = aw_2499_cast_fp16)[name = tensor("op_29739_cast_fp16")]; + tensor var_29740_cast_fp16 = softmax(axis = var_21077, x = aw_2501_cast_fp16)[name = tensor("op_29740_cast_fp16")]; + tensor var_29741_cast_fp16 = softmax(axis = var_21077, x = aw_2503_cast_fp16)[name = tensor("op_29741_cast_fp16")]; + tensor var_29742_cast_fp16 = softmax(axis = var_21077, x = aw_2505_cast_fp16)[name = tensor("op_29742_cast_fp16")]; + tensor var_29743_cast_fp16 = softmax(axis = var_21077, x = aw_2507_cast_fp16)[name = tensor("op_29743_cast_fp16")]; + tensor var_29744_cast_fp16 = softmax(axis = var_21077, x = aw_2509_cast_fp16)[name = tensor("op_29744_cast_fp16")]; + tensor var_29745_cast_fp16 = softmax(axis = var_21077, x = aw_2511_cast_fp16)[name = tensor("op_29745_cast_fp16")]; + tensor var_29746_cast_fp16 = softmax(axis = var_21077, x = aw_2513_cast_fp16)[name = tensor("op_29746_cast_fp16")]; + tensor var_29747_cast_fp16 = softmax(axis = var_21077, x = aw_2515_cast_fp16)[name = tensor("op_29747_cast_fp16")]; + tensor var_29748_cast_fp16 = softmax(axis = var_21077, x = aw_2517_cast_fp16)[name = tensor("op_29748_cast_fp16")]; + tensor var_29749_cast_fp16 = softmax(axis = var_21077, x = aw_2519_cast_fp16)[name = tensor("op_29749_cast_fp16")]; + tensor var_29751_equation_0 = const()[name = tensor("op_29751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29751_cast_fp16 = einsum(equation = var_29751_equation_0, values = (var_29571_cast_fp16, var_29730_cast_fp16))[name = tensor("op_29751_cast_fp16")]; + tensor var_29753_equation_0 = const()[name = tensor("op_29753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29753_cast_fp16 = einsum(equation = var_29753_equation_0, values = (var_29575_cast_fp16, var_29731_cast_fp16))[name = tensor("op_29753_cast_fp16")]; + tensor var_29755_equation_0 = const()[name = tensor("op_29755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29755_cast_fp16 = einsum(equation = var_29755_equation_0, values = (var_29579_cast_fp16, var_29732_cast_fp16))[name = tensor("op_29755_cast_fp16")]; + tensor var_29757_equation_0 = const()[name = tensor("op_29757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29757_cast_fp16 = einsum(equation = var_29757_equation_0, values = (var_29583_cast_fp16, var_29733_cast_fp16))[name = tensor("op_29757_cast_fp16")]; + tensor var_29759_equation_0 = const()[name = tensor("op_29759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29759_cast_fp16 = einsum(equation = var_29759_equation_0, values = (var_29587_cast_fp16, var_29734_cast_fp16))[name = tensor("op_29759_cast_fp16")]; + tensor var_29761_equation_0 = const()[name = tensor("op_29761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29761_cast_fp16 = einsum(equation = var_29761_equation_0, values = (var_29591_cast_fp16, var_29735_cast_fp16))[name = tensor("op_29761_cast_fp16")]; + tensor var_29763_equation_0 = const()[name = tensor("op_29763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29763_cast_fp16 = einsum(equation = var_29763_equation_0, values = (var_29595_cast_fp16, var_29736_cast_fp16))[name = tensor("op_29763_cast_fp16")]; + tensor var_29765_equation_0 = const()[name = tensor("op_29765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29765_cast_fp16 = einsum(equation = var_29765_equation_0, values = (var_29599_cast_fp16, var_29737_cast_fp16))[name = tensor("op_29765_cast_fp16")]; + tensor var_29767_equation_0 = const()[name = tensor("op_29767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29767_cast_fp16 = einsum(equation = var_29767_equation_0, values = (var_29603_cast_fp16, var_29738_cast_fp16))[name = tensor("op_29767_cast_fp16")]; + tensor var_29769_equation_0 = const()[name = tensor("op_29769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29769_cast_fp16 = einsum(equation = var_29769_equation_0, values = (var_29607_cast_fp16, var_29739_cast_fp16))[name = tensor("op_29769_cast_fp16")]; + tensor var_29771_equation_0 = const()[name = tensor("op_29771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29771_cast_fp16 = einsum(equation = var_29771_equation_0, values = (var_29611_cast_fp16, var_29740_cast_fp16))[name = tensor("op_29771_cast_fp16")]; + tensor var_29773_equation_0 = const()[name = tensor("op_29773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29773_cast_fp16 = einsum(equation = var_29773_equation_0, values = (var_29615_cast_fp16, var_29741_cast_fp16))[name = tensor("op_29773_cast_fp16")]; + tensor var_29775_equation_0 = const()[name = tensor("op_29775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29775_cast_fp16 = einsum(equation = var_29775_equation_0, values = (var_29619_cast_fp16, var_29742_cast_fp16))[name = tensor("op_29775_cast_fp16")]; + tensor var_29777_equation_0 = const()[name = tensor("op_29777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29777_cast_fp16 = einsum(equation = var_29777_equation_0, values = (var_29623_cast_fp16, var_29743_cast_fp16))[name = tensor("op_29777_cast_fp16")]; + tensor var_29779_equation_0 = const()[name = tensor("op_29779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29779_cast_fp16 = einsum(equation = var_29779_equation_0, values = (var_29627_cast_fp16, var_29744_cast_fp16))[name = tensor("op_29779_cast_fp16")]; + tensor var_29781_equation_0 = const()[name = tensor("op_29781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29781_cast_fp16 = einsum(equation = var_29781_equation_0, values = (var_29631_cast_fp16, var_29745_cast_fp16))[name = tensor("op_29781_cast_fp16")]; + tensor var_29783_equation_0 = const()[name = tensor("op_29783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29783_cast_fp16 = einsum(equation = var_29783_equation_0, values = (var_29635_cast_fp16, var_29746_cast_fp16))[name = tensor("op_29783_cast_fp16")]; + tensor var_29785_equation_0 = const()[name = tensor("op_29785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29785_cast_fp16 = einsum(equation = var_29785_equation_0, values = (var_29639_cast_fp16, var_29747_cast_fp16))[name = tensor("op_29785_cast_fp16")]; + tensor var_29787_equation_0 = const()[name = tensor("op_29787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29787_cast_fp16 = einsum(equation = var_29787_equation_0, values = (var_29643_cast_fp16, var_29748_cast_fp16))[name = tensor("op_29787_cast_fp16")]; + tensor var_29789_equation_0 = const()[name = tensor("op_29789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29789_cast_fp16 = einsum(equation = var_29789_equation_0, values = (var_29647_cast_fp16, var_29749_cast_fp16))[name = tensor("op_29789_cast_fp16")]; + tensor input_399_interleave_0 = const()[name = tensor("input_399_interleave_0"), val = tensor(false)]; + tensor input_399_cast_fp16 = concat(axis = var_21077, interleave = input_399_interleave_0, values = (var_29751_cast_fp16, var_29753_cast_fp16, var_29755_cast_fp16, var_29757_cast_fp16, var_29759_cast_fp16, var_29761_cast_fp16, var_29763_cast_fp16, var_29765_cast_fp16, var_29767_cast_fp16, var_29769_cast_fp16, var_29771_cast_fp16, var_29773_cast_fp16, var_29775_cast_fp16, var_29777_cast_fp16, var_29779_cast_fp16, var_29781_cast_fp16, var_29783_cast_fp16, var_29785_cast_fp16, var_29787_cast_fp16, var_29789_cast_fp16))[name = tensor("input_399_cast_fp16")]; + tensor var_29799_pad_type_0 = const()[name = tensor("op_29799_pad_type_0"), val = tensor("valid")]; + tensor var_29799_strides_0 = const()[name = tensor("op_29799_strides_0"), val = tensor([1, 1])]; + tensor var_29799_pad_0 = const()[name = tensor("op_29799_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29799_dilations_0 = const()[name = tensor("op_29799_dilations_0"), val = tensor([1, 1])]; + tensor var_29799_groups_0 = const()[name = tensor("op_29799_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886623360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(887852224))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(887852416)))]; + tensor var_29799_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_29799_dilations_0, groups = var_29799_groups_0, pad = var_29799_pad_0, pad_type = var_29799_pad_type_0, strides = var_29799_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized, x = input_399_cast_fp16)[name = tensor("op_29799_cast_fp16")]; + tensor inputs_201_cast_fp16 = add(x = var_29799_cast_fp16, y = inputs_199_cast_fp16)[name = tensor("inputs_201_cast_fp16")]; + tensor hidden_states_265_axes_0 = const()[name = tensor("hidden_states_265_axes_0"), val = tensor([1])]; + tensor hidden_states_265_gamma_0_to_fp16 = const()[name = tensor("hidden_states_265_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(887855040)))]; + tensor hidden_states_265_beta_0_to_fp16 = const()[name = tensor("hidden_states_265_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(887857664)))]; + tensor var_29809_to_fp16 = const()[name = tensor("op_29809_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_265_cast_fp16 = layer_norm(axes = hidden_states_265_axes_0, beta = hidden_states_265_beta_0_to_fp16, epsilon = var_29809_to_fp16, gamma = hidden_states_265_gamma_0_to_fp16, x = inputs_201_cast_fp16)[name = tensor("hidden_states_265_cast_fp16")]; + tensor q_135_pad_type_0 = const()[name = tensor("q_135_pad_type_0"), val = tensor("valid")]; + tensor q_135_strides_0 = const()[name = tensor("q_135_strides_0"), val = tensor([1, 1])]; + tensor q_135_pad_0 = const()[name = tensor("q_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor q_135_dilations_0 = const()[name = tensor("q_135_dilations_0"), val = tensor([1, 1])]; + tensor q_135_groups_0 = const()[name = tensor("q_135_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(887860288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889089152))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_135_cast_fp16 = conv(dilations = q_135_dilations_0, groups = q_135_groups_0, pad = q_135_pad_0, pad_type = q_135_pad_type_0, strides = q_135_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_265_cast_fp16)[name = tensor("q_135_cast_fp16")]; + tensor k_269_pad_type_0 = const()[name = tensor("k_269_pad_type_0"), val = tensor("valid")]; + tensor k_269_strides_0 = const()[name = tensor("k_269_strides_0"), val = tensor([1, 1])]; + tensor k_269_pad_0 = const()[name = tensor("k_269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_269_dilations_0 = const()[name = tensor("k_269_dilations_0"), val = tensor([1, 1])]; + tensor k_269_groups_0 = const()[name = tensor("k_269_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889089344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(891055488))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor k_269_cast_fp16 = conv(dilations = k_269_dilations_0, groups = k_269_groups_0, pad = k_269_pad_0, pad_type = k_269_pad_type_0, strides = k_269_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_269_cast_fp16")]; + tensor v_135_pad_type_0 = const()[name = tensor("v_135_pad_type_0"), val = tensor("valid")]; + tensor v_135_strides_0 = const()[name = tensor("v_135_strides_0"), val = tensor([1, 1])]; + tensor v_135_pad_0 = const()[name = tensor("v_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_135_dilations_0 = const()[name = tensor("v_135_dilations_0"), val = tensor([1, 1])]; + tensor v_135_groups_0 = const()[name = tensor("v_135_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(891055680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893021824))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 2048, 1, 1])]; + tensor v_135_cast_fp16 = conv(dilations = v_135_dilations_0, groups = v_135_groups_0, pad = v_135_pad_0, pad_type = v_135_pad_type_0, strides = v_135_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_135_cast_fp16")]; + tensor var_29842_begin_0 = const()[name = tensor("op_29842_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29842_end_0 = const()[name = tensor("op_29842_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_29842_end_mask_0 = const()[name = tensor("op_29842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29842_cast_fp16 = slice_by_index(begin = var_29842_begin_0, end = var_29842_end_0, end_mask = var_29842_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29842_cast_fp16")]; + tensor var_29846_begin_0 = const()[name = tensor("op_29846_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29846_end_0 = const()[name = tensor("op_29846_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_29846_end_mask_0 = const()[name = tensor("op_29846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29846_cast_fp16 = slice_by_index(begin = var_29846_begin_0, end = var_29846_end_0, end_mask = var_29846_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29846_cast_fp16")]; + tensor var_29850_begin_0 = const()[name = tensor("op_29850_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29850_end_0 = const()[name = tensor("op_29850_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_29850_end_mask_0 = const()[name = tensor("op_29850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29850_cast_fp16 = slice_by_index(begin = var_29850_begin_0, end = var_29850_end_0, end_mask = var_29850_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29850_cast_fp16")]; + tensor var_29854_begin_0 = const()[name = tensor("op_29854_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29854_end_0 = const()[name = tensor("op_29854_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_29854_end_mask_0 = const()[name = tensor("op_29854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29854_cast_fp16 = slice_by_index(begin = var_29854_begin_0, end = var_29854_end_0, end_mask = var_29854_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29854_cast_fp16")]; + tensor var_29858_begin_0 = const()[name = tensor("op_29858_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29858_end_0 = const()[name = tensor("op_29858_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_29858_end_mask_0 = const()[name = tensor("op_29858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29858_cast_fp16 = slice_by_index(begin = var_29858_begin_0, end = var_29858_end_0, end_mask = var_29858_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29858_cast_fp16")]; + tensor var_29862_begin_0 = const()[name = tensor("op_29862_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29862_end_0 = const()[name = tensor("op_29862_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_29862_end_mask_0 = const()[name = tensor("op_29862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29862_cast_fp16 = slice_by_index(begin = var_29862_begin_0, end = var_29862_end_0, end_mask = var_29862_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29862_cast_fp16")]; + tensor var_29866_begin_0 = const()[name = tensor("op_29866_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29866_end_0 = const()[name = tensor("op_29866_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_29866_end_mask_0 = const()[name = tensor("op_29866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29866_cast_fp16 = slice_by_index(begin = var_29866_begin_0, end = var_29866_end_0, end_mask = var_29866_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29866_cast_fp16")]; + tensor var_29870_begin_0 = const()[name = tensor("op_29870_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29870_end_0 = const()[name = tensor("op_29870_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_29870_end_mask_0 = const()[name = tensor("op_29870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29870_cast_fp16 = slice_by_index(begin = var_29870_begin_0, end = var_29870_end_0, end_mask = var_29870_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29870_cast_fp16")]; + tensor var_29874_begin_0 = const()[name = tensor("op_29874_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29874_end_0 = const()[name = tensor("op_29874_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_29874_end_mask_0 = const()[name = tensor("op_29874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29874_cast_fp16 = slice_by_index(begin = var_29874_begin_0, end = var_29874_end_0, end_mask = var_29874_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29874_cast_fp16")]; + tensor var_29878_begin_0 = const()[name = tensor("op_29878_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29878_end_0 = const()[name = tensor("op_29878_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_29878_end_mask_0 = const()[name = tensor("op_29878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29878_cast_fp16 = slice_by_index(begin = var_29878_begin_0, end = var_29878_end_0, end_mask = var_29878_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29878_cast_fp16")]; + tensor var_29882_begin_0 = const()[name = tensor("op_29882_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29882_end_0 = const()[name = tensor("op_29882_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_29882_end_mask_0 = const()[name = tensor("op_29882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29882_cast_fp16 = slice_by_index(begin = var_29882_begin_0, end = var_29882_end_0, end_mask = var_29882_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29882_cast_fp16")]; + tensor var_29886_begin_0 = const()[name = tensor("op_29886_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29886_end_0 = const()[name = tensor("op_29886_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_29886_end_mask_0 = const()[name = tensor("op_29886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29886_cast_fp16 = slice_by_index(begin = var_29886_begin_0, end = var_29886_end_0, end_mask = var_29886_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29886_cast_fp16")]; + tensor var_29890_begin_0 = const()[name = tensor("op_29890_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29890_end_0 = const()[name = tensor("op_29890_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_29890_end_mask_0 = const()[name = tensor("op_29890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29890_cast_fp16 = slice_by_index(begin = var_29890_begin_0, end = var_29890_end_0, end_mask = var_29890_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29890_cast_fp16")]; + tensor var_29894_begin_0 = const()[name = tensor("op_29894_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29894_end_0 = const()[name = tensor("op_29894_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_29894_end_mask_0 = const()[name = tensor("op_29894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29894_cast_fp16 = slice_by_index(begin = var_29894_begin_0, end = var_29894_end_0, end_mask = var_29894_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29894_cast_fp16")]; + tensor var_29898_begin_0 = const()[name = tensor("op_29898_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29898_end_0 = const()[name = tensor("op_29898_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_29898_end_mask_0 = const()[name = tensor("op_29898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29898_cast_fp16 = slice_by_index(begin = var_29898_begin_0, end = var_29898_end_0, end_mask = var_29898_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29898_cast_fp16")]; + tensor var_29902_begin_0 = const()[name = tensor("op_29902_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29902_end_0 = const()[name = tensor("op_29902_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29902_end_mask_0 = const()[name = tensor("op_29902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29902_cast_fp16 = slice_by_index(begin = var_29902_begin_0, end = var_29902_end_0, end_mask = var_29902_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29902_cast_fp16")]; + tensor var_29906_begin_0 = const()[name = tensor("op_29906_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29906_end_0 = const()[name = tensor("op_29906_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_29906_end_mask_0 = const()[name = tensor("op_29906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29906_cast_fp16 = slice_by_index(begin = var_29906_begin_0, end = var_29906_end_0, end_mask = var_29906_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29906_cast_fp16")]; + tensor var_29910_begin_0 = const()[name = tensor("op_29910_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29910_end_0 = const()[name = tensor("op_29910_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_29910_end_mask_0 = const()[name = tensor("op_29910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29910_cast_fp16 = slice_by_index(begin = var_29910_begin_0, end = var_29910_end_0, end_mask = var_29910_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29910_cast_fp16")]; + tensor var_29914_begin_0 = const()[name = tensor("op_29914_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29914_end_0 = const()[name = tensor("op_29914_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29914_end_mask_0 = const()[name = tensor("op_29914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29914_cast_fp16 = slice_by_index(begin = var_29914_begin_0, end = var_29914_end_0, end_mask = var_29914_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29914_cast_fp16")]; + tensor var_29918_begin_0 = const()[name = tensor("op_29918_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29918_end_0 = const()[name = tensor("op_29918_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29918_end_mask_0 = const()[name = tensor("op_29918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29918_cast_fp16 = slice_by_index(begin = var_29918_begin_0, end = var_29918_end_0, end_mask = var_29918_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29918_cast_fp16")]; + tensor k_271_perm_0 = const()[name = tensor("k_271_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29925_begin_0 = const()[name = tensor("op_29925_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29925_end_0 = const()[name = tensor("op_29925_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_29925_end_mask_0 = const()[name = tensor("op_29925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_271_cast_fp16 = transpose(perm = k_271_perm_0, x = k_269_cast_fp16)[name = tensor("transpose_0")]; + tensor var_29925_cast_fp16 = slice_by_index(begin = var_29925_begin_0, end = var_29925_end_0, end_mask = var_29925_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29925_cast_fp16")]; + tensor var_29929_begin_0 = const()[name = tensor("op_29929_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29929_end_0 = const()[name = tensor("op_29929_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_29929_end_mask_0 = const()[name = tensor("op_29929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29929_cast_fp16 = slice_by_index(begin = var_29929_begin_0, end = var_29929_end_0, end_mask = var_29929_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29929_cast_fp16")]; + tensor var_29933_begin_0 = const()[name = tensor("op_29933_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29933_end_0 = const()[name = tensor("op_29933_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_29933_end_mask_0 = const()[name = tensor("op_29933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29933_cast_fp16 = slice_by_index(begin = var_29933_begin_0, end = var_29933_end_0, end_mask = var_29933_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29933_cast_fp16")]; + tensor var_29937_begin_0 = const()[name = tensor("op_29937_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29937_end_0 = const()[name = tensor("op_29937_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_29937_end_mask_0 = const()[name = tensor("op_29937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29937_cast_fp16 = slice_by_index(begin = var_29937_begin_0, end = var_29937_end_0, end_mask = var_29937_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29937_cast_fp16")]; + tensor var_29941_begin_0 = const()[name = tensor("op_29941_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29941_end_0 = const()[name = tensor("op_29941_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_29941_end_mask_0 = const()[name = tensor("op_29941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29941_cast_fp16 = slice_by_index(begin = var_29941_begin_0, end = var_29941_end_0, end_mask = var_29941_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29941_cast_fp16")]; + tensor var_29945_begin_0 = const()[name = tensor("op_29945_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29945_end_0 = const()[name = tensor("op_29945_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_29945_end_mask_0 = const()[name = tensor("op_29945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29945_cast_fp16 = slice_by_index(begin = var_29945_begin_0, end = var_29945_end_0, end_mask = var_29945_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29945_cast_fp16")]; + tensor var_29949_begin_0 = const()[name = tensor("op_29949_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29949_end_0 = const()[name = tensor("op_29949_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_29949_end_mask_0 = const()[name = tensor("op_29949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29949_cast_fp16 = slice_by_index(begin = var_29949_begin_0, end = var_29949_end_0, end_mask = var_29949_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29949_cast_fp16")]; + tensor var_29953_begin_0 = const()[name = tensor("op_29953_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29953_end_0 = const()[name = tensor("op_29953_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_29953_end_mask_0 = const()[name = tensor("op_29953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29953_cast_fp16 = slice_by_index(begin = var_29953_begin_0, end = var_29953_end_0, end_mask = var_29953_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29953_cast_fp16")]; + tensor var_29957_begin_0 = const()[name = tensor("op_29957_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29957_end_0 = const()[name = tensor("op_29957_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_29957_end_mask_0 = const()[name = tensor("op_29957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29957_cast_fp16 = slice_by_index(begin = var_29957_begin_0, end = var_29957_end_0, end_mask = var_29957_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29957_cast_fp16")]; + tensor var_29961_begin_0 = const()[name = tensor("op_29961_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29961_end_0 = const()[name = tensor("op_29961_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_29961_end_mask_0 = const()[name = tensor("op_29961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29961_cast_fp16 = slice_by_index(begin = var_29961_begin_0, end = var_29961_end_0, end_mask = var_29961_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29961_cast_fp16")]; + tensor var_29965_begin_0 = const()[name = tensor("op_29965_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29965_end_0 = const()[name = tensor("op_29965_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_29965_end_mask_0 = const()[name = tensor("op_29965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29965_cast_fp16 = slice_by_index(begin = var_29965_begin_0, end = var_29965_end_0, end_mask = var_29965_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29965_cast_fp16")]; + tensor var_29969_begin_0 = const()[name = tensor("op_29969_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29969_end_0 = const()[name = tensor("op_29969_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_29969_end_mask_0 = const()[name = tensor("op_29969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29969_cast_fp16 = slice_by_index(begin = var_29969_begin_0, end = var_29969_end_0, end_mask = var_29969_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29969_cast_fp16")]; + tensor var_29973_begin_0 = const()[name = tensor("op_29973_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29973_end_0 = const()[name = tensor("op_29973_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_29973_end_mask_0 = const()[name = tensor("op_29973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29973_cast_fp16 = slice_by_index(begin = var_29973_begin_0, end = var_29973_end_0, end_mask = var_29973_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29973_cast_fp16")]; + tensor var_29977_begin_0 = const()[name = tensor("op_29977_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29977_end_0 = const()[name = tensor("op_29977_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_29977_end_mask_0 = const()[name = tensor("op_29977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29977_cast_fp16 = slice_by_index(begin = var_29977_begin_0, end = var_29977_end_0, end_mask = var_29977_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29977_cast_fp16")]; + tensor var_29981_begin_0 = const()[name = tensor("op_29981_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29981_end_0 = const()[name = tensor("op_29981_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_29981_end_mask_0 = const()[name = tensor("op_29981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29981_cast_fp16 = slice_by_index(begin = var_29981_begin_0, end = var_29981_end_0, end_mask = var_29981_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29981_cast_fp16")]; + tensor var_29985_begin_0 = const()[name = tensor("op_29985_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29985_end_0 = const()[name = tensor("op_29985_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_29985_end_mask_0 = const()[name = tensor("op_29985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29985_cast_fp16 = slice_by_index(begin = var_29985_begin_0, end = var_29985_end_0, end_mask = var_29985_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29985_cast_fp16")]; + tensor var_29989_begin_0 = const()[name = tensor("op_29989_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29989_end_0 = const()[name = tensor("op_29989_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_29989_end_mask_0 = const()[name = tensor("op_29989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29989_cast_fp16 = slice_by_index(begin = var_29989_begin_0, end = var_29989_end_0, end_mask = var_29989_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29989_cast_fp16")]; + tensor var_29993_begin_0 = const()[name = tensor("op_29993_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29993_end_0 = const()[name = tensor("op_29993_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_29993_end_mask_0 = const()[name = tensor("op_29993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29993_cast_fp16 = slice_by_index(begin = var_29993_begin_0, end = var_29993_end_0, end_mask = var_29993_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29993_cast_fp16")]; + tensor var_29997_begin_0 = const()[name = tensor("op_29997_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29997_end_0 = const()[name = tensor("op_29997_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_29997_end_mask_0 = const()[name = tensor("op_29997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29997_cast_fp16 = slice_by_index(begin = var_29997_begin_0, end = var_29997_end_0, end_mask = var_29997_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29997_cast_fp16")]; + tensor var_30001_begin_0 = const()[name = tensor("op_30001_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_30001_end_0 = const()[name = tensor("op_30001_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_30001_end_mask_0 = const()[name = tensor("op_30001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30001_cast_fp16 = slice_by_index(begin = var_30001_begin_0, end = var_30001_end_0, end_mask = var_30001_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_30001_cast_fp16")]; + tensor var_30003_begin_0 = const()[name = tensor("op_30003_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30003_end_0 = const()[name = tensor("op_30003_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_30003_end_mask_0 = const()[name = tensor("op_30003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30003_cast_fp16 = slice_by_index(begin = var_30003_begin_0, end = var_30003_end_0, end_mask = var_30003_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30003_cast_fp16")]; + tensor var_30007_begin_0 = const()[name = tensor("op_30007_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30007_end_0 = const()[name = tensor("op_30007_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_30007_end_mask_0 = const()[name = tensor("op_30007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30007_cast_fp16 = slice_by_index(begin = var_30007_begin_0, end = var_30007_end_0, end_mask = var_30007_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30007_cast_fp16")]; + tensor var_30011_begin_0 = const()[name = tensor("op_30011_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30011_end_0 = const()[name = tensor("op_30011_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_30011_end_mask_0 = const()[name = tensor("op_30011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30011_cast_fp16 = slice_by_index(begin = var_30011_begin_0, end = var_30011_end_0, end_mask = var_30011_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30011_cast_fp16")]; + tensor var_30015_begin_0 = const()[name = tensor("op_30015_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30015_end_0 = const()[name = tensor("op_30015_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_30015_end_mask_0 = const()[name = tensor("op_30015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30015_cast_fp16 = slice_by_index(begin = var_30015_begin_0, end = var_30015_end_0, end_mask = var_30015_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30015_cast_fp16")]; + tensor var_30019_begin_0 = const()[name = tensor("op_30019_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30019_end_0 = const()[name = tensor("op_30019_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_30019_end_mask_0 = const()[name = tensor("op_30019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30019_cast_fp16 = slice_by_index(begin = var_30019_begin_0, end = var_30019_end_0, end_mask = var_30019_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30019_cast_fp16")]; + tensor var_30023_begin_0 = const()[name = tensor("op_30023_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30023_end_0 = const()[name = tensor("op_30023_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_30023_end_mask_0 = const()[name = tensor("op_30023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30023_cast_fp16 = slice_by_index(begin = var_30023_begin_0, end = var_30023_end_0, end_mask = var_30023_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30023_cast_fp16")]; + tensor var_30027_begin_0 = const()[name = tensor("op_30027_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30027_end_0 = const()[name = tensor("op_30027_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_30027_end_mask_0 = const()[name = tensor("op_30027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30027_cast_fp16 = slice_by_index(begin = var_30027_begin_0, end = var_30027_end_0, end_mask = var_30027_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30027_cast_fp16")]; + tensor var_30031_begin_0 = const()[name = tensor("op_30031_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30031_end_0 = const()[name = tensor("op_30031_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_30031_end_mask_0 = const()[name = tensor("op_30031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30031_cast_fp16 = slice_by_index(begin = var_30031_begin_0, end = var_30031_end_0, end_mask = var_30031_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30031_cast_fp16")]; + tensor var_30035_begin_0 = const()[name = tensor("op_30035_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30035_end_0 = const()[name = tensor("op_30035_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_30035_end_mask_0 = const()[name = tensor("op_30035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30035_cast_fp16 = slice_by_index(begin = var_30035_begin_0, end = var_30035_end_0, end_mask = var_30035_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30035_cast_fp16")]; + tensor var_30039_begin_0 = const()[name = tensor("op_30039_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30039_end_0 = const()[name = tensor("op_30039_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_30039_end_mask_0 = const()[name = tensor("op_30039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30039_cast_fp16 = slice_by_index(begin = var_30039_begin_0, end = var_30039_end_0, end_mask = var_30039_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30039_cast_fp16")]; + tensor var_30043_begin_0 = const()[name = tensor("op_30043_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30043_end_0 = const()[name = tensor("op_30043_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_30043_end_mask_0 = const()[name = tensor("op_30043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30043_cast_fp16 = slice_by_index(begin = var_30043_begin_0, end = var_30043_end_0, end_mask = var_30043_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30043_cast_fp16")]; + tensor var_30047_begin_0 = const()[name = tensor("op_30047_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30047_end_0 = const()[name = tensor("op_30047_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_30047_end_mask_0 = const()[name = tensor("op_30047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30047_cast_fp16 = slice_by_index(begin = var_30047_begin_0, end = var_30047_end_0, end_mask = var_30047_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30047_cast_fp16")]; + tensor var_30051_begin_0 = const()[name = tensor("op_30051_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30051_end_0 = const()[name = tensor("op_30051_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_30051_end_mask_0 = const()[name = tensor("op_30051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30051_cast_fp16 = slice_by_index(begin = var_30051_begin_0, end = var_30051_end_0, end_mask = var_30051_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30051_cast_fp16")]; + tensor var_30055_begin_0 = const()[name = tensor("op_30055_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30055_end_0 = const()[name = tensor("op_30055_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_30055_end_mask_0 = const()[name = tensor("op_30055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30055_cast_fp16 = slice_by_index(begin = var_30055_begin_0, end = var_30055_end_0, end_mask = var_30055_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30055_cast_fp16")]; + tensor var_30059_begin_0 = const()[name = tensor("op_30059_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30059_end_0 = const()[name = tensor("op_30059_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_30059_end_mask_0 = const()[name = tensor("op_30059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30059_cast_fp16 = slice_by_index(begin = var_30059_begin_0, end = var_30059_end_0, end_mask = var_30059_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30059_cast_fp16")]; + tensor var_30063_begin_0 = const()[name = tensor("op_30063_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30063_end_0 = const()[name = tensor("op_30063_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_30063_end_mask_0 = const()[name = tensor("op_30063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30063_cast_fp16 = slice_by_index(begin = var_30063_begin_0, end = var_30063_end_0, end_mask = var_30063_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30063_cast_fp16")]; + tensor var_30067_begin_0 = const()[name = tensor("op_30067_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30067_end_0 = const()[name = tensor("op_30067_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_30067_end_mask_0 = const()[name = tensor("op_30067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30067_cast_fp16 = slice_by_index(begin = var_30067_begin_0, end = var_30067_end_0, end_mask = var_30067_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30067_cast_fp16")]; + tensor var_30071_begin_0 = const()[name = tensor("op_30071_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30071_end_0 = const()[name = tensor("op_30071_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_30071_end_mask_0 = const()[name = tensor("op_30071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30071_cast_fp16 = slice_by_index(begin = var_30071_begin_0, end = var_30071_end_0, end_mask = var_30071_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30071_cast_fp16")]; + tensor var_30075_begin_0 = const()[name = tensor("op_30075_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30075_end_0 = const()[name = tensor("op_30075_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_30075_end_mask_0 = const()[name = tensor("op_30075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30075_cast_fp16 = slice_by_index(begin = var_30075_begin_0, end = var_30075_end_0, end_mask = var_30075_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30075_cast_fp16")]; + tensor var_30079_begin_0 = const()[name = tensor("op_30079_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30079_end_0 = const()[name = tensor("op_30079_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_30079_end_mask_0 = const()[name = tensor("op_30079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30079_cast_fp16 = slice_by_index(begin = var_30079_begin_0, end = var_30079_end_0, end_mask = var_30079_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30079_cast_fp16")]; + tensor var_30083_equation_0 = const()[name = tensor("op_30083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30083_cast_fp16 = einsum(equation = var_30083_equation_0, values = (var_29925_cast_fp16, var_29842_cast_fp16))[name = tensor("op_30083_cast_fp16")]; + tensor var_30084_to_fp16 = const()[name = tensor("op_30084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2521_cast_fp16 = mul(x = var_30083_cast_fp16, y = var_30084_to_fp16)[name = tensor("aw_2521_cast_fp16")]; + tensor var_30087_equation_0 = const()[name = tensor("op_30087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30087_cast_fp16 = einsum(equation = var_30087_equation_0, values = (var_29929_cast_fp16, var_29846_cast_fp16))[name = tensor("op_30087_cast_fp16")]; + tensor var_30088_to_fp16 = const()[name = tensor("op_30088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2523_cast_fp16 = mul(x = var_30087_cast_fp16, y = var_30088_to_fp16)[name = tensor("aw_2523_cast_fp16")]; + tensor var_30091_equation_0 = const()[name = tensor("op_30091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30091_cast_fp16 = einsum(equation = var_30091_equation_0, values = (var_29933_cast_fp16, var_29850_cast_fp16))[name = tensor("op_30091_cast_fp16")]; + tensor var_30092_to_fp16 = const()[name = tensor("op_30092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2525_cast_fp16 = mul(x = var_30091_cast_fp16, y = var_30092_to_fp16)[name = tensor("aw_2525_cast_fp16")]; + tensor var_30095_equation_0 = const()[name = tensor("op_30095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30095_cast_fp16 = einsum(equation = var_30095_equation_0, values = (var_29937_cast_fp16, var_29854_cast_fp16))[name = tensor("op_30095_cast_fp16")]; + tensor var_30096_to_fp16 = const()[name = tensor("op_30096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2527_cast_fp16 = mul(x = var_30095_cast_fp16, y = var_30096_to_fp16)[name = tensor("aw_2527_cast_fp16")]; + tensor var_30099_equation_0 = const()[name = tensor("op_30099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30099_cast_fp16 = einsum(equation = var_30099_equation_0, values = (var_29941_cast_fp16, var_29858_cast_fp16))[name = tensor("op_30099_cast_fp16")]; + tensor var_30100_to_fp16 = const()[name = tensor("op_30100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2529_cast_fp16 = mul(x = var_30099_cast_fp16, y = var_30100_to_fp16)[name = tensor("aw_2529_cast_fp16")]; + tensor var_30103_equation_0 = const()[name = tensor("op_30103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30103_cast_fp16 = einsum(equation = var_30103_equation_0, values = (var_29945_cast_fp16, var_29862_cast_fp16))[name = tensor("op_30103_cast_fp16")]; + tensor var_30104_to_fp16 = const()[name = tensor("op_30104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2531_cast_fp16 = mul(x = var_30103_cast_fp16, y = var_30104_to_fp16)[name = tensor("aw_2531_cast_fp16")]; + tensor var_30107_equation_0 = const()[name = tensor("op_30107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30107_cast_fp16 = einsum(equation = var_30107_equation_0, values = (var_29949_cast_fp16, var_29866_cast_fp16))[name = tensor("op_30107_cast_fp16")]; + tensor var_30108_to_fp16 = const()[name = tensor("op_30108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2533_cast_fp16 = mul(x = var_30107_cast_fp16, y = var_30108_to_fp16)[name = tensor("aw_2533_cast_fp16")]; + tensor var_30111_equation_0 = const()[name = tensor("op_30111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30111_cast_fp16 = einsum(equation = var_30111_equation_0, values = (var_29953_cast_fp16, var_29870_cast_fp16))[name = tensor("op_30111_cast_fp16")]; + tensor var_30112_to_fp16 = const()[name = tensor("op_30112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2535_cast_fp16 = mul(x = var_30111_cast_fp16, y = var_30112_to_fp16)[name = tensor("aw_2535_cast_fp16")]; + tensor var_30115_equation_0 = const()[name = tensor("op_30115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30115_cast_fp16 = einsum(equation = var_30115_equation_0, values = (var_29957_cast_fp16, var_29874_cast_fp16))[name = tensor("op_30115_cast_fp16")]; + tensor var_30116_to_fp16 = const()[name = tensor("op_30116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2537_cast_fp16 = mul(x = var_30115_cast_fp16, y = var_30116_to_fp16)[name = tensor("aw_2537_cast_fp16")]; + tensor var_30119_equation_0 = const()[name = tensor("op_30119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30119_cast_fp16 = einsum(equation = var_30119_equation_0, values = (var_29961_cast_fp16, var_29878_cast_fp16))[name = tensor("op_30119_cast_fp16")]; + tensor var_30120_to_fp16 = const()[name = tensor("op_30120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2539_cast_fp16 = mul(x = var_30119_cast_fp16, y = var_30120_to_fp16)[name = tensor("aw_2539_cast_fp16")]; + tensor var_30123_equation_0 = const()[name = tensor("op_30123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30123_cast_fp16 = einsum(equation = var_30123_equation_0, values = (var_29965_cast_fp16, var_29882_cast_fp16))[name = tensor("op_30123_cast_fp16")]; + tensor var_30124_to_fp16 = const()[name = tensor("op_30124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2541_cast_fp16 = mul(x = var_30123_cast_fp16, y = var_30124_to_fp16)[name = tensor("aw_2541_cast_fp16")]; + tensor var_30127_equation_0 = const()[name = tensor("op_30127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30127_cast_fp16 = einsum(equation = var_30127_equation_0, values = (var_29969_cast_fp16, var_29886_cast_fp16))[name = tensor("op_30127_cast_fp16")]; + tensor var_30128_to_fp16 = const()[name = tensor("op_30128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2543_cast_fp16 = mul(x = var_30127_cast_fp16, y = var_30128_to_fp16)[name = tensor("aw_2543_cast_fp16")]; + tensor var_30131_equation_0 = const()[name = tensor("op_30131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30131_cast_fp16 = einsum(equation = var_30131_equation_0, values = (var_29973_cast_fp16, var_29890_cast_fp16))[name = tensor("op_30131_cast_fp16")]; + tensor var_30132_to_fp16 = const()[name = tensor("op_30132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2545_cast_fp16 = mul(x = var_30131_cast_fp16, y = var_30132_to_fp16)[name = tensor("aw_2545_cast_fp16")]; + tensor var_30135_equation_0 = const()[name = tensor("op_30135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30135_cast_fp16 = einsum(equation = var_30135_equation_0, values = (var_29977_cast_fp16, var_29894_cast_fp16))[name = tensor("op_30135_cast_fp16")]; + tensor var_30136_to_fp16 = const()[name = tensor("op_30136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2547_cast_fp16 = mul(x = var_30135_cast_fp16, y = var_30136_to_fp16)[name = tensor("aw_2547_cast_fp16")]; + tensor var_30139_equation_0 = const()[name = tensor("op_30139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30139_cast_fp16 = einsum(equation = var_30139_equation_0, values = (var_29981_cast_fp16, var_29898_cast_fp16))[name = tensor("op_30139_cast_fp16")]; + tensor var_30140_to_fp16 = const()[name = tensor("op_30140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2549_cast_fp16 = mul(x = var_30139_cast_fp16, y = var_30140_to_fp16)[name = tensor("aw_2549_cast_fp16")]; + tensor var_30143_equation_0 = const()[name = tensor("op_30143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30143_cast_fp16 = einsum(equation = var_30143_equation_0, values = (var_29985_cast_fp16, var_29902_cast_fp16))[name = tensor("op_30143_cast_fp16")]; + tensor var_30144_to_fp16 = const()[name = tensor("op_30144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2551_cast_fp16 = mul(x = var_30143_cast_fp16, y = var_30144_to_fp16)[name = tensor("aw_2551_cast_fp16")]; + tensor var_30147_equation_0 = const()[name = tensor("op_30147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30147_cast_fp16 = einsum(equation = var_30147_equation_0, values = (var_29989_cast_fp16, var_29906_cast_fp16))[name = tensor("op_30147_cast_fp16")]; + tensor var_30148_to_fp16 = const()[name = tensor("op_30148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2553_cast_fp16 = mul(x = var_30147_cast_fp16, y = var_30148_to_fp16)[name = tensor("aw_2553_cast_fp16")]; + tensor var_30151_equation_0 = const()[name = tensor("op_30151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30151_cast_fp16 = einsum(equation = var_30151_equation_0, values = (var_29993_cast_fp16, var_29910_cast_fp16))[name = tensor("op_30151_cast_fp16")]; + tensor var_30152_to_fp16 = const()[name = tensor("op_30152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2555_cast_fp16 = mul(x = var_30151_cast_fp16, y = var_30152_to_fp16)[name = tensor("aw_2555_cast_fp16")]; + tensor var_30155_equation_0 = const()[name = tensor("op_30155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30155_cast_fp16 = einsum(equation = var_30155_equation_0, values = (var_29997_cast_fp16, var_29914_cast_fp16))[name = tensor("op_30155_cast_fp16")]; + tensor var_30156_to_fp16 = const()[name = tensor("op_30156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2557_cast_fp16 = mul(x = var_30155_cast_fp16, y = var_30156_to_fp16)[name = tensor("aw_2557_cast_fp16")]; + tensor var_30159_equation_0 = const()[name = tensor("op_30159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30159_cast_fp16 = einsum(equation = var_30159_equation_0, values = (var_30001_cast_fp16, var_29918_cast_fp16))[name = tensor("op_30159_cast_fp16")]; + tensor var_30160_to_fp16 = const()[name = tensor("op_30160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2559_cast_fp16 = mul(x = var_30159_cast_fp16, y = var_30160_to_fp16)[name = tensor("aw_2559_cast_fp16")]; + tensor var_30162_cast_fp16 = softmax(axis = var_21077, x = aw_2521_cast_fp16)[name = tensor("op_30162_cast_fp16")]; + tensor var_30163_cast_fp16 = softmax(axis = var_21077, x = aw_2523_cast_fp16)[name = tensor("op_30163_cast_fp16")]; + tensor var_30164_cast_fp16 = softmax(axis = var_21077, x = aw_2525_cast_fp16)[name = tensor("op_30164_cast_fp16")]; + tensor var_30165_cast_fp16 = softmax(axis = var_21077, x = aw_2527_cast_fp16)[name = tensor("op_30165_cast_fp16")]; + tensor var_30166_cast_fp16 = softmax(axis = var_21077, x = aw_2529_cast_fp16)[name = tensor("op_30166_cast_fp16")]; + tensor var_30167_cast_fp16 = softmax(axis = var_21077, x = aw_2531_cast_fp16)[name = tensor("op_30167_cast_fp16")]; + tensor var_30168_cast_fp16 = softmax(axis = var_21077, x = aw_2533_cast_fp16)[name = tensor("op_30168_cast_fp16")]; + tensor var_30169_cast_fp16 = softmax(axis = var_21077, x = aw_2535_cast_fp16)[name = tensor("op_30169_cast_fp16")]; + tensor var_30170_cast_fp16 = softmax(axis = var_21077, x = aw_2537_cast_fp16)[name = tensor("op_30170_cast_fp16")]; + tensor var_30171_cast_fp16 = softmax(axis = var_21077, x = aw_2539_cast_fp16)[name = tensor("op_30171_cast_fp16")]; + tensor var_30172_cast_fp16 = softmax(axis = var_21077, x = aw_2541_cast_fp16)[name = tensor("op_30172_cast_fp16")]; + tensor var_30173_cast_fp16 = softmax(axis = var_21077, x = aw_2543_cast_fp16)[name = tensor("op_30173_cast_fp16")]; + tensor var_30174_cast_fp16 = softmax(axis = var_21077, x = aw_2545_cast_fp16)[name = tensor("op_30174_cast_fp16")]; + tensor var_30175_cast_fp16 = softmax(axis = var_21077, x = aw_2547_cast_fp16)[name = tensor("op_30175_cast_fp16")]; + tensor var_30176_cast_fp16 = softmax(axis = var_21077, x = aw_2549_cast_fp16)[name = tensor("op_30176_cast_fp16")]; + tensor var_30177_cast_fp16 = softmax(axis = var_21077, x = aw_2551_cast_fp16)[name = tensor("op_30177_cast_fp16")]; + tensor var_30178_cast_fp16 = softmax(axis = var_21077, x = aw_2553_cast_fp16)[name = tensor("op_30178_cast_fp16")]; + tensor var_30179_cast_fp16 = softmax(axis = var_21077, x = aw_2555_cast_fp16)[name = tensor("op_30179_cast_fp16")]; + tensor var_30180_cast_fp16 = softmax(axis = var_21077, x = aw_2557_cast_fp16)[name = tensor("op_30180_cast_fp16")]; + tensor var_30181_cast_fp16 = softmax(axis = var_21077, x = aw_2559_cast_fp16)[name = tensor("op_30181_cast_fp16")]; + tensor var_30183_equation_0 = const()[name = tensor("op_30183_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30183_cast_fp16 = einsum(equation = var_30183_equation_0, values = (var_30003_cast_fp16, var_30162_cast_fp16))[name = tensor("op_30183_cast_fp16")]; + tensor var_30185_equation_0 = const()[name = tensor("op_30185_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30185_cast_fp16 = einsum(equation = var_30185_equation_0, values = (var_30007_cast_fp16, var_30163_cast_fp16))[name = tensor("op_30185_cast_fp16")]; + tensor var_30187_equation_0 = const()[name = tensor("op_30187_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30187_cast_fp16 = einsum(equation = var_30187_equation_0, values = (var_30011_cast_fp16, var_30164_cast_fp16))[name = tensor("op_30187_cast_fp16")]; + tensor var_30189_equation_0 = const()[name = tensor("op_30189_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30189_cast_fp16 = einsum(equation = var_30189_equation_0, values = (var_30015_cast_fp16, var_30165_cast_fp16))[name = tensor("op_30189_cast_fp16")]; + tensor var_30191_equation_0 = const()[name = tensor("op_30191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30191_cast_fp16 = einsum(equation = var_30191_equation_0, values = (var_30019_cast_fp16, var_30166_cast_fp16))[name = tensor("op_30191_cast_fp16")]; + tensor var_30193_equation_0 = const()[name = tensor("op_30193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30193_cast_fp16 = einsum(equation = var_30193_equation_0, values = (var_30023_cast_fp16, var_30167_cast_fp16))[name = tensor("op_30193_cast_fp16")]; + tensor var_30195_equation_0 = const()[name = tensor("op_30195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30195_cast_fp16 = einsum(equation = var_30195_equation_0, values = (var_30027_cast_fp16, var_30168_cast_fp16))[name = tensor("op_30195_cast_fp16")]; + tensor var_30197_equation_0 = const()[name = tensor("op_30197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30197_cast_fp16 = einsum(equation = var_30197_equation_0, values = (var_30031_cast_fp16, var_30169_cast_fp16))[name = tensor("op_30197_cast_fp16")]; + tensor var_30199_equation_0 = const()[name = tensor("op_30199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30199_cast_fp16 = einsum(equation = var_30199_equation_0, values = (var_30035_cast_fp16, var_30170_cast_fp16))[name = tensor("op_30199_cast_fp16")]; + tensor var_30201_equation_0 = const()[name = tensor("op_30201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30201_cast_fp16 = einsum(equation = var_30201_equation_0, values = (var_30039_cast_fp16, var_30171_cast_fp16))[name = tensor("op_30201_cast_fp16")]; + tensor var_30203_equation_0 = const()[name = tensor("op_30203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30203_cast_fp16 = einsum(equation = var_30203_equation_0, values = (var_30043_cast_fp16, var_30172_cast_fp16))[name = tensor("op_30203_cast_fp16")]; + tensor var_30205_equation_0 = const()[name = tensor("op_30205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30205_cast_fp16 = einsum(equation = var_30205_equation_0, values = (var_30047_cast_fp16, var_30173_cast_fp16))[name = tensor("op_30205_cast_fp16")]; + tensor var_30207_equation_0 = const()[name = tensor("op_30207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30207_cast_fp16 = einsum(equation = var_30207_equation_0, values = (var_30051_cast_fp16, var_30174_cast_fp16))[name = tensor("op_30207_cast_fp16")]; + tensor var_30209_equation_0 = const()[name = tensor("op_30209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30209_cast_fp16 = einsum(equation = var_30209_equation_0, values = (var_30055_cast_fp16, var_30175_cast_fp16))[name = tensor("op_30209_cast_fp16")]; + tensor var_30211_equation_0 = const()[name = tensor("op_30211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30211_cast_fp16 = einsum(equation = var_30211_equation_0, values = (var_30059_cast_fp16, var_30176_cast_fp16))[name = tensor("op_30211_cast_fp16")]; + tensor var_30213_equation_0 = const()[name = tensor("op_30213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30213_cast_fp16 = einsum(equation = var_30213_equation_0, values = (var_30063_cast_fp16, var_30177_cast_fp16))[name = tensor("op_30213_cast_fp16")]; + tensor var_30215_equation_0 = const()[name = tensor("op_30215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30215_cast_fp16 = einsum(equation = var_30215_equation_0, values = (var_30067_cast_fp16, var_30178_cast_fp16))[name = tensor("op_30215_cast_fp16")]; + tensor var_30217_equation_0 = const()[name = tensor("op_30217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30217_cast_fp16 = einsum(equation = var_30217_equation_0, values = (var_30071_cast_fp16, var_30179_cast_fp16))[name = tensor("op_30217_cast_fp16")]; + tensor var_30219_equation_0 = const()[name = tensor("op_30219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30219_cast_fp16 = einsum(equation = var_30219_equation_0, values = (var_30075_cast_fp16, var_30180_cast_fp16))[name = tensor("op_30219_cast_fp16")]; + tensor var_30221_equation_0 = const()[name = tensor("op_30221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30221_cast_fp16 = einsum(equation = var_30221_equation_0, values = (var_30079_cast_fp16, var_30181_cast_fp16))[name = tensor("op_30221_cast_fp16")]; + tensor input_401_interleave_0 = const()[name = tensor("input_401_interleave_0"), val = tensor(false)]; + tensor input_401_cast_fp16 = concat(axis = var_21077, interleave = input_401_interleave_0, values = (var_30183_cast_fp16, var_30185_cast_fp16, var_30187_cast_fp16, var_30189_cast_fp16, var_30191_cast_fp16, var_30193_cast_fp16, var_30195_cast_fp16, var_30197_cast_fp16, var_30199_cast_fp16, var_30201_cast_fp16, var_30203_cast_fp16, var_30205_cast_fp16, var_30207_cast_fp16, var_30209_cast_fp16, var_30211_cast_fp16, var_30213_cast_fp16, var_30215_cast_fp16, var_30217_cast_fp16, var_30219_cast_fp16, var_30221_cast_fp16))[name = tensor("input_401_cast_fp16")]; + tensor var_30231_pad_type_0 = const()[name = tensor("op_30231_pad_type_0"), val = tensor("valid")]; + tensor var_30231_strides_0 = const()[name = tensor("op_30231_strides_0"), val = tensor([1, 1])]; + tensor var_30231_pad_0 = const()[name = tensor("op_30231_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30231_dilations_0 = const()[name = tensor("op_30231_dilations_0"), val = tensor([1, 1])]; + tensor var_30231_groups_0 = const()[name = tensor("op_30231_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893022016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894250880))), name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894251072)))]; + tensor var_30231_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_30231_dilations_0, groups = var_30231_groups_0, pad = var_30231_pad_0, pad_type = var_30231_pad_type_0, strides = var_30231_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized, x = input_401_cast_fp16)[name = tensor("op_30231_cast_fp16")]; + tensor inputs_203_cast_fp16 = add(x = var_30231_cast_fp16, y = inputs_201_cast_fp16)[name = tensor("inputs_203_cast_fp16")]; + tensor input_403_axes_0 = const()[name = tensor("input_403_axes_0"), val = tensor([1])]; + tensor input_403_gamma_0_to_fp16 = const()[name = tensor("input_403_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894253696)))]; + tensor input_403_beta_0_to_fp16 = const()[name = tensor("input_403_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894256320)))]; + tensor var_30241_to_fp16 = const()[name = tensor("op_30241_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_403_cast_fp16 = layer_norm(axes = input_403_axes_0, beta = input_403_beta_0_to_fp16, epsilon = var_30241_to_fp16, gamma = input_403_gamma_0_to_fp16, x = inputs_203_cast_fp16)[name = tensor("input_403_cast_fp16")]; + tensor var_30261_pad_type_0 = const()[name = tensor("op_30261_pad_type_0"), val = tensor("valid")]; + tensor var_30261_strides_0 = const()[name = tensor("op_30261_strides_0"), val = tensor([1, 1])]; + tensor var_30261_pad_0 = const()[name = tensor("op_30261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30261_dilations_0 = const()[name = tensor("op_30261_dilations_0"), val = tensor([1, 1])]; + tensor var_30261_groups_0 = const()[name = tensor("op_30261_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894258944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(904089408))), name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(904089600)))]; + tensor var_30261_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_30261_dilations_0, groups = var_30261_groups_0, pad = var_30261_pad_0, pad_type = var_30261_pad_type_0, strides = var_30261_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized, x = input_403_cast_fp16)[name = tensor("op_30261_cast_fp16")]; + tensor var_30262_split_sizes_0 = const()[name = tensor("op_30262_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_30262_axis_0 = const()[name = tensor("op_30262_axis_0"), val = tensor(1)]; + tensor var_30262_cast_fp16_0, tensor var_30262_cast_fp16_1 = split(axis = var_30262_axis_0, split_sizes = var_30262_split_sizes_0, x = var_30261_cast_fp16)[name = tensor("op_30262_cast_fp16")]; + tensor var_30264_mode_0 = const()[name = tensor("op_30264_mode_0"), val = tensor("EXACT")]; + tensor var_30264_cast_fp16 = gelu(mode = var_30264_mode_0, x = var_30262_cast_fp16_1)[name = tensor("op_30264_cast_fp16")]; + tensor input_405_cast_fp16 = mul(x = var_30262_cast_fp16_0, y = var_30264_cast_fp16)[name = tensor("input_405_cast_fp16")]; + tensor var_30272_pad_type_0 = const()[name = tensor("op_30272_pad_type_0"), val = tensor("valid")]; + tensor var_30272_strides_0 = const()[name = tensor("op_30272_strides_0"), val = tensor([1, 1])]; + tensor var_30272_pad_0 = const()[name = tensor("op_30272_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30272_dilations_0 = const()[name = tensor("op_30272_dilations_0"), val = tensor([1, 1])]; + tensor var_30272_groups_0 = const()[name = tensor("op_30272_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(904110144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(909025408))), name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(909025600)))]; + tensor var_30272_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_30272_dilations_0, groups = var_30272_groups_0, pad = var_30272_pad_0, pad_type = var_30272_pad_type_0, strides = var_30272_strides_0, weight = mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized, x = input_405_cast_fp16)[name = tensor("op_30272_cast_fp16")]; + tensor hidden_states_269_cast_fp16 = add(x = var_30272_cast_fp16, y = inputs_203_cast_fp16)[name = tensor("hidden_states_269_cast_fp16")]; + tensor var_30274 = const()[name = tensor("op_30274"), val = tensor([2, 1280, 32, 32])]; + tensor input_407_cast_fp16 = reshape(shape = var_30274, x = hidden_states_269_cast_fp16)[name = tensor("input_407_cast_fp16")]; + tensor hidden_states_271_pad_type_0 = const()[name = tensor("hidden_states_271_pad_type_0"), val = tensor("valid")]; + tensor hidden_states_271_strides_0 = const()[name = tensor("hidden_states_271_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_271_pad_0 = const()[name = tensor("hidden_states_271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_271_dilations_0 = const()[name = tensor("hidden_states_271_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_271_groups_0 = const()[name = tensor("hidden_states_271_groups_0"), val = tensor(1)]; + tensor mid_block_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(909028224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(910257088))), name = tensor("mid_block_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(910257280)))]; + tensor hidden_states_271_cast_fp16 = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_271_dilations_0, groups = hidden_states_271_groups_0, pad = hidden_states_271_pad_0, pad_type = hidden_states_271_pad_type_0, strides = hidden_states_271_strides_0, weight = mid_block_attentions_0_proj_out_weight_to_fp16_palettized, x = input_407_cast_fp16)[name = tensor("hidden_states_271_cast_fp16")]; + tensor input_409_cast_fp16 = add(x = hidden_states_271_cast_fp16, y = hidden_states_205_cast_fp16)[name = tensor("input_409_cast_fp16")]; + tensor reshape_76_shape_0 = const()[name = tensor("reshape_76_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_76_cast_fp16 = reshape(shape = reshape_76_shape_0, x = input_409_cast_fp16)[name = tensor("reshape_76_cast_fp16")]; + tensor reduce_mean_57_axes_0 = const()[name = tensor("reduce_mean_57_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_57_keep_dims_0 = const()[name = tensor("reduce_mean_57_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_57_cast_fp16 = reduce_mean(axes = reduce_mean_57_axes_0, keep_dims = reduce_mean_57_keep_dims_0, x = reshape_76_cast_fp16)[name = tensor("reduce_mean_57_cast_fp16")]; + tensor sub_38_cast_fp16 = sub(x = reshape_76_cast_fp16, y = reduce_mean_57_cast_fp16)[name = tensor("sub_38_cast_fp16")]; + tensor square_19_cast_fp16 = square(x = sub_38_cast_fp16)[name = tensor("square_19_cast_fp16")]; + tensor reduce_mean_59_axes_0 = const()[name = tensor("reduce_mean_59_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_59_keep_dims_0 = const()[name = tensor("reduce_mean_59_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_59_cast_fp16 = reduce_mean(axes = reduce_mean_59_axes_0, keep_dims = reduce_mean_59_keep_dims_0, x = square_19_cast_fp16)[name = tensor("reduce_mean_59_cast_fp16")]; + tensor add_38_y_0_to_fp16 = const()[name = tensor("add_38_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_38_cast_fp16 = add(x = reduce_mean_59_cast_fp16, y = add_38_y_0_to_fp16)[name = tensor("add_38_cast_fp16")]; + tensor sqrt_19_cast_fp16 = sqrt(x = add_38_cast_fp16)[name = tensor("sqrt_19_cast_fp16")]; + tensor real_div_19_cast_fp16 = real_div(x = sub_38_cast_fp16, y = sqrt_19_cast_fp16)[name = tensor("real_div_19_cast_fp16")]; + tensor reshape_77_shape_0 = const()[name = tensor("reshape_77_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_77_cast_fp16 = reshape(shape = reshape_77_shape_0, x = real_div_19_cast_fp16)[name = tensor("reshape_77_cast_fp16")]; + tensor add_39_gamma_0_to_fp16 = const()[name = tensor("add_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(910259904)))]; + tensor add_39_beta_0_to_fp16 = const()[name = tensor("add_39_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(910262528)))]; + tensor add_39_epsilon_0_to_fp16 = const()[name = tensor("add_39_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_39_cast_fp16 = batch_norm(beta = add_39_beta_0_to_fp16, epsilon = add_39_epsilon_0_to_fp16, gamma = add_39_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_77_cast_fp16)[name = tensor("add_39_cast_fp16")]; + tensor input_413_cast_fp16 = silu(x = add_39_cast_fp16)[name = tensor("input_413_cast_fp16")]; + tensor hidden_states_273_pad_type_0 = const()[name = tensor("hidden_states_273_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_273_pad_0 = const()[name = tensor("hidden_states_273_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_273_strides_0 = const()[name = tensor("hidden_states_273_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_273_dilations_0 = const()[name = tensor("hidden_states_273_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_273_groups_0 = const()[name = tensor("hidden_states_273_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(910265152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(921324416))), name = tensor("mid_block_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(921324608)))]; + tensor hidden_states_273_cast_fp16 = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_273_dilations_0, groups = hidden_states_273_groups_0, pad = hidden_states_273_pad_0, pad_type = hidden_states_273_pad_type_0, strides = hidden_states_273_strides_0, weight = mid_block_resnets_1_conv1_weight_to_fp16_palettized, x = input_413_cast_fp16)[name = tensor("hidden_states_273_cast_fp16")]; + tensor temb_15_pad_type_0 = const()[name = tensor("temb_15_pad_type_0"), val = tensor("valid")]; + tensor temb_15_strides_0 = const()[name = tensor("temb_15_strides_0"), val = tensor([1, 1])]; + tensor temb_15_pad_0 = const()[name = tensor("temb_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_15_dilations_0 = const()[name = tensor("temb_15_dilations_0"), val = tensor([1, 1])]; + tensor temb_15_groups_0 = const()[name = tensor("temb_15_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(921327232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922556096))), name = tensor("mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922556288)))]; + tensor temb_15_cast_fp16 = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_15_dilations_0, groups = temb_15_groups_0, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = temb_15_strides_0, weight = mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_15_cast_fp16")]; + tensor input_417_cast_fp16 = add(x = hidden_states_273_cast_fp16, y = temb_15_cast_fp16)[name = tensor("input_417_cast_fp16")]; + tensor reshape_80_shape_0 = const()[name = tensor("reshape_80_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_80_cast_fp16 = reshape(shape = reshape_80_shape_0, x = input_417_cast_fp16)[name = tensor("reshape_80_cast_fp16")]; + tensor reduce_mean_60_axes_0 = const()[name = tensor("reduce_mean_60_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_60_keep_dims_0 = const()[name = tensor("reduce_mean_60_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_60_cast_fp16 = reduce_mean(axes = reduce_mean_60_axes_0, keep_dims = reduce_mean_60_keep_dims_0, x = reshape_80_cast_fp16)[name = tensor("reduce_mean_60_cast_fp16")]; + tensor sub_40_cast_fp16 = sub(x = reshape_80_cast_fp16, y = reduce_mean_60_cast_fp16)[name = tensor("sub_40_cast_fp16")]; + tensor square_20_cast_fp16 = square(x = sub_40_cast_fp16)[name = tensor("square_20_cast_fp16")]; + tensor reduce_mean_62_axes_0 = const()[name = tensor("reduce_mean_62_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_62_keep_dims_0 = const()[name = tensor("reduce_mean_62_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_62_cast_fp16 = reduce_mean(axes = reduce_mean_62_axes_0, keep_dims = reduce_mean_62_keep_dims_0, x = square_20_cast_fp16)[name = tensor("reduce_mean_62_cast_fp16")]; + tensor add_40_y_0_to_fp16 = const()[name = tensor("add_40_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_40_cast_fp16 = add(x = reduce_mean_62_cast_fp16, y = add_40_y_0_to_fp16)[name = tensor("add_40_cast_fp16")]; + tensor sqrt_20_cast_fp16 = sqrt(x = add_40_cast_fp16)[name = tensor("sqrt_20_cast_fp16")]; + tensor real_div_20_cast_fp16 = real_div(x = sub_40_cast_fp16, y = sqrt_20_cast_fp16)[name = tensor("real_div_20_cast_fp16")]; + tensor reshape_81_shape_0 = const()[name = tensor("reshape_81_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_81_cast_fp16 = reshape(shape = reshape_81_shape_0, x = real_div_20_cast_fp16)[name = tensor("reshape_81_cast_fp16")]; + tensor add_41_gamma_0_to_fp16 = const()[name = tensor("add_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922558912)))]; + tensor add_41_beta_0_to_fp16 = const()[name = tensor("add_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922561536)))]; + tensor add_41_epsilon_0_to_fp16 = const()[name = tensor("add_41_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_41_cast_fp16 = batch_norm(beta = add_41_beta_0_to_fp16, epsilon = add_41_epsilon_0_to_fp16, gamma = add_41_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_81_cast_fp16)[name = tensor("add_41_cast_fp16")]; + tensor input_421_cast_fp16 = silu(x = add_41_cast_fp16)[name = tensor("input_421_cast_fp16")]; + tensor hidden_states_275_pad_type_0 = const()[name = tensor("hidden_states_275_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_275_pad_0 = const()[name = tensor("hidden_states_275_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_275_strides_0 = const()[name = tensor("hidden_states_275_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_275_dilations_0 = const()[name = tensor("hidden_states_275_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_275_groups_0 = const()[name = tensor("hidden_states_275_groups_0"), val = tensor(1)]; + tensor mid_block_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922564160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(933623424))), name = tensor("mid_block_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(933623616)))]; + tensor hidden_states_275_cast_fp16 = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_275_dilations_0, groups = hidden_states_275_groups_0, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = hidden_states_275_strides_0, weight = mid_block_resnets_1_conv2_weight_to_fp16_palettized, x = input_421_cast_fp16)[name = tensor("hidden_states_275_cast_fp16")]; + tensor hidden_states_277_cast_fp16 = add(x = input_409_cast_fp16, y = hidden_states_275_cast_fp16)[name = tensor("hidden_states_277_cast_fp16")]; + tensor var_30355 = const()[name = tensor("op_30355"), val = tensor(1)]; + tensor input_423_interleave_0 = const()[name = tensor("input_423_interleave_0"), val = tensor(false)]; + tensor input_423_cast_fp16_1 = concat(axis = var_30355, interleave = input_423_interleave_0, values = (hidden_states_277_cast_fp16, input_311_cast_fp16))[name = tensor("input_423_cast_fp16")]; + tensor reshape_84_shape_0 = const()[name = tensor("reshape_84_shape_0"), val = tensor([2, 32, 80, 32, 32])]; + tensor reshape_84_cast_fp16 = reshape(shape = reshape_84_shape_0, x = input_423_cast_fp16_1)[name = tensor("reshape_84_cast_fp16")]; + tensor reduce_mean_63_axes_0 = const()[name = tensor("reduce_mean_63_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_63_keep_dims_0 = const()[name = tensor("reduce_mean_63_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_63_cast_fp16 = reduce_mean(axes = reduce_mean_63_axes_0, keep_dims = reduce_mean_63_keep_dims_0, x = reshape_84_cast_fp16)[name = tensor("reduce_mean_63_cast_fp16")]; + tensor sub_42_cast_fp16 = sub(x = reshape_84_cast_fp16, y = reduce_mean_63_cast_fp16)[name = tensor("sub_42_cast_fp16")]; + tensor square_21_cast_fp16 = square(x = sub_42_cast_fp16)[name = tensor("square_21_cast_fp16")]; + tensor reduce_mean_65_axes_0 = const()[name = tensor("reduce_mean_65_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_65_keep_dims_0 = const()[name = tensor("reduce_mean_65_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_65_cast_fp16 = reduce_mean(axes = reduce_mean_65_axes_0, keep_dims = reduce_mean_65_keep_dims_0, x = square_21_cast_fp16)[name = tensor("reduce_mean_65_cast_fp16")]; + tensor add_42_y_0_to_fp16 = const()[name = tensor("add_42_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_42_cast_fp16 = add(x = reduce_mean_65_cast_fp16, y = add_42_y_0_to_fp16)[name = tensor("add_42_cast_fp16")]; + tensor sqrt_21_cast_fp16 = sqrt(x = add_42_cast_fp16)[name = tensor("sqrt_21_cast_fp16")]; + tensor real_div_21_cast_fp16 = real_div(x = sub_42_cast_fp16, y = sqrt_21_cast_fp16)[name = tensor("real_div_21_cast_fp16")]; + tensor reshape_85_shape_0 = const()[name = tensor("reshape_85_shape_0"), val = tensor([2, 2560, 32, 32])]; + tensor reshape_85_cast_fp16 = reshape(shape = reshape_85_shape_0, x = real_div_21_cast_fp16)[name = tensor("reshape_85_cast_fp16")]; + tensor add_43_mean_0_to_fp16 = const()[name = tensor("add_43_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(933626240)))]; + tensor add_43_variance_0_to_fp16 = const()[name = tensor("add_43_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(933631424)))]; + tensor add_43_gamma_0_to_fp16 = const()[name = tensor("add_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(933636608)))]; + tensor add_43_beta_0_to_fp16 = const()[name = tensor("add_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(933641792)))]; + tensor add_43_epsilon_0_to_fp16 = const()[name = tensor("add_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_43_cast_fp16 = batch_norm(beta = add_43_beta_0_to_fp16, epsilon = add_43_epsilon_0_to_fp16, gamma = add_43_gamma_0_to_fp16, mean = add_43_mean_0_to_fp16, variance = add_43_variance_0_to_fp16, x = reshape_85_cast_fp16)[name = tensor("add_43_cast_fp16")]; + tensor input_427_cast_fp16 = silu(x = add_43_cast_fp16)[name = tensor("input_427_cast_fp16")]; + tensor hidden_states_279_pad_type_0 = const()[name = tensor("hidden_states_279_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_279_pad_0 = const()[name = tensor("hidden_states_279_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_279_strides_0 = const()[name = tensor("hidden_states_279_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_279_dilations_0 = const()[name = tensor("hidden_states_279_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_279_groups_0 = const()[name = tensor("hidden_states_279_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(933646976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(955765440))), name = tensor("up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 3, 3])]; + tensor up_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(955765632)))]; + tensor hidden_states_279_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_279_dilations_0, groups = hidden_states_279_groups_0, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = hidden_states_279_strides_0, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_427_cast_fp16)[name = tensor("hidden_states_279_cast_fp16")]; + tensor temb_17_pad_type_0 = const()[name = tensor("temb_17_pad_type_0"), val = tensor("valid")]; + tensor temb_17_strides_0 = const()[name = tensor("temb_17_strides_0"), val = tensor([1, 1])]; + tensor temb_17_pad_0 = const()[name = tensor("temb_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor temb_17_dilations_0 = const()[name = tensor("temb_17_dilations_0"), val = tensor([1, 1])]; + tensor temb_17_groups_0 = const()[name = tensor("temb_17_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(955768256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(956997120))), name = tensor("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(956997312)))]; + tensor temb_17_cast_fp16 = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_17_dilations_0, groups = temb_17_groups_0, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = temb_17_strides_0, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_21_cast_fp16_1)[name = tensor("temb_17_cast_fp16")]; + tensor input_431_cast_fp16 = add(x = hidden_states_279_cast_fp16, y = temb_17_cast_fp16)[name = tensor("input_431_cast_fp16")]; + tensor reshape_88_shape_0 = const()[name = tensor("reshape_88_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_88_cast_fp16 = reshape(shape = reshape_88_shape_0, x = input_431_cast_fp16)[name = tensor("reshape_88_cast_fp16")]; + tensor reduce_mean_66_axes_0 = const()[name = tensor("reduce_mean_66_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_66_keep_dims_0 = const()[name = tensor("reduce_mean_66_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_66_cast_fp16 = reduce_mean(axes = reduce_mean_66_axes_0, keep_dims = reduce_mean_66_keep_dims_0, x = reshape_88_cast_fp16)[name = tensor("reduce_mean_66_cast_fp16")]; + tensor sub_44_cast_fp16 = sub(x = reshape_88_cast_fp16, y = reduce_mean_66_cast_fp16)[name = tensor("sub_44_cast_fp16")]; + tensor square_22_cast_fp16 = square(x = sub_44_cast_fp16)[name = tensor("square_22_cast_fp16")]; + tensor reduce_mean_68_axes_0 = const()[name = tensor("reduce_mean_68_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_68_keep_dims_0 = const()[name = tensor("reduce_mean_68_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_68_cast_fp16 = reduce_mean(axes = reduce_mean_68_axes_0, keep_dims = reduce_mean_68_keep_dims_0, x = square_22_cast_fp16)[name = tensor("reduce_mean_68_cast_fp16")]; + tensor add_44_y_0_to_fp16 = const()[name = tensor("add_44_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_44_cast_fp16 = add(x = reduce_mean_68_cast_fp16, y = add_44_y_0_to_fp16)[name = tensor("add_44_cast_fp16")]; + tensor sqrt_22_cast_fp16 = sqrt(x = add_44_cast_fp16)[name = tensor("sqrt_22_cast_fp16")]; + tensor real_div_22_cast_fp16 = real_div(x = sub_44_cast_fp16, y = sqrt_22_cast_fp16)[name = tensor("real_div_22_cast_fp16")]; + tensor reshape_89_shape_0 = const()[name = tensor("reshape_89_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_89_cast_fp16 = reshape(shape = reshape_89_shape_0, x = real_div_22_cast_fp16)[name = tensor("reshape_89_cast_fp16")]; + tensor add_45_gamma_0_to_fp16 = const()[name = tensor("add_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(956999936)))]; + tensor add_45_beta_0_to_fp16 = const()[name = tensor("add_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(957002560)))]; + tensor add_45_epsilon_0_to_fp16 = const()[name = tensor("add_45_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_45_cast_fp16 = batch_norm(beta = add_45_beta_0_to_fp16, epsilon = add_45_epsilon_0_to_fp16, gamma = add_45_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_89_cast_fp16)[name = tensor("add_45_cast_fp16")]; + tensor input_435_cast_fp16 = silu(x = add_45_cast_fp16)[name = tensor("input_435_cast_fp16")]; + tensor hidden_states_281_pad_type_0 = const()[name = tensor("hidden_states_281_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_281_pad_0 = const()[name = tensor("hidden_states_281_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor hidden_states_281_strides_0 = const()[name = tensor("hidden_states_281_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_281_dilations_0 = const()[name = tensor("hidden_states_281_dilations_0"), val = tensor([1, 1])]; + tensor hidden_states_281_groups_0 = const()[name = tensor("hidden_states_281_groups_0"), val = tensor(1)]; + tensor up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(957005184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968064448))), name = tensor("up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968064640)))]; + tensor hidden_states_281_cast_fp16_1 = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_281_dilations_0, groups = hidden_states_281_groups_0, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = hidden_states_281_strides_0, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_435_cast_fp16)[name = tensor("hidden_states_281_cast_fp16")]; + tensor input_79_cast_fp16_dtype_0 = const()[name = tensor("input_79_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_45_cast_fp16_dtype_0 = const()[name = tensor("input_45_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_423_cast_fp16_dtype_0 = const()[name = tensor("input_423_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_213_cast_fp16_dtype_0 = const()[name = tensor("input_213_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_113_cast_fp16_dtype_0 = const()[name = tensor("input_113_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_13_cast_fp16_dtype_0 = const()[name = tensor("input_13_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_29_cast_fp16_dtype_0 = const()[name = tensor("input_29_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor hidden_states_281_cast_fp16_dtype_0 = const()[name = tensor("hidden_states_281_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_115_cast_fp16_dtype_0 = const()[name = tensor("input_115_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_43_cast_fp16_dtype_0 = const()[name = tensor("input_43_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_21_cast_fp16_dtype_0 = const()[name = tensor("input_21_cast_fp16_dtype_0"), val = tensor("fp32")]; + tensor input_21_cast_fp16 = cast(dtype = input_21_cast_fp16_dtype_0, x = input_21_cast_fp16_1)[name = tensor("cast_11")]; + tensor input_43_cast_fp16 = cast(dtype = input_43_cast_fp16_dtype_0, x = input_43_cast_fp16_1)[name = tensor("cast_12")]; + tensor input_115_cast_fp16 = cast(dtype = input_115_cast_fp16_dtype_0, x = input_115_cast_fp16_1)[name = tensor("cast_13")]; + tensor hidden_states_281_cast_fp16 = cast(dtype = hidden_states_281_cast_fp16_dtype_0, x = hidden_states_281_cast_fp16_1)[name = tensor("cast_14")]; + tensor input_29_cast_fp16 = cast(dtype = input_29_cast_fp16_dtype_0, x = input_29_cast_fp16_1)[name = tensor("cast_15")]; + tensor input_13_cast_fp16 = cast(dtype = input_13_cast_fp16_dtype_0, x = input_13_cast_fp16_1)[name = tensor("cast_16")]; + tensor input_113_cast_fp16 = cast(dtype = input_113_cast_fp16_dtype_0, x = input_113_cast_fp16_1)[name = tensor("cast_17")]; + tensor input_213_cast_fp16 = cast(dtype = input_213_cast_fp16_dtype_0, x = input_213_cast_fp16_1)[name = tensor("cast_18")]; + tensor input_423_cast_fp16 = cast(dtype = input_423_cast_fp16_dtype_0, x = input_423_cast_fp16_1)[name = tensor("cast_19")]; + tensor input_45_cast_fp16 = cast(dtype = input_45_cast_fp16_dtype_0, x = input_45_cast_fp16_1)[name = tensor("cast_20")]; + tensor input_79_cast_fp16 = cast(dtype = input_79_cast_fp16_dtype_0, x = input_79_cast_fp16_1)[name = tensor("cast_21")]; + } -> (input_79_cast_fp16, input_45_cast_fp16, input_423_cast_fp16, input_213_cast_fp16, input_113_cast_fp16, input_13_cast_fp16, input_29_cast_fp16, hidden_states_281_cast_fp16, input_115_cast_fp16, input_43_cast_fp16, input_21_cast_fp16); +} \ No newline at end of file