diff --git "a/toxic.mlmodelc/model.mil" "b/toxic.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/toxic.mlmodelc/model.mil" @@ -0,0 +1,645 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})] +{ + func main(tensor attention_mask, tensor input_ids) { + int32 var_26 = const()[name = string("op_26"), val = int32(1)]; + tensor var_42 = not_equal(x = input_ids, y = var_26)[name = string("op_42")]; + string mask_1_dtype_0 = const()[name = string("mask_1_dtype_0"), val = string("int32")]; + bool var_44_exclusive_0 = const()[name = string("op_44_exclusive_0"), val = bool(false)]; + bool var_44_reverse_0 = const()[name = string("op_44_reverse_0"), val = bool(false)]; + tensor mask_1 = cast(dtype = mask_1_dtype_0, x = var_42)[name = string("cast_4")]; + tensor var_44 = cumsum(axis = var_26, exclusive = var_44_exclusive_0, reverse = var_44_reverse_0, x = mask_1)[name = string("op_44")]; + tensor incremental_indices = mul(x = var_44, y = mask_1)[name = string("incremental_indices")]; + int32 var_50 = const()[name = string("op_50"), val = int32(1)]; + tensor input_3 = add(x = incremental_indices, y = var_50)[name = string("input_3")]; + int32 inputs_embeds_batch_dims_0 = const()[name = string("inputs_embeds_batch_dims_0"), val = int32(0)]; + bool inputs_embeds_validate_indices_0 = const()[name = string("inputs_embeds_validate_indices_0"), val = bool(false)]; + tensor encoder_embeddings_word_embeddings_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23319680))))[name = string("encoder_embeddings_word_embeddings_weight_to_fp16_palettized")]; + int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)]; + tensor greater_equal_0 = greater_equal(x = input_ids, y = greater_equal_0_y_0)[name = string("greater_equal_0")]; + int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(60728)]; + tensor add_0 = add(x = input_ids, y = slice_by_index_0)[name = string("add_0")]; + tensor select_0 = select(a = input_ids, b = add_0, cond = greater_equal_0)[name = string("select_0")]; + int32 greater_equal_0_y_0_1 = const()[name = string("greater_equal_0_y_0_1"), val = int32(0)]; + tensor greater_equal_0_1 = greater_equal(x = select_0, y = greater_equal_0_y_0_1)[name = string("greater_equal_0_1")]; + int32 slice_by_index_0_1 = const()[name = string("slice_by_index_0_1"), val = int32(60728)]; + tensor add_0_1 = add(x = select_0, y = slice_by_index_0_1)[name = string("add_0_1")]; + tensor select_0_1 = select(a = select_0, b = add_0_1, cond = greater_equal_0_1)[name = string("select_0_1")]; + int32 inputs_embeds_cast_fp16_axis_0 = const()[name = string("inputs_embeds_cast_fp16_axis_0"), val = int32(0)]; + tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_cast_fp16_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = select_0_1, validate_indices = inputs_embeds_validate_indices_0, x = encoder_embeddings_word_embeddings_weight_to_fp16_palettized)[name = string("inputs_embeds_cast_fp16")]; + tensor token_type_embeddings_1_to_fp16 = const()[name = string("token_type_embeddings_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23319808)))]; + tensor embeddings_1_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = token_type_embeddings_1_to_fp16)[name = string("embeddings_1_cast_fp16")]; + int32 position_embeddings_1_axis_0 = const()[name = string("position_embeddings_1_axis_0"), val = int32(0)]; + int32 position_embeddings_1_batch_dims_0 = const()[name = string("position_embeddings_1_batch_dims_0"), val = int32(0)]; + bool position_embeddings_1_validate_indices_0 = const()[name = string("position_embeddings_1_validate_indices_0"), val = bool(false)]; + tensor encoder_embeddings_position_embeddings_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23516480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23713920))))[name = string("encoder_embeddings_position_embeddings_weight_to_fp16_palettized")]; + string input_3_to_uint16_dtype_0 = const()[name = string("input_3_to_uint16_dtype_0"), val = string("uint16")]; + tensor input_3_to_uint16 = cast(dtype = input_3_to_uint16_dtype_0, x = input_3)[name = string("cast_3")]; + tensor position_embeddings_1_cast_fp16_cast_uint16 = gather(axis = position_embeddings_1_axis_0, batch_dims = position_embeddings_1_batch_dims_0, indices = input_3_to_uint16, validate_indices = position_embeddings_1_validate_indices_0, x = encoder_embeddings_position_embeddings_weight_to_fp16_palettized)[name = string("position_embeddings_1_cast_fp16_cast_uint16")]; + tensor input_5_cast_fp16 = add(x = embeddings_1_cast_fp16, y = position_embeddings_1_cast_fp16_cast_uint16)[name = string("input_5_cast_fp16")]; + tensor input_7_axes_0 = const()[name = string("input_7_axes_0"), val = tensor([-1])]; + tensor encoder_embeddings_LayerNorm_weight_to_fp16 = const()[name = string("encoder_embeddings_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23714048)))]; + tensor encoder_embeddings_LayerNorm_bias_to_fp16 = const()[name = string("encoder_embeddings_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23715648)))]; + fp16 var_21_to_fp16 = const()[name = string("op_21_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_7_cast_fp16 = layer_norm(axes = input_7_axes_0, beta = encoder_embeddings_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_embeddings_LayerNorm_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + tensor var_68_axes_0 = const()[name = string("op_68_axes_0"), val = tensor([1])]; + tensor var_68 = expand_dims(axes = var_68_axes_0, x = attention_mask)[name = string("op_68")]; + tensor var_69_axes_0 = const()[name = string("op_69_axes_0"), val = tensor([2])]; + tensor var_69 = expand_dims(axes = var_69_axes_0, x = var_68)[name = string("op_69")]; + tensor var_72_reps_0 = const()[name = string("op_72_reps_0"), val = tensor([1, 1, 128, 1])]; + tensor var_72 = tile(reps = var_72_reps_0, x = var_69)[name = string("op_72")]; + fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x1p+0)]; + string expanded_mask_to_fp16_dtype_0 = const()[name = string("expanded_mask_to_fp16_dtype_0"), val = string("fp16")]; + tensor var_72_to_fp16 = cast(dtype = expanded_mask_to_fp16_dtype_0, x = var_72)[name = string("cast_2")]; + tensor inverted_mask_cast_fp16 = sub(x = const_4_to_fp16, y = var_72_to_fp16)[name = string("inverted_mask_cast_fp16")]; + string var_77_dtype_0 = const()[name = string("op_77_dtype_0"), val = string("bool")]; + fp16 var_10_to_fp16 = const()[name = string("op_10_to_fp16"), val = fp16(-inf)]; + tensor inverted_mask_cast_fp16_to_bool = cast(dtype = var_77_dtype_0, x = inverted_mask_cast_fp16)[name = string("cast_1")]; + tensor attention_mask_cast_fp16 = select(a = var_10_to_fp16, b = inverted_mask_cast_fp16, cond = inverted_mask_cast_fp16_to_bool)[name = string("attention_mask_cast_fp16")]; + tensor encoder_encoder_layer_0_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23717248))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24012224))))[name = string("encoder_encoder_layer_0_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_0_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24013056)))]; + tensor linear_0_cast_fp16 = linear(bias = encoder_encoder_layer_0_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_0_attention_self_query_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = string("linear_0_cast_fp16")]; + tensor var_116 = const()[name = string("op_116"), val = tensor([1, -1, 12, 64])]; + tensor var_117_cast_fp16 = reshape(shape = var_116, x = linear_0_cast_fp16)[name = string("op_117_cast_fp16")]; + tensor encoder_encoder_layer_0_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24014656))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24309632))))[name = string("encoder_encoder_layer_0_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_0_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24310464)))]; + tensor linear_1_cast_fp16 = linear(bias = encoder_encoder_layer_0_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_0_attention_self_key_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = string("linear_1_cast_fp16")]; + tensor var_122 = const()[name = string("op_122"), val = tensor([1, -1, 12, 64])]; + tensor var_123_cast_fp16 = reshape(shape = var_122, x = linear_1_cast_fp16)[name = string("op_123_cast_fp16")]; + tensor encoder_encoder_layer_0_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24312064))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24607040))))[name = string("encoder_encoder_layer_0_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_0_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24607872)))]; + tensor linear_2_cast_fp16 = linear(bias = encoder_encoder_layer_0_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_0_attention_self_value_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = string("linear_2_cast_fp16")]; + tensor var_128 = const()[name = string("op_128"), val = tensor([1, -1, 12, 64])]; + tensor var_129_cast_fp16 = reshape(shape = var_128, x = linear_2_cast_fp16)[name = string("op_129_cast_fp16")]; + tensor transpose_72_perm_0 = const()[name = string("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_73_perm_0 = const()[name = string("transpose_73_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_74_perm_0 = const()[name = string("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_74 = transpose(perm = transpose_74_perm_0, x = var_129_cast_fp16)[name = string("transpose_153")]; + tensor transpose_73 = transpose(perm = transpose_73_perm_0, x = var_123_cast_fp16)[name = string("transpose_154")]; + tensor transpose_72 = transpose(perm = transpose_72_perm_0, x = var_117_cast_fp16)[name = string("transpose_155")]; + tensor attn_output_1_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_73, query = transpose_72, value = transpose_74)[name = string("attn_output_1_cast_fp16")]; + tensor attn_output_3_perm_0 = const()[name = string("attn_output_3_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_133 = const()[name = string("op_133"), val = tensor([1, 128, 768])]; + tensor attn_output_3_cast_fp16 = transpose(perm = attn_output_3_perm_0, x = attn_output_1_cast_fp16)[name = string("transpose_152")]; + tensor input_9_cast_fp16 = reshape(shape = var_133, x = attn_output_3_cast_fp16)[name = string("input_9_cast_fp16")]; + tensor encoder_encoder_layer_0_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24609472))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24904448))))[name = string("encoder_encoder_layer_0_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_0_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24905280)))]; + tensor linear_3_cast_fp16 = linear(bias = encoder_encoder_layer_0_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_0_attention_output_dense_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = string("linear_3_cast_fp16")]; + tensor input_13_cast_fp16 = add(x = linear_3_cast_fp16, y = input_7_cast_fp16)[name = string("input_13_cast_fp16")]; + tensor input_15_axes_0 = const()[name = string("input_15_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24906880)))]; + tensor encoder_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24908480)))]; + tensor input_15_cast_fp16 = layer_norm(axes = input_15_axes_0, beta = encoder_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + tensor encoder_encoder_layer_0_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24910080))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26089792))))[name = string("encoder_encoder_layer_0_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_0_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26092928)))]; + tensor linear_4_cast_fp16 = linear(bias = encoder_encoder_layer_0_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_0_intermediate_dense_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = string("linear_4_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = linear_4_cast_fp16)[name = string("input_19_cast_fp16")]; + tensor encoder_encoder_layer_0_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26099136))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27278848))))[name = string("encoder_encoder_layer_0_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_0_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27279680)))]; + tensor linear_5_cast_fp16 = linear(bias = encoder_encoder_layer_0_output_dense_bias_to_fp16, weight = encoder_encoder_layer_0_output_dense_weight_to_fp16_palettized, x = input_19_cast_fp16)[name = string("linear_5_cast_fp16")]; + tensor input_23_cast_fp16 = add(x = linear_5_cast_fp16, y = input_15_cast_fp16)[name = string("input_23_cast_fp16")]; + tensor hidden_states_7_axes_0 = const()[name = string("hidden_states_7_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_0_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_0_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27281280)))]; + tensor encoder_encoder_layer_0_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_0_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27282880)))]; + tensor hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = encoder_encoder_layer_0_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_0_output_LayerNorm_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor encoder_encoder_layer_1_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27284480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27579456))))[name = string("encoder_encoder_layer_1_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_1_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27580288)))]; + tensor linear_6_cast_fp16 = linear(bias = encoder_encoder_layer_1_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_1_attention_self_query_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_6_cast_fp16")]; + tensor var_175 = const()[name = string("op_175"), val = tensor([1, -1, 12, 64])]; + tensor var_176_cast_fp16 = reshape(shape = var_175, x = linear_6_cast_fp16)[name = string("op_176_cast_fp16")]; + tensor encoder_encoder_layer_1_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27581888))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27876864))))[name = string("encoder_encoder_layer_1_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_1_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27877696)))]; + tensor linear_7_cast_fp16 = linear(bias = encoder_encoder_layer_1_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_1_attention_self_key_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_7_cast_fp16")]; + tensor var_181 = const()[name = string("op_181"), val = tensor([1, -1, 12, 64])]; + tensor var_182_cast_fp16 = reshape(shape = var_181, x = linear_7_cast_fp16)[name = string("op_182_cast_fp16")]; + tensor encoder_encoder_layer_1_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27879296))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28174272))))[name = string("encoder_encoder_layer_1_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_1_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28175104)))]; + tensor linear_8_cast_fp16 = linear(bias = encoder_encoder_layer_1_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_1_attention_self_value_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_8_cast_fp16")]; + tensor var_187 = const()[name = string("op_187"), val = tensor([1, -1, 12, 64])]; + tensor var_188_cast_fp16 = reshape(shape = var_187, x = linear_8_cast_fp16)[name = string("op_188_cast_fp16")]; + tensor transpose_75_perm_0 = const()[name = string("transpose_75_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_76_perm_0 = const()[name = string("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77_perm_0 = const()[name = string("transpose_77_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77 = transpose(perm = transpose_77_perm_0, x = var_188_cast_fp16)[name = string("transpose_149")]; + tensor transpose_76 = transpose(perm = transpose_76_perm_0, x = var_182_cast_fp16)[name = string("transpose_150")]; + tensor transpose_75 = transpose(perm = transpose_75_perm_0, x = var_176_cast_fp16)[name = string("transpose_151")]; + tensor attn_output_5_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_76, query = transpose_75, value = transpose_77)[name = string("attn_output_5_cast_fp16")]; + tensor attn_output_7_perm_0 = const()[name = string("attn_output_7_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_192 = const()[name = string("op_192"), val = tensor([1, 128, 768])]; + tensor attn_output_7_cast_fp16 = transpose(perm = attn_output_7_perm_0, x = attn_output_5_cast_fp16)[name = string("transpose_148")]; + tensor input_25_cast_fp16 = reshape(shape = var_192, x = attn_output_7_cast_fp16)[name = string("input_25_cast_fp16")]; + tensor encoder_encoder_layer_1_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28176704))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28471680))))[name = string("encoder_encoder_layer_1_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_1_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28472512)))]; + tensor linear_9_cast_fp16 = linear(bias = encoder_encoder_layer_1_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_1_attention_output_dense_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = string("linear_9_cast_fp16")]; + tensor input_29_cast_fp16 = add(x = linear_9_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("input_29_cast_fp16")]; + tensor input_31_axes_0 = const()[name = string("input_31_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28474112)))]; + tensor encoder_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28475712)))]; + tensor input_31_cast_fp16 = layer_norm(axes = input_31_axes_0, beta = encoder_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + tensor encoder_encoder_layer_1_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28477312))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29657024))))[name = string("encoder_encoder_layer_1_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_1_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29660160)))]; + tensor linear_10_cast_fp16 = linear(bias = encoder_encoder_layer_1_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_1_intermediate_dense_weight_to_fp16_palettized, x = input_31_cast_fp16)[name = string("linear_10_cast_fp16")]; + string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("EXACT")]; + tensor input_35_cast_fp16 = gelu(mode = input_35_mode_0, x = linear_10_cast_fp16)[name = string("input_35_cast_fp16")]; + tensor encoder_encoder_layer_1_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29666368))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30846080))))[name = string("encoder_encoder_layer_1_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_1_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30846912)))]; + tensor linear_11_cast_fp16 = linear(bias = encoder_encoder_layer_1_output_dense_bias_to_fp16, weight = encoder_encoder_layer_1_output_dense_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = string("linear_11_cast_fp16")]; + tensor input_39_cast_fp16 = add(x = linear_11_cast_fp16, y = input_31_cast_fp16)[name = string("input_39_cast_fp16")]; + tensor hidden_states_13_axes_0 = const()[name = string("hidden_states_13_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_1_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_1_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30848512)))]; + tensor encoder_encoder_layer_1_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_1_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30850112)))]; + tensor hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = encoder_encoder_layer_1_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_1_output_LayerNorm_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor encoder_encoder_layer_2_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30851712))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31146688))))[name = string("encoder_encoder_layer_2_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_2_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31147520)))]; + tensor linear_12_cast_fp16 = linear(bias = encoder_encoder_layer_2_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_2_attention_self_query_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_12_cast_fp16")]; + tensor var_234 = const()[name = string("op_234"), val = tensor([1, -1, 12, 64])]; + tensor var_235_cast_fp16 = reshape(shape = var_234, x = linear_12_cast_fp16)[name = string("op_235_cast_fp16")]; + tensor encoder_encoder_layer_2_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31149120))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31444096))))[name = string("encoder_encoder_layer_2_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_2_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31444928)))]; + tensor linear_13_cast_fp16 = linear(bias = encoder_encoder_layer_2_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_2_attention_self_key_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_13_cast_fp16")]; + tensor var_240 = const()[name = string("op_240"), val = tensor([1, -1, 12, 64])]; + tensor var_241_cast_fp16 = reshape(shape = var_240, x = linear_13_cast_fp16)[name = string("op_241_cast_fp16")]; + tensor encoder_encoder_layer_2_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31446528))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31741504))))[name = string("encoder_encoder_layer_2_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_2_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31742336)))]; + tensor linear_14_cast_fp16 = linear(bias = encoder_encoder_layer_2_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_2_attention_self_value_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_14_cast_fp16")]; + tensor var_246 = const()[name = string("op_246"), val = tensor([1, -1, 12, 64])]; + tensor var_247_cast_fp16 = reshape(shape = var_246, x = linear_14_cast_fp16)[name = string("op_247_cast_fp16")]; + tensor transpose_78_perm_0 = const()[name = string("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_79_perm_0 = const()[name = string("transpose_79_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_80_perm_0 = const()[name = string("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_80 = transpose(perm = transpose_80_perm_0, x = var_247_cast_fp16)[name = string("transpose_145")]; + tensor transpose_79 = transpose(perm = transpose_79_perm_0, x = var_241_cast_fp16)[name = string("transpose_146")]; + tensor transpose_78 = transpose(perm = transpose_78_perm_0, x = var_235_cast_fp16)[name = string("transpose_147")]; + tensor attn_output_9_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_79, query = transpose_78, value = transpose_80)[name = string("attn_output_9_cast_fp16")]; + tensor attn_output_11_perm_0 = const()[name = string("attn_output_11_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_251 = const()[name = string("op_251"), val = tensor([1, 128, 768])]; + tensor attn_output_11_cast_fp16 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = string("transpose_144")]; + tensor input_41_cast_fp16 = reshape(shape = var_251, x = attn_output_11_cast_fp16)[name = string("input_41_cast_fp16")]; + tensor encoder_encoder_layer_2_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31743936))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32038912))))[name = string("encoder_encoder_layer_2_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_2_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32039744)))]; + tensor linear_15_cast_fp16 = linear(bias = encoder_encoder_layer_2_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_2_attention_output_dense_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = string("linear_15_cast_fp16")]; + tensor input_45_cast_fp16 = add(x = linear_15_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("input_45_cast_fp16")]; + tensor input_47_axes_0 = const()[name = string("input_47_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32041344)))]; + tensor encoder_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32042944)))]; + tensor input_47_cast_fp16 = layer_norm(axes = input_47_axes_0, beta = encoder_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + tensor encoder_encoder_layer_2_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32044544))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33224256))))[name = string("encoder_encoder_layer_2_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_2_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33227392)))]; + tensor linear_16_cast_fp16 = linear(bias = encoder_encoder_layer_2_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_2_intermediate_dense_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = string("linear_16_cast_fp16")]; + string input_51_mode_0 = const()[name = string("input_51_mode_0"), val = string("EXACT")]; + tensor input_51_cast_fp16 = gelu(mode = input_51_mode_0, x = linear_16_cast_fp16)[name = string("input_51_cast_fp16")]; + tensor encoder_encoder_layer_2_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33233600))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34413312))))[name = string("encoder_encoder_layer_2_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_2_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34414144)))]; + tensor linear_17_cast_fp16 = linear(bias = encoder_encoder_layer_2_output_dense_bias_to_fp16, weight = encoder_encoder_layer_2_output_dense_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = string("linear_17_cast_fp16")]; + tensor input_55_cast_fp16 = add(x = linear_17_cast_fp16, y = input_47_cast_fp16)[name = string("input_55_cast_fp16")]; + tensor hidden_states_19_axes_0 = const()[name = string("hidden_states_19_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_2_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_2_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34415744)))]; + tensor encoder_encoder_layer_2_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_2_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34417344)))]; + tensor hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = encoder_encoder_layer_2_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_2_output_LayerNorm_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor encoder_encoder_layer_3_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34418944))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34713920))))[name = string("encoder_encoder_layer_3_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_3_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34714752)))]; + tensor linear_18_cast_fp16 = linear(bias = encoder_encoder_layer_3_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_3_attention_self_query_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_18_cast_fp16")]; + tensor var_293 = const()[name = string("op_293"), val = tensor([1, -1, 12, 64])]; + tensor var_294_cast_fp16 = reshape(shape = var_293, x = linear_18_cast_fp16)[name = string("op_294_cast_fp16")]; + tensor encoder_encoder_layer_3_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34716352))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35011328))))[name = string("encoder_encoder_layer_3_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_3_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35012160)))]; + tensor linear_19_cast_fp16 = linear(bias = encoder_encoder_layer_3_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_3_attention_self_key_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_19_cast_fp16")]; + tensor var_299 = const()[name = string("op_299"), val = tensor([1, -1, 12, 64])]; + tensor var_300_cast_fp16 = reshape(shape = var_299, x = linear_19_cast_fp16)[name = string("op_300_cast_fp16")]; + tensor encoder_encoder_layer_3_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35013760))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35308736))))[name = string("encoder_encoder_layer_3_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_3_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35309568)))]; + tensor linear_20_cast_fp16 = linear(bias = encoder_encoder_layer_3_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_3_attention_self_value_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_20_cast_fp16")]; + tensor var_305 = const()[name = string("op_305"), val = tensor([1, -1, 12, 64])]; + tensor var_306_cast_fp16 = reshape(shape = var_305, x = linear_20_cast_fp16)[name = string("op_306_cast_fp16")]; + tensor transpose_81_perm_0 = const()[name = string("transpose_81_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_82_perm_0 = const()[name = string("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83_perm_0 = const()[name = string("transpose_83_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83 = transpose(perm = transpose_83_perm_0, x = var_306_cast_fp16)[name = string("transpose_141")]; + tensor transpose_82 = transpose(perm = transpose_82_perm_0, x = var_300_cast_fp16)[name = string("transpose_142")]; + tensor transpose_81 = transpose(perm = transpose_81_perm_0, x = var_294_cast_fp16)[name = string("transpose_143")]; + tensor attn_output_13_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_82, query = transpose_81, value = transpose_83)[name = string("attn_output_13_cast_fp16")]; + tensor attn_output_15_perm_0 = const()[name = string("attn_output_15_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_310 = const()[name = string("op_310"), val = tensor([1, 128, 768])]; + tensor attn_output_15_cast_fp16 = transpose(perm = attn_output_15_perm_0, x = attn_output_13_cast_fp16)[name = string("transpose_140")]; + tensor input_57_cast_fp16 = reshape(shape = var_310, x = attn_output_15_cast_fp16)[name = string("input_57_cast_fp16")]; + tensor encoder_encoder_layer_3_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35311168))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35606144))))[name = string("encoder_encoder_layer_3_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_3_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35606976)))]; + tensor linear_21_cast_fp16 = linear(bias = encoder_encoder_layer_3_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_3_attention_output_dense_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = string("linear_21_cast_fp16")]; + tensor input_61_cast_fp16 = add(x = linear_21_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("input_61_cast_fp16")]; + tensor input_63_axes_0 = const()[name = string("input_63_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35608576)))]; + tensor encoder_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35610176)))]; + tensor input_63_cast_fp16 = layer_norm(axes = input_63_axes_0, beta = encoder_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; + tensor encoder_encoder_layer_3_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35611776))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36791488))))[name = string("encoder_encoder_layer_3_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_3_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36794624)))]; + tensor linear_22_cast_fp16 = linear(bias = encoder_encoder_layer_3_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_3_intermediate_dense_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = string("linear_22_cast_fp16")]; + string input_67_mode_0 = const()[name = string("input_67_mode_0"), val = string("EXACT")]; + tensor input_67_cast_fp16 = gelu(mode = input_67_mode_0, x = linear_22_cast_fp16)[name = string("input_67_cast_fp16")]; + tensor encoder_encoder_layer_3_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36800832))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37980544))))[name = string("encoder_encoder_layer_3_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_3_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37981376)))]; + tensor linear_23_cast_fp16 = linear(bias = encoder_encoder_layer_3_output_dense_bias_to_fp16, weight = encoder_encoder_layer_3_output_dense_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = string("linear_23_cast_fp16")]; + tensor input_71_cast_fp16 = add(x = linear_23_cast_fp16, y = input_63_cast_fp16)[name = string("input_71_cast_fp16")]; + tensor hidden_states_25_axes_0 = const()[name = string("hidden_states_25_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_3_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_3_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37982976)))]; + tensor encoder_encoder_layer_3_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_3_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37984576)))]; + tensor hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = encoder_encoder_layer_3_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_3_output_LayerNorm_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor encoder_encoder_layer_4_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37986176))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38281152))))[name = string("encoder_encoder_layer_4_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_4_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38281984)))]; + tensor linear_24_cast_fp16 = linear(bias = encoder_encoder_layer_4_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_4_attention_self_query_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_24_cast_fp16")]; + tensor var_352 = const()[name = string("op_352"), val = tensor([1, -1, 12, 64])]; + tensor var_353_cast_fp16 = reshape(shape = var_352, x = linear_24_cast_fp16)[name = string("op_353_cast_fp16")]; + tensor encoder_encoder_layer_4_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38283584))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38578560))))[name = string("encoder_encoder_layer_4_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_4_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38579392)))]; + tensor linear_25_cast_fp16 = linear(bias = encoder_encoder_layer_4_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_4_attention_self_key_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_25_cast_fp16")]; + tensor var_358 = const()[name = string("op_358"), val = tensor([1, -1, 12, 64])]; + tensor var_359_cast_fp16 = reshape(shape = var_358, x = linear_25_cast_fp16)[name = string("op_359_cast_fp16")]; + tensor encoder_encoder_layer_4_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38580992))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38875968))))[name = string("encoder_encoder_layer_4_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_4_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38876800)))]; + tensor linear_26_cast_fp16 = linear(bias = encoder_encoder_layer_4_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_4_attention_self_value_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_26_cast_fp16")]; + tensor var_364 = const()[name = string("op_364"), val = tensor([1, -1, 12, 64])]; + tensor var_365_cast_fp16 = reshape(shape = var_364, x = linear_26_cast_fp16)[name = string("op_365_cast_fp16")]; + tensor transpose_84_perm_0 = const()[name = string("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_85_perm_0 = const()[name = string("transpose_85_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_86_perm_0 = const()[name = string("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_86 = transpose(perm = transpose_86_perm_0, x = var_365_cast_fp16)[name = string("transpose_137")]; + tensor transpose_85 = transpose(perm = transpose_85_perm_0, x = var_359_cast_fp16)[name = string("transpose_138")]; + tensor transpose_84 = transpose(perm = transpose_84_perm_0, x = var_353_cast_fp16)[name = string("transpose_139")]; + tensor attn_output_17_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_85, query = transpose_84, value = transpose_86)[name = string("attn_output_17_cast_fp16")]; + tensor attn_output_19_perm_0 = const()[name = string("attn_output_19_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_369 = const()[name = string("op_369"), val = tensor([1, 128, 768])]; + tensor attn_output_19_cast_fp16 = transpose(perm = attn_output_19_perm_0, x = attn_output_17_cast_fp16)[name = string("transpose_136")]; + tensor input_73_cast_fp16 = reshape(shape = var_369, x = attn_output_19_cast_fp16)[name = string("input_73_cast_fp16")]; + tensor encoder_encoder_layer_4_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38878400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39173376))))[name = string("encoder_encoder_layer_4_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_4_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39174208)))]; + tensor linear_27_cast_fp16 = linear(bias = encoder_encoder_layer_4_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_4_attention_output_dense_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = string("linear_27_cast_fp16")]; + tensor input_77_cast_fp16 = add(x = linear_27_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("input_77_cast_fp16")]; + tensor input_79_axes_0 = const()[name = string("input_79_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39175808)))]; + tensor encoder_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39177408)))]; + tensor input_79_cast_fp16 = layer_norm(axes = input_79_axes_0, beta = encoder_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + tensor encoder_encoder_layer_4_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39179008))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40358720))))[name = string("encoder_encoder_layer_4_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_4_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40361856)))]; + tensor linear_28_cast_fp16 = linear(bias = encoder_encoder_layer_4_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_4_intermediate_dense_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = string("linear_28_cast_fp16")]; + string input_83_mode_0 = const()[name = string("input_83_mode_0"), val = string("EXACT")]; + tensor input_83_cast_fp16 = gelu(mode = input_83_mode_0, x = linear_28_cast_fp16)[name = string("input_83_cast_fp16")]; + tensor encoder_encoder_layer_4_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40368064))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41547776))))[name = string("encoder_encoder_layer_4_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_4_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41548608)))]; + tensor linear_29_cast_fp16 = linear(bias = encoder_encoder_layer_4_output_dense_bias_to_fp16, weight = encoder_encoder_layer_4_output_dense_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = string("linear_29_cast_fp16")]; + tensor input_87_cast_fp16 = add(x = linear_29_cast_fp16, y = input_79_cast_fp16)[name = string("input_87_cast_fp16")]; + tensor hidden_states_31_axes_0 = const()[name = string("hidden_states_31_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_4_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_4_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41550208)))]; + tensor encoder_encoder_layer_4_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_4_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41551808)))]; + tensor hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = encoder_encoder_layer_4_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_4_output_LayerNorm_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; + tensor encoder_encoder_layer_5_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41553408))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41848384))))[name = string("encoder_encoder_layer_5_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_5_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41849216)))]; + tensor linear_30_cast_fp16 = linear(bias = encoder_encoder_layer_5_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_5_attention_self_query_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = string("linear_30_cast_fp16")]; + tensor var_411 = const()[name = string("op_411"), val = tensor([1, -1, 12, 64])]; + tensor var_412_cast_fp16 = reshape(shape = var_411, x = linear_30_cast_fp16)[name = string("op_412_cast_fp16")]; + tensor encoder_encoder_layer_5_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41850816))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42145792))))[name = string("encoder_encoder_layer_5_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_5_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42146624)))]; + tensor linear_31_cast_fp16 = linear(bias = encoder_encoder_layer_5_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_5_attention_self_key_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = string("linear_31_cast_fp16")]; + tensor var_417 = const()[name = string("op_417"), val = tensor([1, -1, 12, 64])]; + tensor var_418_cast_fp16 = reshape(shape = var_417, x = linear_31_cast_fp16)[name = string("op_418_cast_fp16")]; + tensor encoder_encoder_layer_5_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42148224))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42443200))))[name = string("encoder_encoder_layer_5_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_5_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42444032)))]; + tensor linear_32_cast_fp16 = linear(bias = encoder_encoder_layer_5_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_5_attention_self_value_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = string("linear_32_cast_fp16")]; + tensor var_423 = const()[name = string("op_423"), val = tensor([1, -1, 12, 64])]; + tensor var_424_cast_fp16 = reshape(shape = var_423, x = linear_32_cast_fp16)[name = string("op_424_cast_fp16")]; + tensor transpose_87_perm_0 = const()[name = string("transpose_87_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_88_perm_0 = const()[name = string("transpose_88_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89_perm_0 = const()[name = string("transpose_89_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89 = transpose(perm = transpose_89_perm_0, x = var_424_cast_fp16)[name = string("transpose_133")]; + tensor transpose_88 = transpose(perm = transpose_88_perm_0, x = var_418_cast_fp16)[name = string("transpose_134")]; + tensor transpose_87 = transpose(perm = transpose_87_perm_0, x = var_412_cast_fp16)[name = string("transpose_135")]; + tensor attn_output_21_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_88, query = transpose_87, value = transpose_89)[name = string("attn_output_21_cast_fp16")]; + tensor attn_output_23_perm_0 = const()[name = string("attn_output_23_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_428 = const()[name = string("op_428"), val = tensor([1, 128, 768])]; + tensor attn_output_23_cast_fp16 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = string("transpose_132")]; + tensor input_89_cast_fp16 = reshape(shape = var_428, x = attn_output_23_cast_fp16)[name = string("input_89_cast_fp16")]; + tensor encoder_encoder_layer_5_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42445632))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42740608))))[name = string("encoder_encoder_layer_5_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_5_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42741440)))]; + tensor linear_33_cast_fp16 = linear(bias = encoder_encoder_layer_5_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_5_attention_output_dense_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = string("linear_33_cast_fp16")]; + tensor input_93_cast_fp16 = add(x = linear_33_cast_fp16, y = hidden_states_31_cast_fp16)[name = string("input_93_cast_fp16")]; + tensor input_95_axes_0 = const()[name = string("input_95_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_5_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_5_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42743040)))]; + tensor encoder_encoder_layer_5_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42744640)))]; + tensor input_95_cast_fp16 = layer_norm(axes = input_95_axes_0, beta = encoder_encoder_layer_5_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_5_attention_output_LayerNorm_weight_to_fp16, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; + tensor encoder_encoder_layer_5_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42746240))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43925952))))[name = string("encoder_encoder_layer_5_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_5_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43929088)))]; + tensor linear_34_cast_fp16 = linear(bias = encoder_encoder_layer_5_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_5_intermediate_dense_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = string("linear_34_cast_fp16")]; + string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("EXACT")]; + tensor input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = linear_34_cast_fp16)[name = string("input_99_cast_fp16")]; + tensor encoder_encoder_layer_5_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43935296))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45115008))))[name = string("encoder_encoder_layer_5_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_5_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45115840)))]; + tensor linear_35_cast_fp16 = linear(bias = encoder_encoder_layer_5_output_dense_bias_to_fp16, weight = encoder_encoder_layer_5_output_dense_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = string("linear_35_cast_fp16")]; + tensor input_103_cast_fp16 = add(x = linear_35_cast_fp16, y = input_95_cast_fp16)[name = string("input_103_cast_fp16")]; + tensor hidden_states_37_axes_0 = const()[name = string("hidden_states_37_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_5_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_5_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45117440)))]; + tensor encoder_encoder_layer_5_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_5_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45119040)))]; + tensor hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = encoder_encoder_layer_5_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_5_output_LayerNorm_weight_to_fp16, x = input_103_cast_fp16)[name = string("hidden_states_37_cast_fp16")]; + tensor encoder_encoder_layer_6_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45120640))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45415616))))[name = string("encoder_encoder_layer_6_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_6_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45416448)))]; + tensor linear_36_cast_fp16 = linear(bias = encoder_encoder_layer_6_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_6_attention_self_query_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = string("linear_36_cast_fp16")]; + tensor var_470 = const()[name = string("op_470"), val = tensor([1, -1, 12, 64])]; + tensor var_471_cast_fp16 = reshape(shape = var_470, x = linear_36_cast_fp16)[name = string("op_471_cast_fp16")]; + tensor encoder_encoder_layer_6_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45418048))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45713024))))[name = string("encoder_encoder_layer_6_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_6_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45713856)))]; + tensor linear_37_cast_fp16 = linear(bias = encoder_encoder_layer_6_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_6_attention_self_key_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = string("linear_37_cast_fp16")]; + tensor var_476 = const()[name = string("op_476"), val = tensor([1, -1, 12, 64])]; + tensor var_477_cast_fp16 = reshape(shape = var_476, x = linear_37_cast_fp16)[name = string("op_477_cast_fp16")]; + tensor encoder_encoder_layer_6_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45715456))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46010432))))[name = string("encoder_encoder_layer_6_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_6_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46011264)))]; + tensor linear_38_cast_fp16 = linear(bias = encoder_encoder_layer_6_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_6_attention_self_value_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = string("linear_38_cast_fp16")]; + tensor var_482 = const()[name = string("op_482"), val = tensor([1, -1, 12, 64])]; + tensor var_483_cast_fp16 = reshape(shape = var_482, x = linear_38_cast_fp16)[name = string("op_483_cast_fp16")]; + tensor transpose_90_perm_0 = const()[name = string("transpose_90_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_91_perm_0 = const()[name = string("transpose_91_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_92_perm_0 = const()[name = string("transpose_92_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_92 = transpose(perm = transpose_92_perm_0, x = var_483_cast_fp16)[name = string("transpose_129")]; + tensor transpose_91 = transpose(perm = transpose_91_perm_0, x = var_477_cast_fp16)[name = string("transpose_130")]; + tensor transpose_90 = transpose(perm = transpose_90_perm_0, x = var_471_cast_fp16)[name = string("transpose_131")]; + tensor attn_output_25_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_91, query = transpose_90, value = transpose_92)[name = string("attn_output_25_cast_fp16")]; + tensor attn_output_27_perm_0 = const()[name = string("attn_output_27_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_487 = const()[name = string("op_487"), val = tensor([1, 128, 768])]; + tensor attn_output_27_cast_fp16 = transpose(perm = attn_output_27_perm_0, x = attn_output_25_cast_fp16)[name = string("transpose_128")]; + tensor input_105_cast_fp16 = reshape(shape = var_487, x = attn_output_27_cast_fp16)[name = string("input_105_cast_fp16")]; + tensor encoder_encoder_layer_6_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46012864))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46307840))))[name = string("encoder_encoder_layer_6_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_6_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46308672)))]; + tensor linear_39_cast_fp16 = linear(bias = encoder_encoder_layer_6_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_6_attention_output_dense_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("linear_39_cast_fp16")]; + tensor input_109_cast_fp16 = add(x = linear_39_cast_fp16, y = hidden_states_37_cast_fp16)[name = string("input_109_cast_fp16")]; + tensor input_111_axes_0 = const()[name = string("input_111_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_6_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_6_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46310272)))]; + tensor encoder_encoder_layer_6_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46311872)))]; + tensor input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = encoder_encoder_layer_6_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_6_attention_output_LayerNorm_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")]; + tensor encoder_encoder_layer_6_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46313472))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47493184))))[name = string("encoder_encoder_layer_6_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_6_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47496320)))]; + tensor linear_40_cast_fp16 = linear(bias = encoder_encoder_layer_6_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_6_intermediate_dense_weight_to_fp16_palettized, x = input_111_cast_fp16)[name = string("linear_40_cast_fp16")]; + string input_115_mode_0 = const()[name = string("input_115_mode_0"), val = string("EXACT")]; + tensor input_115_cast_fp16 = gelu(mode = input_115_mode_0, x = linear_40_cast_fp16)[name = string("input_115_cast_fp16")]; + tensor encoder_encoder_layer_6_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47502528))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48682240))))[name = string("encoder_encoder_layer_6_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_6_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48683072)))]; + tensor linear_41_cast_fp16 = linear(bias = encoder_encoder_layer_6_output_dense_bias_to_fp16, weight = encoder_encoder_layer_6_output_dense_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = string("linear_41_cast_fp16")]; + tensor input_119_cast_fp16 = add(x = linear_41_cast_fp16, y = input_111_cast_fp16)[name = string("input_119_cast_fp16")]; + tensor hidden_states_43_axes_0 = const()[name = string("hidden_states_43_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_6_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_6_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48684672)))]; + tensor encoder_encoder_layer_6_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_6_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48686272)))]; + tensor hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = encoder_encoder_layer_6_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_6_output_LayerNorm_weight_to_fp16, x = input_119_cast_fp16)[name = string("hidden_states_43_cast_fp16")]; + tensor encoder_encoder_layer_7_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48687872))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48982848))))[name = string("encoder_encoder_layer_7_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_7_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48983680)))]; + tensor linear_42_cast_fp16 = linear(bias = encoder_encoder_layer_7_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_7_attention_self_query_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = string("linear_42_cast_fp16")]; + tensor var_529 = const()[name = string("op_529"), val = tensor([1, -1, 12, 64])]; + tensor var_530_cast_fp16 = reshape(shape = var_529, x = linear_42_cast_fp16)[name = string("op_530_cast_fp16")]; + tensor encoder_encoder_layer_7_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48985280))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49280256))))[name = string("encoder_encoder_layer_7_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_7_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49281088)))]; + tensor linear_43_cast_fp16 = linear(bias = encoder_encoder_layer_7_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_7_attention_self_key_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = string("linear_43_cast_fp16")]; + tensor var_535 = const()[name = string("op_535"), val = tensor([1, -1, 12, 64])]; + tensor var_536_cast_fp16 = reshape(shape = var_535, x = linear_43_cast_fp16)[name = string("op_536_cast_fp16")]; + tensor encoder_encoder_layer_7_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49282688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49577664))))[name = string("encoder_encoder_layer_7_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_7_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49578496)))]; + tensor linear_44_cast_fp16 = linear(bias = encoder_encoder_layer_7_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_7_attention_self_value_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = string("linear_44_cast_fp16")]; + tensor var_541 = const()[name = string("op_541"), val = tensor([1, -1, 12, 64])]; + tensor var_542_cast_fp16 = reshape(shape = var_541, x = linear_44_cast_fp16)[name = string("op_542_cast_fp16")]; + tensor transpose_93_perm_0 = const()[name = string("transpose_93_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_94_perm_0 = const()[name = string("transpose_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95_perm_0 = const()[name = string("transpose_95_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95 = transpose(perm = transpose_95_perm_0, x = var_542_cast_fp16)[name = string("transpose_125")]; + tensor transpose_94 = transpose(perm = transpose_94_perm_0, x = var_536_cast_fp16)[name = string("transpose_126")]; + tensor transpose_93 = transpose(perm = transpose_93_perm_0, x = var_530_cast_fp16)[name = string("transpose_127")]; + tensor attn_output_29_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_94, query = transpose_93, value = transpose_95)[name = string("attn_output_29_cast_fp16")]; + tensor attn_output_31_perm_0 = const()[name = string("attn_output_31_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_546 = const()[name = string("op_546"), val = tensor([1, 128, 768])]; + tensor attn_output_31_cast_fp16 = transpose(perm = attn_output_31_perm_0, x = attn_output_29_cast_fp16)[name = string("transpose_124")]; + tensor input_121_cast_fp16 = reshape(shape = var_546, x = attn_output_31_cast_fp16)[name = string("input_121_cast_fp16")]; + tensor encoder_encoder_layer_7_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49580096))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49875072))))[name = string("encoder_encoder_layer_7_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_7_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49875904)))]; + tensor linear_45_cast_fp16 = linear(bias = encoder_encoder_layer_7_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_7_attention_output_dense_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = string("linear_45_cast_fp16")]; + tensor input_125_cast_fp16 = add(x = linear_45_cast_fp16, y = hidden_states_43_cast_fp16)[name = string("input_125_cast_fp16")]; + tensor input_127_axes_0 = const()[name = string("input_127_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_7_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_7_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49877504)))]; + tensor encoder_encoder_layer_7_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49879104)))]; + tensor input_127_cast_fp16 = layer_norm(axes = input_127_axes_0, beta = encoder_encoder_layer_7_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_7_attention_output_LayerNorm_weight_to_fp16, x = input_125_cast_fp16)[name = string("input_127_cast_fp16")]; + tensor encoder_encoder_layer_7_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49880704))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51060416))))[name = string("encoder_encoder_layer_7_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_7_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51063552)))]; + tensor linear_46_cast_fp16 = linear(bias = encoder_encoder_layer_7_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_7_intermediate_dense_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = string("linear_46_cast_fp16")]; + string input_131_mode_0 = const()[name = string("input_131_mode_0"), val = string("EXACT")]; + tensor input_131_cast_fp16 = gelu(mode = input_131_mode_0, x = linear_46_cast_fp16)[name = string("input_131_cast_fp16")]; + tensor encoder_encoder_layer_7_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51069760))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52249472))))[name = string("encoder_encoder_layer_7_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_7_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52250304)))]; + tensor linear_47_cast_fp16 = linear(bias = encoder_encoder_layer_7_output_dense_bias_to_fp16, weight = encoder_encoder_layer_7_output_dense_weight_to_fp16_palettized, x = input_131_cast_fp16)[name = string("linear_47_cast_fp16")]; + tensor input_135_cast_fp16 = add(x = linear_47_cast_fp16, y = input_127_cast_fp16)[name = string("input_135_cast_fp16")]; + tensor hidden_states_49_axes_0 = const()[name = string("hidden_states_49_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_7_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_7_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52251904)))]; + tensor encoder_encoder_layer_7_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_7_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52253504)))]; + tensor hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = encoder_encoder_layer_7_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_7_output_LayerNorm_weight_to_fp16, x = input_135_cast_fp16)[name = string("hidden_states_49_cast_fp16")]; + tensor encoder_encoder_layer_8_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52255104))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52550080))))[name = string("encoder_encoder_layer_8_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_8_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52550912)))]; + tensor linear_48_cast_fp16 = linear(bias = encoder_encoder_layer_8_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_8_attention_self_query_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = string("linear_48_cast_fp16")]; + tensor var_588 = const()[name = string("op_588"), val = tensor([1, -1, 12, 64])]; + tensor var_589_cast_fp16 = reshape(shape = var_588, x = linear_48_cast_fp16)[name = string("op_589_cast_fp16")]; + tensor encoder_encoder_layer_8_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52552512))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52847488))))[name = string("encoder_encoder_layer_8_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_8_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52848320)))]; + tensor linear_49_cast_fp16 = linear(bias = encoder_encoder_layer_8_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_8_attention_self_key_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = string("linear_49_cast_fp16")]; + tensor var_594 = const()[name = string("op_594"), val = tensor([1, -1, 12, 64])]; + tensor var_595_cast_fp16 = reshape(shape = var_594, x = linear_49_cast_fp16)[name = string("op_595_cast_fp16")]; + tensor encoder_encoder_layer_8_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52849920))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53144896))))[name = string("encoder_encoder_layer_8_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_8_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53145728)))]; + tensor linear_50_cast_fp16 = linear(bias = encoder_encoder_layer_8_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_8_attention_self_value_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = string("linear_50_cast_fp16")]; + tensor var_600 = const()[name = string("op_600"), val = tensor([1, -1, 12, 64])]; + tensor var_601_cast_fp16 = reshape(shape = var_600, x = linear_50_cast_fp16)[name = string("op_601_cast_fp16")]; + tensor transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_98 = transpose(perm = transpose_98_perm_0, x = var_601_cast_fp16)[name = string("transpose_121")]; + tensor transpose_97 = transpose(perm = transpose_97_perm_0, x = var_595_cast_fp16)[name = string("transpose_122")]; + tensor transpose_96 = transpose(perm = transpose_96_perm_0, x = var_589_cast_fp16)[name = string("transpose_123")]; + tensor attn_output_33_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_97, query = transpose_96, value = transpose_98)[name = string("attn_output_33_cast_fp16")]; + tensor attn_output_35_perm_0 = const()[name = string("attn_output_35_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_605 = const()[name = string("op_605"), val = tensor([1, 128, 768])]; + tensor attn_output_35_cast_fp16 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = string("transpose_120")]; + tensor input_137_cast_fp16 = reshape(shape = var_605, x = attn_output_35_cast_fp16)[name = string("input_137_cast_fp16")]; + tensor encoder_encoder_layer_8_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53147328))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53442304))))[name = string("encoder_encoder_layer_8_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_8_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53443136)))]; + tensor linear_51_cast_fp16 = linear(bias = encoder_encoder_layer_8_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_8_attention_output_dense_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = string("linear_51_cast_fp16")]; + tensor input_141_cast_fp16 = add(x = linear_51_cast_fp16, y = hidden_states_49_cast_fp16)[name = string("input_141_cast_fp16")]; + tensor input_143_axes_0 = const()[name = string("input_143_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_8_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_8_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53444736)))]; + tensor encoder_encoder_layer_8_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53446336)))]; + tensor input_143_cast_fp16 = layer_norm(axes = input_143_axes_0, beta = encoder_encoder_layer_8_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_8_attention_output_LayerNorm_weight_to_fp16, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")]; + tensor encoder_encoder_layer_8_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53447936))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54627648))))[name = string("encoder_encoder_layer_8_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_8_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54630784)))]; + tensor linear_52_cast_fp16 = linear(bias = encoder_encoder_layer_8_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_8_intermediate_dense_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = string("linear_52_cast_fp16")]; + string input_147_mode_0 = const()[name = string("input_147_mode_0"), val = string("EXACT")]; + tensor input_147_cast_fp16 = gelu(mode = input_147_mode_0, x = linear_52_cast_fp16)[name = string("input_147_cast_fp16")]; + tensor encoder_encoder_layer_8_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54636992))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55816704))))[name = string("encoder_encoder_layer_8_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_8_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55817536)))]; + tensor linear_53_cast_fp16 = linear(bias = encoder_encoder_layer_8_output_dense_bias_to_fp16, weight = encoder_encoder_layer_8_output_dense_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = string("linear_53_cast_fp16")]; + tensor input_151_cast_fp16 = add(x = linear_53_cast_fp16, y = input_143_cast_fp16)[name = string("input_151_cast_fp16")]; + tensor hidden_states_55_axes_0 = const()[name = string("hidden_states_55_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_8_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_8_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55819136)))]; + tensor encoder_encoder_layer_8_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_8_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55820736)))]; + tensor hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = encoder_encoder_layer_8_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_8_output_LayerNorm_weight_to_fp16, x = input_151_cast_fp16)[name = string("hidden_states_55_cast_fp16")]; + tensor encoder_encoder_layer_9_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55822336))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56117312))))[name = string("encoder_encoder_layer_9_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_9_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56118144)))]; + tensor linear_54_cast_fp16 = linear(bias = encoder_encoder_layer_9_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_9_attention_self_query_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = string("linear_54_cast_fp16")]; + tensor var_647 = const()[name = string("op_647"), val = tensor([1, -1, 12, 64])]; + tensor var_648_cast_fp16 = reshape(shape = var_647, x = linear_54_cast_fp16)[name = string("op_648_cast_fp16")]; + tensor encoder_encoder_layer_9_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56119744))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56414720))))[name = string("encoder_encoder_layer_9_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_9_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56415552)))]; + tensor linear_55_cast_fp16 = linear(bias = encoder_encoder_layer_9_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_9_attention_self_key_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = string("linear_55_cast_fp16")]; + tensor var_653 = const()[name = string("op_653"), val = tensor([1, -1, 12, 64])]; + tensor var_654_cast_fp16 = reshape(shape = var_653, x = linear_55_cast_fp16)[name = string("op_654_cast_fp16")]; + tensor encoder_encoder_layer_9_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56417152))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56712128))))[name = string("encoder_encoder_layer_9_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_9_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56712960)))]; + tensor linear_56_cast_fp16 = linear(bias = encoder_encoder_layer_9_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_9_attention_self_value_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = string("linear_56_cast_fp16")]; + tensor var_659 = const()[name = string("op_659"), val = tensor([1, -1, 12, 64])]; + tensor var_660_cast_fp16 = reshape(shape = var_659, x = linear_56_cast_fp16)[name = string("op_660_cast_fp16")]; + tensor transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101 = transpose(perm = transpose_101_perm_0, x = var_660_cast_fp16)[name = string("transpose_117")]; + tensor transpose_100 = transpose(perm = transpose_100_perm_0, x = var_654_cast_fp16)[name = string("transpose_118")]; + tensor transpose_99 = transpose(perm = transpose_99_perm_0, x = var_648_cast_fp16)[name = string("transpose_119")]; + tensor attn_output_37_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_100, query = transpose_99, value = transpose_101)[name = string("attn_output_37_cast_fp16")]; + tensor attn_output_39_perm_0 = const()[name = string("attn_output_39_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_664 = const()[name = string("op_664"), val = tensor([1, 128, 768])]; + tensor attn_output_39_cast_fp16 = transpose(perm = attn_output_39_perm_0, x = attn_output_37_cast_fp16)[name = string("transpose_116")]; + tensor input_153_cast_fp16 = reshape(shape = var_664, x = attn_output_39_cast_fp16)[name = string("input_153_cast_fp16")]; + tensor encoder_encoder_layer_9_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56714560))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57009536))))[name = string("encoder_encoder_layer_9_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_9_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57010368)))]; + tensor linear_57_cast_fp16 = linear(bias = encoder_encoder_layer_9_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_9_attention_output_dense_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = string("linear_57_cast_fp16")]; + tensor input_157_cast_fp16 = add(x = linear_57_cast_fp16, y = hidden_states_55_cast_fp16)[name = string("input_157_cast_fp16")]; + tensor input_159_axes_0 = const()[name = string("input_159_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_9_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_9_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57011968)))]; + tensor encoder_encoder_layer_9_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57013568)))]; + tensor input_159_cast_fp16 = layer_norm(axes = input_159_axes_0, beta = encoder_encoder_layer_9_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_9_attention_output_LayerNorm_weight_to_fp16, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")]; + tensor encoder_encoder_layer_9_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57015168))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58194880))))[name = string("encoder_encoder_layer_9_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_9_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58198016)))]; + tensor linear_58_cast_fp16 = linear(bias = encoder_encoder_layer_9_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_9_intermediate_dense_weight_to_fp16_palettized, x = input_159_cast_fp16)[name = string("linear_58_cast_fp16")]; + string input_163_mode_0 = const()[name = string("input_163_mode_0"), val = string("EXACT")]; + tensor input_163_cast_fp16 = gelu(mode = input_163_mode_0, x = linear_58_cast_fp16)[name = string("input_163_cast_fp16")]; + tensor encoder_encoder_layer_9_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58204224))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59383936))))[name = string("encoder_encoder_layer_9_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_9_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59384768)))]; + tensor linear_59_cast_fp16 = linear(bias = encoder_encoder_layer_9_output_dense_bias_to_fp16, weight = encoder_encoder_layer_9_output_dense_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = string("linear_59_cast_fp16")]; + tensor input_167_cast_fp16 = add(x = linear_59_cast_fp16, y = input_159_cast_fp16)[name = string("input_167_cast_fp16")]; + tensor hidden_states_61_axes_0 = const()[name = string("hidden_states_61_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_9_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_9_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59386368)))]; + tensor encoder_encoder_layer_9_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_9_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59387968)))]; + tensor hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = encoder_encoder_layer_9_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_9_output_LayerNorm_weight_to_fp16, x = input_167_cast_fp16)[name = string("hidden_states_61_cast_fp16")]; + tensor encoder_encoder_layer_10_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59389568))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59684544))))[name = string("encoder_encoder_layer_10_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_10_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59685376)))]; + tensor linear_60_cast_fp16 = linear(bias = encoder_encoder_layer_10_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_10_attention_self_query_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = string("linear_60_cast_fp16")]; + tensor var_706 = const()[name = string("op_706"), val = tensor([1, -1, 12, 64])]; + tensor var_707_cast_fp16 = reshape(shape = var_706, x = linear_60_cast_fp16)[name = string("op_707_cast_fp16")]; + tensor encoder_encoder_layer_10_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59686976))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59981952))))[name = string("encoder_encoder_layer_10_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_10_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59982784)))]; + tensor linear_61_cast_fp16 = linear(bias = encoder_encoder_layer_10_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_10_attention_self_key_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = string("linear_61_cast_fp16")]; + tensor var_712 = const()[name = string("op_712"), val = tensor([1, -1, 12, 64])]; + tensor var_713_cast_fp16 = reshape(shape = var_712, x = linear_61_cast_fp16)[name = string("op_713_cast_fp16")]; + tensor encoder_encoder_layer_10_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59984384))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60279360))))[name = string("encoder_encoder_layer_10_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_10_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60280192)))]; + tensor linear_62_cast_fp16 = linear(bias = encoder_encoder_layer_10_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_10_attention_self_value_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = string("linear_62_cast_fp16")]; + tensor var_718 = const()[name = string("op_718"), val = tensor([1, -1, 12, 64])]; + tensor var_719_cast_fp16 = reshape(shape = var_718, x = linear_62_cast_fp16)[name = string("op_719_cast_fp16")]; + tensor transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_104 = transpose(perm = transpose_104_perm_0, x = var_719_cast_fp16)[name = string("transpose_113")]; + tensor transpose_103 = transpose(perm = transpose_103_perm_0, x = var_713_cast_fp16)[name = string("transpose_114")]; + tensor transpose_102 = transpose(perm = transpose_102_perm_0, x = var_707_cast_fp16)[name = string("transpose_115")]; + tensor attn_output_41_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_103, query = transpose_102, value = transpose_104)[name = string("attn_output_41_cast_fp16")]; + tensor attn_output_43_perm_0 = const()[name = string("attn_output_43_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_723 = const()[name = string("op_723"), val = tensor([1, 128, 768])]; + tensor attn_output_43_cast_fp16 = transpose(perm = attn_output_43_perm_0, x = attn_output_41_cast_fp16)[name = string("transpose_112")]; + tensor input_169_cast_fp16 = reshape(shape = var_723, x = attn_output_43_cast_fp16)[name = string("input_169_cast_fp16")]; + tensor encoder_encoder_layer_10_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60281792))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60576768))))[name = string("encoder_encoder_layer_10_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_10_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60577600)))]; + tensor linear_63_cast_fp16 = linear(bias = encoder_encoder_layer_10_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_10_attention_output_dense_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = string("linear_63_cast_fp16")]; + tensor input_173_cast_fp16 = add(x = linear_63_cast_fp16, y = hidden_states_61_cast_fp16)[name = string("input_173_cast_fp16")]; + tensor input_175_axes_0 = const()[name = string("input_175_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_10_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_10_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60579200)))]; + tensor encoder_encoder_layer_10_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60580800)))]; + tensor input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = encoder_encoder_layer_10_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_10_attention_output_LayerNorm_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")]; + tensor encoder_encoder_layer_10_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60582400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61762112))))[name = string("encoder_encoder_layer_10_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_10_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61765248)))]; + tensor linear_64_cast_fp16 = linear(bias = encoder_encoder_layer_10_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_10_intermediate_dense_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = string("linear_64_cast_fp16")]; + string input_179_mode_0 = const()[name = string("input_179_mode_0"), val = string("EXACT")]; + tensor input_179_cast_fp16 = gelu(mode = input_179_mode_0, x = linear_64_cast_fp16)[name = string("input_179_cast_fp16")]; + tensor encoder_encoder_layer_10_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61771456))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62951168))))[name = string("encoder_encoder_layer_10_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_10_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62952000)))]; + tensor linear_65_cast_fp16 = linear(bias = encoder_encoder_layer_10_output_dense_bias_to_fp16, weight = encoder_encoder_layer_10_output_dense_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = string("linear_65_cast_fp16")]; + tensor input_183_cast_fp16 = add(x = linear_65_cast_fp16, y = input_175_cast_fp16)[name = string("input_183_cast_fp16")]; + tensor hidden_states_67_axes_0 = const()[name = string("hidden_states_67_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_10_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_10_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62953600)))]; + tensor encoder_encoder_layer_10_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_10_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62955200)))]; + tensor hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = encoder_encoder_layer_10_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_10_output_LayerNorm_weight_to_fp16, x = input_183_cast_fp16)[name = string("hidden_states_67_cast_fp16")]; + tensor encoder_encoder_layer_11_attention_self_query_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62956800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63251776))))[name = string("encoder_encoder_layer_11_attention_self_query_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_11_attention_self_query_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63252608)))]; + tensor linear_66_cast_fp16 = linear(bias = encoder_encoder_layer_11_attention_self_query_bias_to_fp16, weight = encoder_encoder_layer_11_attention_self_query_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = string("linear_66_cast_fp16")]; + tensor var_765 = const()[name = string("op_765"), val = tensor([1, -1, 12, 64])]; + tensor var_766_cast_fp16 = reshape(shape = var_765, x = linear_66_cast_fp16)[name = string("op_766_cast_fp16")]; + tensor encoder_encoder_layer_11_attention_self_key_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63254208))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63549184))))[name = string("encoder_encoder_layer_11_attention_self_key_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_11_attention_self_key_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63550016)))]; + tensor linear_67_cast_fp16 = linear(bias = encoder_encoder_layer_11_attention_self_key_bias_to_fp16, weight = encoder_encoder_layer_11_attention_self_key_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = string("linear_67_cast_fp16")]; + tensor var_771 = const()[name = string("op_771"), val = tensor([1, -1, 12, 64])]; + tensor var_772_cast_fp16 = reshape(shape = var_771, x = linear_67_cast_fp16)[name = string("op_772_cast_fp16")]; + tensor encoder_encoder_layer_11_attention_self_value_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63551616))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63846592))))[name = string("encoder_encoder_layer_11_attention_self_value_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_11_attention_self_value_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63847424)))]; + tensor linear_68_cast_fp16 = linear(bias = encoder_encoder_layer_11_attention_self_value_bias_to_fp16, weight = encoder_encoder_layer_11_attention_self_value_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = string("linear_68_cast_fp16")]; + tensor var_777 = const()[name = string("op_777"), val = tensor([1, -1, 12, 64])]; + tensor var_778_cast_fp16 = reshape(shape = var_777, x = linear_68_cast_fp16)[name = string("op_778_cast_fp16")]; + tensor transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107 = transpose(perm = transpose_107_perm_0, x = var_778_cast_fp16)[name = string("transpose_109")]; + tensor transpose_106 = transpose(perm = transpose_106_perm_0, x = var_772_cast_fp16)[name = string("transpose_110")]; + tensor transpose_105 = transpose(perm = transpose_105_perm_0, x = var_766_cast_fp16)[name = string("transpose_111")]; + tensor attn_output_45_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_106, query = transpose_105, value = transpose_107)[name = string("attn_output_45_cast_fp16")]; + tensor attn_output_perm_0 = const()[name = string("attn_output_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_782 = const()[name = string("op_782"), val = tensor([1, 128, 768])]; + tensor attn_output_cast_fp16 = transpose(perm = attn_output_perm_0, x = attn_output_45_cast_fp16)[name = string("transpose_108")]; + tensor input_185_cast_fp16 = reshape(shape = var_782, x = attn_output_cast_fp16)[name = string("input_185_cast_fp16")]; + tensor encoder_encoder_layer_11_attention_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63849024))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64144000))))[name = string("encoder_encoder_layer_11_attention_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_11_attention_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64144832)))]; + tensor linear_69_cast_fp16 = linear(bias = encoder_encoder_layer_11_attention_output_dense_bias_to_fp16, weight = encoder_encoder_layer_11_attention_output_dense_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = string("linear_69_cast_fp16")]; + tensor input_189_cast_fp16 = add(x = linear_69_cast_fp16, y = hidden_states_67_cast_fp16)[name = string("input_189_cast_fp16")]; + tensor input_191_axes_0 = const()[name = string("input_191_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_11_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_11_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64146432)))]; + tensor encoder_encoder_layer_11_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64148032)))]; + tensor input_191_cast_fp16 = layer_norm(axes = input_191_axes_0, beta = encoder_encoder_layer_11_attention_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_11_attention_output_LayerNorm_weight_to_fp16, x = input_189_cast_fp16)[name = string("input_191_cast_fp16")]; + tensor encoder_encoder_layer_11_intermediate_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64149632))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65329344))))[name = string("encoder_encoder_layer_11_intermediate_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_11_intermediate_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65332480)))]; + tensor linear_70_cast_fp16 = linear(bias = encoder_encoder_layer_11_intermediate_dense_bias_to_fp16, weight = encoder_encoder_layer_11_intermediate_dense_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = string("linear_70_cast_fp16")]; + string input_195_mode_0 = const()[name = string("input_195_mode_0"), val = string("EXACT")]; + tensor input_195_cast_fp16 = gelu(mode = input_195_mode_0, x = linear_70_cast_fp16)[name = string("input_195_cast_fp16")]; + tensor encoder_encoder_layer_11_output_dense_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65338688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66518400))))[name = string("encoder_encoder_layer_11_output_dense_weight_to_fp16_palettized")]; + tensor encoder_encoder_layer_11_output_dense_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66519232)))]; + tensor linear_71_cast_fp16 = linear(bias = encoder_encoder_layer_11_output_dense_bias_to_fp16, weight = encoder_encoder_layer_11_output_dense_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = string("linear_71_cast_fp16")]; + tensor input_199_cast_fp16 = add(x = linear_71_cast_fp16, y = input_191_cast_fp16)[name = string("input_199_cast_fp16")]; + tensor hidden_states_axes_0 = const()[name = string("hidden_states_axes_0"), val = tensor([-1])]; + tensor encoder_encoder_layer_11_output_LayerNorm_weight_to_fp16 = const()[name = string("encoder_encoder_layer_11_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66520832)))]; + tensor encoder_encoder_layer_11_output_LayerNorm_bias_to_fp16 = const()[name = string("encoder_encoder_layer_11_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66522432)))]; + tensor hidden_states_cast_fp16 = layer_norm(axes = hidden_states_axes_0, beta = encoder_encoder_layer_11_output_LayerNorm_bias_to_fp16, epsilon = var_21_to_fp16, gamma = encoder_encoder_layer_11_output_LayerNorm_weight_to_fp16, x = input_199_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_812_axes_0 = const()[name = string("op_812_axes_0"), val = tensor([-1])]; + tensor var_812 = expand_dims(axes = var_812_axes_0, x = attention_mask)[name = string("op_812")]; + string mask_to_fp16_dtype_0 = const()[name = string("mask_to_fp16_dtype_0"), val = string("fp16")]; + tensor var_812_to_fp16 = cast(dtype = mask_to_fp16_dtype_0, x = var_812)[name = string("cast_0")]; + tensor var_818_cast_fp16 = mul(x = hidden_states_cast_fp16, y = var_812_to_fp16)[name = string("op_818_cast_fp16")]; + tensor var_823_axes_0 = const()[name = string("op_823_axes_0"), val = tensor([1])]; + bool var_823_keep_dims_0 = const()[name = string("op_823_keep_dims_0"), val = bool(false)]; + tensor var_823_cast_fp16 = reduce_sum(axes = var_823_axes_0, keep_dims = var_823_keep_dims_0, x = var_818_cast_fp16)[name = string("op_823_cast_fp16")]; + tensor var_828_axes_0 = const()[name = string("op_828_axes_0"), val = tensor([1])]; + bool var_828_keep_dims_0 = const()[name = string("op_828_keep_dims_0"), val = bool(false)]; + tensor var_828_cast_fp16 = reduce_sum(axes = var_828_axes_0, keep_dims = var_828_keep_dims_0, x = var_812_to_fp16)[name = string("op_828_cast_fp16")]; + fp16 var_829_promoted_to_fp16 = const()[name = string("op_829_promoted_to_fp16"), val = fp16(0x1p+0)]; + fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(inf)]; + tensor clip_0_cast_fp16 = clip(alpha = var_829_promoted_to_fp16, beta = const_29_to_fp16, x = var_828_cast_fp16)[name = string("clip_0_cast_fp16")]; + tensor input_205_cast_fp16 = real_div(x = var_823_cast_fp16, y = clip_0_cast_fp16)[name = string("input_205_cast_fp16")]; + tensor content_head_weight_to_fp16 = const()[name = string("content_head_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66524032)))]; + tensor content_head_bias_to_fp16 = const()[name = string("content_head_bias_to_fp16"), val = tensor([-0x1.3ecp-7, -0x1.c4p-6, 0x1.068p-5, 0x1.088p-7, -0x1.474p-6])]; + tensor content_logits = linear(bias = content_head_bias_to_fp16, weight = content_head_weight_to_fp16, x = input_205_cast_fp16)[name = string("linear_72_cast_fp16")]; + tensor target_head_weight_to_fp16 = const()[name = string("target_head_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66531776)))]; + tensor target_head_bias_to_fp16 = const()[name = string("target_head_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66547200)))]; + tensor target_logits = linear(bias = target_head_bias_to_fp16, weight = target_head_weight_to_fp16, x = input_205_cast_fp16)[name = string("linear_73_cast_fp16")]; + } -> (content_logits, target_logits); +} \ No newline at end of file