diff --git "a/llm.mnn.json" "b/llm.mnn.json" new file mode 100644--- /dev/null +++ "b/llm.mnn.json" @@ -0,0 +1,35826 @@ + +{ "bizCode": "MNNTest", "extraInfo": +{ "version": "3.4.0" } +, "oplists": +[ +{ "main_type": "Input", "main": +{ "dims": +[ 1, 1, 4096 ] +, "dtype": "DT_FLOAT", "dformat": "NCHW" } +, "name": "input_ids", "outputIndexes": +[ 2 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Input", "main": +{ "dims": +[ 1, 1, -1, -1 ] +, "dtype": "DT_FLOAT", "dformat": "NCHW" } +, "name": "attention_mask", "outputIndexes": +[ 128 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Input", "main": +{ "dims": +[ 1, -1 ] +, "dtype": "DT_INT32", "dformat": "NCHW" } +, "name": "position_ids", "outputIndexes": +[ 0 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Input", "main": +{ "dims": +[ 1 ] +, "dtype": "DT_INT32", "dformat": "NCHW" } +, "name": "logits_index", "outputIndexes": +[ 1 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 3 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1, -1, 4096 ] + } +, "name": "/blocks.0/Constant_output_0", "outputIndexes": +[ 3 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.0/Reshape_output_0", "outputIndexes": +[ 4 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 0, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.0/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 5 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 5 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3397 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3397 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3398 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3398 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 32768, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.0/self_attn/q_proj/Linear", "outputIndexes": +[ 3399 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3399 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3400 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3400 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 6 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 5 ] +, "main_type": "NONE", "name": "/blocks.0/self_attn/Shape_output_0", "outputIndexes": +[ 7 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 7 ] +, "main_type": "NONE", "name": "Shape22", "outputIndexes": +[ 8 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 7 ] +, "main_type": "NONE", "name": "Rank24", "outputIndexes": +[ 9 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 9, 9 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp26", "outputIndexes": +[ 10 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 10 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze27", "outputIndexes": +[ 11 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1 ] + } +, "name": "Const28", "outputIndexes": +[ 12 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 10, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp29", "outputIndexes": +[ 13 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 13 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze30", "outputIndexes": +[ 14 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1 ] + } +, "name": "Unsqueeze32", "outputIndexes": +[ 15 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 8, 11, 14, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice33", "outputIndexes": +[ 16 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 16 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze34", "outputIndexes": +[ 17 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 17, 17 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp36", "outputIndexes": +[ 18 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 7, 18, 10 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.0/self_attn/Gather_output_0", "outputIndexes": +[ 19 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 0 ] + } +, "name": "/rotary/Constant_3_output_0", "outputIndexes": +[ 20 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 19, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 21 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 10 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze46", "outputIndexes": +[ 22 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 13 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze49", "outputIndexes": +[ 23 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 8, 22, 23, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice52", "outputIndexes": +[ 24 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 24 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze53", "outputIndexes": +[ 25 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 25 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp54", "outputIndexes": +[ 26 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 26, 25 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp55", "outputIndexes": +[ 27 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 7, 27, 10 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.0/self_attn/Gather_1_output_0", "outputIndexes": +[ 28 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 28, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 29 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 32 ] + } +, "name": "/blocks.0/self_attn/Constant_2_output_0", "outputIndexes": +[ 30 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 128 ] + } +, "name": "/blocks.0/self_attn/Constant_3_output_0", "outputIndexes": +[ 31 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 21, 29, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.0/self_attn/Concat_output_0", "outputIndexes": +[ 32 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 6, 32 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.0/self_attn/Reshape_output_0", "outputIndexes": +[ 33 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 33 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 10518550, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.0/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 35 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 0 ] +, "main_type": "CastParam", "main": +{ "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" } +, "name": "/rotary/Cast_output_0", "outputIndexes": +[ 36 ] +, "type": "Cast", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 2 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ -1, 1 ] + } +, "name": "/rotary/Constant_output_0", "outputIndexes": +[ 37 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 36, 37 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/rotary/Reshape_output_0", "outputIndexes": +[ 38 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 64 ] +, "dataFormat": "NCHW", "dataType": "DT_FLOAT", "float32s": +[ 1.0, 0.805842, 0.649382, 0.523299, 0.421697, 0.339821, 0.273842, 0.220673, 0.177828, 0.143301, 0.115478, 0.093057, 0.074989, 0.06043, 0.048697, 0.039242, 0.031623, 0.025483, 0.020535, 0.016548, 0.013335, 0.010746, 0.00866, 0.006978, 0.005623, 0.004532, 0.003652, 0.002943, 0.002371, 0.001911, 0.00154, 0.001241, 0.001, 0.000806, 0.000649, 0.000523, 0.000422, 0.00034, 0.000274, 0.000221, 0.000178, 0.000143, 0.000115, 0.000093, 0.000075, 0.00006, 0.000049, 0.000039, 0.000032, 0.000025, 0.000021, 0.000017, 0.000013, 0.000011, 0.000009, 0.000007, 0.000006, 0.000005, 0.000004, 0.000003, 0.000002, 0.000002, 0.000002, 0.000001 ] + } +, "name": "/rotary/Constant_1_output_0", "outputIndexes": +[ 39 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 38, 39 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/rotary/Mul_output_0", "outputIndexes": +[ 40 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 40 ] +, "main_type": "UnaryOp", "main": +{ "opType": "COS", "T": "DT_FLOAT" } +, "name": "/rotary/Cos_output_0", "outputIndexes": +[ 41 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 41, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_output_0", "outputIndexes": +[ 42 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 40 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SIN", "T": "DT_FLOAT" } +, "name": "/rotary/Sin_output_0", "outputIndexes": +[ 43 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 43, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_1_output_0", "outputIndexes": +[ 44 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 42, 44 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/rotary/Concat_output_0", "outputIndexes": +[ 45 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 45, 45 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/rotary/Concat_1_output_0", "outputIndexes": +[ 46 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 2 ] + } +, "name": "/rotary/Constant_5_output_0", "outputIndexes": +[ 47 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 46, 47 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_2_output_0", "outputIndexes": +[ 48 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 48, 15 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_3_output_0", "outputIndexes": +[ 49 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 49 ] +, "main_type": "NONE", "name": "Shape85", "outputIndexes": +[ 50 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 49 ] +, "main_type": "NONE", "name": "Rank87", "outputIndexes": +[ 51 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 51, 51 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp89", "outputIndexes": +[ 52 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 52 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze90", "outputIndexes": +[ 53 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 52, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp92", "outputIndexes": +[ 54 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 54 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze93", "outputIndexes": +[ 55 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 50, 53, 55, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice96", "outputIndexes": +[ 56 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 56 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze97", "outputIndexes": +[ 57 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 57, 57 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp99", "outputIndexes": +[ 58 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 49, 58, 52 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.0/self_attn/Gather_2_output_0", "outputIndexes": +[ 59 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 35, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Mul_output_0", "outputIndexes": +[ 60 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 35 ] +, "main_type": "NONE", "name": "/blocks.0/self_attn/Shape_2_output_0", "outputIndexes": +[ 61 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "main_type": "Blob", "main": +{ "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 3 ] + } +, "name": "/blocks.0/self_attn/Constant_8_output_0", "outputIndexes": +[ 62 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 61 ] +, "main_type": "NONE", "name": "Shape104", "outputIndexes": +[ 63 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 61 ] +, "main_type": "NONE", "name": "Rank106", "outputIndexes": +[ 64 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 64 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp108", "outputIndexes": +[ 65 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 65 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze109", "outputIndexes": +[ 66 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 65, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp111", "outputIndexes": +[ 67 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 67 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze112", "outputIndexes": +[ 68 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 63, 66, 68, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice115", "outputIndexes": +[ 69 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 69 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze116", "outputIndexes": +[ 70 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 70 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp117", "outputIndexes": +[ 71 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 71, 70 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp118", "outputIndexes": +[ 72 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 61, 72, 65 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.0/self_attn/Gather_4_output_0", "outputIndexes": +[ 73 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 2 ] + } +, "name": "/blocks.0/self_attn/Constant_9_output_0", "outputIndexes": +[ 74 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 73, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Div_output_0", "outputIndexes": +[ 75 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 75, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 76 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1073741824 ] + } +, "name": "/blocks.0/self_attn/Constant_16_output_0", "outputIndexes": +[ 77 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 3 ] + } +, "name": "/blocks.0/self_attn/Constant_10_output_0", "outputIndexes": +[ 78 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 35, 76, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.0/self_attn/Slice_1_output_0", "outputIndexes": +[ 79 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 79 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.0/self_attn/Neg_output_0", "outputIndexes": +[ 80 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 75, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 81 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 35, 20, 81, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.0/self_attn/Slice_output_0", "outputIndexes": +[ 82 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 80, 82 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.0/self_attn/Concat_3_output_0", "outputIndexes": +[ 83 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 52 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze136", "outputIndexes": +[ 84 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 54 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze139", "outputIndexes": +[ 85 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 50, 84, 85, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice142", "outputIndexes": +[ 86 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 86 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze143", "outputIndexes": +[ 87 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 87 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp144", "outputIndexes": +[ 88 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 88, 87 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp145", "outputIndexes": +[ 89 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 49, 89, 52 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.0/self_attn/Gather_3_output_0", "outputIndexes": +[ 90 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 83, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Mul_1_output_0", "outputIndexes": +[ 91 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 60, 91 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Add_output_0", "outputIndexes": +[ 92 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3398 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 10519574, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.0/self_attn/k_proj/Linear", "outputIndexes": +[ 3403 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3403 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3404 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3404 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 93 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 19, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 94 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 28, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 95 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 8 ] + } +, "name": "/blocks.0/self_attn/Constant_4_output_0", "outputIndexes": +[ 96 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 94, 95, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.0/self_attn/Concat_1_output_0", "outputIndexes": +[ 97 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 93, 97 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.0/self_attn/Reshape_1_output_0", "outputIndexes": +[ 98 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 98 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 13141036, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.0/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 100 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 100, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Mul_2_output_0", "outputIndexes": +[ 101 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 100 ] +, "main_type": "NONE", "name": "/blocks.0/self_attn/Shape_3_output_0", "outputIndexes": +[ 102 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 102 ] +, "main_type": "NONE", "name": "Shape166", "outputIndexes": +[ 103 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 102 ] +, "main_type": "NONE", "name": "Rank168", "outputIndexes": +[ 104 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 104, 104 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp170", "outputIndexes": +[ 105 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 105 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze171", "outputIndexes": +[ 106 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 105, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp173", "outputIndexes": +[ 107 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 107 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze174", "outputIndexes": +[ 108 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 103, 106, 108, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice177", "outputIndexes": +[ 109 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 109 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze178", "outputIndexes": +[ 110 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 110 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp179", "outputIndexes": +[ 111 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 111, 110 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp180", "outputIndexes": +[ 112 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 102, 112, 105 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.0/self_attn/Gather_5_output_0", "outputIndexes": +[ 113 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 113, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Div_1_output_0", "outputIndexes": +[ 114 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 114, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 115 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 100, 115, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.0/self_attn/Slice_3_output_0", "outputIndexes": +[ 116 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 116 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.0/self_attn/Neg_1_output_0", "outputIndexes": +[ 117 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 114, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 118 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 100, 20, 118, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.0/self_attn/Slice_2_output_0", "outputIndexes": +[ 119 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 117, 119 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.0/self_attn/Concat_4_output_0", "outputIndexes": +[ 120 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 120, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Mul_3_output_0", "outputIndexes": +[ 121 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 101, 121 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/self_attn/Add_1_output_0", "outputIndexes": +[ 122 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3398 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 13142060, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.0/self_attn/v_proj/Linear", "outputIndexes": +[ 3407 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3407 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3408 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3408 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 123 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 19, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 124 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 28, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.0/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 125 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 124, 125, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.0/self_attn/Concat_2_output_0", "outputIndexes": +[ 126 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 123, 126 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.0/self_attn/Reshape_2_output_0", "outputIndexes": +[ 127 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 92, 122, 127, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.0/self_attn/FusedAttention", "outputIndexes": +[ 129 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 129 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3409 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3409 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3410 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3410 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 15763522, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.0/self_attn/o_proj/Linear", "outputIndexes": +[ 3411 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3411 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3412 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3412 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 130 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4, 130 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/Add_output_0", "outputIndexes": +[ 131 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 131 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 26249304, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.0/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 132 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 132 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3413 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3413 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3414 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3414 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 26282072, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.0/mlp/gate_proj/Linear", "outputIndexes": +[ 3415 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3415 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3416 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3416 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 133 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 133 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.0/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 134 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3414 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 57739374, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.0/mlp/up_proj/Linear", "outputIndexes": +[ 3419 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3419 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3420 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3420 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 135 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 134, 135 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/mlp/Mul_output_0", "outputIndexes": +[ 136 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 136 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3421 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3421 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3422 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3422 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 89196676, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.0/mlp/down_proj/Linear", "outputIndexes": +[ 3423 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3423 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3424 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3424 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 137 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 131, 137 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.0/Add_1_output_0", "outputIndexes": +[ 138 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 138, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.1/Reshape_output_0", "outputIndexes": +[ 139 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 139 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 120653978, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.1/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 140 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 140 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3425 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3425 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3426 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3426 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 120686746, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.1/self_attn/q_proj/Linear", "outputIndexes": +[ 3427 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3427 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3428 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3428 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 141 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 140 ] +, "main_type": "NONE", "name": "/blocks.1/self_attn/Shape_output_0", "outputIndexes": +[ 142 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 142 ] +, "main_type": "NONE", "name": "Shape228", "outputIndexes": +[ 143 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 142 ] +, "main_type": "NONE", "name": "Rank230", "outputIndexes": +[ 144 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 144, 144 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp232", "outputIndexes": +[ 145 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 145 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze233", "outputIndexes": +[ 146 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 145, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp235", "outputIndexes": +[ 147 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 147 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze236", "outputIndexes": +[ 148 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 143, 146, 148, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice239", "outputIndexes": +[ 149 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 149 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze240", "outputIndexes": +[ 150 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 150, 150 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp242", "outputIndexes": +[ 151 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 142, 151, 145 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.1/self_attn/Gather_output_0", "outputIndexes": +[ 152 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 152, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 153 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 145 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze250", "outputIndexes": +[ 154 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 147 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze253", "outputIndexes": +[ 155 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 143, 154, 155, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice256", "outputIndexes": +[ 156 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 156 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze257", "outputIndexes": +[ 157 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 157 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp258", "outputIndexes": +[ 158 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 158, 157 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp259", "outputIndexes": +[ 159 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 142, 159, 145 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.1/self_attn/Gather_1_output_0", "outputIndexes": +[ 160 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 160, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 161 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 153, 161, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.1/self_attn/Concat_output_0", "outputIndexes": +[ 162 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 141, 162 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.1/self_attn/Reshape_output_0", "outputIndexes": +[ 163 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 163 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 131172528, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.1/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 165 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 165, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Mul_output_0", "outputIndexes": +[ 166 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 165 ] +, "main_type": "NONE", "name": "/blocks.1/self_attn/Shape_2_output_0", "outputIndexes": +[ 167 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 167 ] +, "main_type": "NONE", "name": "Shape274", "outputIndexes": +[ 168 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 167 ] +, "main_type": "NONE", "name": "Rank276", "outputIndexes": +[ 169 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 169, 169 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp278", "outputIndexes": +[ 170 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 170 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze279", "outputIndexes": +[ 171 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 170, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp281", "outputIndexes": +[ 172 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 172 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze282", "outputIndexes": +[ 173 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 168, 171, 173, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice285", "outputIndexes": +[ 174 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 174 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze286", "outputIndexes": +[ 175 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 175 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp287", "outputIndexes": +[ 176 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 176, 175 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp288", "outputIndexes": +[ 177 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 167, 177, 170 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.1/self_attn/Gather_2_output_0", "outputIndexes": +[ 178 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 178, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Div_output_0", "outputIndexes": +[ 179 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 179, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 180 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 165, 180, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.1/self_attn/Slice_1_output_0", "outputIndexes": +[ 181 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 181 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.1/self_attn/Neg_output_0", "outputIndexes": +[ 182 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 179, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 183 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 165, 20, 183, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.1/self_attn/Slice_output_0", "outputIndexes": +[ 184 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 182, 184 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.1/self_attn/Concat_3_output_0", "outputIndexes": +[ 185 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 185, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Mul_1_output_0", "outputIndexes": +[ 186 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 166, 186 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Add_output_0", "outputIndexes": +[ 187 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3426 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 131173552, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.1/self_attn/k_proj/Linear", "outputIndexes": +[ 3431 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3431 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3432 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3432 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 188 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 152, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 189 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 160, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 190 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 189, 190, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.1/self_attn/Concat_1_output_0", "outputIndexes": +[ 191 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 188, 191 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.1/self_attn/Reshape_1_output_0", "outputIndexes": +[ 192 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 192 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 133795014, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.1/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 194 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 194, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Mul_2_output_0", "outputIndexes": +[ 195 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 194 ] +, "main_type": "NONE", "name": "/blocks.1/self_attn/Shape_3_output_0", "outputIndexes": +[ 196 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 196 ] +, "main_type": "NONE", "name": "Shape316", "outputIndexes": +[ 197 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 196 ] +, "main_type": "NONE", "name": "Rank318", "outputIndexes": +[ 198 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 198, 198 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp320", "outputIndexes": +[ 199 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 199 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze321", "outputIndexes": +[ 200 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 199, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp323", "outputIndexes": +[ 201 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 201 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze324", "outputIndexes": +[ 202 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 197, 200, 202, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice327", "outputIndexes": +[ 203 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 203 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze328", "outputIndexes": +[ 204 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 204 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp329", "outputIndexes": +[ 205 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 205, 204 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp330", "outputIndexes": +[ 206 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 196, 206, 199 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.1/self_attn/Gather_3_output_0", "outputIndexes": +[ 207 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 207, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Div_1_output_0", "outputIndexes": +[ 208 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 208, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 209 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 194, 209, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.1/self_attn/Slice_3_output_0", "outputIndexes": +[ 210 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 210 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.1/self_attn/Neg_1_output_0", "outputIndexes": +[ 211 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 208, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 212 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 194, 20, 212, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.1/self_attn/Slice_2_output_0", "outputIndexes": +[ 213 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 211, 213 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.1/self_attn/Concat_4_output_0", "outputIndexes": +[ 214 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 214, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Mul_3_output_0", "outputIndexes": +[ 215 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 195, 215 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/self_attn/Add_1_output_0", "outputIndexes": +[ 216 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3426 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 133796038, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.1/self_attn/v_proj/Linear", "outputIndexes": +[ 3435 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3435 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3436 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3436 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 217 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 152, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 218 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 160, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.1/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 219 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 218, 219, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.1/self_attn/Concat_2_output_0", "outputIndexes": +[ 220 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 217, 220 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.1/self_attn/Reshape_2_output_0", "outputIndexes": +[ 221 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 187, 216, 221, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.1/self_attn/FusedAttention", "outputIndexes": +[ 222 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 222 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3437 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3437 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3438 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3438 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 136417500, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.1/self_attn/o_proj/Linear", "outputIndexes": +[ 3439 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3439 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3440 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3440 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 223 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 139, 223 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/Add_output_0", "outputIndexes": +[ 224 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 224 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 146903282, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.1/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 225 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 225 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3441 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3441 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3442 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3442 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 146936050, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.1/mlp/gate_proj/Linear", "outputIndexes": +[ 3443 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3443 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3444 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3444 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 226 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 226 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.1/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 227 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3442 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 178393352, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.1/mlp/up_proj/Linear", "outputIndexes": +[ 3447 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3447 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3448 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3448 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 228 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 227, 228 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/mlp/Mul_output_0", "outputIndexes": +[ 229 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 229 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3449 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3449 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3450 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3450 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 209850654, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.1/mlp/down_proj/Linear", "outputIndexes": +[ 3451 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3451 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3452 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3452 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 230 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 224, 230 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.1/Add_1_output_0", "outputIndexes": +[ 231 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 231, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.2/Reshape_output_0", "outputIndexes": +[ 232 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 232 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 241307956, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.2/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 233 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 233 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3453 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3453 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3454 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3454 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 241340724, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.2/self_attn/q_proj/Linear", "outputIndexes": +[ 3455 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3455 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3456 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3456 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 234 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 233 ] +, "main_type": "NONE", "name": "/blocks.2/self_attn/Shape_output_0", "outputIndexes": +[ 235 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 235 ] +, "main_type": "NONE", "name": "Shape377", "outputIndexes": +[ 236 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 235 ] +, "main_type": "NONE", "name": "Rank379", "outputIndexes": +[ 237 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 237, 237 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp381", "outputIndexes": +[ 238 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 238 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze382", "outputIndexes": +[ 239 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 238, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp384", "outputIndexes": +[ 240 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 240 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze385", "outputIndexes": +[ 241 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 236, 239, 241, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice388", "outputIndexes": +[ 242 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 242 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze389", "outputIndexes": +[ 243 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 243, 243 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp391", "outputIndexes": +[ 244 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 235, 244, 238 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.2/self_attn/Gather_output_0", "outputIndexes": +[ 245 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 245, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 246 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 238 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze399", "outputIndexes": +[ 247 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 240 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze402", "outputIndexes": +[ 248 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 236, 247, 248, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice405", "outputIndexes": +[ 249 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 249 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze406", "outputIndexes": +[ 250 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 250 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp407", "outputIndexes": +[ 251 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 251, 250 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp408", "outputIndexes": +[ 252 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 235, 252, 238 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.2/self_attn/Gather_1_output_0", "outputIndexes": +[ 253 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 253, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 254 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 246, 254, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.2/self_attn/Concat_output_0", "outputIndexes": +[ 255 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 234, 255 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.2/self_attn/Reshape_output_0", "outputIndexes": +[ 256 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 256 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 251826506, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.2/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 258 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 258, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Mul_output_0", "outputIndexes": +[ 259 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 258 ] +, "main_type": "NONE", "name": "/blocks.2/self_attn/Shape_2_output_0", "outputIndexes": +[ 260 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 260 ] +, "main_type": "NONE", "name": "Shape423", "outputIndexes": +[ 261 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 260 ] +, "main_type": "NONE", "name": "Rank425", "outputIndexes": +[ 262 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 262, 262 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp427", "outputIndexes": +[ 263 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 263 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze428", "outputIndexes": +[ 264 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 263, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp430", "outputIndexes": +[ 265 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 265 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze431", "outputIndexes": +[ 266 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 261, 264, 266, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice434", "outputIndexes": +[ 267 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 267 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze435", "outputIndexes": +[ 268 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 268 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp436", "outputIndexes": +[ 269 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 269, 268 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp437", "outputIndexes": +[ 270 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 260, 270, 263 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.2/self_attn/Gather_2_output_0", "outputIndexes": +[ 271 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 271, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Div_output_0", "outputIndexes": +[ 272 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 272, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 273 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 258, 273, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.2/self_attn/Slice_1_output_0", "outputIndexes": +[ 274 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 274 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.2/self_attn/Neg_output_0", "outputIndexes": +[ 275 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 272, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 276 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 258, 20, 276, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.2/self_attn/Slice_output_0", "outputIndexes": +[ 277 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 275, 277 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.2/self_attn/Concat_3_output_0", "outputIndexes": +[ 278 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 278, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Mul_1_output_0", "outputIndexes": +[ 279 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 259, 279 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Add_output_0", "outputIndexes": +[ 280 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3454 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 251827530, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.2/self_attn/k_proj/Linear", "outputIndexes": +[ 3459 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3459 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3460 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3460 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 281 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 245, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 282 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 253, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 283 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 282, 283, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.2/self_attn/Concat_1_output_0", "outputIndexes": +[ 284 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 281, 284 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.2/self_attn/Reshape_1_output_0", "outputIndexes": +[ 285 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 285 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 254448992, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.2/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 287 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 287, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Mul_2_output_0", "outputIndexes": +[ 288 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 287 ] +, "main_type": "NONE", "name": "/blocks.2/self_attn/Shape_3_output_0", "outputIndexes": +[ 289 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 289 ] +, "main_type": "NONE", "name": "Shape465", "outputIndexes": +[ 290 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 289 ] +, "main_type": "NONE", "name": "Rank467", "outputIndexes": +[ 291 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 291, 291 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp469", "outputIndexes": +[ 292 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 292 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze470", "outputIndexes": +[ 293 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 292, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp472", "outputIndexes": +[ 294 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 294 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze473", "outputIndexes": +[ 295 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 290, 293, 295, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice476", "outputIndexes": +[ 296 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 296 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze477", "outputIndexes": +[ 297 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 297 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp478", "outputIndexes": +[ 298 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 298, 297 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp479", "outputIndexes": +[ 299 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 289, 299, 292 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.2/self_attn/Gather_3_output_0", "outputIndexes": +[ 300 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 300, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Div_1_output_0", "outputIndexes": +[ 301 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 301, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 302 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 287, 302, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.2/self_attn/Slice_3_output_0", "outputIndexes": +[ 303 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 303 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.2/self_attn/Neg_1_output_0", "outputIndexes": +[ 304 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 301, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 305 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 287, 20, 305, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.2/self_attn/Slice_2_output_0", "outputIndexes": +[ 306 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 304, 306 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.2/self_attn/Concat_4_output_0", "outputIndexes": +[ 307 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 307, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Mul_3_output_0", "outputIndexes": +[ 308 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 288, 308 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/self_attn/Add_1_output_0", "outputIndexes": +[ 309 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3454 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 254450016, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.2/self_attn/v_proj/Linear", "outputIndexes": +[ 3463 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3463 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3464 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3464 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 310 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 245, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 311 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 253, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.2/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 312 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 311, 312, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.2/self_attn/Concat_2_output_0", "outputIndexes": +[ 313 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 310, 313 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.2/self_attn/Reshape_2_output_0", "outputIndexes": +[ 314 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 280, 309, 314, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.2/self_attn/FusedAttention", "outputIndexes": +[ 315 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 315 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3465 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3465 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3466 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3466 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 257071478, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.2/self_attn/o_proj/Linear", "outputIndexes": +[ 3467 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3467 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3468 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3468 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 316 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 232, 316 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/Add_output_0", "outputIndexes": +[ 317 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 317 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 267557260, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.2/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 318 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 318 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3469 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3469 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3470 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3470 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 267590028, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.2/mlp/gate_proj/Linear", "outputIndexes": +[ 3471 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3471 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3472 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3472 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 319 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 319 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.2/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 320 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3470 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 299047330, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.2/mlp/up_proj/Linear", "outputIndexes": +[ 3475 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3475 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3476 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3476 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 321 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 320, 321 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/mlp/Mul_output_0", "outputIndexes": +[ 322 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 322 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3477 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3477 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3478 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3478 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 330504632, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.2/mlp/down_proj/Linear", "outputIndexes": +[ 3479 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3479 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3480 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3480 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 323 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 317, 323 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.2/Add_1_output_0", "outputIndexes": +[ 324 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 324, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.3/Reshape_output_0", "outputIndexes": +[ 325 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 325 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 361961934, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.3/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 326 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 326 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3481 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3481 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3482 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3482 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 361994702, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.3/self_attn/q_proj/Linear", "outputIndexes": +[ 3483 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3483 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3484 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3484 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 327 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 326 ] +, "main_type": "NONE", "name": "/blocks.3/self_attn/Shape_output_0", "outputIndexes": +[ 328 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 328 ] +, "main_type": "NONE", "name": "Shape526", "outputIndexes": +[ 329 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 328 ] +, "main_type": "NONE", "name": "Rank528", "outputIndexes": +[ 330 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 330, 330 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp530", "outputIndexes": +[ 331 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 331 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze531", "outputIndexes": +[ 332 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 331, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp533", "outputIndexes": +[ 333 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 333 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze534", "outputIndexes": +[ 334 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 329, 332, 334, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice537", "outputIndexes": +[ 335 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 335 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze538", "outputIndexes": +[ 336 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 336, 336 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp540", "outputIndexes": +[ 337 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 328, 337, 331 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.3/self_attn/Gather_output_0", "outputIndexes": +[ 338 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 338, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 339 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 331 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze548", "outputIndexes": +[ 340 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 333 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze551", "outputIndexes": +[ 341 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 329, 340, 341, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice554", "outputIndexes": +[ 342 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 342 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze555", "outputIndexes": +[ 343 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 343 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp556", "outputIndexes": +[ 344 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 344, 343 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp557", "outputIndexes": +[ 345 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 328, 345, 331 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.3/self_attn/Gather_1_output_0", "outputIndexes": +[ 346 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 346, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 347 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 339, 347, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.3/self_attn/Concat_output_0", "outputIndexes": +[ 348 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 327, 348 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.3/self_attn/Reshape_output_0", "outputIndexes": +[ 349 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 349 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 372480484, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.3/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 351 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 351, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Mul_output_0", "outputIndexes": +[ 352 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 351 ] +, "main_type": "NONE", "name": "/blocks.3/self_attn/Shape_2_output_0", "outputIndexes": +[ 353 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 353 ] +, "main_type": "NONE", "name": "Shape572", "outputIndexes": +[ 354 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 353 ] +, "main_type": "NONE", "name": "Rank574", "outputIndexes": +[ 355 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 355, 355 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp576", "outputIndexes": +[ 356 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 356 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze577", "outputIndexes": +[ 357 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 356, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp579", "outputIndexes": +[ 358 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 358 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze580", "outputIndexes": +[ 359 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 354, 357, 359, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice583", "outputIndexes": +[ 360 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 360 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze584", "outputIndexes": +[ 361 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 361 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp585", "outputIndexes": +[ 362 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 362, 361 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp586", "outputIndexes": +[ 363 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 353, 363, 356 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.3/self_attn/Gather_2_output_0", "outputIndexes": +[ 364 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 364, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Div_output_0", "outputIndexes": +[ 365 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 365, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 366 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 351, 366, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.3/self_attn/Slice_1_output_0", "outputIndexes": +[ 367 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 367 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.3/self_attn/Neg_output_0", "outputIndexes": +[ 368 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 365, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 369 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 351, 20, 369, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.3/self_attn/Slice_output_0", "outputIndexes": +[ 370 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 368, 370 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.3/self_attn/Concat_3_output_0", "outputIndexes": +[ 371 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 371, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Mul_1_output_0", "outputIndexes": +[ 372 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 352, 372 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Add_output_0", "outputIndexes": +[ 373 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3482 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 372481508, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.3/self_attn/k_proj/Linear", "outputIndexes": +[ 3487 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3487 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3488 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3488 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 374 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 338, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 375 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 346, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 376 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 375, 376, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.3/self_attn/Concat_1_output_0", "outputIndexes": +[ 377 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 374, 377 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.3/self_attn/Reshape_1_output_0", "outputIndexes": +[ 378 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 378 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 375102970, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.3/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 380 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 380, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Mul_2_output_0", "outputIndexes": +[ 381 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 380 ] +, "main_type": "NONE", "name": "/blocks.3/self_attn/Shape_3_output_0", "outputIndexes": +[ 382 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 382 ] +, "main_type": "NONE", "name": "Shape614", "outputIndexes": +[ 383 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 382 ] +, "main_type": "NONE", "name": "Rank616", "outputIndexes": +[ 384 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 384, 384 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp618", "outputIndexes": +[ 385 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 385 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze619", "outputIndexes": +[ 386 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 385, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp621", "outputIndexes": +[ 387 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 387 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze622", "outputIndexes": +[ 388 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 383, 386, 388, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice625", "outputIndexes": +[ 389 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 389 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze626", "outputIndexes": +[ 390 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 390 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp627", "outputIndexes": +[ 391 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 391, 390 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp628", "outputIndexes": +[ 392 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 382, 392, 385 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.3/self_attn/Gather_3_output_0", "outputIndexes": +[ 393 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 393, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Div_1_output_0", "outputIndexes": +[ 394 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 394, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 395 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 380, 395, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.3/self_attn/Slice_3_output_0", "outputIndexes": +[ 396 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 396 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.3/self_attn/Neg_1_output_0", "outputIndexes": +[ 397 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 394, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 398 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 380, 20, 398, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.3/self_attn/Slice_2_output_0", "outputIndexes": +[ 399 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 397, 399 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.3/self_attn/Concat_4_output_0", "outputIndexes": +[ 400 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 400, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Mul_3_output_0", "outputIndexes": +[ 401 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 381, 401 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/self_attn/Add_1_output_0", "outputIndexes": +[ 402 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3482 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 375103994, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.3/self_attn/v_proj/Linear", "outputIndexes": +[ 3491 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3491 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3492 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3492 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 403 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 338, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 404 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 346, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.3/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 405 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 404, 405, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.3/self_attn/Concat_2_output_0", "outputIndexes": +[ 406 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 403, 406 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.3/self_attn/Reshape_2_output_0", "outputIndexes": +[ 407 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 373, 402, 407, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.3/self_attn/FusedAttention", "outputIndexes": +[ 408 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 408 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3493 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3493 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3494 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3494 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 377725456, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.3/self_attn/o_proj/Linear", "outputIndexes": +[ 3495 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3495 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3496 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3496 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 409 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 325, 409 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/Add_output_0", "outputIndexes": +[ 410 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 410 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 388211238, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.3/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 411 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 411 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3497 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3497 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3498 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3498 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 388244006, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.3/mlp/gate_proj/Linear", "outputIndexes": +[ 3499 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3499 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3500 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3500 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 412 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 412 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.3/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 413 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3498 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 419701308, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.3/mlp/up_proj/Linear", "outputIndexes": +[ 3503 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3503 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3504 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3504 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 414 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 413, 414 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/mlp/Mul_output_0", "outputIndexes": +[ 415 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 415 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3505 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3505 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3506 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3506 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 451158610, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.3/mlp/down_proj/Linear", "outputIndexes": +[ 3507 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3507 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3508 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3508 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 416 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 410, 416 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.3/Add_1_output_0", "outputIndexes": +[ 417 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 417, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.4/Reshape_output_0", "outputIndexes": +[ 418 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 418 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 482615912, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.4/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 419 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 419 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3509 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3509 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3510 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3510 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 482648680, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.4/self_attn/q_proj/Linear", "outputIndexes": +[ 3511 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3511 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3512 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3512 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 420 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 419 ] +, "main_type": "NONE", "name": "/blocks.4/self_attn/Shape_output_0", "outputIndexes": +[ 421 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 421 ] +, "main_type": "NONE", "name": "Shape675", "outputIndexes": +[ 422 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 421 ] +, "main_type": "NONE", "name": "Rank677", "outputIndexes": +[ 423 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 423, 423 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp679", "outputIndexes": +[ 424 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 424 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze680", "outputIndexes": +[ 425 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 424, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp682", "outputIndexes": +[ 426 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 426 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze683", "outputIndexes": +[ 427 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 422, 425, 427, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice686", "outputIndexes": +[ 428 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 428 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze687", "outputIndexes": +[ 429 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 429, 429 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp689", "outputIndexes": +[ 430 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 421, 430, 424 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.4/self_attn/Gather_output_0", "outputIndexes": +[ 431 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 431, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 432 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 424 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze697", "outputIndexes": +[ 433 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 426 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze700", "outputIndexes": +[ 434 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 422, 433, 434, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice703", "outputIndexes": +[ 435 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 435 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze704", "outputIndexes": +[ 436 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 436 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp705", "outputIndexes": +[ 437 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 437, 436 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp706", "outputIndexes": +[ 438 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 421, 438, 424 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.4/self_attn/Gather_1_output_0", "outputIndexes": +[ 439 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 439, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 440 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 432, 440, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.4/self_attn/Concat_output_0", "outputIndexes": +[ 441 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 420, 441 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.4/self_attn/Reshape_output_0", "outputIndexes": +[ 442 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 442 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 493134462, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.4/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 444 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 444, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Mul_output_0", "outputIndexes": +[ 445 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 444 ] +, "main_type": "NONE", "name": "/blocks.4/self_attn/Shape_2_output_0", "outputIndexes": +[ 446 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 446 ] +, "main_type": "NONE", "name": "Shape721", "outputIndexes": +[ 447 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 446 ] +, "main_type": "NONE", "name": "Rank723", "outputIndexes": +[ 448 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 448, 448 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp725", "outputIndexes": +[ 449 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 449 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze726", "outputIndexes": +[ 450 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 449, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp728", "outputIndexes": +[ 451 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 451 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze729", "outputIndexes": +[ 452 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 447, 450, 452, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice732", "outputIndexes": +[ 453 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 453 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze733", "outputIndexes": +[ 454 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 454 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp734", "outputIndexes": +[ 455 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 455, 454 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp735", "outputIndexes": +[ 456 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 446, 456, 449 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.4/self_attn/Gather_2_output_0", "outputIndexes": +[ 457 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 457, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Div_output_0", "outputIndexes": +[ 458 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 458, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 459 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 444, 459, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.4/self_attn/Slice_1_output_0", "outputIndexes": +[ 460 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 460 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.4/self_attn/Neg_output_0", "outputIndexes": +[ 461 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 458, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 462 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 444, 20, 462, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.4/self_attn/Slice_output_0", "outputIndexes": +[ 463 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 461, 463 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.4/self_attn/Concat_3_output_0", "outputIndexes": +[ 464 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 464, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Mul_1_output_0", "outputIndexes": +[ 465 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 445, 465 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Add_output_0", "outputIndexes": +[ 466 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3510 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 493135486, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.4/self_attn/k_proj/Linear", "outputIndexes": +[ 3515 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3515 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3516 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3516 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 467 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 431, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 468 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 439, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 469 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 468, 469, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.4/self_attn/Concat_1_output_0", "outputIndexes": +[ 470 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 467, 470 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.4/self_attn/Reshape_1_output_0", "outputIndexes": +[ 471 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 471 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 495756948, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.4/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 473 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 473, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Mul_2_output_0", "outputIndexes": +[ 474 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 473 ] +, "main_type": "NONE", "name": "/blocks.4/self_attn/Shape_3_output_0", "outputIndexes": +[ 475 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 475 ] +, "main_type": "NONE", "name": "Shape763", "outputIndexes": +[ 476 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 475 ] +, "main_type": "NONE", "name": "Rank765", "outputIndexes": +[ 477 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 477, 477 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp767", "outputIndexes": +[ 478 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 478 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze768", "outputIndexes": +[ 479 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 478, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp770", "outputIndexes": +[ 480 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 480 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze771", "outputIndexes": +[ 481 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 476, 479, 481, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice774", "outputIndexes": +[ 482 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 482 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze775", "outputIndexes": +[ 483 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 483 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp776", "outputIndexes": +[ 484 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 484, 483 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp777", "outputIndexes": +[ 485 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 475, 485, 478 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.4/self_attn/Gather_3_output_0", "outputIndexes": +[ 486 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 486, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Div_1_output_0", "outputIndexes": +[ 487 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 487, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 488 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 473, 488, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.4/self_attn/Slice_3_output_0", "outputIndexes": +[ 489 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 489 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.4/self_attn/Neg_1_output_0", "outputIndexes": +[ 490 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 487, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 491 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 473, 20, 491, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.4/self_attn/Slice_2_output_0", "outputIndexes": +[ 492 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 490, 492 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.4/self_attn/Concat_4_output_0", "outputIndexes": +[ 493 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 493, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Mul_3_output_0", "outputIndexes": +[ 494 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 474, 494 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/self_attn/Add_1_output_0", "outputIndexes": +[ 495 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3510 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 495757972, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.4/self_attn/v_proj/Linear", "outputIndexes": +[ 3519 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3519 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3520 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3520 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 496 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 431, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 497 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 439, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.4/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 498 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 497, 498, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.4/self_attn/Concat_2_output_0", "outputIndexes": +[ 499 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 496, 499 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.4/self_attn/Reshape_2_output_0", "outputIndexes": +[ 500 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 466, 495, 500, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.4/self_attn/FusedAttention", "outputIndexes": +[ 501 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 501 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3521 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3521 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3522 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3522 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 498379434, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.4/self_attn/o_proj/Linear", "outputIndexes": +[ 3523 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3523 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3524 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3524 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 502 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 418, 502 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/Add_output_0", "outputIndexes": +[ 503 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 503 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 508865216, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.4/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 504 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 504 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3525 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3525 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3526 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3526 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 508897984, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.4/mlp/gate_proj/Linear", "outputIndexes": +[ 3527 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3527 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3528 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3528 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 505 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 505 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.4/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 506 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3526 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 540355286, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.4/mlp/up_proj/Linear", "outputIndexes": +[ 3531 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3531 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3532 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3532 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 507 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 506, 507 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/mlp/Mul_output_0", "outputIndexes": +[ 508 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 508 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3533 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3533 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3534 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3534 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 571812588, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.4/mlp/down_proj/Linear", "outputIndexes": +[ 3535 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3535 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3536 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3536 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 509 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 503, 509 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.4/Add_1_output_0", "outputIndexes": +[ 510 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 510, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.5/Reshape_output_0", "outputIndexes": +[ 511 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 511 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 603269890, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.5/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 512 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 512 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3537 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3537 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3538 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3538 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 603302658, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.5/self_attn/q_proj/Linear", "outputIndexes": +[ 3539 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3539 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3540 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3540 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 513 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 512 ] +, "main_type": "NONE", "name": "/blocks.5/self_attn/Shape_output_0", "outputIndexes": +[ 514 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 514 ] +, "main_type": "NONE", "name": "Shape824", "outputIndexes": +[ 515 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 514 ] +, "main_type": "NONE", "name": "Rank826", "outputIndexes": +[ 516 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 516, 516 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp828", "outputIndexes": +[ 517 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 517 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze829", "outputIndexes": +[ 518 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 517, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp831", "outputIndexes": +[ 519 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 519 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze832", "outputIndexes": +[ 520 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 515, 518, 520, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice835", "outputIndexes": +[ 521 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 521 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze836", "outputIndexes": +[ 522 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 522, 522 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp838", "outputIndexes": +[ 523 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 514, 523, 517 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.5/self_attn/Gather_output_0", "outputIndexes": +[ 524 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 524, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 525 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 517 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze846", "outputIndexes": +[ 526 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 519 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze849", "outputIndexes": +[ 527 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 515, 526, 527, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice852", "outputIndexes": +[ 528 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 528 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze853", "outputIndexes": +[ 529 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 529 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp854", "outputIndexes": +[ 530 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 530, 529 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp855", "outputIndexes": +[ 531 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 514, 531, 517 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.5/self_attn/Gather_1_output_0", "outputIndexes": +[ 532 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 532, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 533 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 525, 533, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.5/self_attn/Concat_output_0", "outputIndexes": +[ 534 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 513, 534 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.5/self_attn/Reshape_output_0", "outputIndexes": +[ 535 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 535 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 613788440, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.5/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 537 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 537, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Mul_output_0", "outputIndexes": +[ 538 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 537 ] +, "main_type": "NONE", "name": "/blocks.5/self_attn/Shape_2_output_0", "outputIndexes": +[ 539 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 539 ] +, "main_type": "NONE", "name": "Shape870", "outputIndexes": +[ 540 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 539 ] +, "main_type": "NONE", "name": "Rank872", "outputIndexes": +[ 541 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 541, 541 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp874", "outputIndexes": +[ 542 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 542 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze875", "outputIndexes": +[ 543 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 542, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp877", "outputIndexes": +[ 544 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 544 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze878", "outputIndexes": +[ 545 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 540, 543, 545, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice881", "outputIndexes": +[ 546 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 546 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze882", "outputIndexes": +[ 547 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 547 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp883", "outputIndexes": +[ 548 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 548, 547 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp884", "outputIndexes": +[ 549 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 539, 549, 542 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.5/self_attn/Gather_2_output_0", "outputIndexes": +[ 550 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 550, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Div_output_0", "outputIndexes": +[ 551 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 551, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 552 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 537, 552, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.5/self_attn/Slice_1_output_0", "outputIndexes": +[ 553 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 553 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.5/self_attn/Neg_output_0", "outputIndexes": +[ 554 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 551, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 555 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 537, 20, 555, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.5/self_attn/Slice_output_0", "outputIndexes": +[ 556 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 554, 556 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.5/self_attn/Concat_3_output_0", "outputIndexes": +[ 557 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 557, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Mul_1_output_0", "outputIndexes": +[ 558 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 538, 558 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Add_output_0", "outputIndexes": +[ 559 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3538 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 613789464, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.5/self_attn/k_proj/Linear", "outputIndexes": +[ 3543 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3543 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3544 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3544 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 560 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 524, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 561 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 532, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 562 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 561, 562, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.5/self_attn/Concat_1_output_0", "outputIndexes": +[ 563 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 560, 563 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.5/self_attn/Reshape_1_output_0", "outputIndexes": +[ 564 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 564 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 616410926, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.5/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 566 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 566, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Mul_2_output_0", "outputIndexes": +[ 567 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 566 ] +, "main_type": "NONE", "name": "/blocks.5/self_attn/Shape_3_output_0", "outputIndexes": +[ 568 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 568 ] +, "main_type": "NONE", "name": "Shape912", "outputIndexes": +[ 569 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 568 ] +, "main_type": "NONE", "name": "Rank914", "outputIndexes": +[ 570 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 570, 570 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp916", "outputIndexes": +[ 571 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 571 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze917", "outputIndexes": +[ 572 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 571, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp919", "outputIndexes": +[ 573 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 573 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze920", "outputIndexes": +[ 574 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 569, 572, 574, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice923", "outputIndexes": +[ 575 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 575 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze924", "outputIndexes": +[ 576 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 576 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp925", "outputIndexes": +[ 577 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 577, 576 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp926", "outputIndexes": +[ 578 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 568, 578, 571 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.5/self_attn/Gather_3_output_0", "outputIndexes": +[ 579 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 579, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Div_1_output_0", "outputIndexes": +[ 580 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 580, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 581 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 566, 581, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.5/self_attn/Slice_3_output_0", "outputIndexes": +[ 582 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 582 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.5/self_attn/Neg_1_output_0", "outputIndexes": +[ 583 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 580, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 584 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 566, 20, 584, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.5/self_attn/Slice_2_output_0", "outputIndexes": +[ 585 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 583, 585 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.5/self_attn/Concat_4_output_0", "outputIndexes": +[ 586 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 586, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Mul_3_output_0", "outputIndexes": +[ 587 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 567, 587 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/self_attn/Add_1_output_0", "outputIndexes": +[ 588 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3538 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 616411950, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.5/self_attn/v_proj/Linear", "outputIndexes": +[ 3547 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3547 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3548 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3548 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 589 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 524, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 590 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 532, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.5/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 591 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 590, 591, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.5/self_attn/Concat_2_output_0", "outputIndexes": +[ 592 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 589, 592 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.5/self_attn/Reshape_2_output_0", "outputIndexes": +[ 593 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 559, 588, 593, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.5/self_attn/FusedAttention", "outputIndexes": +[ 594 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 594 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3549 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3549 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3550 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3550 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 619033412, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.5/self_attn/o_proj/Linear", "outputIndexes": +[ 3551 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3551 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3552 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3552 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 595 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 511, 595 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/Add_output_0", "outputIndexes": +[ 596 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 596 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 629519194, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.5/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 597 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 597 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3553 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3553 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3554 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3554 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 629551962, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.5/mlp/gate_proj/Linear", "outputIndexes": +[ 3555 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3555 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3556 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3556 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 598 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 598 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.5/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 599 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3554 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 661009264, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.5/mlp/up_proj/Linear", "outputIndexes": +[ 3559 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3559 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3560 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3560 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 600 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 599, 600 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/mlp/Mul_output_0", "outputIndexes": +[ 601 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 601 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3561 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3561 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3562 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3562 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 692466566, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.5/mlp/down_proj/Linear", "outputIndexes": +[ 3563 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3563 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3564 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3564 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 602 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 596, 602 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.5/Add_1_output_0", "outputIndexes": +[ 603 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 603, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.6/Reshape_output_0", "outputIndexes": +[ 604 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 604 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 723923868, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.6/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 605 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 605 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3565 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3565 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3566 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3566 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 723956636, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.6/self_attn/q_proj/Linear", "outputIndexes": +[ 3567 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3567 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3568 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3568 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 606 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 605 ] +, "main_type": "NONE", "name": "/blocks.6/self_attn/Shape_output_0", "outputIndexes": +[ 607 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 607 ] +, "main_type": "NONE", "name": "Shape973", "outputIndexes": +[ 608 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 607 ] +, "main_type": "NONE", "name": "Rank975", "outputIndexes": +[ 609 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 609, 609 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp977", "outputIndexes": +[ 610 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 610 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze978", "outputIndexes": +[ 611 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 610, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp980", "outputIndexes": +[ 612 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 612 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze981", "outputIndexes": +[ 613 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 608, 611, 613, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice984", "outputIndexes": +[ 614 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 614 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze985", "outputIndexes": +[ 615 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 615, 615 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp987", "outputIndexes": +[ 616 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 607, 616, 610 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.6/self_attn/Gather_output_0", "outputIndexes": +[ 617 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 617, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 618 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 610 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze995", "outputIndexes": +[ 619 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 612 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze998", "outputIndexes": +[ 620 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 608, 619, 620, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1001", "outputIndexes": +[ 621 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 621 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1002", "outputIndexes": +[ 622 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 622 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1003", "outputIndexes": +[ 623 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 623, 622 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1004", "outputIndexes": +[ 624 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 607, 624, 610 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.6/self_attn/Gather_1_output_0", "outputIndexes": +[ 625 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 625, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 626 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 618, 626, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.6/self_attn/Concat_output_0", "outputIndexes": +[ 627 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 606, 627 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.6/self_attn/Reshape_output_0", "outputIndexes": +[ 628 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 628 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 734442418, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.6/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 630 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 630, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Mul_output_0", "outputIndexes": +[ 631 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 630 ] +, "main_type": "NONE", "name": "/blocks.6/self_attn/Shape_2_output_0", "outputIndexes": +[ 632 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 632 ] +, "main_type": "NONE", "name": "Shape1019", "outputIndexes": +[ 633 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 632 ] +, "main_type": "NONE", "name": "Rank1021", "outputIndexes": +[ 634 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 634, 634 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1023", "outputIndexes": +[ 635 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 635 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1024", "outputIndexes": +[ 636 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 635, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1026", "outputIndexes": +[ 637 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 637 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1027", "outputIndexes": +[ 638 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 633, 636, 638, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1030", "outputIndexes": +[ 639 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 639 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1031", "outputIndexes": +[ 640 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 640 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1032", "outputIndexes": +[ 641 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 641, 640 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1033", "outputIndexes": +[ 642 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 632, 642, 635 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.6/self_attn/Gather_2_output_0", "outputIndexes": +[ 643 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 643, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Div_output_0", "outputIndexes": +[ 644 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 644, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 645 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 630, 645, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.6/self_attn/Slice_1_output_0", "outputIndexes": +[ 646 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 646 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.6/self_attn/Neg_output_0", "outputIndexes": +[ 647 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 644, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 648 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 630, 20, 648, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.6/self_attn/Slice_output_0", "outputIndexes": +[ 649 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 647, 649 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.6/self_attn/Concat_3_output_0", "outputIndexes": +[ 650 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 650, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Mul_1_output_0", "outputIndexes": +[ 651 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 631, 651 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Add_output_0", "outputIndexes": +[ 652 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3566 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 734443442, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.6/self_attn/k_proj/Linear", "outputIndexes": +[ 3571 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3571 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3572 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3572 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 653 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 617, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 654 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 625, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 655 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 654, 655, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.6/self_attn/Concat_1_output_0", "outputIndexes": +[ 656 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 653, 656 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.6/self_attn/Reshape_1_output_0", "outputIndexes": +[ 657 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 657 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 737064904, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.6/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 659 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 659, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Mul_2_output_0", "outputIndexes": +[ 660 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 659 ] +, "main_type": "NONE", "name": "/blocks.6/self_attn/Shape_3_output_0", "outputIndexes": +[ 661 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 661 ] +, "main_type": "NONE", "name": "Shape1061", "outputIndexes": +[ 662 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 661 ] +, "main_type": "NONE", "name": "Rank1063", "outputIndexes": +[ 663 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 663, 663 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1065", "outputIndexes": +[ 664 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 664 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1066", "outputIndexes": +[ 665 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 664, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1068", "outputIndexes": +[ 666 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 666 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1069", "outputIndexes": +[ 667 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 662, 665, 667, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1072", "outputIndexes": +[ 668 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 668 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1073", "outputIndexes": +[ 669 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 669 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1074", "outputIndexes": +[ 670 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 670, 669 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1075", "outputIndexes": +[ 671 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 661, 671, 664 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.6/self_attn/Gather_3_output_0", "outputIndexes": +[ 672 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 672, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Div_1_output_0", "outputIndexes": +[ 673 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 673, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 674 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 659, 674, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.6/self_attn/Slice_3_output_0", "outputIndexes": +[ 675 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 675 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.6/self_attn/Neg_1_output_0", "outputIndexes": +[ 676 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 673, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 677 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 659, 20, 677, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.6/self_attn/Slice_2_output_0", "outputIndexes": +[ 678 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 676, 678 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.6/self_attn/Concat_4_output_0", "outputIndexes": +[ 679 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 679, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Mul_3_output_0", "outputIndexes": +[ 680 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 660, 680 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/self_attn/Add_1_output_0", "outputIndexes": +[ 681 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3566 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 737065928, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.6/self_attn/v_proj/Linear", "outputIndexes": +[ 3575 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3575 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3576 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3576 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 682 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 617, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 683 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 625, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.6/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 684 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 683, 684, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.6/self_attn/Concat_2_output_0", "outputIndexes": +[ 685 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 682, 685 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.6/self_attn/Reshape_2_output_0", "outputIndexes": +[ 686 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 652, 681, 686, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.6/self_attn/FusedAttention", "outputIndexes": +[ 687 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 687 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3577 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3577 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3578 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3578 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 739687390, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.6/self_attn/o_proj/Linear", "outputIndexes": +[ 3579 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3579 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3580 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3580 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 688 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 604, 688 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/Add_output_0", "outputIndexes": +[ 689 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 689 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 750173172, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.6/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 690 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 690 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3581 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3581 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3582 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3582 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 750205940, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.6/mlp/gate_proj/Linear", "outputIndexes": +[ 3583 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3583 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3584 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3584 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 691 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 691 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.6/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 692 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3582 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 781663242, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.6/mlp/up_proj/Linear", "outputIndexes": +[ 3587 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3587 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3588 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3588 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 693 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 692, 693 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/mlp/Mul_output_0", "outputIndexes": +[ 694 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 694 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3589 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3589 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3590 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3590 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 813120544, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.6/mlp/down_proj/Linear", "outputIndexes": +[ 3591 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3591 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3592 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3592 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 695 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 689, 695 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.6/Add_1_output_0", "outputIndexes": +[ 696 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 696, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.7/Reshape_output_0", "outputIndexes": +[ 697 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 697 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 844577846, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.7/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 698 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 698 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3593 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3593 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3594 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3594 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 844610614, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.7/self_attn/q_proj/Linear", "outputIndexes": +[ 3595 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3595 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3596 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3596 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 699 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 698 ] +, "main_type": "NONE", "name": "/blocks.7/self_attn/Shape_output_0", "outputIndexes": +[ 700 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 700 ] +, "main_type": "NONE", "name": "Shape1122", "outputIndexes": +[ 701 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 700 ] +, "main_type": "NONE", "name": "Rank1124", "outputIndexes": +[ 702 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 702, 702 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1126", "outputIndexes": +[ 703 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 703 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1127", "outputIndexes": +[ 704 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 703, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1129", "outputIndexes": +[ 705 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 705 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1130", "outputIndexes": +[ 706 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 701, 704, 706, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1133", "outputIndexes": +[ 707 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 707 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1134", "outputIndexes": +[ 708 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 708, 708 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1136", "outputIndexes": +[ 709 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 700, 709, 703 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.7/self_attn/Gather_output_0", "outputIndexes": +[ 710 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 710, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 711 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 703 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1144", "outputIndexes": +[ 712 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 705 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1147", "outputIndexes": +[ 713 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 701, 712, 713, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1150", "outputIndexes": +[ 714 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 714 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1151", "outputIndexes": +[ 715 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 715 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1152", "outputIndexes": +[ 716 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 716, 715 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1153", "outputIndexes": +[ 717 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 700, 717, 703 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.7/self_attn/Gather_1_output_0", "outputIndexes": +[ 718 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 718, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 719 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 711, 719, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.7/self_attn/Concat_output_0", "outputIndexes": +[ 720 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 699, 720 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.7/self_attn/Reshape_output_0", "outputIndexes": +[ 721 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 721 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 855096396, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.7/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 723 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 723, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Mul_output_0", "outputIndexes": +[ 724 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 723 ] +, "main_type": "NONE", "name": "/blocks.7/self_attn/Shape_2_output_0", "outputIndexes": +[ 725 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 725 ] +, "main_type": "NONE", "name": "Shape1168", "outputIndexes": +[ 726 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 725 ] +, "main_type": "NONE", "name": "Rank1170", "outputIndexes": +[ 727 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 727, 727 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1172", "outputIndexes": +[ 728 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 728 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1173", "outputIndexes": +[ 729 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 728, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1175", "outputIndexes": +[ 730 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 730 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1176", "outputIndexes": +[ 731 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 726, 729, 731, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1179", "outputIndexes": +[ 732 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 732 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1180", "outputIndexes": +[ 733 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 733 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1181", "outputIndexes": +[ 734 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 734, 733 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1182", "outputIndexes": +[ 735 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 725, 735, 728 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.7/self_attn/Gather_2_output_0", "outputIndexes": +[ 736 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 736, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Div_output_0", "outputIndexes": +[ 737 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 737, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 738 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 723, 738, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.7/self_attn/Slice_1_output_0", "outputIndexes": +[ 739 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 739 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.7/self_attn/Neg_output_0", "outputIndexes": +[ 740 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 737, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 741 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 723, 20, 741, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.7/self_attn/Slice_output_0", "outputIndexes": +[ 742 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 740, 742 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.7/self_attn/Concat_3_output_0", "outputIndexes": +[ 743 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 743, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Mul_1_output_0", "outputIndexes": +[ 744 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 724, 744 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Add_output_0", "outputIndexes": +[ 745 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3594 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 855097420, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.7/self_attn/k_proj/Linear", "outputIndexes": +[ 3599 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3599 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3600 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3600 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 746 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 710, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 747 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 718, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 748 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 747, 748, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.7/self_attn/Concat_1_output_0", "outputIndexes": +[ 749 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 746, 749 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.7/self_attn/Reshape_1_output_0", "outputIndexes": +[ 750 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 750 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 857718882, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.7/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 752 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 752, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Mul_2_output_0", "outputIndexes": +[ 753 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 752 ] +, "main_type": "NONE", "name": "/blocks.7/self_attn/Shape_3_output_0", "outputIndexes": +[ 754 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 754 ] +, "main_type": "NONE", "name": "Shape1210", "outputIndexes": +[ 755 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 754 ] +, "main_type": "NONE", "name": "Rank1212", "outputIndexes": +[ 756 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 756, 756 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1214", "outputIndexes": +[ 757 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 757 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1215", "outputIndexes": +[ 758 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 757, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1217", "outputIndexes": +[ 759 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 759 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1218", "outputIndexes": +[ 760 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 755, 758, 760, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1221", "outputIndexes": +[ 761 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 761 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1222", "outputIndexes": +[ 762 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 762 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1223", "outputIndexes": +[ 763 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 763, 762 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1224", "outputIndexes": +[ 764 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 754, 764, 757 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.7/self_attn/Gather_3_output_0", "outputIndexes": +[ 765 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 765, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Div_1_output_0", "outputIndexes": +[ 766 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 766, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 767 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 752, 767, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.7/self_attn/Slice_3_output_0", "outputIndexes": +[ 768 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 768 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.7/self_attn/Neg_1_output_0", "outputIndexes": +[ 769 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 766, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 770 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 752, 20, 770, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.7/self_attn/Slice_2_output_0", "outputIndexes": +[ 771 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 769, 771 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.7/self_attn/Concat_4_output_0", "outputIndexes": +[ 772 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 772, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Mul_3_output_0", "outputIndexes": +[ 773 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 753, 773 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/self_attn/Add_1_output_0", "outputIndexes": +[ 774 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3594 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 857719906, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.7/self_attn/v_proj/Linear", "outputIndexes": +[ 3603 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3603 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3604 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3604 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 775 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 710, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 776 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 718, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.7/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 777 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 776, 777, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.7/self_attn/Concat_2_output_0", "outputIndexes": +[ 778 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 775, 778 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.7/self_attn/Reshape_2_output_0", "outputIndexes": +[ 779 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 745, 774, 779, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.7/self_attn/FusedAttention", "outputIndexes": +[ 780 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 780 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3605 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3605 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3606 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3606 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 860341368, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.7/self_attn/o_proj/Linear", "outputIndexes": +[ 3607 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3607 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3608 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3608 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 781 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 697, 781 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/Add_output_0", "outputIndexes": +[ 782 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 782 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 870827150, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.7/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 783 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 783 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3609 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3609 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3610 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3610 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 870859918, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.7/mlp/gate_proj/Linear", "outputIndexes": +[ 3611 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3611 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3612 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3612 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 784 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 784 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.7/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 785 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3610 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 902317220, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.7/mlp/up_proj/Linear", "outputIndexes": +[ 3615 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3615 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3616 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3616 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 786 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 785, 786 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/mlp/Mul_output_0", "outputIndexes": +[ 787 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 787 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3617 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3617 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3618 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3618 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 933774522, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.7/mlp/down_proj/Linear", "outputIndexes": +[ 3619 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3619 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3620 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3620 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 788 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 782, 788 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.7/Add_1_output_0", "outputIndexes": +[ 789 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 789, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.8/Reshape_output_0", "outputIndexes": +[ 790 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 790 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 965231824, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.8/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 791 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 791 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3621 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3621 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3622 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3622 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 965264592, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.8/self_attn/q_proj/Linear", "outputIndexes": +[ 3623 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3623 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3624 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3624 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 792 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 791 ] +, "main_type": "NONE", "name": "/blocks.8/self_attn/Shape_output_0", "outputIndexes": +[ 793 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 793 ] +, "main_type": "NONE", "name": "Shape1271", "outputIndexes": +[ 794 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 793 ] +, "main_type": "NONE", "name": "Rank1273", "outputIndexes": +[ 795 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 795, 795 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1275", "outputIndexes": +[ 796 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 796 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1276", "outputIndexes": +[ 797 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 796, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1278", "outputIndexes": +[ 798 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 798 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1279", "outputIndexes": +[ 799 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 794, 797, 799, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1282", "outputIndexes": +[ 800 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 800 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1283", "outputIndexes": +[ 801 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 801, 801 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1285", "outputIndexes": +[ 802 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 793, 802, 796 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.8/self_attn/Gather_output_0", "outputIndexes": +[ 803 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 803, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 804 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 796 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1293", "outputIndexes": +[ 805 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 798 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1296", "outputIndexes": +[ 806 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 794, 805, 806, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1299", "outputIndexes": +[ 807 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 807 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1300", "outputIndexes": +[ 808 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 808 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1301", "outputIndexes": +[ 809 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 809, 808 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1302", "outputIndexes": +[ 810 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 793, 810, 796 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.8/self_attn/Gather_1_output_0", "outputIndexes": +[ 811 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 811, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 812 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 804, 812, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.8/self_attn/Concat_output_0", "outputIndexes": +[ 813 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 792, 813 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.8/self_attn/Reshape_output_0", "outputIndexes": +[ 814 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 814 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 975750374, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.8/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 816 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 816, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Mul_output_0", "outputIndexes": +[ 817 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 816 ] +, "main_type": "NONE", "name": "/blocks.8/self_attn/Shape_2_output_0", "outputIndexes": +[ 818 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 818 ] +, "main_type": "NONE", "name": "Shape1317", "outputIndexes": +[ 819 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 818 ] +, "main_type": "NONE", "name": "Rank1319", "outputIndexes": +[ 820 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 820, 820 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1321", "outputIndexes": +[ 821 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 821 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1322", "outputIndexes": +[ 822 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 821, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1324", "outputIndexes": +[ 823 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 823 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1325", "outputIndexes": +[ 824 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 819, 822, 824, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1328", "outputIndexes": +[ 825 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 825 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1329", "outputIndexes": +[ 826 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 826 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1330", "outputIndexes": +[ 827 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 827, 826 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1331", "outputIndexes": +[ 828 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 818, 828, 821 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.8/self_attn/Gather_2_output_0", "outputIndexes": +[ 829 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 829, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Div_output_0", "outputIndexes": +[ 830 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 830, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 831 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 816, 831, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.8/self_attn/Slice_1_output_0", "outputIndexes": +[ 832 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 832 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.8/self_attn/Neg_output_0", "outputIndexes": +[ 833 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 830, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 834 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 816, 20, 834, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.8/self_attn/Slice_output_0", "outputIndexes": +[ 835 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 833, 835 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.8/self_attn/Concat_3_output_0", "outputIndexes": +[ 836 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 836, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Mul_1_output_0", "outputIndexes": +[ 837 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 817, 837 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Add_output_0", "outputIndexes": +[ 838 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3622 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 975751398, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.8/self_attn/k_proj/Linear", "outputIndexes": +[ 3627 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3627 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3628 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3628 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 839 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 803, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 840 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 811, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 841 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 840, 841, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.8/self_attn/Concat_1_output_0", "outputIndexes": +[ 842 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 839, 842 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.8/self_attn/Reshape_1_output_0", "outputIndexes": +[ 843 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 843 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 978372860, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.8/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 845 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 845, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Mul_2_output_0", "outputIndexes": +[ 846 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 845 ] +, "main_type": "NONE", "name": "/blocks.8/self_attn/Shape_3_output_0", "outputIndexes": +[ 847 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 847 ] +, "main_type": "NONE", "name": "Shape1359", "outputIndexes": +[ 848 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 847 ] +, "main_type": "NONE", "name": "Rank1361", "outputIndexes": +[ 849 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 849, 849 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1363", "outputIndexes": +[ 850 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 850 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1364", "outputIndexes": +[ 851 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 850, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1366", "outputIndexes": +[ 852 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 852 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1367", "outputIndexes": +[ 853 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 848, 851, 853, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1370", "outputIndexes": +[ 854 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 854 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1371", "outputIndexes": +[ 855 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 855 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1372", "outputIndexes": +[ 856 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 856, 855 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1373", "outputIndexes": +[ 857 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 847, 857, 850 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.8/self_attn/Gather_3_output_0", "outputIndexes": +[ 858 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 858, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Div_1_output_0", "outputIndexes": +[ 859 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 859, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 860 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 845, 860, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.8/self_attn/Slice_3_output_0", "outputIndexes": +[ 861 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 861 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.8/self_attn/Neg_1_output_0", "outputIndexes": +[ 862 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 859, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 863 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 845, 20, 863, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.8/self_attn/Slice_2_output_0", "outputIndexes": +[ 864 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 862, 864 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.8/self_attn/Concat_4_output_0", "outputIndexes": +[ 865 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 865, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Mul_3_output_0", "outputIndexes": +[ 866 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 846, 866 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/self_attn/Add_1_output_0", "outputIndexes": +[ 867 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3622 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 978373884, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.8/self_attn/v_proj/Linear", "outputIndexes": +[ 3631 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3631 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3632 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3632 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 868 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 803, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 869 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 811, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.8/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 870 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 869, 870, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.8/self_attn/Concat_2_output_0", "outputIndexes": +[ 871 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 868, 871 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.8/self_attn/Reshape_2_output_0", "outputIndexes": +[ 872 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 838, 867, 872, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.8/self_attn/FusedAttention", "outputIndexes": +[ 873 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 873 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3633 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3633 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3634 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3634 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 980995346, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.8/self_attn/o_proj/Linear", "outputIndexes": +[ 3635 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3635 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3636 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3636 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 874 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 790, 874 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/Add_output_0", "outputIndexes": +[ 875 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 875 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 991481128, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.8/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 876 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 876 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3637 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3637 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3638 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3638 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 991513896, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.8/mlp/gate_proj/Linear", "outputIndexes": +[ 3639 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3639 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3640 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3640 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 877 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 877 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.8/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 878 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3638 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1022971198, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.8/mlp/up_proj/Linear", "outputIndexes": +[ 3643 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3643 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3644 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3644 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 879 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 878, 879 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/mlp/Mul_output_0", "outputIndexes": +[ 880 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 880 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3645 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3645 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3646 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3646 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1054428500, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.8/mlp/down_proj/Linear", "outputIndexes": +[ 3647 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3647 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3648 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3648 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 881 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 875, 881 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.8/Add_1_output_0", "outputIndexes": +[ 882 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 882, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.9/Reshape_output_0", "outputIndexes": +[ 883 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 883 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1085885802, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.9/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 884 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 884 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3649 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3649 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3650 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3650 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1085918570, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.9/self_attn/q_proj/Linear", "outputIndexes": +[ 3651 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3651 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3652 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3652 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 885 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 884 ] +, "main_type": "NONE", "name": "/blocks.9/self_attn/Shape_output_0", "outputIndexes": +[ 886 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 886 ] +, "main_type": "NONE", "name": "Shape1420", "outputIndexes": +[ 887 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 886 ] +, "main_type": "NONE", "name": "Rank1422", "outputIndexes": +[ 888 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 888, 888 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1424", "outputIndexes": +[ 889 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 889 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1425", "outputIndexes": +[ 890 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 889, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1427", "outputIndexes": +[ 891 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 891 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1428", "outputIndexes": +[ 892 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 887, 890, 892, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1431", "outputIndexes": +[ 893 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 893 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1432", "outputIndexes": +[ 894 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 894, 894 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1434", "outputIndexes": +[ 895 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 886, 895, 889 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.9/self_attn/Gather_output_0", "outputIndexes": +[ 896 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 896, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 897 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 889 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1442", "outputIndexes": +[ 898 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 891 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1445", "outputIndexes": +[ 899 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 887, 898, 899, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1448", "outputIndexes": +[ 900 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 900 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1449", "outputIndexes": +[ 901 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 901 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1450", "outputIndexes": +[ 902 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 902, 901 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1451", "outputIndexes": +[ 903 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 886, 903, 889 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.9/self_attn/Gather_1_output_0", "outputIndexes": +[ 904 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 904, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 905 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 897, 905, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.9/self_attn/Concat_output_0", "outputIndexes": +[ 906 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 885, 906 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.9/self_attn/Reshape_output_0", "outputIndexes": +[ 907 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 907 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1096404352, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.9/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 909 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 909, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Mul_output_0", "outputIndexes": +[ 910 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 909 ] +, "main_type": "NONE", "name": "/blocks.9/self_attn/Shape_2_output_0", "outputIndexes": +[ 911 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 911 ] +, "main_type": "NONE", "name": "Shape1466", "outputIndexes": +[ 912 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 911 ] +, "main_type": "NONE", "name": "Rank1468", "outputIndexes": +[ 913 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 913, 913 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1470", "outputIndexes": +[ 914 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 914 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1471", "outputIndexes": +[ 915 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 914, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1473", "outputIndexes": +[ 916 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 916 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1474", "outputIndexes": +[ 917 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 912, 915, 917, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1477", "outputIndexes": +[ 918 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 918 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1478", "outputIndexes": +[ 919 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 919 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1479", "outputIndexes": +[ 920 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 920, 919 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1480", "outputIndexes": +[ 921 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 911, 921, 914 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.9/self_attn/Gather_2_output_0", "outputIndexes": +[ 922 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 922, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Div_output_0", "outputIndexes": +[ 923 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 923, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 924 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 909, 924, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.9/self_attn/Slice_1_output_0", "outputIndexes": +[ 925 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 925 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.9/self_attn/Neg_output_0", "outputIndexes": +[ 926 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 923, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 927 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 909, 20, 927, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.9/self_attn/Slice_output_0", "outputIndexes": +[ 928 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 926, 928 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.9/self_attn/Concat_3_output_0", "outputIndexes": +[ 929 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 929, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Mul_1_output_0", "outputIndexes": +[ 930 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 910, 930 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Add_output_0", "outputIndexes": +[ 931 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3650 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1096405376, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.9/self_attn/k_proj/Linear", "outputIndexes": +[ 3655 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3655 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3656 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3656 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 932 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 896, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 933 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 904, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 934 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 933, 934, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.9/self_attn/Concat_1_output_0", "outputIndexes": +[ 935 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 932, 935 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.9/self_attn/Reshape_1_output_0", "outputIndexes": +[ 936 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 936 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1099026838, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.9/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 938 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 938, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Mul_2_output_0", "outputIndexes": +[ 939 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 938 ] +, "main_type": "NONE", "name": "/blocks.9/self_attn/Shape_3_output_0", "outputIndexes": +[ 940 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 940 ] +, "main_type": "NONE", "name": "Shape1508", "outputIndexes": +[ 941 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 940 ] +, "main_type": "NONE", "name": "Rank1510", "outputIndexes": +[ 942 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 942, 942 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1512", "outputIndexes": +[ 943 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 943 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1513", "outputIndexes": +[ 944 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 943, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1515", "outputIndexes": +[ 945 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 945 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1516", "outputIndexes": +[ 946 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 941, 944, 946, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1519", "outputIndexes": +[ 947 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 947 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1520", "outputIndexes": +[ 948 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 948 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1521", "outputIndexes": +[ 949 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 949, 948 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1522", "outputIndexes": +[ 950 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 940, 950, 943 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.9/self_attn/Gather_3_output_0", "outputIndexes": +[ 951 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 951, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Div_1_output_0", "outputIndexes": +[ 952 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 952, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 953 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 938, 953, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.9/self_attn/Slice_3_output_0", "outputIndexes": +[ 954 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 954 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.9/self_attn/Neg_1_output_0", "outputIndexes": +[ 955 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 952, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 956 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 938, 20, 956, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.9/self_attn/Slice_2_output_0", "outputIndexes": +[ 957 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 955, 957 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.9/self_attn/Concat_4_output_0", "outputIndexes": +[ 958 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 958, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Mul_3_output_0", "outputIndexes": +[ 959 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 939, 959 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/self_attn/Add_1_output_0", "outputIndexes": +[ 960 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3650 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1099027862, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.9/self_attn/v_proj/Linear", "outputIndexes": +[ 3659 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3659 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3660 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3660 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 961 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 896, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 962 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 904, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.9/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 963 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 962, 963, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.9/self_attn/Concat_2_output_0", "outputIndexes": +[ 964 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 961, 964 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.9/self_attn/Reshape_2_output_0", "outputIndexes": +[ 965 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 931, 960, 965, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.9/self_attn/FusedAttention", "outputIndexes": +[ 966 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 966 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3661 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3661 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3662 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3662 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1101649324, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.9/self_attn/o_proj/Linear", "outputIndexes": +[ 3663 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3663 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3664 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3664 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 967 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 883, 967 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/Add_output_0", "outputIndexes": +[ 968 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 968 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1112135106, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.9/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 969 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 969 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3665 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3665 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3666 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3666 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1112167874, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.9/mlp/gate_proj/Linear", "outputIndexes": +[ 3667 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3667 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3668 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3668 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 970 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 970 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.9/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 971 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3666 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1143625176, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.9/mlp/up_proj/Linear", "outputIndexes": +[ 3671 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3671 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3672 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3672 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 972 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 971, 972 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/mlp/Mul_output_0", "outputIndexes": +[ 973 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 973 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3673 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3673 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3674 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3674 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1175082478, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.9/mlp/down_proj/Linear", "outputIndexes": +[ 3675 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3675 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3676 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3676 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 974 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 968, 974 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.9/Add_1_output_0", "outputIndexes": +[ 975 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 975, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.10/Reshape_output_0", "outputIndexes": +[ 976 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 976 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1206539780, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.10/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 977 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 977 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3677 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3677 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3678 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3678 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1206572548, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.10/self_attn/q_proj/Linear", "outputIndexes": +[ 3679 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3679 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3680 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3680 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 978 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 977 ] +, "main_type": "NONE", "name": "/blocks.10/self_attn/Shape_output_0", "outputIndexes": +[ 979 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 979 ] +, "main_type": "NONE", "name": "Shape1569", "outputIndexes": +[ 980 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 979 ] +, "main_type": "NONE", "name": "Rank1571", "outputIndexes": +[ 981 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 981, 981 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1573", "outputIndexes": +[ 982 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 982 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1574", "outputIndexes": +[ 983 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 982, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1576", "outputIndexes": +[ 984 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 984 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1577", "outputIndexes": +[ 985 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 980, 983, 985, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1580", "outputIndexes": +[ 986 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 986 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1581", "outputIndexes": +[ 987 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 987, 987 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1583", "outputIndexes": +[ 988 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 979, 988, 982 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.10/self_attn/Gather_output_0", "outputIndexes": +[ 989 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 989, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 990 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 982 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1591", "outputIndexes": +[ 991 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 984 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1594", "outputIndexes": +[ 992 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 980, 991, 992, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1597", "outputIndexes": +[ 993 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 993 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1598", "outputIndexes": +[ 994 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 994 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1599", "outputIndexes": +[ 995 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 995, 994 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1600", "outputIndexes": +[ 996 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 979, 996, 982 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.10/self_attn/Gather_1_output_0", "outputIndexes": +[ 997 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 997, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 998 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 990, 998, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.10/self_attn/Concat_output_0", "outputIndexes": +[ 999 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 978, 999 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.10/self_attn/Reshape_output_0", "outputIndexes": +[ 1000 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1000 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1217058330, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.10/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1002 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1002, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Mul_output_0", "outputIndexes": +[ 1003 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1002 ] +, "main_type": "NONE", "name": "/blocks.10/self_attn/Shape_2_output_0", "outputIndexes": +[ 1004 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1004 ] +, "main_type": "NONE", "name": "Shape1615", "outputIndexes": +[ 1005 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1004 ] +, "main_type": "NONE", "name": "Rank1617", "outputIndexes": +[ 1006 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1006, 1006 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1619", "outputIndexes": +[ 1007 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1007 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1620", "outputIndexes": +[ 1008 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1007, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1622", "outputIndexes": +[ 1009 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1009 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1623", "outputIndexes": +[ 1010 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1005, 1008, 1010, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1626", "outputIndexes": +[ 1011 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1011 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1627", "outputIndexes": +[ 1012 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1012 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1628", "outputIndexes": +[ 1013 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1013, 1012 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1629", "outputIndexes": +[ 1014 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1004, 1014, 1007 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.10/self_attn/Gather_2_output_0", "outputIndexes": +[ 1015 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1015, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Div_output_0", "outputIndexes": +[ 1016 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1016, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1017 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1002, 1017, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.10/self_attn/Slice_1_output_0", "outputIndexes": +[ 1018 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1018 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.10/self_attn/Neg_output_0", "outputIndexes": +[ 1019 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1016, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1020 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1002, 20, 1020, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.10/self_attn/Slice_output_0", "outputIndexes": +[ 1021 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1019, 1021 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.10/self_attn/Concat_3_output_0", "outputIndexes": +[ 1022 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1022, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Mul_1_output_0", "outputIndexes": +[ 1023 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1003, 1023 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Add_output_0", "outputIndexes": +[ 1024 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3678 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1217059354, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.10/self_attn/k_proj/Linear", "outputIndexes": +[ 3683 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3683 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3684 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3684 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1025 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 989, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1026 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 997, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1027 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1026, 1027, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.10/self_attn/Concat_1_output_0", "outputIndexes": +[ 1028 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1025, 1028 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.10/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1029 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1029 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1219680816, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.10/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1031 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1031, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Mul_2_output_0", "outputIndexes": +[ 1032 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1031 ] +, "main_type": "NONE", "name": "/blocks.10/self_attn/Shape_3_output_0", "outputIndexes": +[ 1033 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1033 ] +, "main_type": "NONE", "name": "Shape1657", "outputIndexes": +[ 1034 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1033 ] +, "main_type": "NONE", "name": "Rank1659", "outputIndexes": +[ 1035 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1035, 1035 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1661", "outputIndexes": +[ 1036 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1036 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1662", "outputIndexes": +[ 1037 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1036, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1664", "outputIndexes": +[ 1038 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1038 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1665", "outputIndexes": +[ 1039 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1034, 1037, 1039, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1668", "outputIndexes": +[ 1040 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1040 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1669", "outputIndexes": +[ 1041 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1041 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1670", "outputIndexes": +[ 1042 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1042, 1041 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1671", "outputIndexes": +[ 1043 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1033, 1043, 1036 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.10/self_attn/Gather_3_output_0", "outputIndexes": +[ 1044 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1044, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Div_1_output_0", "outputIndexes": +[ 1045 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1045, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1046 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1031, 1046, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.10/self_attn/Slice_3_output_0", "outputIndexes": +[ 1047 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1047 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.10/self_attn/Neg_1_output_0", "outputIndexes": +[ 1048 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1045, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1049 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1031, 20, 1049, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.10/self_attn/Slice_2_output_0", "outputIndexes": +[ 1050 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1048, 1050 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.10/self_attn/Concat_4_output_0", "outputIndexes": +[ 1051 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1051, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Mul_3_output_0", "outputIndexes": +[ 1052 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1032, 1052 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/self_attn/Add_1_output_0", "outputIndexes": +[ 1053 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3678 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1219681840, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.10/self_attn/v_proj/Linear", "outputIndexes": +[ 3687 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3687 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3688 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3688 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1054 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 989, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1055 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 997, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.10/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1056 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1055, 1056, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.10/self_attn/Concat_2_output_0", "outputIndexes": +[ 1057 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1054, 1057 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.10/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1058 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1024, 1053, 1058, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.10/self_attn/FusedAttention", "outputIndexes": +[ 1059 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1059 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3689 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3689 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3690 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3690 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1222303302, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.10/self_attn/o_proj/Linear", "outputIndexes": +[ 3691 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3691 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3692 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3692 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1060 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 976, 1060 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/Add_output_0", "outputIndexes": +[ 1061 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1061 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1232789084, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.10/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1062 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1062 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3693 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3693 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3694 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3694 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1232821852, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.10/mlp/gate_proj/Linear", "outputIndexes": +[ 3695 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3695 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3696 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3696 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1063 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1063 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.10/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1064 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3694 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1264279154, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.10/mlp/up_proj/Linear", "outputIndexes": +[ 3699 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3699 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3700 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3700 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1065 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1064, 1065 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/mlp/Mul_output_0", "outputIndexes": +[ 1066 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1066 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3701 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3701 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3702 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3702 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1295736456, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.10/mlp/down_proj/Linear", "outputIndexes": +[ 3703 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3703 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3704 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3704 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1067 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1061, 1067 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.10/Add_1_output_0", "outputIndexes": +[ 1068 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1068, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.11/Reshape_output_0", "outputIndexes": +[ 1069 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1069 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1327193758, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.11/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1070 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1070 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3705 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3705 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3706 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3706 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1327226526, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.11/self_attn/q_proj/Linear", "outputIndexes": +[ 3707 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3707 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3708 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3708 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1071 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1070 ] +, "main_type": "NONE", "name": "/blocks.11/self_attn/Shape_output_0", "outputIndexes": +[ 1072 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1072 ] +, "main_type": "NONE", "name": "Shape1718", "outputIndexes": +[ 1073 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1072 ] +, "main_type": "NONE", "name": "Rank1720", "outputIndexes": +[ 1074 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1074, 1074 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1722", "outputIndexes": +[ 1075 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1075 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1723", "outputIndexes": +[ 1076 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1075, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1725", "outputIndexes": +[ 1077 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1077 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1726", "outputIndexes": +[ 1078 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1073, 1076, 1078, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1729", "outputIndexes": +[ 1079 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1079 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1730", "outputIndexes": +[ 1080 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1080, 1080 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1732", "outputIndexes": +[ 1081 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1072, 1081, 1075 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.11/self_attn/Gather_output_0", "outputIndexes": +[ 1082 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1082, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1083 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1075 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1740", "outputIndexes": +[ 1084 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1077 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1743", "outputIndexes": +[ 1085 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1073, 1084, 1085, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1746", "outputIndexes": +[ 1086 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1086 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1747", "outputIndexes": +[ 1087 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1087 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1748", "outputIndexes": +[ 1088 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1088, 1087 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1749", "outputIndexes": +[ 1089 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1072, 1089, 1075 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.11/self_attn/Gather_1_output_0", "outputIndexes": +[ 1090 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1090, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1091 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1083, 1091, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.11/self_attn/Concat_output_0", "outputIndexes": +[ 1092 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1071, 1092 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.11/self_attn/Reshape_output_0", "outputIndexes": +[ 1093 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1093 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1337712308, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.11/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1095 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1095, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Mul_output_0", "outputIndexes": +[ 1096 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1095 ] +, "main_type": "NONE", "name": "/blocks.11/self_attn/Shape_2_output_0", "outputIndexes": +[ 1097 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1097 ] +, "main_type": "NONE", "name": "Shape1764", "outputIndexes": +[ 1098 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1097 ] +, "main_type": "NONE", "name": "Rank1766", "outputIndexes": +[ 1099 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1099, 1099 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1768", "outputIndexes": +[ 1100 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1100 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1769", "outputIndexes": +[ 1101 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1100, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1771", "outputIndexes": +[ 1102 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1102 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1772", "outputIndexes": +[ 1103 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1098, 1101, 1103, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1775", "outputIndexes": +[ 1104 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1104 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1776", "outputIndexes": +[ 1105 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1105 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1777", "outputIndexes": +[ 1106 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1106, 1105 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1778", "outputIndexes": +[ 1107 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1097, 1107, 1100 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.11/self_attn/Gather_2_output_0", "outputIndexes": +[ 1108 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1108, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Div_output_0", "outputIndexes": +[ 1109 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1109, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1110 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1095, 1110, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.11/self_attn/Slice_1_output_0", "outputIndexes": +[ 1111 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1111 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.11/self_attn/Neg_output_0", "outputIndexes": +[ 1112 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1109, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1113 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1095, 20, 1113, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.11/self_attn/Slice_output_0", "outputIndexes": +[ 1114 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1112, 1114 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.11/self_attn/Concat_3_output_0", "outputIndexes": +[ 1115 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1115, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Mul_1_output_0", "outputIndexes": +[ 1116 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1096, 1116 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Add_output_0", "outputIndexes": +[ 1117 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3706 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1337713332, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.11/self_attn/k_proj/Linear", "outputIndexes": +[ 3711 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3711 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3712 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3712 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1118 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1082, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1119 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1090, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1120 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1119, 1120, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.11/self_attn/Concat_1_output_0", "outputIndexes": +[ 1121 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1118, 1121 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.11/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1122 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1122 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1340334794, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.11/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1124 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1124, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Mul_2_output_0", "outputIndexes": +[ 1125 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1124 ] +, "main_type": "NONE", "name": "/blocks.11/self_attn/Shape_3_output_0", "outputIndexes": +[ 1126 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1126 ] +, "main_type": "NONE", "name": "Shape1806", "outputIndexes": +[ 1127 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1126 ] +, "main_type": "NONE", "name": "Rank1808", "outputIndexes": +[ 1128 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1128, 1128 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1810", "outputIndexes": +[ 1129 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1129 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1811", "outputIndexes": +[ 1130 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1129, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1813", "outputIndexes": +[ 1131 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1131 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1814", "outputIndexes": +[ 1132 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1127, 1130, 1132, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1817", "outputIndexes": +[ 1133 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1133 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1818", "outputIndexes": +[ 1134 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1134 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1819", "outputIndexes": +[ 1135 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1135, 1134 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1820", "outputIndexes": +[ 1136 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1126, 1136, 1129 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.11/self_attn/Gather_3_output_0", "outputIndexes": +[ 1137 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1137, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Div_1_output_0", "outputIndexes": +[ 1138 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1138, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1139 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1124, 1139, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.11/self_attn/Slice_3_output_0", "outputIndexes": +[ 1140 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1140 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.11/self_attn/Neg_1_output_0", "outputIndexes": +[ 1141 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1138, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1142 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1124, 20, 1142, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.11/self_attn/Slice_2_output_0", "outputIndexes": +[ 1143 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1141, 1143 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.11/self_attn/Concat_4_output_0", "outputIndexes": +[ 1144 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1144, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Mul_3_output_0", "outputIndexes": +[ 1145 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1125, 1145 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/self_attn/Add_1_output_0", "outputIndexes": +[ 1146 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3706 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1340335818, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.11/self_attn/v_proj/Linear", "outputIndexes": +[ 3715 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3715 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3716 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3716 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1147 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1082, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1148 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1090, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.11/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1149 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1148, 1149, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.11/self_attn/Concat_2_output_0", "outputIndexes": +[ 1150 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1147, 1150 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.11/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1151 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1117, 1146, 1151, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.11/self_attn/FusedAttention", "outputIndexes": +[ 1152 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1152 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3717 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3717 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3718 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3718 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1342957280, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.11/self_attn/o_proj/Linear", "outputIndexes": +[ 3719 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3719 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3720 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3720 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1153 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1069, 1153 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/Add_output_0", "outputIndexes": +[ 1154 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1154 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1353443062, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.11/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1155 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1155 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3721 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3721 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3722 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3722 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1353475830, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.11/mlp/gate_proj/Linear", "outputIndexes": +[ 3723 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3723 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3724 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3724 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1156 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1156 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.11/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1157 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3722 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1384933132, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.11/mlp/up_proj/Linear", "outputIndexes": +[ 3727 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3727 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3728 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3728 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1158 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1157, 1158 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/mlp/Mul_output_0", "outputIndexes": +[ 1159 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1159 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3729 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3729 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3730 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3730 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1416390434, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.11/mlp/down_proj/Linear", "outputIndexes": +[ 3731 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3731 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3732 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3732 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1160 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1154, 1160 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.11/Add_1_output_0", "outputIndexes": +[ 1161 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1161, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.12/Reshape_output_0", "outputIndexes": +[ 1162 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1162 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1447847736, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.12/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1163 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1163 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3733 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3733 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3734 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3734 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1447880504, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.12/self_attn/q_proj/Linear", "outputIndexes": +[ 3735 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3735 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3736 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3736 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1164 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1163 ] +, "main_type": "NONE", "name": "/blocks.12/self_attn/Shape_output_0", "outputIndexes": +[ 1165 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1165 ] +, "main_type": "NONE", "name": "Shape1867", "outputIndexes": +[ 1166 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1165 ] +, "main_type": "NONE", "name": "Rank1869", "outputIndexes": +[ 1167 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1167, 1167 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1871", "outputIndexes": +[ 1168 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1168 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1872", "outputIndexes": +[ 1169 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1168, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1874", "outputIndexes": +[ 1170 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1170 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1875", "outputIndexes": +[ 1171 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1166, 1169, 1171, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1878", "outputIndexes": +[ 1172 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1172 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1879", "outputIndexes": +[ 1173 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1173, 1173 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1881", "outputIndexes": +[ 1174 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1165, 1174, 1168 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.12/self_attn/Gather_output_0", "outputIndexes": +[ 1175 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1175, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1176 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1168 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1889", "outputIndexes": +[ 1177 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1170 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1892", "outputIndexes": +[ 1178 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1166, 1177, 1178, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1895", "outputIndexes": +[ 1179 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1179 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1896", "outputIndexes": +[ 1180 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1180 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1897", "outputIndexes": +[ 1181 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1181, 1180 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1898", "outputIndexes": +[ 1182 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1165, 1182, 1168 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.12/self_attn/Gather_1_output_0", "outputIndexes": +[ 1183 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1183, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1184 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1176, 1184, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.12/self_attn/Concat_output_0", "outputIndexes": +[ 1185 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1164, 1185 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.12/self_attn/Reshape_output_0", "outputIndexes": +[ 1186 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1186 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1458366286, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.12/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1188 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1188, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Mul_output_0", "outputIndexes": +[ 1189 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1188 ] +, "main_type": "NONE", "name": "/blocks.12/self_attn/Shape_2_output_0", "outputIndexes": +[ 1190 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1190 ] +, "main_type": "NONE", "name": "Shape1913", "outputIndexes": +[ 1191 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1190 ] +, "main_type": "NONE", "name": "Rank1915", "outputIndexes": +[ 1192 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1192, 1192 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1917", "outputIndexes": +[ 1193 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1193 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1918", "outputIndexes": +[ 1194 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1193, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1920", "outputIndexes": +[ 1195 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1195 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1921", "outputIndexes": +[ 1196 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1191, 1194, 1196, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1924", "outputIndexes": +[ 1197 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1197 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1925", "outputIndexes": +[ 1198 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1198 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1926", "outputIndexes": +[ 1199 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1199, 1198 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1927", "outputIndexes": +[ 1200 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1190, 1200, 1193 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.12/self_attn/Gather_2_output_0", "outputIndexes": +[ 1201 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1201, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Div_output_0", "outputIndexes": +[ 1202 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1202, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1203 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1188, 1203, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.12/self_attn/Slice_1_output_0", "outputIndexes": +[ 1204 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1204 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.12/self_attn/Neg_output_0", "outputIndexes": +[ 1205 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1202, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1206 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1188, 20, 1206, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.12/self_attn/Slice_output_0", "outputIndexes": +[ 1207 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1205, 1207 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.12/self_attn/Concat_3_output_0", "outputIndexes": +[ 1208 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1208, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Mul_1_output_0", "outputIndexes": +[ 1209 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1189, 1209 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Add_output_0", "outputIndexes": +[ 1210 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3734 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1458367310, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.12/self_attn/k_proj/Linear", "outputIndexes": +[ 3739 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3739 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3740 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3740 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1211 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1175, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1212 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1183, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1213 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1212, 1213, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.12/self_attn/Concat_1_output_0", "outputIndexes": +[ 1214 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1211, 1214 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.12/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1215 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1215 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1460988772, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.12/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1217 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1217, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Mul_2_output_0", "outputIndexes": +[ 1218 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1217 ] +, "main_type": "NONE", "name": "/blocks.12/self_attn/Shape_3_output_0", "outputIndexes": +[ 1219 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1219 ] +, "main_type": "NONE", "name": "Shape1955", "outputIndexes": +[ 1220 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1219 ] +, "main_type": "NONE", "name": "Rank1957", "outputIndexes": +[ 1221 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1221, 1221 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1959", "outputIndexes": +[ 1222 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1222 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1960", "outputIndexes": +[ 1223 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1222, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1962", "outputIndexes": +[ 1224 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1224 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1963", "outputIndexes": +[ 1225 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1220, 1223, 1225, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1966", "outputIndexes": +[ 1226 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1226 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1967", "outputIndexes": +[ 1227 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1227 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1968", "outputIndexes": +[ 1228 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1228, 1227 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1969", "outputIndexes": +[ 1229 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1219, 1229, 1222 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.12/self_attn/Gather_3_output_0", "outputIndexes": +[ 1230 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1230, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Div_1_output_0", "outputIndexes": +[ 1231 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1231, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1232 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1217, 1232, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.12/self_attn/Slice_3_output_0", "outputIndexes": +[ 1233 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1233 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.12/self_attn/Neg_1_output_0", "outputIndexes": +[ 1234 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1231, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1235 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1217, 20, 1235, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.12/self_attn/Slice_2_output_0", "outputIndexes": +[ 1236 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1234, 1236 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.12/self_attn/Concat_4_output_0", "outputIndexes": +[ 1237 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1237, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Mul_3_output_0", "outputIndexes": +[ 1238 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1218, 1238 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/self_attn/Add_1_output_0", "outputIndexes": +[ 1239 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3734 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1460989796, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.12/self_attn/v_proj/Linear", "outputIndexes": +[ 3743 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3743 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3744 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3744 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1240 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1175, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1241 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1183, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.12/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1242 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1241, 1242, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.12/self_attn/Concat_2_output_0", "outputIndexes": +[ 1243 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1240, 1243 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.12/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1244 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1210, 1239, 1244, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.12/self_attn/FusedAttention", "outputIndexes": +[ 1245 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1245 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3745 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3745 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3746 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3746 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1463611258, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.12/self_attn/o_proj/Linear", "outputIndexes": +[ 3747 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3747 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3748 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3748 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1246 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1162, 1246 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/Add_output_0", "outputIndexes": +[ 1247 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1247 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1474097040, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.12/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1248 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1248 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3749 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3749 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3750 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3750 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1474129808, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.12/mlp/gate_proj/Linear", "outputIndexes": +[ 3751 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3751 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3752 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3752 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1249 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1249 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.12/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1250 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3750 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1505587110, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.12/mlp/up_proj/Linear", "outputIndexes": +[ 3755 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3755 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3756 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3756 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1251 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1250, 1251 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/mlp/Mul_output_0", "outputIndexes": +[ 1252 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1252 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3757 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3757 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3758 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3758 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1537044412, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.12/mlp/down_proj/Linear", "outputIndexes": +[ 3759 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3759 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3760 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3760 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1253 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1247, 1253 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.12/Add_1_output_0", "outputIndexes": +[ 1254 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1254, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.13/Reshape_output_0", "outputIndexes": +[ 1255 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1255 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1568501714, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.13/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1256 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1256 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3761 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3761 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3762 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3762 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1568534482, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.13/self_attn/q_proj/Linear", "outputIndexes": +[ 3763 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3763 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3764 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3764 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1257 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1256 ] +, "main_type": "NONE", "name": "/blocks.13/self_attn/Shape_output_0", "outputIndexes": +[ 1258 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1258 ] +, "main_type": "NONE", "name": "Shape2016", "outputIndexes": +[ 1259 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1258 ] +, "main_type": "NONE", "name": "Rank2018", "outputIndexes": +[ 1260 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1260, 1260 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2020", "outputIndexes": +[ 1261 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1261 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2021", "outputIndexes": +[ 1262 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1261, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2023", "outputIndexes": +[ 1263 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1263 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2024", "outputIndexes": +[ 1264 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1259, 1262, 1264, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2027", "outputIndexes": +[ 1265 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1265 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2028", "outputIndexes": +[ 1266 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1266, 1266 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2030", "outputIndexes": +[ 1267 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1258, 1267, 1261 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.13/self_attn/Gather_output_0", "outputIndexes": +[ 1268 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1268, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1269 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1261 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2038", "outputIndexes": +[ 1270 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1263 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2041", "outputIndexes": +[ 1271 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1259, 1270, 1271, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2044", "outputIndexes": +[ 1272 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1272 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2045", "outputIndexes": +[ 1273 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1273 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2046", "outputIndexes": +[ 1274 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1274, 1273 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2047", "outputIndexes": +[ 1275 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1258, 1275, 1261 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.13/self_attn/Gather_1_output_0", "outputIndexes": +[ 1276 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1276, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1277 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1269, 1277, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.13/self_attn/Concat_output_0", "outputIndexes": +[ 1278 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1257, 1278 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.13/self_attn/Reshape_output_0", "outputIndexes": +[ 1279 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1279 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1579020264, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.13/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1281 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1281, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Mul_output_0", "outputIndexes": +[ 1282 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1281 ] +, "main_type": "NONE", "name": "/blocks.13/self_attn/Shape_2_output_0", "outputIndexes": +[ 1283 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1283 ] +, "main_type": "NONE", "name": "Shape2062", "outputIndexes": +[ 1284 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1283 ] +, "main_type": "NONE", "name": "Rank2064", "outputIndexes": +[ 1285 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1285, 1285 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2066", "outputIndexes": +[ 1286 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1286 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2067", "outputIndexes": +[ 1287 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1286, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2069", "outputIndexes": +[ 1288 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1288 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2070", "outputIndexes": +[ 1289 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1284, 1287, 1289, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2073", "outputIndexes": +[ 1290 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1290 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2074", "outputIndexes": +[ 1291 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1291 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2075", "outputIndexes": +[ 1292 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1292, 1291 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2076", "outputIndexes": +[ 1293 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1283, 1293, 1286 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.13/self_attn/Gather_2_output_0", "outputIndexes": +[ 1294 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1294, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Div_output_0", "outputIndexes": +[ 1295 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1295, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1296 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1281, 1296, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.13/self_attn/Slice_1_output_0", "outputIndexes": +[ 1297 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1297 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.13/self_attn/Neg_output_0", "outputIndexes": +[ 1298 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1295, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1299 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1281, 20, 1299, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.13/self_attn/Slice_output_0", "outputIndexes": +[ 1300 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1298, 1300 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.13/self_attn/Concat_3_output_0", "outputIndexes": +[ 1301 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1301, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Mul_1_output_0", "outputIndexes": +[ 1302 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1282, 1302 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Add_output_0", "outputIndexes": +[ 1303 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3762 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1579021288, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.13/self_attn/k_proj/Linear", "outputIndexes": +[ 3767 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3767 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3768 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3768 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1304 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1268, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1305 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1276, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1306 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1305, 1306, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.13/self_attn/Concat_1_output_0", "outputIndexes": +[ 1307 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1304, 1307 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.13/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1308 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1308 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1581642750, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.13/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1310 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1310, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Mul_2_output_0", "outputIndexes": +[ 1311 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1310 ] +, "main_type": "NONE", "name": "/blocks.13/self_attn/Shape_3_output_0", "outputIndexes": +[ 1312 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1312 ] +, "main_type": "NONE", "name": "Shape2104", "outputIndexes": +[ 1313 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1312 ] +, "main_type": "NONE", "name": "Rank2106", "outputIndexes": +[ 1314 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1314, 1314 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2108", "outputIndexes": +[ 1315 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1315 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2109", "outputIndexes": +[ 1316 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1315, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2111", "outputIndexes": +[ 1317 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1317 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2112", "outputIndexes": +[ 1318 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1313, 1316, 1318, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2115", "outputIndexes": +[ 1319 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1319 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2116", "outputIndexes": +[ 1320 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1320 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2117", "outputIndexes": +[ 1321 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1321, 1320 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2118", "outputIndexes": +[ 1322 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1312, 1322, 1315 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.13/self_attn/Gather_3_output_0", "outputIndexes": +[ 1323 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1323, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Div_1_output_0", "outputIndexes": +[ 1324 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1324, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1325 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1310, 1325, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.13/self_attn/Slice_3_output_0", "outputIndexes": +[ 1326 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1326 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.13/self_attn/Neg_1_output_0", "outputIndexes": +[ 1327 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1324, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1328 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1310, 20, 1328, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.13/self_attn/Slice_2_output_0", "outputIndexes": +[ 1329 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1327, 1329 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.13/self_attn/Concat_4_output_0", "outputIndexes": +[ 1330 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1330, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Mul_3_output_0", "outputIndexes": +[ 1331 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1311, 1331 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/self_attn/Add_1_output_0", "outputIndexes": +[ 1332 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3762 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1581643774, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.13/self_attn/v_proj/Linear", "outputIndexes": +[ 3771 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3771 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3772 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3772 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1333 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1268, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1334 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1276, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.13/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1335 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1334, 1335, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.13/self_attn/Concat_2_output_0", "outputIndexes": +[ 1336 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1333, 1336 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.13/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1337 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1303, 1332, 1337, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.13/self_attn/FusedAttention", "outputIndexes": +[ 1338 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1338 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3773 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3773 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3774 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3774 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1584265236, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.13/self_attn/o_proj/Linear", "outputIndexes": +[ 3775 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3775 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3776 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3776 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1339 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1255, 1339 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/Add_output_0", "outputIndexes": +[ 1340 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1340 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1594751018, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.13/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1341 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1341 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3777 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3777 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3778 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3778 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1594783786, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.13/mlp/gate_proj/Linear", "outputIndexes": +[ 3779 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3779 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3780 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3780 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1342 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1342 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.13/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1343 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3778 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1626241088, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.13/mlp/up_proj/Linear", "outputIndexes": +[ 3783 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3783 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3784 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3784 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1344 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1343, 1344 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/mlp/Mul_output_0", "outputIndexes": +[ 1345 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1345 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3785 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3785 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3786 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3786 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1657698390, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.13/mlp/down_proj/Linear", "outputIndexes": +[ 3787 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3787 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3788 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3788 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1346 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1340, 1346 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.13/Add_1_output_0", "outputIndexes": +[ 1347 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1347, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.14/Reshape_output_0", "outputIndexes": +[ 1348 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1348 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1689155692, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.14/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1349 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1349 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3789 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3789 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3790 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3790 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1689188460, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.14/self_attn/q_proj/Linear", "outputIndexes": +[ 3791 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3791 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3792 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3792 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1350 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1349 ] +, "main_type": "NONE", "name": "/blocks.14/self_attn/Shape_output_0", "outputIndexes": +[ 1351 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1351 ] +, "main_type": "NONE", "name": "Shape2165", "outputIndexes": +[ 1352 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1351 ] +, "main_type": "NONE", "name": "Rank2167", "outputIndexes": +[ 1353 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1353, 1353 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2169", "outputIndexes": +[ 1354 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1354 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2170", "outputIndexes": +[ 1355 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1354, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2172", "outputIndexes": +[ 1356 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1356 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2173", "outputIndexes": +[ 1357 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1352, 1355, 1357, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2176", "outputIndexes": +[ 1358 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1358 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2177", "outputIndexes": +[ 1359 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1359, 1359 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2179", "outputIndexes": +[ 1360 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1351, 1360, 1354 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.14/self_attn/Gather_output_0", "outputIndexes": +[ 1361 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1361, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1362 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1354 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2187", "outputIndexes": +[ 1363 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1356 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2190", "outputIndexes": +[ 1364 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1352, 1363, 1364, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2193", "outputIndexes": +[ 1365 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1365 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2194", "outputIndexes": +[ 1366 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1366 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2195", "outputIndexes": +[ 1367 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1367, 1366 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2196", "outputIndexes": +[ 1368 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1351, 1368, 1354 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.14/self_attn/Gather_1_output_0", "outputIndexes": +[ 1369 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1369, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1370 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1362, 1370, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.14/self_attn/Concat_output_0", "outputIndexes": +[ 1371 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1350, 1371 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.14/self_attn/Reshape_output_0", "outputIndexes": +[ 1372 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1372 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1699674242, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.14/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1374 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1374, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Mul_output_0", "outputIndexes": +[ 1375 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1374 ] +, "main_type": "NONE", "name": "/blocks.14/self_attn/Shape_2_output_0", "outputIndexes": +[ 1376 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1376 ] +, "main_type": "NONE", "name": "Shape2211", "outputIndexes": +[ 1377 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1376 ] +, "main_type": "NONE", "name": "Rank2213", "outputIndexes": +[ 1378 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1378, 1378 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2215", "outputIndexes": +[ 1379 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1379 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2216", "outputIndexes": +[ 1380 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1379, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2218", "outputIndexes": +[ 1381 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1381 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2219", "outputIndexes": +[ 1382 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1377, 1380, 1382, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2222", "outputIndexes": +[ 1383 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1383 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2223", "outputIndexes": +[ 1384 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1384 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2224", "outputIndexes": +[ 1385 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1385, 1384 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2225", "outputIndexes": +[ 1386 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1376, 1386, 1379 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.14/self_attn/Gather_2_output_0", "outputIndexes": +[ 1387 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1387, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Div_output_0", "outputIndexes": +[ 1388 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1388, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1389 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1374, 1389, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.14/self_attn/Slice_1_output_0", "outputIndexes": +[ 1390 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1390 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.14/self_attn/Neg_output_0", "outputIndexes": +[ 1391 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1388, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1392 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1374, 20, 1392, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.14/self_attn/Slice_output_0", "outputIndexes": +[ 1393 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1391, 1393 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.14/self_attn/Concat_3_output_0", "outputIndexes": +[ 1394 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1394, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Mul_1_output_0", "outputIndexes": +[ 1395 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1375, 1395 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Add_output_0", "outputIndexes": +[ 1396 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3790 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1699675266, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.14/self_attn/k_proj/Linear", "outputIndexes": +[ 3795 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3795 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3796 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3796 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1397 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1361, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1398 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1369, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1399 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1398, 1399, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.14/self_attn/Concat_1_output_0", "outputIndexes": +[ 1400 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1397, 1400 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.14/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1401 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1401 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1702296728, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.14/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1403 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1403, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Mul_2_output_0", "outputIndexes": +[ 1404 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1403 ] +, "main_type": "NONE", "name": "/blocks.14/self_attn/Shape_3_output_0", "outputIndexes": +[ 1405 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1405 ] +, "main_type": "NONE", "name": "Shape2253", "outputIndexes": +[ 1406 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1405 ] +, "main_type": "NONE", "name": "Rank2255", "outputIndexes": +[ 1407 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1407, 1407 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2257", "outputIndexes": +[ 1408 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1408 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2258", "outputIndexes": +[ 1409 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1408, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2260", "outputIndexes": +[ 1410 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1410 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2261", "outputIndexes": +[ 1411 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1406, 1409, 1411, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2264", "outputIndexes": +[ 1412 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1412 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2265", "outputIndexes": +[ 1413 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1413 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2266", "outputIndexes": +[ 1414 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1414, 1413 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2267", "outputIndexes": +[ 1415 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1405, 1415, 1408 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.14/self_attn/Gather_3_output_0", "outputIndexes": +[ 1416 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1416, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Div_1_output_0", "outputIndexes": +[ 1417 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1417, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1418 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1403, 1418, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.14/self_attn/Slice_3_output_0", "outputIndexes": +[ 1419 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1419 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.14/self_attn/Neg_1_output_0", "outputIndexes": +[ 1420 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1417, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1421 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1403, 20, 1421, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.14/self_attn/Slice_2_output_0", "outputIndexes": +[ 1422 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1420, 1422 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.14/self_attn/Concat_4_output_0", "outputIndexes": +[ 1423 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1423, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Mul_3_output_0", "outputIndexes": +[ 1424 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1404, 1424 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/self_attn/Add_1_output_0", "outputIndexes": +[ 1425 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3790 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1702297752, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.14/self_attn/v_proj/Linear", "outputIndexes": +[ 3799 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3799 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3800 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3800 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1426 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1361, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1427 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1369, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.14/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1428 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1427, 1428, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.14/self_attn/Concat_2_output_0", "outputIndexes": +[ 1429 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1426, 1429 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.14/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1430 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1396, 1425, 1430, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.14/self_attn/FusedAttention", "outputIndexes": +[ 1431 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1431 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3801 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3801 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3802 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3802 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1704919214, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.14/self_attn/o_proj/Linear", "outputIndexes": +[ 3803 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3803 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3804 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3804 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1432 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1348, 1432 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/Add_output_0", "outputIndexes": +[ 1433 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1433 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1715404996, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.14/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1434 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1434 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3805 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3805 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3806 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3806 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1715437764, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.14/mlp/gate_proj/Linear", "outputIndexes": +[ 3807 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3807 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3808 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3808 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1435 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1435 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.14/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1436 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3806 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1746895066, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.14/mlp/up_proj/Linear", "outputIndexes": +[ 3811 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3811 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3812 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3812 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1437 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1436, 1437 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/mlp/Mul_output_0", "outputIndexes": +[ 1438 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1438 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3813 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3813 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3814 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3814 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1778352368, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.14/mlp/down_proj/Linear", "outputIndexes": +[ 3815 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3815 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3816 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3816 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1439 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1433, 1439 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.14/Add_1_output_0", "outputIndexes": +[ 1440 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1440, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.15/Reshape_output_0", "outputIndexes": +[ 1441 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1441 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1809809670, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.15/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1442 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1442 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3817 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3817 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3818 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3818 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1809842438, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.15/self_attn/q_proj/Linear", "outputIndexes": +[ 3819 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3819 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3820 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3820 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1443 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1442 ] +, "main_type": "NONE", "name": "/blocks.15/self_attn/Shape_output_0", "outputIndexes": +[ 1444 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1444 ] +, "main_type": "NONE", "name": "Shape2314", "outputIndexes": +[ 1445 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1444 ] +, "main_type": "NONE", "name": "Rank2316", "outputIndexes": +[ 1446 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1446, 1446 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2318", "outputIndexes": +[ 1447 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1447 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2319", "outputIndexes": +[ 1448 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1447, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2321", "outputIndexes": +[ 1449 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1449 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2322", "outputIndexes": +[ 1450 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1445, 1448, 1450, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2325", "outputIndexes": +[ 1451 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1451 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2326", "outputIndexes": +[ 1452 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1452, 1452 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2328", "outputIndexes": +[ 1453 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1444, 1453, 1447 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.15/self_attn/Gather_output_0", "outputIndexes": +[ 1454 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1454, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1455 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1447 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2336", "outputIndexes": +[ 1456 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1449 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2339", "outputIndexes": +[ 1457 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1445, 1456, 1457, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2342", "outputIndexes": +[ 1458 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1458 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2343", "outputIndexes": +[ 1459 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1459 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2344", "outputIndexes": +[ 1460 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1460, 1459 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2345", "outputIndexes": +[ 1461 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1444, 1461, 1447 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.15/self_attn/Gather_1_output_0", "outputIndexes": +[ 1462 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1462, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1463 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1455, 1463, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.15/self_attn/Concat_output_0", "outputIndexes": +[ 1464 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1443, 1464 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.15/self_attn/Reshape_output_0", "outputIndexes": +[ 1465 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1465 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1820328220, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.15/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1467 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1467, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Mul_output_0", "outputIndexes": +[ 1468 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1467 ] +, "main_type": "NONE", "name": "/blocks.15/self_attn/Shape_2_output_0", "outputIndexes": +[ 1469 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1469 ] +, "main_type": "NONE", "name": "Shape2360", "outputIndexes": +[ 1470 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1469 ] +, "main_type": "NONE", "name": "Rank2362", "outputIndexes": +[ 1471 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1471, 1471 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2364", "outputIndexes": +[ 1472 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1472 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2365", "outputIndexes": +[ 1473 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1472, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2367", "outputIndexes": +[ 1474 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1474 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2368", "outputIndexes": +[ 1475 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1470, 1473, 1475, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2371", "outputIndexes": +[ 1476 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1476 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2372", "outputIndexes": +[ 1477 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1477 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2373", "outputIndexes": +[ 1478 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1478, 1477 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2374", "outputIndexes": +[ 1479 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1469, 1479, 1472 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.15/self_attn/Gather_2_output_0", "outputIndexes": +[ 1480 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1480, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Div_output_0", "outputIndexes": +[ 1481 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1481, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1482 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1467, 1482, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.15/self_attn/Slice_1_output_0", "outputIndexes": +[ 1483 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1483 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.15/self_attn/Neg_output_0", "outputIndexes": +[ 1484 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1481, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1485 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1467, 20, 1485, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.15/self_attn/Slice_output_0", "outputIndexes": +[ 1486 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1484, 1486 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.15/self_attn/Concat_3_output_0", "outputIndexes": +[ 1487 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1487, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Mul_1_output_0", "outputIndexes": +[ 1488 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1468, 1488 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Add_output_0", "outputIndexes": +[ 1489 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3818 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1820329244, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.15/self_attn/k_proj/Linear", "outputIndexes": +[ 3823 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3823 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3824 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3824 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1490 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1454, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1491 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1462, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1492 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1491, 1492, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.15/self_attn/Concat_1_output_0", "outputIndexes": +[ 1493 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1490, 1493 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.15/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1494 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1494 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1822950706, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.15/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1496 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1496, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Mul_2_output_0", "outputIndexes": +[ 1497 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1496 ] +, "main_type": "NONE", "name": "/blocks.15/self_attn/Shape_3_output_0", "outputIndexes": +[ 1498 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1498 ] +, "main_type": "NONE", "name": "Shape2402", "outputIndexes": +[ 1499 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1498 ] +, "main_type": "NONE", "name": "Rank2404", "outputIndexes": +[ 1500 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1500, 1500 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2406", "outputIndexes": +[ 1501 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1501 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2407", "outputIndexes": +[ 1502 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1501, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2409", "outputIndexes": +[ 1503 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1503 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2410", "outputIndexes": +[ 1504 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1499, 1502, 1504, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2413", "outputIndexes": +[ 1505 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1505 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2414", "outputIndexes": +[ 1506 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1506 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2415", "outputIndexes": +[ 1507 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1507, 1506 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2416", "outputIndexes": +[ 1508 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1498, 1508, 1501 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.15/self_attn/Gather_3_output_0", "outputIndexes": +[ 1509 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1509, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Div_1_output_0", "outputIndexes": +[ 1510 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1510, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1511 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1496, 1511, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.15/self_attn/Slice_3_output_0", "outputIndexes": +[ 1512 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1512 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.15/self_attn/Neg_1_output_0", "outputIndexes": +[ 1513 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1510, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1514 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1496, 20, 1514, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.15/self_attn/Slice_2_output_0", "outputIndexes": +[ 1515 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1513, 1515 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.15/self_attn/Concat_4_output_0", "outputIndexes": +[ 1516 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1516, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Mul_3_output_0", "outputIndexes": +[ 1517 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1497, 1517 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/self_attn/Add_1_output_0", "outputIndexes": +[ 1518 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3818 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1822951730, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.15/self_attn/v_proj/Linear", "outputIndexes": +[ 3827 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3827 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3828 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3828 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1519 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1454, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1520 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1462, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.15/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1521 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1520, 1521, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.15/self_attn/Concat_2_output_0", "outputIndexes": +[ 1522 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1519, 1522 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.15/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1523 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1489, 1518, 1523, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.15/self_attn/FusedAttention", "outputIndexes": +[ 1524 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1524 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3829 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3829 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3830 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3830 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1825573192, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.15/self_attn/o_proj/Linear", "outputIndexes": +[ 3831 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3831 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3832 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3832 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1525 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1441, 1525 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/Add_output_0", "outputIndexes": +[ 1526 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1526 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1836058974, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.15/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1527 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1527 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3833 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3833 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3834 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3834 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1836091742, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.15/mlp/gate_proj/Linear", "outputIndexes": +[ 3835 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3835 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3836 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3836 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1528 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1528 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.15/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1529 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3834 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1867549044, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.15/mlp/up_proj/Linear", "outputIndexes": +[ 3839 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3839 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3840 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3840 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1530 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1529, 1530 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/mlp/Mul_output_0", "outputIndexes": +[ 1531 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1531 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3841 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3841 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3842 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3842 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1899006346, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.15/mlp/down_proj/Linear", "outputIndexes": +[ 3843 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3843 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3844 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3844 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1532 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1526, 1532 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.15/Add_1_output_0", "outputIndexes": +[ 1533 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1533, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.16/Reshape_output_0", "outputIndexes": +[ 1534 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1534 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1930463648, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.16/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1535 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1535 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3845 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3845 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3846 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3846 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1930496416, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.16/self_attn/q_proj/Linear", "outputIndexes": +[ 3847 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3847 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3848 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3848 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1536 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1535 ] +, "main_type": "NONE", "name": "/blocks.16/self_attn/Shape_output_0", "outputIndexes": +[ 1537 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1537 ] +, "main_type": "NONE", "name": "Shape2463", "outputIndexes": +[ 1538 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1537 ] +, "main_type": "NONE", "name": "Rank2465", "outputIndexes": +[ 1539 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1539, 1539 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2467", "outputIndexes": +[ 1540 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1540 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2468", "outputIndexes": +[ 1541 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1540, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2470", "outputIndexes": +[ 1542 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1542 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2471", "outputIndexes": +[ 1543 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1538, 1541, 1543, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2474", "outputIndexes": +[ 1544 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1544 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2475", "outputIndexes": +[ 1545 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1545, 1545 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2477", "outputIndexes": +[ 1546 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1537, 1546, 1540 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.16/self_attn/Gather_output_0", "outputIndexes": +[ 1547 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1547, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1548 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1540 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2485", "outputIndexes": +[ 1549 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1542 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2488", "outputIndexes": +[ 1550 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1538, 1549, 1550, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2491", "outputIndexes": +[ 1551 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1551 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2492", "outputIndexes": +[ 1552 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1552 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2493", "outputIndexes": +[ 1553 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1553, 1552 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2494", "outputIndexes": +[ 1554 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1537, 1554, 1540 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.16/self_attn/Gather_1_output_0", "outputIndexes": +[ 1555 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1555, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1556 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1548, 1556, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.16/self_attn/Concat_output_0", "outputIndexes": +[ 1557 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1536, 1557 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.16/self_attn/Reshape_output_0", "outputIndexes": +[ 1558 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1558 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1940982198, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.16/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1560 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1560, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Mul_output_0", "outputIndexes": +[ 1561 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1560 ] +, "main_type": "NONE", "name": "/blocks.16/self_attn/Shape_2_output_0", "outputIndexes": +[ 1562 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1562 ] +, "main_type": "NONE", "name": "Shape2509", "outputIndexes": +[ 1563 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1562 ] +, "main_type": "NONE", "name": "Rank2511", "outputIndexes": +[ 1564 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1564, 1564 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2513", "outputIndexes": +[ 1565 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1565 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2514", "outputIndexes": +[ 1566 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1565, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2516", "outputIndexes": +[ 1567 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1567 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2517", "outputIndexes": +[ 1568 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1563, 1566, 1568, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2520", "outputIndexes": +[ 1569 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1569 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2521", "outputIndexes": +[ 1570 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1570 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2522", "outputIndexes": +[ 1571 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1571, 1570 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2523", "outputIndexes": +[ 1572 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1562, 1572, 1565 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.16/self_attn/Gather_2_output_0", "outputIndexes": +[ 1573 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1573, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Div_output_0", "outputIndexes": +[ 1574 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1574, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1575 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1560, 1575, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.16/self_attn/Slice_1_output_0", "outputIndexes": +[ 1576 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1576 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.16/self_attn/Neg_output_0", "outputIndexes": +[ 1577 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1574, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1578 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1560, 20, 1578, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.16/self_attn/Slice_output_0", "outputIndexes": +[ 1579 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1577, 1579 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.16/self_attn/Concat_3_output_0", "outputIndexes": +[ 1580 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1580, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Mul_1_output_0", "outputIndexes": +[ 1581 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1561, 1581 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Add_output_0", "outputIndexes": +[ 1582 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3846 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1940983222, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.16/self_attn/k_proj/Linear", "outputIndexes": +[ 3851 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3851 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3852 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3852 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1583 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1547, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1584 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1555, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1585 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1584, 1585, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.16/self_attn/Concat_1_output_0", "outputIndexes": +[ 1586 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1583, 1586 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.16/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1587 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1587 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1943604684, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.16/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1589 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1589, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Mul_2_output_0", "outputIndexes": +[ 1590 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1589 ] +, "main_type": "NONE", "name": "/blocks.16/self_attn/Shape_3_output_0", "outputIndexes": +[ 1591 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1591 ] +, "main_type": "NONE", "name": "Shape2551", "outputIndexes": +[ 1592 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1591 ] +, "main_type": "NONE", "name": "Rank2553", "outputIndexes": +[ 1593 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1593, 1593 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2555", "outputIndexes": +[ 1594 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1594 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2556", "outputIndexes": +[ 1595 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1594, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2558", "outputIndexes": +[ 1596 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1596 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2559", "outputIndexes": +[ 1597 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1592, 1595, 1597, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2562", "outputIndexes": +[ 1598 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1598 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2563", "outputIndexes": +[ 1599 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1599 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2564", "outputIndexes": +[ 1600 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1600, 1599 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2565", "outputIndexes": +[ 1601 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1591, 1601, 1594 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.16/self_attn/Gather_3_output_0", "outputIndexes": +[ 1602 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1602, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Div_1_output_0", "outputIndexes": +[ 1603 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1603, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1604 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1589, 1604, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.16/self_attn/Slice_3_output_0", "outputIndexes": +[ 1605 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1605 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.16/self_attn/Neg_1_output_0", "outputIndexes": +[ 1606 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1603, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1607 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1589, 20, 1607, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.16/self_attn/Slice_2_output_0", "outputIndexes": +[ 1608 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1606, 1608 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.16/self_attn/Concat_4_output_0", "outputIndexes": +[ 1609 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1609, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Mul_3_output_0", "outputIndexes": +[ 1610 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1590, 1610 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/self_attn/Add_1_output_0", "outputIndexes": +[ 1611 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3846 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1943605708, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.16/self_attn/v_proj/Linear", "outputIndexes": +[ 3855 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3855 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3856 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3856 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1612 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1547, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1613 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1555, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.16/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1614 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1613, 1614, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.16/self_attn/Concat_2_output_0", "outputIndexes": +[ 1615 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1612, 1615 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.16/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1616 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1582, 1611, 1616, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.16/self_attn/FusedAttention", "outputIndexes": +[ 1617 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1617 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3857 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3857 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3858 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3858 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1946227170, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.16/self_attn/o_proj/Linear", "outputIndexes": +[ 3859 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3859 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3860 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3860 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1618 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1534, 1618 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/Add_output_0", "outputIndexes": +[ 1619 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1619 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1956712952, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.16/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1620 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1620 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3861 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3861 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3862 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3862 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1956745720, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.16/mlp/gate_proj/Linear", "outputIndexes": +[ 3863 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3863 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3864 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3864 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1621 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1621 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.16/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1622 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3862 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 1988203022, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.16/mlp/up_proj/Linear", "outputIndexes": +[ 3867 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3867 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3868 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3868 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1623 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1622, 1623 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/mlp/Mul_output_0", "outputIndexes": +[ 1624 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1624 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3869 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3869 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3870 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3870 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2019660324, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.16/mlp/down_proj/Linear", "outputIndexes": +[ 3871 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3871 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3872 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3872 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1625 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1619, 1625 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.16/Add_1_output_0", "outputIndexes": +[ 1626 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1626, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.17/Reshape_output_0", "outputIndexes": +[ 1627 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1627 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2051117626, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.17/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1628 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1628 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3873 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3873 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3874 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3874 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2051150394, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.17/self_attn/q_proj/Linear", "outputIndexes": +[ 3875 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3875 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3876 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3876 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1629 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1628 ] +, "main_type": "NONE", "name": "/blocks.17/self_attn/Shape_output_0", "outputIndexes": +[ 1630 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1630 ] +, "main_type": "NONE", "name": "Shape2612", "outputIndexes": +[ 1631 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1630 ] +, "main_type": "NONE", "name": "Rank2614", "outputIndexes": +[ 1632 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1632, 1632 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2616", "outputIndexes": +[ 1633 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1633 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2617", "outputIndexes": +[ 1634 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1633, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2619", "outputIndexes": +[ 1635 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1635 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2620", "outputIndexes": +[ 1636 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1631, 1634, 1636, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2623", "outputIndexes": +[ 1637 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1637 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2624", "outputIndexes": +[ 1638 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1638, 1638 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2626", "outputIndexes": +[ 1639 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1630, 1639, 1633 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.17/self_attn/Gather_output_0", "outputIndexes": +[ 1640 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1640, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1641 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1633 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2634", "outputIndexes": +[ 1642 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1635 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2637", "outputIndexes": +[ 1643 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1631, 1642, 1643, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2640", "outputIndexes": +[ 1644 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1644 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2641", "outputIndexes": +[ 1645 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1645 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2642", "outputIndexes": +[ 1646 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1646, 1645 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2643", "outputIndexes": +[ 1647 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1630, 1647, 1633 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.17/self_attn/Gather_1_output_0", "outputIndexes": +[ 1648 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1648, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1649 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1641, 1649, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.17/self_attn/Concat_output_0", "outputIndexes": +[ 1650 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1629, 1650 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.17/self_attn/Reshape_output_0", "outputIndexes": +[ 1651 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1651 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2061636176, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.17/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1653 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1653, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Mul_output_0", "outputIndexes": +[ 1654 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1653 ] +, "main_type": "NONE", "name": "/blocks.17/self_attn/Shape_2_output_0", "outputIndexes": +[ 1655 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1655 ] +, "main_type": "NONE", "name": "Shape2658", "outputIndexes": +[ 1656 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1655 ] +, "main_type": "NONE", "name": "Rank2660", "outputIndexes": +[ 1657 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1657, 1657 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2662", "outputIndexes": +[ 1658 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1658 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2663", "outputIndexes": +[ 1659 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1658, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2665", "outputIndexes": +[ 1660 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1660 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2666", "outputIndexes": +[ 1661 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1656, 1659, 1661, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2669", "outputIndexes": +[ 1662 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1662 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2670", "outputIndexes": +[ 1663 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1663 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2671", "outputIndexes": +[ 1664 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1664, 1663 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2672", "outputIndexes": +[ 1665 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1655, 1665, 1658 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.17/self_attn/Gather_2_output_0", "outputIndexes": +[ 1666 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1666, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Div_output_0", "outputIndexes": +[ 1667 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1667, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1668 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1653, 1668, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.17/self_attn/Slice_1_output_0", "outputIndexes": +[ 1669 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1669 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.17/self_attn/Neg_output_0", "outputIndexes": +[ 1670 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1667, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1671 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1653, 20, 1671, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.17/self_attn/Slice_output_0", "outputIndexes": +[ 1672 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1670, 1672 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.17/self_attn/Concat_3_output_0", "outputIndexes": +[ 1673 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1673, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Mul_1_output_0", "outputIndexes": +[ 1674 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1654, 1674 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Add_output_0", "outputIndexes": +[ 1675 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3874 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2061637200, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.17/self_attn/k_proj/Linear", "outputIndexes": +[ 3879 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3879 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3880 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3880 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1676 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1640, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1677 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1648, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1678 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1677, 1678, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.17/self_attn/Concat_1_output_0", "outputIndexes": +[ 1679 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1676, 1679 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.17/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1680 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1680 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2064258662, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.17/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1682 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1682, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Mul_2_output_0", "outputIndexes": +[ 1683 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1682 ] +, "main_type": "NONE", "name": "/blocks.17/self_attn/Shape_3_output_0", "outputIndexes": +[ 1684 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1684 ] +, "main_type": "NONE", "name": "Shape2700", "outputIndexes": +[ 1685 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1684 ] +, "main_type": "NONE", "name": "Rank2702", "outputIndexes": +[ 1686 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1686, 1686 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2704", "outputIndexes": +[ 1687 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1687 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2705", "outputIndexes": +[ 1688 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1687, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2707", "outputIndexes": +[ 1689 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1689 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2708", "outputIndexes": +[ 1690 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1685, 1688, 1690, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2711", "outputIndexes": +[ 1691 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1691 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2712", "outputIndexes": +[ 1692 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1692 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2713", "outputIndexes": +[ 1693 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1693, 1692 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2714", "outputIndexes": +[ 1694 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1684, 1694, 1687 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.17/self_attn/Gather_3_output_0", "outputIndexes": +[ 1695 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1695, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Div_1_output_0", "outputIndexes": +[ 1696 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1696, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1697 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1682, 1697, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.17/self_attn/Slice_3_output_0", "outputIndexes": +[ 1698 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1698 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.17/self_attn/Neg_1_output_0", "outputIndexes": +[ 1699 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1696, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1700 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1682, 20, 1700, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.17/self_attn/Slice_2_output_0", "outputIndexes": +[ 1701 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1699, 1701 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.17/self_attn/Concat_4_output_0", "outputIndexes": +[ 1702 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1702, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Mul_3_output_0", "outputIndexes": +[ 1703 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1683, 1703 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/self_attn/Add_1_output_0", "outputIndexes": +[ 1704 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3874 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2064259686, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.17/self_attn/v_proj/Linear", "outputIndexes": +[ 3883 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3883 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3884 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3884 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1705 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1640, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1706 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1648, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.17/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1707 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1706, 1707, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.17/self_attn/Concat_2_output_0", "outputIndexes": +[ 1708 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1705, 1708 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.17/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1709 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1675, 1704, 1709, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.17/self_attn/FusedAttention", "outputIndexes": +[ 1710 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1710 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3885 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3885 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3886 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3886 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2066881148, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.17/self_attn/o_proj/Linear", "outputIndexes": +[ 3887 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3887 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3888 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3888 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1711 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1627, 1711 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/Add_output_0", "outputIndexes": +[ 1712 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1712 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2077366930, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.17/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1713 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1713 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3889 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3889 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3890 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3890 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2077399698, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.17/mlp/gate_proj/Linear", "outputIndexes": +[ 3891 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3891 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3892 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3892 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1714 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1714 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.17/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1715 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3890 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2108857000, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.17/mlp/up_proj/Linear", "outputIndexes": +[ 3895 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3895 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3896 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3896 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1716 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1715, 1716 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/mlp/Mul_output_0", "outputIndexes": +[ 1717 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1717 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3897 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3897 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3898 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3898 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2140314302, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.17/mlp/down_proj/Linear", "outputIndexes": +[ 3899 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3899 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3900 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3900 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1718 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1712, 1718 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.17/Add_1_output_0", "outputIndexes": +[ 1719 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1719, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.18/Reshape_output_0", "outputIndexes": +[ 1720 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1720 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2171771604, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.18/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1721 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1721 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3901 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3901 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3902 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3902 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2171804372, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.18/self_attn/q_proj/Linear", "outputIndexes": +[ 3903 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3903 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3904 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3904 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1722 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1721 ] +, "main_type": "NONE", "name": "/blocks.18/self_attn/Shape_output_0", "outputIndexes": +[ 1723 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1723 ] +, "main_type": "NONE", "name": "Shape2761", "outputIndexes": +[ 1724 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1723 ] +, "main_type": "NONE", "name": "Rank2763", "outputIndexes": +[ 1725 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1725, 1725 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2765", "outputIndexes": +[ 1726 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1726 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2766", "outputIndexes": +[ 1727 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1726, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2768", "outputIndexes": +[ 1728 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1728 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2769", "outputIndexes": +[ 1729 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1724, 1727, 1729, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2772", "outputIndexes": +[ 1730 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1730 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2773", "outputIndexes": +[ 1731 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1731, 1731 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2775", "outputIndexes": +[ 1732 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1723, 1732, 1726 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.18/self_attn/Gather_output_0", "outputIndexes": +[ 1733 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1733, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1734 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1726 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2783", "outputIndexes": +[ 1735 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1728 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2786", "outputIndexes": +[ 1736 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1724, 1735, 1736, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2789", "outputIndexes": +[ 1737 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1737 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2790", "outputIndexes": +[ 1738 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1738 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2791", "outputIndexes": +[ 1739 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1739, 1738 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2792", "outputIndexes": +[ 1740 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1723, 1740, 1726 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.18/self_attn/Gather_1_output_0", "outputIndexes": +[ 1741 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1741, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1742 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1734, 1742, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.18/self_attn/Concat_output_0", "outputIndexes": +[ 1743 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1722, 1743 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.18/self_attn/Reshape_output_0", "outputIndexes": +[ 1744 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1744 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2182290154, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.18/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1746 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1746, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Mul_output_0", "outputIndexes": +[ 1747 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1746 ] +, "main_type": "NONE", "name": "/blocks.18/self_attn/Shape_2_output_0", "outputIndexes": +[ 1748 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1748 ] +, "main_type": "NONE", "name": "Shape2807", "outputIndexes": +[ 1749 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1748 ] +, "main_type": "NONE", "name": "Rank2809", "outputIndexes": +[ 1750 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1750, 1750 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2811", "outputIndexes": +[ 1751 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1751 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2812", "outputIndexes": +[ 1752 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1751, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2814", "outputIndexes": +[ 1753 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1753 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2815", "outputIndexes": +[ 1754 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1749, 1752, 1754, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2818", "outputIndexes": +[ 1755 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1755 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2819", "outputIndexes": +[ 1756 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1756 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2820", "outputIndexes": +[ 1757 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1757, 1756 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2821", "outputIndexes": +[ 1758 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1748, 1758, 1751 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.18/self_attn/Gather_2_output_0", "outputIndexes": +[ 1759 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1759, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Div_output_0", "outputIndexes": +[ 1760 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1760, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1761 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1746, 1761, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.18/self_attn/Slice_1_output_0", "outputIndexes": +[ 1762 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1762 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.18/self_attn/Neg_output_0", "outputIndexes": +[ 1763 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1760, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1764 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1746, 20, 1764, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.18/self_attn/Slice_output_0", "outputIndexes": +[ 1765 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1763, 1765 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.18/self_attn/Concat_3_output_0", "outputIndexes": +[ 1766 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1766, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Mul_1_output_0", "outputIndexes": +[ 1767 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1747, 1767 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Add_output_0", "outputIndexes": +[ 1768 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3902 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2182291178, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.18/self_attn/k_proj/Linear", "outputIndexes": +[ 3907 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3907 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3908 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3908 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1769 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1733, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1770 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1741, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1771 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1770, 1771, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.18/self_attn/Concat_1_output_0", "outputIndexes": +[ 1772 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1769, 1772 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.18/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1773 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1773 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2184912640, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.18/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1775 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1775, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Mul_2_output_0", "outputIndexes": +[ 1776 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1775 ] +, "main_type": "NONE", "name": "/blocks.18/self_attn/Shape_3_output_0", "outputIndexes": +[ 1777 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1777 ] +, "main_type": "NONE", "name": "Shape2849", "outputIndexes": +[ 1778 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1777 ] +, "main_type": "NONE", "name": "Rank2851", "outputIndexes": +[ 1779 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1779, 1779 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2853", "outputIndexes": +[ 1780 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1780 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2854", "outputIndexes": +[ 1781 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1780, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2856", "outputIndexes": +[ 1782 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1782 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2857", "outputIndexes": +[ 1783 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1778, 1781, 1783, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2860", "outputIndexes": +[ 1784 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1784 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2861", "outputIndexes": +[ 1785 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1785 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2862", "outputIndexes": +[ 1786 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1786, 1785 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2863", "outputIndexes": +[ 1787 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1777, 1787, 1780 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.18/self_attn/Gather_3_output_0", "outputIndexes": +[ 1788 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1788, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Div_1_output_0", "outputIndexes": +[ 1789 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1789, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1790 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1775, 1790, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.18/self_attn/Slice_3_output_0", "outputIndexes": +[ 1791 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1791 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.18/self_attn/Neg_1_output_0", "outputIndexes": +[ 1792 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1789, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1793 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1775, 20, 1793, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.18/self_attn/Slice_2_output_0", "outputIndexes": +[ 1794 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1792, 1794 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.18/self_attn/Concat_4_output_0", "outputIndexes": +[ 1795 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1795, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Mul_3_output_0", "outputIndexes": +[ 1796 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1776, 1796 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/self_attn/Add_1_output_0", "outputIndexes": +[ 1797 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3902 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2184913664, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.18/self_attn/v_proj/Linear", "outputIndexes": +[ 3911 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3911 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3912 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3912 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1798 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1733, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1799 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1741, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.18/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1800 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1799, 1800, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.18/self_attn/Concat_2_output_0", "outputIndexes": +[ 1801 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1798, 1801 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.18/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1802 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1768, 1797, 1802, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.18/self_attn/FusedAttention", "outputIndexes": +[ 1803 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1803 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3913 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3913 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3914 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3914 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2187535126, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.18/self_attn/o_proj/Linear", "outputIndexes": +[ 3915 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3915 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3916 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3916 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1804 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1720, 1804 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/Add_output_0", "outputIndexes": +[ 1805 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1805 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2198020908, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.18/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1806 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1806 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3917 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3917 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3918 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3918 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2198053676, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.18/mlp/gate_proj/Linear", "outputIndexes": +[ 3919 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3919 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3920 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3920 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1807 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1807 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.18/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1808 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3918 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2229510978, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.18/mlp/up_proj/Linear", "outputIndexes": +[ 3923 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3923 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3924 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3924 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1809 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1808, 1809 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/mlp/Mul_output_0", "outputIndexes": +[ 1810 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1810 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3925 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3925 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3926 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3926 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2260968280, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.18/mlp/down_proj/Linear", "outputIndexes": +[ 3927 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3927 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3928 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3928 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1811 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1805, 1811 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.18/Add_1_output_0", "outputIndexes": +[ 1812 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1812, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.19/Reshape_output_0", "outputIndexes": +[ 1813 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1813 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2292425582, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.19/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1814 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1814 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3929 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3929 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3930 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3930 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2292458350, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.19/self_attn/q_proj/Linear", "outputIndexes": +[ 3931 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3931 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3932 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3932 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1815 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1814 ] +, "main_type": "NONE", "name": "/blocks.19/self_attn/Shape_output_0", "outputIndexes": +[ 1816 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1816 ] +, "main_type": "NONE", "name": "Shape2910", "outputIndexes": +[ 1817 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1816 ] +, "main_type": "NONE", "name": "Rank2912", "outputIndexes": +[ 1818 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1818, 1818 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2914", "outputIndexes": +[ 1819 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1819 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2915", "outputIndexes": +[ 1820 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1819, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2917", "outputIndexes": +[ 1821 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1821 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2918", "outputIndexes": +[ 1822 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1817, 1820, 1822, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2921", "outputIndexes": +[ 1823 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1823 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2922", "outputIndexes": +[ 1824 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1824, 1824 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2924", "outputIndexes": +[ 1825 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1816, 1825, 1819 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.19/self_attn/Gather_output_0", "outputIndexes": +[ 1826 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1826, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1827 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1819 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2932", "outputIndexes": +[ 1828 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1821 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2935", "outputIndexes": +[ 1829 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1817, 1828, 1829, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2938", "outputIndexes": +[ 1830 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1830 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2939", "outputIndexes": +[ 1831 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1831 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2940", "outputIndexes": +[ 1832 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1832, 1831 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2941", "outputIndexes": +[ 1833 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1816, 1833, 1819 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.19/self_attn/Gather_1_output_0", "outputIndexes": +[ 1834 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1834, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1835 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1827, 1835, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.19/self_attn/Concat_output_0", "outputIndexes": +[ 1836 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1815, 1836 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.19/self_attn/Reshape_output_0", "outputIndexes": +[ 1837 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1837 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2302944132, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.19/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1839 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1839, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Mul_output_0", "outputIndexes": +[ 1840 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1839 ] +, "main_type": "NONE", "name": "/blocks.19/self_attn/Shape_2_output_0", "outputIndexes": +[ 1841 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1841 ] +, "main_type": "NONE", "name": "Shape2956", "outputIndexes": +[ 1842 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1841 ] +, "main_type": "NONE", "name": "Rank2958", "outputIndexes": +[ 1843 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1843, 1843 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2960", "outputIndexes": +[ 1844 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1844 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2961", "outputIndexes": +[ 1845 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1844, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2963", "outputIndexes": +[ 1846 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1846 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2964", "outputIndexes": +[ 1847 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1842, 1845, 1847, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2967", "outputIndexes": +[ 1848 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1848 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2968", "outputIndexes": +[ 1849 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1849 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2969", "outputIndexes": +[ 1850 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1850, 1849 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2970", "outputIndexes": +[ 1851 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1841, 1851, 1844 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.19/self_attn/Gather_2_output_0", "outputIndexes": +[ 1852 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1852, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Div_output_0", "outputIndexes": +[ 1853 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1853, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1854 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1839, 1854, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.19/self_attn/Slice_1_output_0", "outputIndexes": +[ 1855 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1855 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.19/self_attn/Neg_output_0", "outputIndexes": +[ 1856 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1853, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1857 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1839, 20, 1857, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.19/self_attn/Slice_output_0", "outputIndexes": +[ 1858 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1856, 1858 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.19/self_attn/Concat_3_output_0", "outputIndexes": +[ 1859 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1859, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Mul_1_output_0", "outputIndexes": +[ 1860 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1840, 1860 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Add_output_0", "outputIndexes": +[ 1861 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3930 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2302945156, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.19/self_attn/k_proj/Linear", "outputIndexes": +[ 3935 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3935 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3936 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3936 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1862 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1826, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1863 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1834, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1864 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1863, 1864, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.19/self_attn/Concat_1_output_0", "outputIndexes": +[ 1865 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1862, 1865 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.19/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1866 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1866 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2305566618, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.19/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1868 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1868, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Mul_2_output_0", "outputIndexes": +[ 1869 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1868 ] +, "main_type": "NONE", "name": "/blocks.19/self_attn/Shape_3_output_0", "outputIndexes": +[ 1870 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1870 ] +, "main_type": "NONE", "name": "Shape2998", "outputIndexes": +[ 1871 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1870 ] +, "main_type": "NONE", "name": "Rank3000", "outputIndexes": +[ 1872 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1872, 1872 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3002", "outputIndexes": +[ 1873 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1873 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3003", "outputIndexes": +[ 1874 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1873, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3005", "outputIndexes": +[ 1875 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1875 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3006", "outputIndexes": +[ 1876 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1871, 1874, 1876, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3009", "outputIndexes": +[ 1877 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1877 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3010", "outputIndexes": +[ 1878 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1878 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3011", "outputIndexes": +[ 1879 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1879, 1878 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3012", "outputIndexes": +[ 1880 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1870, 1880, 1873 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.19/self_attn/Gather_3_output_0", "outputIndexes": +[ 1881 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1881, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Div_1_output_0", "outputIndexes": +[ 1882 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1882, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1883 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1868, 1883, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.19/self_attn/Slice_3_output_0", "outputIndexes": +[ 1884 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1884 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.19/self_attn/Neg_1_output_0", "outputIndexes": +[ 1885 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1882, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1886 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1868, 20, 1886, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.19/self_attn/Slice_2_output_0", "outputIndexes": +[ 1887 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1885, 1887 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.19/self_attn/Concat_4_output_0", "outputIndexes": +[ 1888 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1888, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Mul_3_output_0", "outputIndexes": +[ 1889 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1869, 1889 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/self_attn/Add_1_output_0", "outputIndexes": +[ 1890 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3930 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2305567642, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.19/self_attn/v_proj/Linear", "outputIndexes": +[ 3939 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3939 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3940 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3940 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1891 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1826, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1892 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1834, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.19/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1893 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1892, 1893, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.19/self_attn/Concat_2_output_0", "outputIndexes": +[ 1894 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1891, 1894 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.19/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1895 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1861, 1890, 1895, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.19/self_attn/FusedAttention", "outputIndexes": +[ 1896 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1896 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3941 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3941 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3942 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3942 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2308189104, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.19/self_attn/o_proj/Linear", "outputIndexes": +[ 3943 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3943 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3944 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3944 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1897 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1813, 1897 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/Add_output_0", "outputIndexes": +[ 1898 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1898 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2318674886, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.19/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1899 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1899 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3945 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3945 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3946 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3946 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2318707654, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.19/mlp/gate_proj/Linear", "outputIndexes": +[ 3947 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3947 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3948 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3948 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1900 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1900 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.19/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1901 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3946 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2350164956, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.19/mlp/up_proj/Linear", "outputIndexes": +[ 3951 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3951 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3952 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3952 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1902 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1901, 1902 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/mlp/Mul_output_0", "outputIndexes": +[ 1903 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1903 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3953 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3953 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3954 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3954 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2381622258, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.19/mlp/down_proj/Linear", "outputIndexes": +[ 3955 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3955 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3956 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3956 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1904 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1898, 1904 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.19/Add_1_output_0", "outputIndexes": +[ 1905 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1905, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.20/Reshape_output_0", "outputIndexes": +[ 1906 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1906 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2413079560, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.20/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 1907 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1907 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3957 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3957 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3958 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3958 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2413112328, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.20/self_attn/q_proj/Linear", "outputIndexes": +[ 3959 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3959 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3960 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3960 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1908 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1907 ] +, "main_type": "NONE", "name": "/blocks.20/self_attn/Shape_output_0", "outputIndexes": +[ 1909 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1909 ] +, "main_type": "NONE", "name": "Shape3059", "outputIndexes": +[ 1910 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1909 ] +, "main_type": "NONE", "name": "Rank3061", "outputIndexes": +[ 1911 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1911, 1911 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3063", "outputIndexes": +[ 1912 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1912 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3064", "outputIndexes": +[ 1913 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1912, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3066", "outputIndexes": +[ 1914 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1914 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3067", "outputIndexes": +[ 1915 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1910, 1913, 1915, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3070", "outputIndexes": +[ 1916 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1916 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3071", "outputIndexes": +[ 1917 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1917, 1917 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3073", "outputIndexes": +[ 1918 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1909, 1918, 1912 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.20/self_attn/Gather_output_0", "outputIndexes": +[ 1919 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1919, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 1920 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1912 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3081", "outputIndexes": +[ 1921 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1914 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3084", "outputIndexes": +[ 1922 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1910, 1921, 1922, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3087", "outputIndexes": +[ 1923 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1923 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3088", "outputIndexes": +[ 1924 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 1924 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3089", "outputIndexes": +[ 1925 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1925, 1924 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3090", "outputIndexes": +[ 1926 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1909, 1926, 1912 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.20/self_attn/Gather_1_output_0", "outputIndexes": +[ 1927 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1927, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 1928 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1920, 1928, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.20/self_attn/Concat_output_0", "outputIndexes": +[ 1929 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1908, 1929 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.20/self_attn/Reshape_output_0", "outputIndexes": +[ 1930 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1930 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2423598110, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.20/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 1932 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1932, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Mul_output_0", "outputIndexes": +[ 1933 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1932 ] +, "main_type": "NONE", "name": "/blocks.20/self_attn/Shape_2_output_0", "outputIndexes": +[ 1934 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1934 ] +, "main_type": "NONE", "name": "Shape3105", "outputIndexes": +[ 1935 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1934 ] +, "main_type": "NONE", "name": "Rank3107", "outputIndexes": +[ 1936 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1936, 1936 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3109", "outputIndexes": +[ 1937 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1937 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3110", "outputIndexes": +[ 1938 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1937, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3112", "outputIndexes": +[ 1939 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1939 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3113", "outputIndexes": +[ 1940 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1935, 1938, 1940, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3116", "outputIndexes": +[ 1941 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1941 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3117", "outputIndexes": +[ 1942 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1942 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3118", "outputIndexes": +[ 1943 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1943, 1942 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3119", "outputIndexes": +[ 1944 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1934, 1944, 1937 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.20/self_attn/Gather_2_output_0", "outputIndexes": +[ 1945 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1945, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Div_output_0", "outputIndexes": +[ 1946 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1946, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 1947 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1932, 1947, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.20/self_attn/Slice_1_output_0", "outputIndexes": +[ 1948 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1948 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.20/self_attn/Neg_output_0", "outputIndexes": +[ 1949 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1946, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 1950 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1932, 20, 1950, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.20/self_attn/Slice_output_0", "outputIndexes": +[ 1951 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1949, 1951 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.20/self_attn/Concat_3_output_0", "outputIndexes": +[ 1952 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1952, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Mul_1_output_0", "outputIndexes": +[ 1953 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1933, 1953 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Add_output_0", "outputIndexes": +[ 1954 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3958 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2423599134, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.20/self_attn/k_proj/Linear", "outputIndexes": +[ 3963 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3963 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3964 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3964 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1955 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1919, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 1956 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1927, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 1957 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1956, 1957, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.20/self_attn/Concat_1_output_0", "outputIndexes": +[ 1958 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1955, 1958 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.20/self_attn/Reshape_1_output_0", "outputIndexes": +[ 1959 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1959 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2426220596, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.20/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 1961 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1961, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Mul_2_output_0", "outputIndexes": +[ 1962 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1961 ] +, "main_type": "NONE", "name": "/blocks.20/self_attn/Shape_3_output_0", "outputIndexes": +[ 1963 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1963 ] +, "main_type": "NONE", "name": "Shape3147", "outputIndexes": +[ 1964 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1963 ] +, "main_type": "NONE", "name": "Rank3149", "outputIndexes": +[ 1965 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1965, 1965 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3151", "outputIndexes": +[ 1966 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1966 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3152", "outputIndexes": +[ 1967 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1966, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3154", "outputIndexes": +[ 1968 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1968 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3155", "outputIndexes": +[ 1969 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1964, 1967, 1969, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3158", "outputIndexes": +[ 1970 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1970 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3159", "outputIndexes": +[ 1971 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 1971 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3160", "outputIndexes": +[ 1972 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1972, 1971 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3161", "outputIndexes": +[ 1973 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1963, 1973, 1966 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.20/self_attn/Gather_3_output_0", "outputIndexes": +[ 1974 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1974, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Div_1_output_0", "outputIndexes": +[ 1975 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1975, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 1976 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1961, 1976, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.20/self_attn/Slice_3_output_0", "outputIndexes": +[ 1977 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1977 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.20/self_attn/Neg_1_output_0", "outputIndexes": +[ 1978 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1975, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 1979 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1961, 20, 1979, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.20/self_attn/Slice_2_output_0", "outputIndexes": +[ 1980 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1978, 1980 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.20/self_attn/Concat_4_output_0", "outputIndexes": +[ 1981 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1981, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Mul_3_output_0", "outputIndexes": +[ 1982 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1962, 1982 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/self_attn/Add_1_output_0", "outputIndexes": +[ 1983 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3958 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2426221620, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.20/self_attn/v_proj/Linear", "outputIndexes": +[ 3967 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3967 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3968 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3968 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1984 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1919, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 1985 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1927, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.20/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 1986 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1985, 1986, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.20/self_attn/Concat_2_output_0", "outputIndexes": +[ 1987 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1984, 1987 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.20/self_attn/Reshape_2_output_0", "outputIndexes": +[ 1988 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1954, 1983, 1988, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.20/self_attn/FusedAttention", "outputIndexes": +[ 1989 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1989 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3969 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3969 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3970 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3970 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2428843082, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.20/self_attn/o_proj/Linear", "outputIndexes": +[ 3971 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3971 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3972 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3972 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1990 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1906, 1990 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/Add_output_0", "outputIndexes": +[ 1991 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1991 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2439328864, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.20/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 1992 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1992 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3973 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3973 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3974 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3974 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2439361632, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.20/mlp/gate_proj/Linear", "outputIndexes": +[ 3975 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3975 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3976 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3976 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1993 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1993 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.20/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 1994 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3974 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2470818934, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.20/mlp/up_proj/Linear", "outputIndexes": +[ 3979 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3979 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3980 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3980 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1995 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1994, 1995 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/mlp/Mul_output_0", "outputIndexes": +[ 1996 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1996 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3981 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3981 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3982 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3982 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2502276236, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.20/mlp/down_proj/Linear", "outputIndexes": +[ 3983 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3983 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3984 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3984 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1997 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1991, 1997 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.20/Add_1_output_0", "outputIndexes": +[ 1998 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1998, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.21/Reshape_output_0", "outputIndexes": +[ 1999 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1999 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2533733538, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.21/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2000 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2000 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3985 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3985 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3986 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3986 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2533766306, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.21/self_attn/q_proj/Linear", "outputIndexes": +[ 3987 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3987 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3988 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3988 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2001 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2000 ] +, "main_type": "NONE", "name": "/blocks.21/self_attn/Shape_output_0", "outputIndexes": +[ 2002 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2002 ] +, "main_type": "NONE", "name": "Shape3208", "outputIndexes": +[ 2003 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2002 ] +, "main_type": "NONE", "name": "Rank3210", "outputIndexes": +[ 2004 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2004, 2004 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3212", "outputIndexes": +[ 2005 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2005 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3213", "outputIndexes": +[ 2006 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2005, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3215", "outputIndexes": +[ 2007 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2007 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3216", "outputIndexes": +[ 2008 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2003, 2006, 2008, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3219", "outputIndexes": +[ 2009 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2009 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3220", "outputIndexes": +[ 2010 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2010, 2010 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3222", "outputIndexes": +[ 2011 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2002, 2011, 2005 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.21/self_attn/Gather_output_0", "outputIndexes": +[ 2012 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2012, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2013 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2005 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3230", "outputIndexes": +[ 2014 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2007 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3233", "outputIndexes": +[ 2015 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2003, 2014, 2015, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3236", "outputIndexes": +[ 2016 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2016 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3237", "outputIndexes": +[ 2017 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2017 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3238", "outputIndexes": +[ 2018 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2018, 2017 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3239", "outputIndexes": +[ 2019 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2002, 2019, 2005 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.21/self_attn/Gather_1_output_0", "outputIndexes": +[ 2020 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2020, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2021 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2013, 2021, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.21/self_attn/Concat_output_0", "outputIndexes": +[ 2022 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2001, 2022 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.21/self_attn/Reshape_output_0", "outputIndexes": +[ 2023 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2023 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2544252088, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.21/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2025 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2025, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Mul_output_0", "outputIndexes": +[ 2026 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2025 ] +, "main_type": "NONE", "name": "/blocks.21/self_attn/Shape_2_output_0", "outputIndexes": +[ 2027 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2027 ] +, "main_type": "NONE", "name": "Shape3254", "outputIndexes": +[ 2028 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2027 ] +, "main_type": "NONE", "name": "Rank3256", "outputIndexes": +[ 2029 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2029, 2029 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3258", "outputIndexes": +[ 2030 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2030 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3259", "outputIndexes": +[ 2031 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2030, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3261", "outputIndexes": +[ 2032 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2032 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3262", "outputIndexes": +[ 2033 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2028, 2031, 2033, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3265", "outputIndexes": +[ 2034 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2034 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3266", "outputIndexes": +[ 2035 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2035 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3267", "outputIndexes": +[ 2036 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2036, 2035 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3268", "outputIndexes": +[ 2037 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2027, 2037, 2030 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.21/self_attn/Gather_2_output_0", "outputIndexes": +[ 2038 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2038, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Div_output_0", "outputIndexes": +[ 2039 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2039, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2040 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2025, 2040, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.21/self_attn/Slice_1_output_0", "outputIndexes": +[ 2041 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2041 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.21/self_attn/Neg_output_0", "outputIndexes": +[ 2042 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2039, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2043 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2025, 20, 2043, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.21/self_attn/Slice_output_0", "outputIndexes": +[ 2044 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2042, 2044 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.21/self_attn/Concat_3_output_0", "outputIndexes": +[ 2045 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2045, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Mul_1_output_0", "outputIndexes": +[ 2046 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2026, 2046 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Add_output_0", "outputIndexes": +[ 2047 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3986 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2544253112, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.21/self_attn/k_proj/Linear", "outputIndexes": +[ 3991 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3991 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3992 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3992 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2048 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2012, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2049 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2020, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2050 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2049, 2050, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.21/self_attn/Concat_1_output_0", "outputIndexes": +[ 2051 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2048, 2051 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.21/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2052 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2052 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2546874574, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.21/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2054 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2054, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Mul_2_output_0", "outputIndexes": +[ 2055 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2054 ] +, "main_type": "NONE", "name": "/blocks.21/self_attn/Shape_3_output_0", "outputIndexes": +[ 2056 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2056 ] +, "main_type": "NONE", "name": "Shape3296", "outputIndexes": +[ 2057 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2056 ] +, "main_type": "NONE", "name": "Rank3298", "outputIndexes": +[ 2058 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2058, 2058 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3300", "outputIndexes": +[ 2059 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2059 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3301", "outputIndexes": +[ 2060 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2059, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3303", "outputIndexes": +[ 2061 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2061 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3304", "outputIndexes": +[ 2062 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2057, 2060, 2062, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3307", "outputIndexes": +[ 2063 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2063 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3308", "outputIndexes": +[ 2064 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2064 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3309", "outputIndexes": +[ 2065 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2065, 2064 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3310", "outputIndexes": +[ 2066 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2056, 2066, 2059 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.21/self_attn/Gather_3_output_0", "outputIndexes": +[ 2067 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2067, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Div_1_output_0", "outputIndexes": +[ 2068 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2068, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2069 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2054, 2069, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.21/self_attn/Slice_3_output_0", "outputIndexes": +[ 2070 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2070 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.21/self_attn/Neg_1_output_0", "outputIndexes": +[ 2071 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2068, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2072 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2054, 20, 2072, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.21/self_attn/Slice_2_output_0", "outputIndexes": +[ 2073 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2071, 2073 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.21/self_attn/Concat_4_output_0", "outputIndexes": +[ 2074 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2074, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Mul_3_output_0", "outputIndexes": +[ 2075 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2055, 2075 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/self_attn/Add_1_output_0", "outputIndexes": +[ 2076 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3986 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2546875598, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.21/self_attn/v_proj/Linear", "outputIndexes": +[ 3995 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3995 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3996 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3996 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2077 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2012, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2078 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2020, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.21/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2079 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2078, 2079, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.21/self_attn/Concat_2_output_0", "outputIndexes": +[ 2080 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2077, 2080 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.21/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2081 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2047, 2076, 2081, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.21/self_attn/FusedAttention", "outputIndexes": +[ 2082 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2082 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3997 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3997 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3998 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3998 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2549497060, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.21/self_attn/o_proj/Linear", "outputIndexes": +[ 3999 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3999 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4000 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4000 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2083 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1999, 2083 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/Add_output_0", "outputIndexes": +[ 2084 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2084 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2559982842, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.21/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2085 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2085 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4001 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4001 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4002 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4002 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2560015610, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.21/mlp/gate_proj/Linear", "outputIndexes": +[ 4003 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4003 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4004 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4004 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2086 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2086 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.21/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2087 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4002 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2591472912, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.21/mlp/up_proj/Linear", "outputIndexes": +[ 4007 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4007 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4008 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4008 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2088 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2087, 2088 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/mlp/Mul_output_0", "outputIndexes": +[ 2089 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2089 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4009 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4009 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4010 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4010 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2622930214, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.21/mlp/down_proj/Linear", "outputIndexes": +[ 4011 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4011 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4012 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4012 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2090 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2084, 2090 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.21/Add_1_output_0", "outputIndexes": +[ 2091 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2091, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.22/Reshape_output_0", "outputIndexes": +[ 2092 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2092 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2654387516, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.22/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2093 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2093 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4013 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4013 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4014 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4014 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2654420284, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.22/self_attn/q_proj/Linear", "outputIndexes": +[ 4015 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4015 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4016 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4016 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2094 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2093 ] +, "main_type": "NONE", "name": "/blocks.22/self_attn/Shape_output_0", "outputIndexes": +[ 2095 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2095 ] +, "main_type": "NONE", "name": "Shape3357", "outputIndexes": +[ 2096 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2095 ] +, "main_type": "NONE", "name": "Rank3359", "outputIndexes": +[ 2097 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2097, 2097 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3361", "outputIndexes": +[ 2098 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2098 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3362", "outputIndexes": +[ 2099 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2098, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3364", "outputIndexes": +[ 2100 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2100 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3365", "outputIndexes": +[ 2101 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2096, 2099, 2101, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3368", "outputIndexes": +[ 2102 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2102 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3369", "outputIndexes": +[ 2103 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2103, 2103 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3371", "outputIndexes": +[ 2104 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2095, 2104, 2098 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.22/self_attn/Gather_output_0", "outputIndexes": +[ 2105 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2105, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2106 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2098 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3379", "outputIndexes": +[ 2107 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2100 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3382", "outputIndexes": +[ 2108 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2096, 2107, 2108, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3385", "outputIndexes": +[ 2109 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2109 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3386", "outputIndexes": +[ 2110 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2110 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3387", "outputIndexes": +[ 2111 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2111, 2110 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3388", "outputIndexes": +[ 2112 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2095, 2112, 2098 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.22/self_attn/Gather_1_output_0", "outputIndexes": +[ 2113 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2113, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2114 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2106, 2114, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.22/self_attn/Concat_output_0", "outputIndexes": +[ 2115 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2094, 2115 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.22/self_attn/Reshape_output_0", "outputIndexes": +[ 2116 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2116 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2664906066, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.22/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2118 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2118, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Mul_output_0", "outputIndexes": +[ 2119 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2118 ] +, "main_type": "NONE", "name": "/blocks.22/self_attn/Shape_2_output_0", "outputIndexes": +[ 2120 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2120 ] +, "main_type": "NONE", "name": "Shape3403", "outputIndexes": +[ 2121 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2120 ] +, "main_type": "NONE", "name": "Rank3405", "outputIndexes": +[ 2122 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2122, 2122 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3407", "outputIndexes": +[ 2123 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2123 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3408", "outputIndexes": +[ 2124 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2123, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3410", "outputIndexes": +[ 2125 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2125 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3411", "outputIndexes": +[ 2126 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2121, 2124, 2126, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3414", "outputIndexes": +[ 2127 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2127 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3415", "outputIndexes": +[ 2128 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2128 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3416", "outputIndexes": +[ 2129 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2129, 2128 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3417", "outputIndexes": +[ 2130 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2120, 2130, 2123 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.22/self_attn/Gather_2_output_0", "outputIndexes": +[ 2131 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2131, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Div_output_0", "outputIndexes": +[ 2132 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2132, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2133 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2118, 2133, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.22/self_attn/Slice_1_output_0", "outputIndexes": +[ 2134 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2134 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.22/self_attn/Neg_output_0", "outputIndexes": +[ 2135 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2132, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2136 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2118, 20, 2136, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.22/self_attn/Slice_output_0", "outputIndexes": +[ 2137 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2135, 2137 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.22/self_attn/Concat_3_output_0", "outputIndexes": +[ 2138 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2138, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Mul_1_output_0", "outputIndexes": +[ 2139 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2119, 2139 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Add_output_0", "outputIndexes": +[ 2140 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4014 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2664907090, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.22/self_attn/k_proj/Linear", "outputIndexes": +[ 4019 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4019 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4020 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4020 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2141 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2105, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2142 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2113, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2143 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2142, 2143, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.22/self_attn/Concat_1_output_0", "outputIndexes": +[ 2144 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2141, 2144 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.22/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2145 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2145 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2667528552, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.22/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2147 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2147, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Mul_2_output_0", "outputIndexes": +[ 2148 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2147 ] +, "main_type": "NONE", "name": "/blocks.22/self_attn/Shape_3_output_0", "outputIndexes": +[ 2149 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2149 ] +, "main_type": "NONE", "name": "Shape3445", "outputIndexes": +[ 2150 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2149 ] +, "main_type": "NONE", "name": "Rank3447", "outputIndexes": +[ 2151 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2151, 2151 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3449", "outputIndexes": +[ 2152 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2152 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3450", "outputIndexes": +[ 2153 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2152, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3452", "outputIndexes": +[ 2154 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2154 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3453", "outputIndexes": +[ 2155 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2150, 2153, 2155, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3456", "outputIndexes": +[ 2156 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2156 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3457", "outputIndexes": +[ 2157 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2157 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3458", "outputIndexes": +[ 2158 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2158, 2157 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3459", "outputIndexes": +[ 2159 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2149, 2159, 2152 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.22/self_attn/Gather_3_output_0", "outputIndexes": +[ 2160 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2160, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Div_1_output_0", "outputIndexes": +[ 2161 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2161, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2162 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2147, 2162, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.22/self_attn/Slice_3_output_0", "outputIndexes": +[ 2163 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2163 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.22/self_attn/Neg_1_output_0", "outputIndexes": +[ 2164 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2161, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2165 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2147, 20, 2165, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.22/self_attn/Slice_2_output_0", "outputIndexes": +[ 2166 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2164, 2166 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.22/self_attn/Concat_4_output_0", "outputIndexes": +[ 2167 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2167, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Mul_3_output_0", "outputIndexes": +[ 2168 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2148, 2168 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/self_attn/Add_1_output_0", "outputIndexes": +[ 2169 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4014 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2667529576, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.22/self_attn/v_proj/Linear", "outputIndexes": +[ 4023 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4023 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4024 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4024 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2170 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2105, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2171 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2113, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.22/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2172 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2171, 2172, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.22/self_attn/Concat_2_output_0", "outputIndexes": +[ 2173 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2170, 2173 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.22/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2174 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2140, 2169, 2174, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.22/self_attn/FusedAttention", "outputIndexes": +[ 2175 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2175 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4025 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4025 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4026 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4026 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2670151038, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.22/self_attn/o_proj/Linear", "outputIndexes": +[ 4027 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4027 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4028 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4028 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2176 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2092, 2176 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/Add_output_0", "outputIndexes": +[ 2177 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2177 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2680636820, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.22/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2178 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2178 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4029 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4029 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4030 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4030 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2680669588, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.22/mlp/gate_proj/Linear", "outputIndexes": +[ 4031 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4031 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4032 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4032 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2179 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2179 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.22/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2180 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4030 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2712126890, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.22/mlp/up_proj/Linear", "outputIndexes": +[ 4035 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4035 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4036 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4036 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2181 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2180, 2181 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/mlp/Mul_output_0", "outputIndexes": +[ 2182 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2182 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4037 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4037 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4038 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4038 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2743584192, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.22/mlp/down_proj/Linear", "outputIndexes": +[ 4039 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4039 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4040 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4040 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2183 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2177, 2183 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.22/Add_1_output_0", "outputIndexes": +[ 2184 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2184, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.23/Reshape_output_0", "outputIndexes": +[ 2185 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2185 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2775041494, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.23/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2186 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2186 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4041 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4041 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4042 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4042 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2775074262, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.23/self_attn/q_proj/Linear", "outputIndexes": +[ 4043 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4043 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4044 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4044 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2187 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2186 ] +, "main_type": "NONE", "name": "/blocks.23/self_attn/Shape_output_0", "outputIndexes": +[ 2188 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2188 ] +, "main_type": "NONE", "name": "Shape3506", "outputIndexes": +[ 2189 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2188 ] +, "main_type": "NONE", "name": "Rank3508", "outputIndexes": +[ 2190 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2190, 2190 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3510", "outputIndexes": +[ 2191 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2191 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3511", "outputIndexes": +[ 2192 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2191, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3513", "outputIndexes": +[ 2193 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2193 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3514", "outputIndexes": +[ 2194 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2189, 2192, 2194, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3517", "outputIndexes": +[ 2195 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2195 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3518", "outputIndexes": +[ 2196 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2196, 2196 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3520", "outputIndexes": +[ 2197 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2188, 2197, 2191 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.23/self_attn/Gather_output_0", "outputIndexes": +[ 2198 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2198, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2199 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2191 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3528", "outputIndexes": +[ 2200 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2193 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3531", "outputIndexes": +[ 2201 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2189, 2200, 2201, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3534", "outputIndexes": +[ 2202 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2202 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3535", "outputIndexes": +[ 2203 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2203 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3536", "outputIndexes": +[ 2204 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2204, 2203 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3537", "outputIndexes": +[ 2205 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2188, 2205, 2191 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.23/self_attn/Gather_1_output_0", "outputIndexes": +[ 2206 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2206, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2207 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2199, 2207, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.23/self_attn/Concat_output_0", "outputIndexes": +[ 2208 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2187, 2208 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.23/self_attn/Reshape_output_0", "outputIndexes": +[ 2209 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2209 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2785560044, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.23/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2211 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2211, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Mul_output_0", "outputIndexes": +[ 2212 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2211 ] +, "main_type": "NONE", "name": "/blocks.23/self_attn/Shape_2_output_0", "outputIndexes": +[ 2213 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2213 ] +, "main_type": "NONE", "name": "Shape3552", "outputIndexes": +[ 2214 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2213 ] +, "main_type": "NONE", "name": "Rank3554", "outputIndexes": +[ 2215 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2215, 2215 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3556", "outputIndexes": +[ 2216 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2216 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3557", "outputIndexes": +[ 2217 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2216, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3559", "outputIndexes": +[ 2218 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2218 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3560", "outputIndexes": +[ 2219 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2214, 2217, 2219, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3563", "outputIndexes": +[ 2220 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2220 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3564", "outputIndexes": +[ 2221 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2221 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3565", "outputIndexes": +[ 2222 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2222, 2221 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3566", "outputIndexes": +[ 2223 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2213, 2223, 2216 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.23/self_attn/Gather_2_output_0", "outputIndexes": +[ 2224 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2224, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Div_output_0", "outputIndexes": +[ 2225 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2225, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2226 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2211, 2226, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.23/self_attn/Slice_1_output_0", "outputIndexes": +[ 2227 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2227 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.23/self_attn/Neg_output_0", "outputIndexes": +[ 2228 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2225, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2229 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2211, 20, 2229, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.23/self_attn/Slice_output_0", "outputIndexes": +[ 2230 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2228, 2230 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.23/self_attn/Concat_3_output_0", "outputIndexes": +[ 2231 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2231, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Mul_1_output_0", "outputIndexes": +[ 2232 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2212, 2232 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Add_output_0", "outputIndexes": +[ 2233 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4042 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2785561068, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.23/self_attn/k_proj/Linear", "outputIndexes": +[ 4047 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4047 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4048 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4048 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2234 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2198, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2235 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2206, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2236 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2235, 2236, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.23/self_attn/Concat_1_output_0", "outputIndexes": +[ 2237 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2234, 2237 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.23/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2238 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2238 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2788182530, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.23/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2240 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2240, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Mul_2_output_0", "outputIndexes": +[ 2241 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2240 ] +, "main_type": "NONE", "name": "/blocks.23/self_attn/Shape_3_output_0", "outputIndexes": +[ 2242 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2242 ] +, "main_type": "NONE", "name": "Shape3594", "outputIndexes": +[ 2243 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2242 ] +, "main_type": "NONE", "name": "Rank3596", "outputIndexes": +[ 2244 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2244, 2244 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3598", "outputIndexes": +[ 2245 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2245 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3599", "outputIndexes": +[ 2246 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2245, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3601", "outputIndexes": +[ 2247 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2247 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3602", "outputIndexes": +[ 2248 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2243, 2246, 2248, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3605", "outputIndexes": +[ 2249 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2249 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3606", "outputIndexes": +[ 2250 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2250 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3607", "outputIndexes": +[ 2251 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2251, 2250 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3608", "outputIndexes": +[ 2252 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2242, 2252, 2245 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.23/self_attn/Gather_3_output_0", "outputIndexes": +[ 2253 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2253, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Div_1_output_0", "outputIndexes": +[ 2254 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2254, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2255 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2240, 2255, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.23/self_attn/Slice_3_output_0", "outputIndexes": +[ 2256 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2256 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.23/self_attn/Neg_1_output_0", "outputIndexes": +[ 2257 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2254, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2258 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2240, 20, 2258, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.23/self_attn/Slice_2_output_0", "outputIndexes": +[ 2259 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2257, 2259 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.23/self_attn/Concat_4_output_0", "outputIndexes": +[ 2260 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2260, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Mul_3_output_0", "outputIndexes": +[ 2261 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2241, 2261 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/self_attn/Add_1_output_0", "outputIndexes": +[ 2262 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4042 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2788183554, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.23/self_attn/v_proj/Linear", "outputIndexes": +[ 4051 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4051 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4052 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4052 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2263 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2198, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2264 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2206, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.23/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2265 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2264, 2265, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.23/self_attn/Concat_2_output_0", "outputIndexes": +[ 2266 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2263, 2266 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.23/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2267 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2233, 2262, 2267, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.23/self_attn/FusedAttention", "outputIndexes": +[ 2268 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2268 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4053 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4053 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4054 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4054 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2790805016, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.23/self_attn/o_proj/Linear", "outputIndexes": +[ 4055 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4055 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4056 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4056 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2269 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2185, 2269 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/Add_output_0", "outputIndexes": +[ 2270 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2270 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2801290798, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.23/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2271 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2271 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4057 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4057 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4058 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4058 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2801323566, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.23/mlp/gate_proj/Linear", "outputIndexes": +[ 4059 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4059 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4060 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4060 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2272 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2272 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.23/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2273 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4058 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2832780868, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.23/mlp/up_proj/Linear", "outputIndexes": +[ 4063 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4063 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4064 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4064 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2274 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2273, 2274 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/mlp/Mul_output_0", "outputIndexes": +[ 2275 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2275 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4065 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4065 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4066 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4066 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2864238170, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.23/mlp/down_proj/Linear", "outputIndexes": +[ 4067 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4067 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4068 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4068 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2276 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2270, 2276 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.23/Add_1_output_0", "outputIndexes": +[ 2277 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2277, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.24/Reshape_output_0", "outputIndexes": +[ 2278 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2278 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2895695472, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.24/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2279 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2279 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4069 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4069 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4070 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4070 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2895728240, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.24/self_attn/q_proj/Linear", "outputIndexes": +[ 4071 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4071 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4072 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4072 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2280 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2279 ] +, "main_type": "NONE", "name": "/blocks.24/self_attn/Shape_output_0", "outputIndexes": +[ 2281 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2281 ] +, "main_type": "NONE", "name": "Shape3655", "outputIndexes": +[ 2282 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2281 ] +, "main_type": "NONE", "name": "Rank3657", "outputIndexes": +[ 2283 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2283, 2283 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3659", "outputIndexes": +[ 2284 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2284 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3660", "outputIndexes": +[ 2285 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2284, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3662", "outputIndexes": +[ 2286 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2286 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3663", "outputIndexes": +[ 2287 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2282, 2285, 2287, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3666", "outputIndexes": +[ 2288 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2288 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3667", "outputIndexes": +[ 2289 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2289, 2289 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3669", "outputIndexes": +[ 2290 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2281, 2290, 2284 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.24/self_attn/Gather_output_0", "outputIndexes": +[ 2291 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2291, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2292 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2284 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3677", "outputIndexes": +[ 2293 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2286 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3680", "outputIndexes": +[ 2294 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2282, 2293, 2294, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3683", "outputIndexes": +[ 2295 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2295 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3684", "outputIndexes": +[ 2296 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2296 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3685", "outputIndexes": +[ 2297 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2297, 2296 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3686", "outputIndexes": +[ 2298 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2281, 2298, 2284 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.24/self_attn/Gather_1_output_0", "outputIndexes": +[ 2299 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2299, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2300 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2292, 2300, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.24/self_attn/Concat_output_0", "outputIndexes": +[ 2301 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2280, 2301 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.24/self_attn/Reshape_output_0", "outputIndexes": +[ 2302 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2302 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2906214022, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.24/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2304 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2304, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Mul_output_0", "outputIndexes": +[ 2305 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2304 ] +, "main_type": "NONE", "name": "/blocks.24/self_attn/Shape_2_output_0", "outputIndexes": +[ 2306 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2306 ] +, "main_type": "NONE", "name": "Shape3701", "outputIndexes": +[ 2307 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2306 ] +, "main_type": "NONE", "name": "Rank3703", "outputIndexes": +[ 2308 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2308, 2308 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3705", "outputIndexes": +[ 2309 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2309 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3706", "outputIndexes": +[ 2310 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2309, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3708", "outputIndexes": +[ 2311 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2311 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3709", "outputIndexes": +[ 2312 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2307, 2310, 2312, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3712", "outputIndexes": +[ 2313 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2313 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3713", "outputIndexes": +[ 2314 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2314 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3714", "outputIndexes": +[ 2315 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2315, 2314 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3715", "outputIndexes": +[ 2316 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2306, 2316, 2309 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.24/self_attn/Gather_2_output_0", "outputIndexes": +[ 2317 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2317, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Div_output_0", "outputIndexes": +[ 2318 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2318, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2319 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2304, 2319, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.24/self_attn/Slice_1_output_0", "outputIndexes": +[ 2320 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2320 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.24/self_attn/Neg_output_0", "outputIndexes": +[ 2321 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2318, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2322 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2304, 20, 2322, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.24/self_attn/Slice_output_0", "outputIndexes": +[ 2323 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2321, 2323 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.24/self_attn/Concat_3_output_0", "outputIndexes": +[ 2324 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2324, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Mul_1_output_0", "outputIndexes": +[ 2325 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2305, 2325 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Add_output_0", "outputIndexes": +[ 2326 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4070 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2906215046, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.24/self_attn/k_proj/Linear", "outputIndexes": +[ 4075 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4075 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4076 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4076 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2327 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2291, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2328 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2299, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2329 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2328, 2329, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.24/self_attn/Concat_1_output_0", "outputIndexes": +[ 2330 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2327, 2330 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.24/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2331 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2331 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2908836508, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.24/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2333 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2333, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Mul_2_output_0", "outputIndexes": +[ 2334 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2333 ] +, "main_type": "NONE", "name": "/blocks.24/self_attn/Shape_3_output_0", "outputIndexes": +[ 2335 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2335 ] +, "main_type": "NONE", "name": "Shape3743", "outputIndexes": +[ 2336 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2335 ] +, "main_type": "NONE", "name": "Rank3745", "outputIndexes": +[ 2337 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2337, 2337 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3747", "outputIndexes": +[ 2338 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2338 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3748", "outputIndexes": +[ 2339 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2338, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3750", "outputIndexes": +[ 2340 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2340 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3751", "outputIndexes": +[ 2341 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2336, 2339, 2341, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3754", "outputIndexes": +[ 2342 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2342 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3755", "outputIndexes": +[ 2343 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2343 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3756", "outputIndexes": +[ 2344 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2344, 2343 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3757", "outputIndexes": +[ 2345 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2335, 2345, 2338 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.24/self_attn/Gather_3_output_0", "outputIndexes": +[ 2346 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2346, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Div_1_output_0", "outputIndexes": +[ 2347 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2347, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2348 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2333, 2348, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.24/self_attn/Slice_3_output_0", "outputIndexes": +[ 2349 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2349 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.24/self_attn/Neg_1_output_0", "outputIndexes": +[ 2350 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2347, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2351 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2333, 20, 2351, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.24/self_attn/Slice_2_output_0", "outputIndexes": +[ 2352 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2350, 2352 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.24/self_attn/Concat_4_output_0", "outputIndexes": +[ 2353 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2353, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Mul_3_output_0", "outputIndexes": +[ 2354 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2334, 2354 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/self_attn/Add_1_output_0", "outputIndexes": +[ 2355 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4070 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2908837532, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.24/self_attn/v_proj/Linear", "outputIndexes": +[ 4079 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4079 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4080 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4080 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2356 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2291, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2357 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2299, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.24/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2358 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2357, 2358, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.24/self_attn/Concat_2_output_0", "outputIndexes": +[ 2359 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2356, 2359 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.24/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2360 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2326, 2355, 2360, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.24/self_attn/FusedAttention", "outputIndexes": +[ 2361 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2361 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4081 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4081 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4082 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4082 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2911458994, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.24/self_attn/o_proj/Linear", "outputIndexes": +[ 4083 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4083 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4084 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4084 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2362 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2278, 2362 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/Add_output_0", "outputIndexes": +[ 2363 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2363 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2921944776, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.24/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2364 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2364 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4085 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4085 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4086 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4086 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2921977544, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.24/mlp/gate_proj/Linear", "outputIndexes": +[ 4087 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4087 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4088 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4088 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2365 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2365 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.24/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2366 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4086 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2953434846, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.24/mlp/up_proj/Linear", "outputIndexes": +[ 4091 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4091 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4092 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4092 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2367 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2366, 2367 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/mlp/Mul_output_0", "outputIndexes": +[ 2368 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2368 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4093 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4093 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4094 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4094 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 2984892148, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.24/mlp/down_proj/Linear", "outputIndexes": +[ 4095 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4095 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4096 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4096 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2369 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2363, 2369 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.24/Add_1_output_0", "outputIndexes": +[ 2370 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2370, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.25/Reshape_output_0", "outputIndexes": +[ 2371 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2371 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3016349450, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.25/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2372 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2372 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4097 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4097 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4098 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4098 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3016382218, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.25/self_attn/q_proj/Linear", "outputIndexes": +[ 4099 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4099 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4100 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4100 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2373 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2372 ] +, "main_type": "NONE", "name": "/blocks.25/self_attn/Shape_output_0", "outputIndexes": +[ 2374 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2374 ] +, "main_type": "NONE", "name": "Shape3804", "outputIndexes": +[ 2375 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2374 ] +, "main_type": "NONE", "name": "Rank3806", "outputIndexes": +[ 2376 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2376, 2376 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3808", "outputIndexes": +[ 2377 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2377 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3809", "outputIndexes": +[ 2378 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2377, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3811", "outputIndexes": +[ 2379 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2379 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3812", "outputIndexes": +[ 2380 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2375, 2378, 2380, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3815", "outputIndexes": +[ 2381 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2381 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3816", "outputIndexes": +[ 2382 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2382, 2382 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3818", "outputIndexes": +[ 2383 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2374, 2383, 2377 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.25/self_attn/Gather_output_0", "outputIndexes": +[ 2384 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2384, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2385 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2377 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3826", "outputIndexes": +[ 2386 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2379 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3829", "outputIndexes": +[ 2387 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2375, 2386, 2387, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3832", "outputIndexes": +[ 2388 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2388 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3833", "outputIndexes": +[ 2389 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2389 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3834", "outputIndexes": +[ 2390 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2390, 2389 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3835", "outputIndexes": +[ 2391 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2374, 2391, 2377 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.25/self_attn/Gather_1_output_0", "outputIndexes": +[ 2392 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2392, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2393 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2385, 2393, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.25/self_attn/Concat_output_0", "outputIndexes": +[ 2394 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2373, 2394 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.25/self_attn/Reshape_output_0", "outputIndexes": +[ 2395 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2395 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3026868000, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.25/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2397 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2397, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Mul_output_0", "outputIndexes": +[ 2398 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2397 ] +, "main_type": "NONE", "name": "/blocks.25/self_attn/Shape_2_output_0", "outputIndexes": +[ 2399 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2399 ] +, "main_type": "NONE", "name": "Shape3850", "outputIndexes": +[ 2400 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2399 ] +, "main_type": "NONE", "name": "Rank3852", "outputIndexes": +[ 2401 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2401, 2401 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3854", "outputIndexes": +[ 2402 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2402 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3855", "outputIndexes": +[ 2403 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2402, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3857", "outputIndexes": +[ 2404 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2404 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3858", "outputIndexes": +[ 2405 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2400, 2403, 2405, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3861", "outputIndexes": +[ 2406 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2406 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3862", "outputIndexes": +[ 2407 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2407 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3863", "outputIndexes": +[ 2408 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2408, 2407 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3864", "outputIndexes": +[ 2409 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2399, 2409, 2402 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.25/self_attn/Gather_2_output_0", "outputIndexes": +[ 2410 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2410, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Div_output_0", "outputIndexes": +[ 2411 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2411, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2412 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2397, 2412, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.25/self_attn/Slice_1_output_0", "outputIndexes": +[ 2413 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2413 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.25/self_attn/Neg_output_0", "outputIndexes": +[ 2414 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2411, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2415 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2397, 20, 2415, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.25/self_attn/Slice_output_0", "outputIndexes": +[ 2416 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2414, 2416 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.25/self_attn/Concat_3_output_0", "outputIndexes": +[ 2417 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2417, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Mul_1_output_0", "outputIndexes": +[ 2418 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2398, 2418 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Add_output_0", "outputIndexes": +[ 2419 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4098 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3026869024, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.25/self_attn/k_proj/Linear", "outputIndexes": +[ 4103 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4103 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4104 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4104 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2420 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2384, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2421 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2392, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2422 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2421, 2422, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.25/self_attn/Concat_1_output_0", "outputIndexes": +[ 2423 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2420, 2423 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.25/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2424 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2424 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3029490486, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.25/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2426 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2426, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Mul_2_output_0", "outputIndexes": +[ 2427 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2426 ] +, "main_type": "NONE", "name": "/blocks.25/self_attn/Shape_3_output_0", "outputIndexes": +[ 2428 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2428 ] +, "main_type": "NONE", "name": "Shape3892", "outputIndexes": +[ 2429 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2428 ] +, "main_type": "NONE", "name": "Rank3894", "outputIndexes": +[ 2430 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2430, 2430 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3896", "outputIndexes": +[ 2431 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2431 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3897", "outputIndexes": +[ 2432 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2431, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3899", "outputIndexes": +[ 2433 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2433 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3900", "outputIndexes": +[ 2434 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2429, 2432, 2434, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3903", "outputIndexes": +[ 2435 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2435 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3904", "outputIndexes": +[ 2436 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2436 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3905", "outputIndexes": +[ 2437 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2437, 2436 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3906", "outputIndexes": +[ 2438 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2428, 2438, 2431 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.25/self_attn/Gather_3_output_0", "outputIndexes": +[ 2439 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2439, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Div_1_output_0", "outputIndexes": +[ 2440 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2440, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2441 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2426, 2441, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.25/self_attn/Slice_3_output_0", "outputIndexes": +[ 2442 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2442 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.25/self_attn/Neg_1_output_0", "outputIndexes": +[ 2443 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2440, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2444 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2426, 20, 2444, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.25/self_attn/Slice_2_output_0", "outputIndexes": +[ 2445 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2443, 2445 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.25/self_attn/Concat_4_output_0", "outputIndexes": +[ 2446 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2446, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Mul_3_output_0", "outputIndexes": +[ 2447 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2427, 2447 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/self_attn/Add_1_output_0", "outputIndexes": +[ 2448 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4098 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3029491510, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.25/self_attn/v_proj/Linear", "outputIndexes": +[ 4107 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4107 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4108 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4108 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2449 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2384, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2450 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2392, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.25/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2451 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2450, 2451, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.25/self_attn/Concat_2_output_0", "outputIndexes": +[ 2452 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2449, 2452 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.25/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2453 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2419, 2448, 2453, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.25/self_attn/FusedAttention", "outputIndexes": +[ 2454 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2454 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4109 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4109 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4110 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4110 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3032112972, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.25/self_attn/o_proj/Linear", "outputIndexes": +[ 4111 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4111 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4112 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4112 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2455 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2371, 2455 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/Add_output_0", "outputIndexes": +[ 2456 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2456 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3042598754, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.25/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2457 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2457 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4113 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4113 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4114 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4114 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3042631522, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.25/mlp/gate_proj/Linear", "outputIndexes": +[ 4115 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4115 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4116 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4116 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2458 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2458 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.25/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2459 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4114 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3074088824, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.25/mlp/up_proj/Linear", "outputIndexes": +[ 4119 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4119 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4120 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4120 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2460 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2459, 2460 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/mlp/Mul_output_0", "outputIndexes": +[ 2461 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2461 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4121 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4121 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4122 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4122 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3105546126, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.25/mlp/down_proj/Linear", "outputIndexes": +[ 4123 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4123 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4124 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4124 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2462 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2456, 2462 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.25/Add_1_output_0", "outputIndexes": +[ 2463 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2463, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.26/Reshape_output_0", "outputIndexes": +[ 2464 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2464 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3137003428, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.26/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2465 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2465 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4125 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4125 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4126 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4126 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3137036196, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.26/self_attn/q_proj/Linear", "outputIndexes": +[ 4127 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4127 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4128 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4128 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2466 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2465 ] +, "main_type": "NONE", "name": "/blocks.26/self_attn/Shape_output_0", "outputIndexes": +[ 2467 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2467 ] +, "main_type": "NONE", "name": "Shape3953", "outputIndexes": +[ 2468 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2467 ] +, "main_type": "NONE", "name": "Rank3955", "outputIndexes": +[ 2469 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2469, 2469 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3957", "outputIndexes": +[ 2470 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2470 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3958", "outputIndexes": +[ 2471 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2470, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3960", "outputIndexes": +[ 2472 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2472 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3961", "outputIndexes": +[ 2473 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2468, 2471, 2473, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3964", "outputIndexes": +[ 2474 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2474 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3965", "outputIndexes": +[ 2475 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2475, 2475 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3967", "outputIndexes": +[ 2476 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2467, 2476, 2470 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.26/self_attn/Gather_output_0", "outputIndexes": +[ 2477 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2477, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2478 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2470 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3975", "outputIndexes": +[ 2479 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2472 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3978", "outputIndexes": +[ 2480 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2468, 2479, 2480, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3981", "outputIndexes": +[ 2481 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2481 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3982", "outputIndexes": +[ 2482 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2482 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3983", "outputIndexes": +[ 2483 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2483, 2482 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3984", "outputIndexes": +[ 2484 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2467, 2484, 2470 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.26/self_attn/Gather_1_output_0", "outputIndexes": +[ 2485 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2485, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2486 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2478, 2486, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.26/self_attn/Concat_output_0", "outputIndexes": +[ 2487 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2466, 2487 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.26/self_attn/Reshape_output_0", "outputIndexes": +[ 2488 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2488 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3147521978, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.26/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2490 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2490, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Mul_output_0", "outputIndexes": +[ 2491 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2490 ] +, "main_type": "NONE", "name": "/blocks.26/self_attn/Shape_2_output_0", "outputIndexes": +[ 2492 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2492 ] +, "main_type": "NONE", "name": "Shape3999", "outputIndexes": +[ 2493 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2492 ] +, "main_type": "NONE", "name": "Rank4001", "outputIndexes": +[ 2494 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2494, 2494 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4003", "outputIndexes": +[ 2495 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2495 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4004", "outputIndexes": +[ 2496 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2495, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4006", "outputIndexes": +[ 2497 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2497 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4007", "outputIndexes": +[ 2498 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2493, 2496, 2498, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4010", "outputIndexes": +[ 2499 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2499 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4011", "outputIndexes": +[ 2500 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2500 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4012", "outputIndexes": +[ 2501 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2501, 2500 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4013", "outputIndexes": +[ 2502 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2492, 2502, 2495 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.26/self_attn/Gather_2_output_0", "outputIndexes": +[ 2503 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2503, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Div_output_0", "outputIndexes": +[ 2504 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2504, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2505 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2490, 2505, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.26/self_attn/Slice_1_output_0", "outputIndexes": +[ 2506 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2506 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.26/self_attn/Neg_output_0", "outputIndexes": +[ 2507 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2504, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2508 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2490, 20, 2508, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.26/self_attn/Slice_output_0", "outputIndexes": +[ 2509 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2507, 2509 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.26/self_attn/Concat_3_output_0", "outputIndexes": +[ 2510 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2510, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Mul_1_output_0", "outputIndexes": +[ 2511 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2491, 2511 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Add_output_0", "outputIndexes": +[ 2512 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4126 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3147523002, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.26/self_attn/k_proj/Linear", "outputIndexes": +[ 4131 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4131 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4132 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4132 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2513 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2477, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2514 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2485, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2515 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2514, 2515, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.26/self_attn/Concat_1_output_0", "outputIndexes": +[ 2516 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2513, 2516 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.26/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2517 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2517 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3150144464, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.26/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2519 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2519, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Mul_2_output_0", "outputIndexes": +[ 2520 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2519 ] +, "main_type": "NONE", "name": "/blocks.26/self_attn/Shape_3_output_0", "outputIndexes": +[ 2521 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2521 ] +, "main_type": "NONE", "name": "Shape4041", "outputIndexes": +[ 2522 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2521 ] +, "main_type": "NONE", "name": "Rank4043", "outputIndexes": +[ 2523 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2523, 2523 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4045", "outputIndexes": +[ 2524 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2524 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4046", "outputIndexes": +[ 2525 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2524, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4048", "outputIndexes": +[ 2526 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2526 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4049", "outputIndexes": +[ 2527 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2522, 2525, 2527, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4052", "outputIndexes": +[ 2528 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2528 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4053", "outputIndexes": +[ 2529 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2529 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4054", "outputIndexes": +[ 2530 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2530, 2529 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4055", "outputIndexes": +[ 2531 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2521, 2531, 2524 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.26/self_attn/Gather_3_output_0", "outputIndexes": +[ 2532 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2532, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Div_1_output_0", "outputIndexes": +[ 2533 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2533, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2534 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2519, 2534, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.26/self_attn/Slice_3_output_0", "outputIndexes": +[ 2535 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2535 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.26/self_attn/Neg_1_output_0", "outputIndexes": +[ 2536 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2533, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2537 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2519, 20, 2537, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.26/self_attn/Slice_2_output_0", "outputIndexes": +[ 2538 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2536, 2538 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.26/self_attn/Concat_4_output_0", "outputIndexes": +[ 2539 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2539, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Mul_3_output_0", "outputIndexes": +[ 2540 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2520, 2540 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/self_attn/Add_1_output_0", "outputIndexes": +[ 2541 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4126 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3150145488, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.26/self_attn/v_proj/Linear", "outputIndexes": +[ 4135 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4135 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4136 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4136 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2542 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2477, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2543 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2485, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.26/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2544 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2543, 2544, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.26/self_attn/Concat_2_output_0", "outputIndexes": +[ 2545 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2542, 2545 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.26/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2546 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2512, 2541, 2546, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.26/self_attn/FusedAttention", "outputIndexes": +[ 2547 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2547 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4137 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4137 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4138 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4138 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3152766950, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.26/self_attn/o_proj/Linear", "outputIndexes": +[ 4139 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4139 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4140 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4140 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2548 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2464, 2548 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/Add_output_0", "outputIndexes": +[ 2549 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2549 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3163252732, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.26/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2550 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2550 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4141 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4141 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4142 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4142 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3163285500, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.26/mlp/gate_proj/Linear", "outputIndexes": +[ 4143 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4143 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4144 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4144 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2551 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2551 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.26/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2552 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4142 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3194742802, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.26/mlp/up_proj/Linear", "outputIndexes": +[ 4147 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4147 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4148 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4148 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2553 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2552, 2553 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/mlp/Mul_output_0", "outputIndexes": +[ 2554 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2554 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4149 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4149 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4150 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4150 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3226200104, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.26/mlp/down_proj/Linear", "outputIndexes": +[ 4151 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4151 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4152 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4152 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2555 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2549, 2555 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.26/Add_1_output_0", "outputIndexes": +[ 2556 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2556, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.27/Reshape_output_0", "outputIndexes": +[ 2557 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2557 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3257657406, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.27/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2558 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2558 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4153 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4153 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4154 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4154 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3257690174, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.27/self_attn/q_proj/Linear", "outputIndexes": +[ 4155 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4155 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4156 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4156 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2559 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2558 ] +, "main_type": "NONE", "name": "/blocks.27/self_attn/Shape_output_0", "outputIndexes": +[ 2560 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2560 ] +, "main_type": "NONE", "name": "Shape4102", "outputIndexes": +[ 2561 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2560 ] +, "main_type": "NONE", "name": "Rank4104", "outputIndexes": +[ 2562 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2562, 2562 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4106", "outputIndexes": +[ 2563 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2563 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4107", "outputIndexes": +[ 2564 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2563, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4109", "outputIndexes": +[ 2565 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2565 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4110", "outputIndexes": +[ 2566 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2561, 2564, 2566, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4113", "outputIndexes": +[ 2567 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2567 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4114", "outputIndexes": +[ 2568 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2568, 2568 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4116", "outputIndexes": +[ 2569 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2560, 2569, 2563 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.27/self_attn/Gather_output_0", "outputIndexes": +[ 2570 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2570, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2571 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2563 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4124", "outputIndexes": +[ 2572 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2565 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4127", "outputIndexes": +[ 2573 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2561, 2572, 2573, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4130", "outputIndexes": +[ 2574 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2574 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4131", "outputIndexes": +[ 2575 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2575 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4132", "outputIndexes": +[ 2576 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2576, 2575 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4133", "outputIndexes": +[ 2577 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2560, 2577, 2563 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.27/self_attn/Gather_1_output_0", "outputIndexes": +[ 2578 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2578, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2579 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2571, 2579, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.27/self_attn/Concat_output_0", "outputIndexes": +[ 2580 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2559, 2580 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.27/self_attn/Reshape_output_0", "outputIndexes": +[ 2581 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2581 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3268175956, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.27/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2583 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2583, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Mul_output_0", "outputIndexes": +[ 2584 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2583 ] +, "main_type": "NONE", "name": "/blocks.27/self_attn/Shape_2_output_0", "outputIndexes": +[ 2585 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2585 ] +, "main_type": "NONE", "name": "Shape4148", "outputIndexes": +[ 2586 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2585 ] +, "main_type": "NONE", "name": "Rank4150", "outputIndexes": +[ 2587 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2587, 2587 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4152", "outputIndexes": +[ 2588 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2588 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4153", "outputIndexes": +[ 2589 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2588, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4155", "outputIndexes": +[ 2590 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2590 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4156", "outputIndexes": +[ 2591 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2586, 2589, 2591, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4159", "outputIndexes": +[ 2592 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2592 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4160", "outputIndexes": +[ 2593 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2593 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4161", "outputIndexes": +[ 2594 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2594, 2593 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4162", "outputIndexes": +[ 2595 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2585, 2595, 2588 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.27/self_attn/Gather_2_output_0", "outputIndexes": +[ 2596 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2596, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Div_output_0", "outputIndexes": +[ 2597 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2597, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2598 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2583, 2598, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.27/self_attn/Slice_1_output_0", "outputIndexes": +[ 2599 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2599 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.27/self_attn/Neg_output_0", "outputIndexes": +[ 2600 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2597, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2601 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2583, 20, 2601, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.27/self_attn/Slice_output_0", "outputIndexes": +[ 2602 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2600, 2602 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.27/self_attn/Concat_3_output_0", "outputIndexes": +[ 2603 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2603, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Mul_1_output_0", "outputIndexes": +[ 2604 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2584, 2604 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Add_output_0", "outputIndexes": +[ 2605 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4154 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3268176980, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.27/self_attn/k_proj/Linear", "outputIndexes": +[ 4159 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4159 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4160 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4160 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2606 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2570, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2607 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2578, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2608 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2607, 2608, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.27/self_attn/Concat_1_output_0", "outputIndexes": +[ 2609 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2606, 2609 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.27/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2610 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2610 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3270798442, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.27/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2612 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2612, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Mul_2_output_0", "outputIndexes": +[ 2613 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2612 ] +, "main_type": "NONE", "name": "/blocks.27/self_attn/Shape_3_output_0", "outputIndexes": +[ 2614 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2614 ] +, "main_type": "NONE", "name": "Shape4190", "outputIndexes": +[ 2615 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2614 ] +, "main_type": "NONE", "name": "Rank4192", "outputIndexes": +[ 2616 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2616, 2616 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4194", "outputIndexes": +[ 2617 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2617 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4195", "outputIndexes": +[ 2618 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2617, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4197", "outputIndexes": +[ 2619 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2619 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4198", "outputIndexes": +[ 2620 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2615, 2618, 2620, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4201", "outputIndexes": +[ 2621 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2621 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4202", "outputIndexes": +[ 2622 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2622 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4203", "outputIndexes": +[ 2623 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2623, 2622 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4204", "outputIndexes": +[ 2624 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2614, 2624, 2617 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.27/self_attn/Gather_3_output_0", "outputIndexes": +[ 2625 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2625, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Div_1_output_0", "outputIndexes": +[ 2626 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2626, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2627 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2612, 2627, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.27/self_attn/Slice_3_output_0", "outputIndexes": +[ 2628 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2628 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.27/self_attn/Neg_1_output_0", "outputIndexes": +[ 2629 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2626, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2630 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2612, 20, 2630, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.27/self_attn/Slice_2_output_0", "outputIndexes": +[ 2631 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2629, 2631 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.27/self_attn/Concat_4_output_0", "outputIndexes": +[ 2632 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2632, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Mul_3_output_0", "outputIndexes": +[ 2633 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2613, 2633 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/self_attn/Add_1_output_0", "outputIndexes": +[ 2634 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4154 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3270799466, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.27/self_attn/v_proj/Linear", "outputIndexes": +[ 4163 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4163 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4164 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4164 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2635 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2570, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2636 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2578, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.27/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2637 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2636, 2637, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.27/self_attn/Concat_2_output_0", "outputIndexes": +[ 2638 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2635, 2638 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.27/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2639 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2605, 2634, 2639, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.27/self_attn/FusedAttention", "outputIndexes": +[ 2640 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2640 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4165 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4165 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4166 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4166 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3273420928, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.27/self_attn/o_proj/Linear", "outputIndexes": +[ 4167 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4167 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4168 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4168 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2641 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2557, 2641 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/Add_output_0", "outputIndexes": +[ 2642 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2642 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3283906710, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.27/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2643 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2643 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4169 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4169 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4170 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4170 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3283939478, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.27/mlp/gate_proj/Linear", "outputIndexes": +[ 4171 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4171 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4172 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4172 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2644 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2644 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.27/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2645 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4170 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3315396780, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.27/mlp/up_proj/Linear", "outputIndexes": +[ 4175 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4175 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4176 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4176 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2646 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2645, 2646 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/mlp/Mul_output_0", "outputIndexes": +[ 2647 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2647 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4177 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4177 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4178 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4178 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3346854082, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.27/mlp/down_proj/Linear", "outputIndexes": +[ 4179 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4179 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4180 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4180 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2648 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2642, 2648 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.27/Add_1_output_0", "outputIndexes": +[ 2649 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2649, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.28/Reshape_output_0", "outputIndexes": +[ 2650 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2650 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3378311384, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.28/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2651 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2651 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4181 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4181 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4182 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4182 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3378344152, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.28/self_attn/q_proj/Linear", "outputIndexes": +[ 4183 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4183 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4184 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4184 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2652 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2651 ] +, "main_type": "NONE", "name": "/blocks.28/self_attn/Shape_output_0", "outputIndexes": +[ 2653 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2653 ] +, "main_type": "NONE", "name": "Shape4251", "outputIndexes": +[ 2654 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2653 ] +, "main_type": "NONE", "name": "Rank4253", "outputIndexes": +[ 2655 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2655, 2655 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4255", "outputIndexes": +[ 2656 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2656 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4256", "outputIndexes": +[ 2657 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2656, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4258", "outputIndexes": +[ 2658 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2658 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4259", "outputIndexes": +[ 2659 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2654, 2657, 2659, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4262", "outputIndexes": +[ 2660 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2660 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4263", "outputIndexes": +[ 2661 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2661, 2661 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4265", "outputIndexes": +[ 2662 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2653, 2662, 2656 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.28/self_attn/Gather_output_0", "outputIndexes": +[ 2663 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2663, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2664 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2656 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4273", "outputIndexes": +[ 2665 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2658 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4276", "outputIndexes": +[ 2666 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2654, 2665, 2666, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4279", "outputIndexes": +[ 2667 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2667 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4280", "outputIndexes": +[ 2668 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2668 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4281", "outputIndexes": +[ 2669 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2669, 2668 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4282", "outputIndexes": +[ 2670 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2653, 2670, 2656 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.28/self_attn/Gather_1_output_0", "outputIndexes": +[ 2671 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2671, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2672 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2664, 2672, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.28/self_attn/Concat_output_0", "outputIndexes": +[ 2673 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2652, 2673 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.28/self_attn/Reshape_output_0", "outputIndexes": +[ 2674 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2674 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3388829934, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.28/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2676 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2676, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Mul_output_0", "outputIndexes": +[ 2677 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2676 ] +, "main_type": "NONE", "name": "/blocks.28/self_attn/Shape_2_output_0", "outputIndexes": +[ 2678 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2678 ] +, "main_type": "NONE", "name": "Shape4297", "outputIndexes": +[ 2679 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2678 ] +, "main_type": "NONE", "name": "Rank4299", "outputIndexes": +[ 2680 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2680, 2680 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4301", "outputIndexes": +[ 2681 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2681 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4302", "outputIndexes": +[ 2682 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2681, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4304", "outputIndexes": +[ 2683 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2683 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4305", "outputIndexes": +[ 2684 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2679, 2682, 2684, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4308", "outputIndexes": +[ 2685 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2685 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4309", "outputIndexes": +[ 2686 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2686 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4310", "outputIndexes": +[ 2687 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2687, 2686 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4311", "outputIndexes": +[ 2688 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2678, 2688, 2681 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.28/self_attn/Gather_2_output_0", "outputIndexes": +[ 2689 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2689, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Div_output_0", "outputIndexes": +[ 2690 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2690, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2691 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2676, 2691, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.28/self_attn/Slice_1_output_0", "outputIndexes": +[ 2692 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2692 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.28/self_attn/Neg_output_0", "outputIndexes": +[ 2693 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2690, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2694 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2676, 20, 2694, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.28/self_attn/Slice_output_0", "outputIndexes": +[ 2695 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2693, 2695 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.28/self_attn/Concat_3_output_0", "outputIndexes": +[ 2696 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2696, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Mul_1_output_0", "outputIndexes": +[ 2697 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2677, 2697 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Add_output_0", "outputIndexes": +[ 2698 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4182 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3388830958, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.28/self_attn/k_proj/Linear", "outputIndexes": +[ 4187 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4187 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4188 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4188 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2699 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2663, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2700 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2671, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2701 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2700, 2701, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.28/self_attn/Concat_1_output_0", "outputIndexes": +[ 2702 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2699, 2702 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.28/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2703 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2703 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3391452420, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.28/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2705 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2705, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Mul_2_output_0", "outputIndexes": +[ 2706 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2705 ] +, "main_type": "NONE", "name": "/blocks.28/self_attn/Shape_3_output_0", "outputIndexes": +[ 2707 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2707 ] +, "main_type": "NONE", "name": "Shape4339", "outputIndexes": +[ 2708 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2707 ] +, "main_type": "NONE", "name": "Rank4341", "outputIndexes": +[ 2709 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2709, 2709 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4343", "outputIndexes": +[ 2710 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2710 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4344", "outputIndexes": +[ 2711 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2710, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4346", "outputIndexes": +[ 2712 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2712 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4347", "outputIndexes": +[ 2713 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2708, 2711, 2713, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4350", "outputIndexes": +[ 2714 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2714 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4351", "outputIndexes": +[ 2715 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2715 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4352", "outputIndexes": +[ 2716 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2716, 2715 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4353", "outputIndexes": +[ 2717 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2707, 2717, 2710 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.28/self_attn/Gather_3_output_0", "outputIndexes": +[ 2718 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2718, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Div_1_output_0", "outputIndexes": +[ 2719 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2719, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2720 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2705, 2720, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.28/self_attn/Slice_3_output_0", "outputIndexes": +[ 2721 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2721 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.28/self_attn/Neg_1_output_0", "outputIndexes": +[ 2722 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2719, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2723 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2705, 20, 2723, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.28/self_attn/Slice_2_output_0", "outputIndexes": +[ 2724 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2722, 2724 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.28/self_attn/Concat_4_output_0", "outputIndexes": +[ 2725 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2725, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Mul_3_output_0", "outputIndexes": +[ 2726 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2706, 2726 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/self_attn/Add_1_output_0", "outputIndexes": +[ 2727 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4182 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3391453444, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.28/self_attn/v_proj/Linear", "outputIndexes": +[ 4191 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4191 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4192 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4192 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2728 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2663, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2729 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2671, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.28/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2730 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2729, 2730, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.28/self_attn/Concat_2_output_0", "outputIndexes": +[ 2731 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2728, 2731 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.28/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2732 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2698, 2727, 2732, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.28/self_attn/FusedAttention", "outputIndexes": +[ 2733 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2733 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4193 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4193 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4194 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4194 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3394074906, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.28/self_attn/o_proj/Linear", "outputIndexes": +[ 4195 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4195 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4196 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4196 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2734 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2650, 2734 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/Add_output_0", "outputIndexes": +[ 2735 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2735 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3404560688, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.28/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2736 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2736 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4197 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4197 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4198 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4198 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3404593456, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.28/mlp/gate_proj/Linear", "outputIndexes": +[ 4199 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4199 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4200 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4200 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2737 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2737 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.28/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2738 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4198 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3436050758, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.28/mlp/up_proj/Linear", "outputIndexes": +[ 4203 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4203 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4204 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4204 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2739 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2738, 2739 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/mlp/Mul_output_0", "outputIndexes": +[ 2740 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2740 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4205 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4205 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4206 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4206 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3467508060, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.28/mlp/down_proj/Linear", "outputIndexes": +[ 4207 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4207 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4208 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4208 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2741 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2735, 2741 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.28/Add_1_output_0", "outputIndexes": +[ 2742 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2742, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.29/Reshape_output_0", "outputIndexes": +[ 2743 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2743 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3498965362, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.29/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2744 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2744 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4209 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4209 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4210 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4210 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3498998130, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.29/self_attn/q_proj/Linear", "outputIndexes": +[ 4211 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4211 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4212 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4212 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2745 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2744 ] +, "main_type": "NONE", "name": "/blocks.29/self_attn/Shape_output_0", "outputIndexes": +[ 2746 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2746 ] +, "main_type": "NONE", "name": "Shape4400", "outputIndexes": +[ 2747 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2746 ] +, "main_type": "NONE", "name": "Rank4402", "outputIndexes": +[ 2748 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2748, 2748 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4404", "outputIndexes": +[ 2749 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2749 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4405", "outputIndexes": +[ 2750 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2749, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4407", "outputIndexes": +[ 2751 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2751 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4408", "outputIndexes": +[ 2752 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2747, 2750, 2752, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4411", "outputIndexes": +[ 2753 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2753 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4412", "outputIndexes": +[ 2754 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2754, 2754 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4414", "outputIndexes": +[ 2755 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2746, 2755, 2749 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.29/self_attn/Gather_output_0", "outputIndexes": +[ 2756 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2756, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2757 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2749 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4422", "outputIndexes": +[ 2758 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2751 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4425", "outputIndexes": +[ 2759 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2747, 2758, 2759, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4428", "outputIndexes": +[ 2760 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2760 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4429", "outputIndexes": +[ 2761 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2761 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4430", "outputIndexes": +[ 2762 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2762, 2761 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4431", "outputIndexes": +[ 2763 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2746, 2763, 2749 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.29/self_attn/Gather_1_output_0", "outputIndexes": +[ 2764 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2764, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2765 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2757, 2765, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.29/self_attn/Concat_output_0", "outputIndexes": +[ 2766 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2745, 2766 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.29/self_attn/Reshape_output_0", "outputIndexes": +[ 2767 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2767 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3509483912, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.29/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2769 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2769, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Mul_output_0", "outputIndexes": +[ 2770 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2769 ] +, "main_type": "NONE", "name": "/blocks.29/self_attn/Shape_2_output_0", "outputIndexes": +[ 2771 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2771 ] +, "main_type": "NONE", "name": "Shape4446", "outputIndexes": +[ 2772 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2771 ] +, "main_type": "NONE", "name": "Rank4448", "outputIndexes": +[ 2773 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2773, 2773 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4450", "outputIndexes": +[ 2774 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2774 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4451", "outputIndexes": +[ 2775 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2774, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4453", "outputIndexes": +[ 2776 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2776 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4454", "outputIndexes": +[ 2777 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2772, 2775, 2777, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4457", "outputIndexes": +[ 2778 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2778 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4458", "outputIndexes": +[ 2779 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2779 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4459", "outputIndexes": +[ 2780 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2780, 2779 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4460", "outputIndexes": +[ 2781 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2771, 2781, 2774 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.29/self_attn/Gather_2_output_0", "outputIndexes": +[ 2782 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2782, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Div_output_0", "outputIndexes": +[ 2783 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2783, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2784 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2769, 2784, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.29/self_attn/Slice_1_output_0", "outputIndexes": +[ 2785 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2785 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.29/self_attn/Neg_output_0", "outputIndexes": +[ 2786 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2783, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2787 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2769, 20, 2787, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.29/self_attn/Slice_output_0", "outputIndexes": +[ 2788 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2786, 2788 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.29/self_attn/Concat_3_output_0", "outputIndexes": +[ 2789 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2789, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Mul_1_output_0", "outputIndexes": +[ 2790 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2770, 2790 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Add_output_0", "outputIndexes": +[ 2791 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4210 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3509484936, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.29/self_attn/k_proj/Linear", "outputIndexes": +[ 4215 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4215 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4216 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4216 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2792 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2756, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2793 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2764, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2794 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2793, 2794, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.29/self_attn/Concat_1_output_0", "outputIndexes": +[ 2795 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2792, 2795 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.29/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2796 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2796 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3512106398, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.29/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2798 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2798, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Mul_2_output_0", "outputIndexes": +[ 2799 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2798 ] +, "main_type": "NONE", "name": "/blocks.29/self_attn/Shape_3_output_0", "outputIndexes": +[ 2800 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2800 ] +, "main_type": "NONE", "name": "Shape4488", "outputIndexes": +[ 2801 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2800 ] +, "main_type": "NONE", "name": "Rank4490", "outputIndexes": +[ 2802 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2802, 2802 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4492", "outputIndexes": +[ 2803 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2803 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4493", "outputIndexes": +[ 2804 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2803, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4495", "outputIndexes": +[ 2805 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2805 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4496", "outputIndexes": +[ 2806 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2801, 2804, 2806, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4499", "outputIndexes": +[ 2807 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2807 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4500", "outputIndexes": +[ 2808 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2808 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4501", "outputIndexes": +[ 2809 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2809, 2808 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4502", "outputIndexes": +[ 2810 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2800, 2810, 2803 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.29/self_attn/Gather_3_output_0", "outputIndexes": +[ 2811 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2811, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Div_1_output_0", "outputIndexes": +[ 2812 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2812, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2813 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2798, 2813, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.29/self_attn/Slice_3_output_0", "outputIndexes": +[ 2814 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2814 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.29/self_attn/Neg_1_output_0", "outputIndexes": +[ 2815 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2812, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2816 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2798, 20, 2816, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.29/self_attn/Slice_2_output_0", "outputIndexes": +[ 2817 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2815, 2817 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.29/self_attn/Concat_4_output_0", "outputIndexes": +[ 2818 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2818, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Mul_3_output_0", "outputIndexes": +[ 2819 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2799, 2819 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/self_attn/Add_1_output_0", "outputIndexes": +[ 2820 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4210 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3512107422, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.29/self_attn/v_proj/Linear", "outputIndexes": +[ 4219 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4219 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4220 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4220 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2821 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2756, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2822 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2764, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.29/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2823 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2822, 2823, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.29/self_attn/Concat_2_output_0", "outputIndexes": +[ 2824 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2821, 2824 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.29/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2825 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2791, 2820, 2825, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.29/self_attn/FusedAttention", "outputIndexes": +[ 2826 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2826 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4221 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4221 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4222 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4222 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3514728884, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.29/self_attn/o_proj/Linear", "outputIndexes": +[ 4223 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4223 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4224 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4224 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2827 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2743, 2827 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/Add_output_0", "outputIndexes": +[ 2828 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2828 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3525214666, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.29/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2829 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2829 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4225 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4225 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4226 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4226 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3525247434, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.29/mlp/gate_proj/Linear", "outputIndexes": +[ 4227 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4227 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4228 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4228 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2830 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2830 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.29/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2831 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4226 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3556704736, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.29/mlp/up_proj/Linear", "outputIndexes": +[ 4231 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4231 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4232 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4232 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2832 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2831, 2832 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/mlp/Mul_output_0", "outputIndexes": +[ 2833 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2833 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4233 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4233 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4234 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4234 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3588162038, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.29/mlp/down_proj/Linear", "outputIndexes": +[ 4235 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4235 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4236 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4236 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2834 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2828, 2834 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.29/Add_1_output_0", "outputIndexes": +[ 2835 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2835, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.30/Reshape_output_0", "outputIndexes": +[ 2836 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2836 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3619619340, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.30/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2837 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2837 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4237 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4237 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4238 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4238 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3619652108, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.30/self_attn/q_proj/Linear", "outputIndexes": +[ 4239 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4239 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4240 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4240 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2838 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2837 ] +, "main_type": "NONE", "name": "/blocks.30/self_attn/Shape_output_0", "outputIndexes": +[ 2839 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2839 ] +, "main_type": "NONE", "name": "Shape4549", "outputIndexes": +[ 2840 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2839 ] +, "main_type": "NONE", "name": "Rank4551", "outputIndexes": +[ 2841 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2841, 2841 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4553", "outputIndexes": +[ 2842 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2842 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4554", "outputIndexes": +[ 2843 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2842, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4556", "outputIndexes": +[ 2844 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2844 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4557", "outputIndexes": +[ 2845 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2840, 2843, 2845, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4560", "outputIndexes": +[ 2846 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2846 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4561", "outputIndexes": +[ 2847 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2847, 2847 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4563", "outputIndexes": +[ 2848 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2839, 2848, 2842 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.30/self_attn/Gather_output_0", "outputIndexes": +[ 2849 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2849, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2850 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2842 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4571", "outputIndexes": +[ 2851 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2844 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4574", "outputIndexes": +[ 2852 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2840, 2851, 2852, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4577", "outputIndexes": +[ 2853 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2853 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4578", "outputIndexes": +[ 2854 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2854 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4579", "outputIndexes": +[ 2855 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2855, 2854 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4580", "outputIndexes": +[ 2856 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2839, 2856, 2842 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.30/self_attn/Gather_1_output_0", "outputIndexes": +[ 2857 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2857, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2858 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2850, 2858, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.30/self_attn/Concat_output_0", "outputIndexes": +[ 2859 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2838, 2859 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.30/self_attn/Reshape_output_0", "outputIndexes": +[ 2860 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2860 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3630137890, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.30/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2862 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2862, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Mul_output_0", "outputIndexes": +[ 2863 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2862 ] +, "main_type": "NONE", "name": "/blocks.30/self_attn/Shape_2_output_0", "outputIndexes": +[ 2864 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2864 ] +, "main_type": "NONE", "name": "Shape4595", "outputIndexes": +[ 2865 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2864 ] +, "main_type": "NONE", "name": "Rank4597", "outputIndexes": +[ 2866 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2866, 2866 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4599", "outputIndexes": +[ 2867 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2867 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4600", "outputIndexes": +[ 2868 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2867, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4602", "outputIndexes": +[ 2869 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2869 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4603", "outputIndexes": +[ 2870 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2865, 2868, 2870, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4606", "outputIndexes": +[ 2871 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2871 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4607", "outputIndexes": +[ 2872 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2872 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4608", "outputIndexes": +[ 2873 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2873, 2872 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4609", "outputIndexes": +[ 2874 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2864, 2874, 2867 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.30/self_attn/Gather_2_output_0", "outputIndexes": +[ 2875 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2875, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Div_output_0", "outputIndexes": +[ 2876 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2876, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2877 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2862, 2877, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.30/self_attn/Slice_1_output_0", "outputIndexes": +[ 2878 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2878 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.30/self_attn/Neg_output_0", "outputIndexes": +[ 2879 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2876, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2880 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2862, 20, 2880, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.30/self_attn/Slice_output_0", "outputIndexes": +[ 2881 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2879, 2881 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.30/self_attn/Concat_3_output_0", "outputIndexes": +[ 2882 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2882, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Mul_1_output_0", "outputIndexes": +[ 2883 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2863, 2883 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Add_output_0", "outputIndexes": +[ 2884 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4238 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3630138914, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.30/self_attn/k_proj/Linear", "outputIndexes": +[ 4243 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4243 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4244 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4244 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2885 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2849, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2886 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2857, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2887 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2886, 2887, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.30/self_attn/Concat_1_output_0", "outputIndexes": +[ 2888 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2885, 2888 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.30/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2889 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2889 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3632760376, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.30/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2891 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2891, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Mul_2_output_0", "outputIndexes": +[ 2892 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2891 ] +, "main_type": "NONE", "name": "/blocks.30/self_attn/Shape_3_output_0", "outputIndexes": +[ 2893 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2893 ] +, "main_type": "NONE", "name": "Shape4637", "outputIndexes": +[ 2894 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2893 ] +, "main_type": "NONE", "name": "Rank4639", "outputIndexes": +[ 2895 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2895, 2895 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4641", "outputIndexes": +[ 2896 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2896 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4642", "outputIndexes": +[ 2897 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2896, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4644", "outputIndexes": +[ 2898 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2898 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4645", "outputIndexes": +[ 2899 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2894, 2897, 2899, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4648", "outputIndexes": +[ 2900 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2900 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4649", "outputIndexes": +[ 2901 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2901 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4650", "outputIndexes": +[ 2902 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2902, 2901 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4651", "outputIndexes": +[ 2903 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2893, 2903, 2896 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.30/self_attn/Gather_3_output_0", "outputIndexes": +[ 2904 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2904, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Div_1_output_0", "outputIndexes": +[ 2905 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2905, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2906 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2891, 2906, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.30/self_attn/Slice_3_output_0", "outputIndexes": +[ 2907 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2907 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.30/self_attn/Neg_1_output_0", "outputIndexes": +[ 2908 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2905, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 2909 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2891, 20, 2909, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.30/self_attn/Slice_2_output_0", "outputIndexes": +[ 2910 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2908, 2910 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.30/self_attn/Concat_4_output_0", "outputIndexes": +[ 2911 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2911, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Mul_3_output_0", "outputIndexes": +[ 2912 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2892, 2912 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/self_attn/Add_1_output_0", "outputIndexes": +[ 2913 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4238 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3632761400, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.30/self_attn/v_proj/Linear", "outputIndexes": +[ 4247 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4247 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4248 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4248 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2914 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2849, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 2915 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2857, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.30/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 2916 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2915, 2916, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.30/self_attn/Concat_2_output_0", "outputIndexes": +[ 2917 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2914, 2917 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.30/self_attn/Reshape_2_output_0", "outputIndexes": +[ 2918 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2884, 2913, 2918, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.30/self_attn/FusedAttention", "outputIndexes": +[ 2919 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2919 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4249 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4249 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4250 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4250 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3635382862, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.30/self_attn/o_proj/Linear", "outputIndexes": +[ 4251 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4251 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4252 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4252 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2920 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2836, 2920 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/Add_output_0", "outputIndexes": +[ 2921 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2921 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3645868644, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.30/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 2922 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2922 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4253 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4253 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4254 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4254 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3645901412, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.30/mlp/gate_proj/Linear", "outputIndexes": +[ 4255 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4255 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4256 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4256 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2923 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2923 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.30/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 2924 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4254 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3677358714, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.30/mlp/up_proj/Linear", "outputIndexes": +[ 4259 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4259 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4260 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4260 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2925 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2924, 2925 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/mlp/Mul_output_0", "outputIndexes": +[ 2926 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2926 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4261 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4261 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4262 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4262 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3708816016, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.30/mlp/down_proj/Linear", "outputIndexes": +[ 4263 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4263 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4264 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4264 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2927 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2921, 2927 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.30/Add_1_output_0", "outputIndexes": +[ 2928 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2928, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.31/Reshape_output_0", "outputIndexes": +[ 2929 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2929 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3740273318, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.31/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 2930 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2930 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4265 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4265 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4266 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4266 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3740306086, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.31/self_attn/q_proj/Linear", "outputIndexes": +[ 4267 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4267 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4268 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4268 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2931 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2930 ] +, "main_type": "NONE", "name": "/blocks.31/self_attn/Shape_output_0", "outputIndexes": +[ 2932 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2932 ] +, "main_type": "NONE", "name": "Shape4698", "outputIndexes": +[ 2933 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2932 ] +, "main_type": "NONE", "name": "Rank4700", "outputIndexes": +[ 2934 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2934, 2934 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4702", "outputIndexes": +[ 2935 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2935 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4703", "outputIndexes": +[ 2936 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2935, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4705", "outputIndexes": +[ 2937 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2937 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4706", "outputIndexes": +[ 2938 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2933, 2936, 2938, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4709", "outputIndexes": +[ 2939 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2939 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4710", "outputIndexes": +[ 2940 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2940, 2940 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4712", "outputIndexes": +[ 2941 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2932, 2941, 2935 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.31/self_attn/Gather_output_0", "outputIndexes": +[ 2942 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2942, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 2943 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2935 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4720", "outputIndexes": +[ 2944 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2937 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4723", "outputIndexes": +[ 2945 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2933, 2944, 2945, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4726", "outputIndexes": +[ 2946 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2946 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4727", "outputIndexes": +[ 2947 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 2947 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4728", "outputIndexes": +[ 2948 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2948, 2947 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4729", "outputIndexes": +[ 2949 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2932, 2949, 2935 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.31/self_attn/Gather_1_output_0", "outputIndexes": +[ 2950 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2950, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 2951 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2943, 2951, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.31/self_attn/Concat_output_0", "outputIndexes": +[ 2952 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2931, 2952 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.31/self_attn/Reshape_output_0", "outputIndexes": +[ 2953 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2953 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3750791868, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.31/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 2955 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2955, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Mul_output_0", "outputIndexes": +[ 2956 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2955 ] +, "main_type": "NONE", "name": "/blocks.31/self_attn/Shape_2_output_0", "outputIndexes": +[ 2957 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2957 ] +, "main_type": "NONE", "name": "Shape4744", "outputIndexes": +[ 2958 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2957 ] +, "main_type": "NONE", "name": "Rank4746", "outputIndexes": +[ 2959 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2959, 2959 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4748", "outputIndexes": +[ 2960 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2960 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4749", "outputIndexes": +[ 2961 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2960, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4751", "outputIndexes": +[ 2962 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2962 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4752", "outputIndexes": +[ 2963 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2958, 2961, 2963, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4755", "outputIndexes": +[ 2964 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2964 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4756", "outputIndexes": +[ 2965 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2965 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4757", "outputIndexes": +[ 2966 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2966, 2965 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4758", "outputIndexes": +[ 2967 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2957, 2967, 2960 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.31/self_attn/Gather_2_output_0", "outputIndexes": +[ 2968 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2968, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Div_output_0", "outputIndexes": +[ 2969 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2969, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 2970 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2955, 2970, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.31/self_attn/Slice_1_output_0", "outputIndexes": +[ 2971 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2971 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.31/self_attn/Neg_output_0", "outputIndexes": +[ 2972 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2969, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 2973 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2955, 20, 2973, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.31/self_attn/Slice_output_0", "outputIndexes": +[ 2974 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2972, 2974 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.31/self_attn/Concat_3_output_0", "outputIndexes": +[ 2975 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2975, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Mul_1_output_0", "outputIndexes": +[ 2976 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2956, 2976 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Add_output_0", "outputIndexes": +[ 2977 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4266 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3750792892, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.31/self_attn/k_proj/Linear", "outputIndexes": +[ 4271 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4271 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4272 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4272 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2978 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2942, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 2979 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2950, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 2980 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2979, 2980, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.31/self_attn/Concat_1_output_0", "outputIndexes": +[ 2981 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2978, 2981 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.31/self_attn/Reshape_1_output_0", "outputIndexes": +[ 2982 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2982 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3753414354, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.31/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 2984 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2984, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Mul_2_output_0", "outputIndexes": +[ 2985 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2984 ] +, "main_type": "NONE", "name": "/blocks.31/self_attn/Shape_3_output_0", "outputIndexes": +[ 2986 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2986 ] +, "main_type": "NONE", "name": "Shape4786", "outputIndexes": +[ 2987 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2986 ] +, "main_type": "NONE", "name": "Rank4788", "outputIndexes": +[ 2988 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2988, 2988 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4790", "outputIndexes": +[ 2989 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2989 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4791", "outputIndexes": +[ 2990 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2989, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4793", "outputIndexes": +[ 2991 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2991 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4794", "outputIndexes": +[ 2992 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2987, 2990, 2992, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4797", "outputIndexes": +[ 2993 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2993 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4798", "outputIndexes": +[ 2994 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 2994 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4799", "outputIndexes": +[ 2995 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2995, 2994 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4800", "outputIndexes": +[ 2996 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2986, 2996, 2989 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.31/self_attn/Gather_3_output_0", "outputIndexes": +[ 2997 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2997, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Div_1_output_0", "outputIndexes": +[ 2998 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2998, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 2999 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2984, 2999, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.31/self_attn/Slice_3_output_0", "outputIndexes": +[ 3000 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3000 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.31/self_attn/Neg_1_output_0", "outputIndexes": +[ 3001 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2998, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 3002 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2984, 20, 3002, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.31/self_attn/Slice_2_output_0", "outputIndexes": +[ 3003 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3001, 3003 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.31/self_attn/Concat_4_output_0", "outputIndexes": +[ 3004 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3004, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Mul_3_output_0", "outputIndexes": +[ 3005 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2985, 3005 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/self_attn/Add_1_output_0", "outputIndexes": +[ 3006 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4266 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3753415378, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.31/self_attn/v_proj/Linear", "outputIndexes": +[ 4275 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4275 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4276 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4276 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3007 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2942, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 3008 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2950, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.31/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 3009 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3008, 3009, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.31/self_attn/Concat_2_output_0", "outputIndexes": +[ 3010 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3007, 3010 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.31/self_attn/Reshape_2_output_0", "outputIndexes": +[ 3011 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2977, 3006, 3011, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.31/self_attn/FusedAttention", "outputIndexes": +[ 3012 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3012 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4277 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4277 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4278 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4278 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3756036840, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.31/self_attn/o_proj/Linear", "outputIndexes": +[ 4279 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4279 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4280 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4280 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3013 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2929, 3013 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/Add_output_0", "outputIndexes": +[ 3014 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3014 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3766522622, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.31/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 3015 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3015 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4281 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4281 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4282 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4282 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3766555390, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.31/mlp/gate_proj/Linear", "outputIndexes": +[ 4283 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4283 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4284 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4284 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3016 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3016 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.31/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 3017 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4282 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3798012692, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.31/mlp/up_proj/Linear", "outputIndexes": +[ 4287 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4287 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4288 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4288 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3018 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3017, 3018 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/mlp/Mul_output_0", "outputIndexes": +[ 3019 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3019 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4289 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4289 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4290 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4290 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3829469994, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.31/mlp/down_proj/Linear", "outputIndexes": +[ 4291 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4291 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4292 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4292 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3020 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3014, 3020 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.31/Add_1_output_0", "outputIndexes": +[ 3021 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3021, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.32/Reshape_output_0", "outputIndexes": +[ 3022 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3022 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3860927296, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.32/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 3023 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3023 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4293 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4293 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4294 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4294 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3860960064, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.32/self_attn/q_proj/Linear", "outputIndexes": +[ 4295 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4295 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4296 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4296 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3024 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3023 ] +, "main_type": "NONE", "name": "/blocks.32/self_attn/Shape_output_0", "outputIndexes": +[ 3025 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3025 ] +, "main_type": "NONE", "name": "Shape4847", "outputIndexes": +[ 3026 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3025 ] +, "main_type": "NONE", "name": "Rank4849", "outputIndexes": +[ 3027 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3027, 3027 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4851", "outputIndexes": +[ 3028 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3028 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4852", "outputIndexes": +[ 3029 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3028, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4854", "outputIndexes": +[ 3030 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3030 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4855", "outputIndexes": +[ 3031 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3026, 3029, 3031, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4858", "outputIndexes": +[ 3032 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3032 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4859", "outputIndexes": +[ 3033 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3033, 3033 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4861", "outputIndexes": +[ 3034 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3025, 3034, 3028 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.32/self_attn/Gather_output_0", "outputIndexes": +[ 3035 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3035, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 3036 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3028 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4869", "outputIndexes": +[ 3037 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3030 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4872", "outputIndexes": +[ 3038 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3026, 3037, 3038, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4875", "outputIndexes": +[ 3039 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3039 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4876", "outputIndexes": +[ 3040 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 3040 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4877", "outputIndexes": +[ 3041 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3041, 3040 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4878", "outputIndexes": +[ 3042 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3025, 3042, 3028 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.32/self_attn/Gather_1_output_0", "outputIndexes": +[ 3043 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3043, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 3044 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3036, 3044, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.32/self_attn/Concat_output_0", "outputIndexes": +[ 3045 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3024, 3045 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.32/self_attn/Reshape_output_0", "outputIndexes": +[ 3046 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3046 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3871445846, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.32/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 3048 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3048, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Mul_output_0", "outputIndexes": +[ 3049 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3048 ] +, "main_type": "NONE", "name": "/blocks.32/self_attn/Shape_2_output_0", "outputIndexes": +[ 3050 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3050 ] +, "main_type": "NONE", "name": "Shape4893", "outputIndexes": +[ 3051 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3050 ] +, "main_type": "NONE", "name": "Rank4895", "outputIndexes": +[ 3052 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3052, 3052 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4897", "outputIndexes": +[ 3053 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3053 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4898", "outputIndexes": +[ 3054 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3053, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4900", "outputIndexes": +[ 3055 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3055 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4901", "outputIndexes": +[ 3056 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3051, 3054, 3056, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4904", "outputIndexes": +[ 3057 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3057 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4905", "outputIndexes": +[ 3058 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3058 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4906", "outputIndexes": +[ 3059 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3059, 3058 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4907", "outputIndexes": +[ 3060 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3050, 3060, 3053 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.32/self_attn/Gather_2_output_0", "outputIndexes": +[ 3061 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3061, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Div_output_0", "outputIndexes": +[ 3062 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3062, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 3063 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3048, 3063, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.32/self_attn/Slice_1_output_0", "outputIndexes": +[ 3064 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3064 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.32/self_attn/Neg_output_0", "outputIndexes": +[ 3065 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3062, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 3066 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3048, 20, 3066, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.32/self_attn/Slice_output_0", "outputIndexes": +[ 3067 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3065, 3067 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.32/self_attn/Concat_3_output_0", "outputIndexes": +[ 3068 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3068, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Mul_1_output_0", "outputIndexes": +[ 3069 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3049, 3069 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Add_output_0", "outputIndexes": +[ 3070 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4294 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3871446870, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.32/self_attn/k_proj/Linear", "outputIndexes": +[ 4299 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4299 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4300 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4300 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3071 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3035, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 3072 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3043, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 3073 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3072, 3073, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.32/self_attn/Concat_1_output_0", "outputIndexes": +[ 3074 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3071, 3074 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.32/self_attn/Reshape_1_output_0", "outputIndexes": +[ 3075 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3075 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3874068332, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.32/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 3077 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3077, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Mul_2_output_0", "outputIndexes": +[ 3078 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3077 ] +, "main_type": "NONE", "name": "/blocks.32/self_attn/Shape_3_output_0", "outputIndexes": +[ 3079 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3079 ] +, "main_type": "NONE", "name": "Shape4935", "outputIndexes": +[ 3080 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3079 ] +, "main_type": "NONE", "name": "Rank4937", "outputIndexes": +[ 3081 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3081, 3081 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4939", "outputIndexes": +[ 3082 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3082 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4940", "outputIndexes": +[ 3083 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3082, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4942", "outputIndexes": +[ 3084 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3084 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4943", "outputIndexes": +[ 3085 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3080, 3083, 3085, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4946", "outputIndexes": +[ 3086 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3086 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4947", "outputIndexes": +[ 3087 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3087 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4948", "outputIndexes": +[ 3088 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3088, 3087 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4949", "outputIndexes": +[ 3089 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3079, 3089, 3082 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.32/self_attn/Gather_3_output_0", "outputIndexes": +[ 3090 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3090, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Div_1_output_0", "outputIndexes": +[ 3091 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3091, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 3092 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3077, 3092, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.32/self_attn/Slice_3_output_0", "outputIndexes": +[ 3093 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3093 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.32/self_attn/Neg_1_output_0", "outputIndexes": +[ 3094 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3091, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 3095 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3077, 20, 3095, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.32/self_attn/Slice_2_output_0", "outputIndexes": +[ 3096 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3094, 3096 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.32/self_attn/Concat_4_output_0", "outputIndexes": +[ 3097 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3097, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Mul_3_output_0", "outputIndexes": +[ 3098 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3078, 3098 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/self_attn/Add_1_output_0", "outputIndexes": +[ 3099 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4294 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3874069356, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.32/self_attn/v_proj/Linear", "outputIndexes": +[ 4303 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4303 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4304 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4304 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3100 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3035, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 3101 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3043, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.32/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 3102 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3101, 3102, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.32/self_attn/Concat_2_output_0", "outputIndexes": +[ 3103 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3100, 3103 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.32/self_attn/Reshape_2_output_0", "outputIndexes": +[ 3104 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3070, 3099, 3104, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.32/self_attn/FusedAttention", "outputIndexes": +[ 3105 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3105 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4305 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4305 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4306 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4306 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3876690818, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.32/self_attn/o_proj/Linear", "outputIndexes": +[ 4307 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4307 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4308 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4308 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3106 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3022, 3106 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/Add_output_0", "outputIndexes": +[ 3107 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3107 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3887176600, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.32/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 3108 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3108 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4309 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4309 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4310 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4310 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3887209368, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.32/mlp/gate_proj/Linear", "outputIndexes": +[ 4311 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4311 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4312 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4312 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3109 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3109 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.32/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 3110 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4310 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3918666670, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.32/mlp/up_proj/Linear", "outputIndexes": +[ 4315 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4315 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4316 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4316 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3111 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3110, 3111 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/mlp/Mul_output_0", "outputIndexes": +[ 3112 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3112 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4317 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4317 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4318 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4318 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3950123972, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.32/mlp/down_proj/Linear", "outputIndexes": +[ 4319 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4319 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4320 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4320 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3113 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3107, 3113 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.32/Add_1_output_0", "outputIndexes": +[ 3114 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3114, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.33/Reshape_output_0", "outputIndexes": +[ 3115 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3115 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3981581274, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.33/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 3116 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3116 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4321 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4321 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4322 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4322 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3981614042, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.33/self_attn/q_proj/Linear", "outputIndexes": +[ 4323 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4323 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4324 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4324 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3117 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3116 ] +, "main_type": "NONE", "name": "/blocks.33/self_attn/Shape_output_0", "outputIndexes": +[ 3118 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3118 ] +, "main_type": "NONE", "name": "Shape4996", "outputIndexes": +[ 3119 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3118 ] +, "main_type": "NONE", "name": "Rank4998", "outputIndexes": +[ 3120 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3120, 3120 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5000", "outputIndexes": +[ 3121 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3121 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5001", "outputIndexes": +[ 3122 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3121, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5003", "outputIndexes": +[ 3123 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3123 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5004", "outputIndexes": +[ 3124 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3119, 3122, 3124, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5007", "outputIndexes": +[ 3125 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3125 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5008", "outputIndexes": +[ 3126 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3126, 3126 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5010", "outputIndexes": +[ 3127 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3118, 3127, 3121 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.33/self_attn/Gather_output_0", "outputIndexes": +[ 3128 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3128, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 3129 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3121 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5018", "outputIndexes": +[ 3130 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3123 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5021", "outputIndexes": +[ 3131 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3119, 3130, 3131, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5024", "outputIndexes": +[ 3132 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3132 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5025", "outputIndexes": +[ 3133 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 3133 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5026", "outputIndexes": +[ 3134 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3134, 3133 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5027", "outputIndexes": +[ 3135 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3118, 3135, 3121 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.33/self_attn/Gather_1_output_0", "outputIndexes": +[ 3136 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3136, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 3137 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3129, 3137, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.33/self_attn/Concat_output_0", "outputIndexes": +[ 3138 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3117, 3138 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.33/self_attn/Reshape_output_0", "outputIndexes": +[ 3139 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3139 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3992099824, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.33/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 3141 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3141, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Mul_output_0", "outputIndexes": +[ 3142 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3141 ] +, "main_type": "NONE", "name": "/blocks.33/self_attn/Shape_2_output_0", "outputIndexes": +[ 3143 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3143 ] +, "main_type": "NONE", "name": "Shape5042", "outputIndexes": +[ 3144 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3143 ] +, "main_type": "NONE", "name": "Rank5044", "outputIndexes": +[ 3145 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3145, 3145 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5046", "outputIndexes": +[ 3146 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3146 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5047", "outputIndexes": +[ 3147 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3146, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5049", "outputIndexes": +[ 3148 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3148 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5050", "outputIndexes": +[ 3149 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3144, 3147, 3149, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5053", "outputIndexes": +[ 3150 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3150 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5054", "outputIndexes": +[ 3151 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3151 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5055", "outputIndexes": +[ 3152 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3152, 3151 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5056", "outputIndexes": +[ 3153 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3143, 3153, 3146 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.33/self_attn/Gather_2_output_0", "outputIndexes": +[ 3154 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3154, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Div_output_0", "outputIndexes": +[ 3155 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3155, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 3156 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3141, 3156, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.33/self_attn/Slice_1_output_0", "outputIndexes": +[ 3157 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3157 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.33/self_attn/Neg_output_0", "outputIndexes": +[ 3158 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3155, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 3159 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3141, 20, 3159, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.33/self_attn/Slice_output_0", "outputIndexes": +[ 3160 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3158, 3160 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.33/self_attn/Concat_3_output_0", "outputIndexes": +[ 3161 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3161, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Mul_1_output_0", "outputIndexes": +[ 3162 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3142, 3162 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Add_output_0", "outputIndexes": +[ 3163 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4322 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3992100848, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.33/self_attn/k_proj/Linear", "outputIndexes": +[ 4327 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4327 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4328 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4328 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3164 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3128, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 3165 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3136, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 3166 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3165, 3166, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.33/self_attn/Concat_1_output_0", "outputIndexes": +[ 3167 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3164, 3167 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.33/self_attn/Reshape_1_output_0", "outputIndexes": +[ 3168 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3168 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3994722310, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.33/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 3170 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3170, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Mul_2_output_0", "outputIndexes": +[ 3171 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3170 ] +, "main_type": "NONE", "name": "/blocks.33/self_attn/Shape_3_output_0", "outputIndexes": +[ 3172 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3172 ] +, "main_type": "NONE", "name": "Shape5084", "outputIndexes": +[ 3173 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3172 ] +, "main_type": "NONE", "name": "Rank5086", "outputIndexes": +[ 3174 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3174, 3174 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5088", "outputIndexes": +[ 3175 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3175 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5089", "outputIndexes": +[ 3176 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3175, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5091", "outputIndexes": +[ 3177 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3177 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5092", "outputIndexes": +[ 3178 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3173, 3176, 3178, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5095", "outputIndexes": +[ 3179 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3179 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5096", "outputIndexes": +[ 3180 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3180 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5097", "outputIndexes": +[ 3181 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3181, 3180 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5098", "outputIndexes": +[ 3182 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3172, 3182, 3175 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.33/self_attn/Gather_3_output_0", "outputIndexes": +[ 3183 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3183, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Div_1_output_0", "outputIndexes": +[ 3184 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3184, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 3185 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3170, 3185, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.33/self_attn/Slice_3_output_0", "outputIndexes": +[ 3186 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3186 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.33/self_attn/Neg_1_output_0", "outputIndexes": +[ 3187 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3184, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 3188 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3170, 20, 3188, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.33/self_attn/Slice_2_output_0", "outputIndexes": +[ 3189 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3187, 3189 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.33/self_attn/Concat_4_output_0", "outputIndexes": +[ 3190 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3190, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Mul_3_output_0", "outputIndexes": +[ 3191 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3171, 3191 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/self_attn/Add_1_output_0", "outputIndexes": +[ 3192 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4322 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3994723334, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.33/self_attn/v_proj/Linear", "outputIndexes": +[ 4331 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4331 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4332 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4332 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3193 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3128, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 3194 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3136, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.33/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 3195 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3194, 3195, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.33/self_attn/Concat_2_output_0", "outputIndexes": +[ 3196 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3193, 3196 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.33/self_attn/Reshape_2_output_0", "outputIndexes": +[ 3197 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3163, 3192, 3197, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.33/self_attn/FusedAttention", "outputIndexes": +[ 3198 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3198 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4333 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4333 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4334 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4334 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 3997344796, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.33/self_attn/o_proj/Linear", "outputIndexes": +[ 4335 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4335 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4336 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4336 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3199 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3115, 3199 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/Add_output_0", "outputIndexes": +[ 3200 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3200 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4007830578, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.33/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 3201 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3201 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4337 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4337 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4338 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4338 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4007863346, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.33/mlp/gate_proj/Linear", "outputIndexes": +[ 4339 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4339 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4340 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4340 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3202 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3202 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.33/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 3203 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4338 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4039320648, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.33/mlp/up_proj/Linear", "outputIndexes": +[ 4343 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4343 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4344 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4344 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3204 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3203, 3204 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/mlp/Mul_output_0", "outputIndexes": +[ 3205 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3205 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4345 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4345 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4346 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4346 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4070777950, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.33/mlp/down_proj/Linear", "outputIndexes": +[ 4347 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4347 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4348 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4348 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3206 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3200, 3206 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.33/Add_1_output_0", "outputIndexes": +[ 3207 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3207, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.34/Reshape_output_0", "outputIndexes": +[ 3208 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3208 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4102235252, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.34/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 3209 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3209 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4349 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4349 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4350 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4350 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4102268020, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.34/self_attn/q_proj/Linear", "outputIndexes": +[ 4351 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4351 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4352 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4352 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3210 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3209 ] +, "main_type": "NONE", "name": "/blocks.34/self_attn/Shape_output_0", "outputIndexes": +[ 3211 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3211 ] +, "main_type": "NONE", "name": "Shape5145", "outputIndexes": +[ 3212 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3211 ] +, "main_type": "NONE", "name": "Rank5147", "outputIndexes": +[ 3213 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3213, 3213 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5149", "outputIndexes": +[ 3214 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3214 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5150", "outputIndexes": +[ 3215 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3214, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5152", "outputIndexes": +[ 3216 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3216 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5153", "outputIndexes": +[ 3217 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3212, 3215, 3217, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5156", "outputIndexes": +[ 3218 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3218 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5157", "outputIndexes": +[ 3219 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3219, 3219 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5159", "outputIndexes": +[ 3220 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3211, 3220, 3214 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.34/self_attn/Gather_output_0", "outputIndexes": +[ 3221 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3221, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 3222 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3214 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5167", "outputIndexes": +[ 3223 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3216 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5170", "outputIndexes": +[ 3224 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3212, 3223, 3224, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5173", "outputIndexes": +[ 3225 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3225 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5174", "outputIndexes": +[ 3226 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 3226 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5175", "outputIndexes": +[ 3227 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3227, 3226 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5176", "outputIndexes": +[ 3228 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3211, 3228, 3214 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.34/self_attn/Gather_1_output_0", "outputIndexes": +[ 3229 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3229, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 3230 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3222, 3230, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.34/self_attn/Concat_output_0", "outputIndexes": +[ 3231 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3210, 3231 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.34/self_attn/Reshape_output_0", "outputIndexes": +[ 3232 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3232 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4112753802, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.34/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 3234 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3234, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Mul_output_0", "outputIndexes": +[ 3235 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3234 ] +, "main_type": "NONE", "name": "/blocks.34/self_attn/Shape_2_output_0", "outputIndexes": +[ 3236 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3236 ] +, "main_type": "NONE", "name": "Shape5191", "outputIndexes": +[ 3237 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3236 ] +, "main_type": "NONE", "name": "Rank5193", "outputIndexes": +[ 3238 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3238, 3238 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5195", "outputIndexes": +[ 3239 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3239 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5196", "outputIndexes": +[ 3240 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3239, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5198", "outputIndexes": +[ 3241 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3241 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5199", "outputIndexes": +[ 3242 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3237, 3240, 3242, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5202", "outputIndexes": +[ 3243 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3243 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5203", "outputIndexes": +[ 3244 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3244 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5204", "outputIndexes": +[ 3245 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3245, 3244 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5205", "outputIndexes": +[ 3246 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3236, 3246, 3239 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.34/self_attn/Gather_2_output_0", "outputIndexes": +[ 3247 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3247, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Div_output_0", "outputIndexes": +[ 3248 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3248, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 3249 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3234, 3249, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.34/self_attn/Slice_1_output_0", "outputIndexes": +[ 3250 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3250 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.34/self_attn/Neg_output_0", "outputIndexes": +[ 3251 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3248, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 3252 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3234, 20, 3252, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.34/self_attn/Slice_output_0", "outputIndexes": +[ 3253 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3251, 3253 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.34/self_attn/Concat_3_output_0", "outputIndexes": +[ 3254 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3254, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Mul_1_output_0", "outputIndexes": +[ 3255 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3235, 3255 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Add_output_0", "outputIndexes": +[ 3256 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4350 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4112754826, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.34/self_attn/k_proj/Linear", "outputIndexes": +[ 4355 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4355 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4356 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4356 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3257 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3221, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 3258 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3229, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 3259 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3258, 3259, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.34/self_attn/Concat_1_output_0", "outputIndexes": +[ 3260 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3257, 3260 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.34/self_attn/Reshape_1_output_0", "outputIndexes": +[ 3261 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3261 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4115376288, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.34/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 3263 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3263, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Mul_2_output_0", "outputIndexes": +[ 3264 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3263 ] +, "main_type": "NONE", "name": "/blocks.34/self_attn/Shape_3_output_0", "outputIndexes": +[ 3265 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3265 ] +, "main_type": "NONE", "name": "Shape5233", "outputIndexes": +[ 3266 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3265 ] +, "main_type": "NONE", "name": "Rank5235", "outputIndexes": +[ 3267 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3267, 3267 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5237", "outputIndexes": +[ 3268 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3268 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5238", "outputIndexes": +[ 3269 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3268, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5240", "outputIndexes": +[ 3270 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3270 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5241", "outputIndexes": +[ 3271 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3266, 3269, 3271, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5244", "outputIndexes": +[ 3272 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3272 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5245", "outputIndexes": +[ 3273 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3273 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5246", "outputIndexes": +[ 3274 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3274, 3273 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5247", "outputIndexes": +[ 3275 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3265, 3275, 3268 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.34/self_attn/Gather_3_output_0", "outputIndexes": +[ 3276 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3276, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Div_1_output_0", "outputIndexes": +[ 3277 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3277, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 3278 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3263, 3278, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.34/self_attn/Slice_3_output_0", "outputIndexes": +[ 3279 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3279 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.34/self_attn/Neg_1_output_0", "outputIndexes": +[ 3280 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3277, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 3281 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3263, 20, 3281, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.34/self_attn/Slice_2_output_0", "outputIndexes": +[ 3282 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3280, 3282 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.34/self_attn/Concat_4_output_0", "outputIndexes": +[ 3283 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3283, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Mul_3_output_0", "outputIndexes": +[ 3284 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3264, 3284 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/self_attn/Add_1_output_0", "outputIndexes": +[ 3285 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4350 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4115377312, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.34/self_attn/v_proj/Linear", "outputIndexes": +[ 4359 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4359 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4360 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4360 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3286 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3221, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 3287 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3229, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.34/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 3288 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3287, 3288, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.34/self_attn/Concat_2_output_0", "outputIndexes": +[ 3289 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3286, 3289 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.34/self_attn/Reshape_2_output_0", "outputIndexes": +[ 3290 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3256, 3285, 3290, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.34/self_attn/FusedAttention", "outputIndexes": +[ 3291 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3291 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4361 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4361 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4362 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4362 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4117998774, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.34/self_attn/o_proj/Linear", "outputIndexes": +[ 4363 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4363 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4364 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4364 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3292 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3208, 3292 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/Add_output_0", "outputIndexes": +[ 3293 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3293 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4128484556, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.34/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 3294 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3294 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4365 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4365 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4366 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4366 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4128517324, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.34/mlp/gate_proj/Linear", "outputIndexes": +[ 4367 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4367 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4368 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4368 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3295 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3295 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.34/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 3296 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4366 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4159974626, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.34/mlp/up_proj/Linear", "outputIndexes": +[ 4371 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4371 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4372 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4372 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3297 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3296, 3297 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/mlp/Mul_output_0", "outputIndexes": +[ 3298 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3298 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4373 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4373 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4374 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4374 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4191431928, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.34/mlp/down_proj/Linear", "outputIndexes": +[ 4375 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4375 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4376 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4376 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3299 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3293, 3299 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.34/Add_1_output_0", "outputIndexes": +[ 3300 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3300, 3 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.35/Reshape_output_0", "outputIndexes": +[ 3301 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3301 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4222889230, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.35/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 3302 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3302 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4377 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4377 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4378 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4378 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4222921998, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.35/self_attn/q_proj/Linear", "outputIndexes": +[ 4379 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4379 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4380 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4380 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3303 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3302 ] +, "main_type": "NONE", "name": "/blocks.35/self_attn/Shape_output_0", "outputIndexes": +[ 3304 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3304 ] +, "main_type": "NONE", "name": "Shape5294", "outputIndexes": +[ 3305 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3304 ] +, "main_type": "NONE", "name": "Rank5296", "outputIndexes": +[ 3306 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3306, 3306 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5298", "outputIndexes": +[ 3307 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3307 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5299", "outputIndexes": +[ 3308 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3307, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5301", "outputIndexes": +[ 3309 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3309 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5302", "outputIndexes": +[ 3310 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3305, 3308, 3310, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5305", "outputIndexes": +[ 3311 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3311 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5306", "outputIndexes": +[ 3312 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3312, 3312 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5308", "outputIndexes": +[ 3313 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3304, 3313, 3307 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.35/self_attn/Gather_output_0", "outputIndexes": +[ 3314 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3314, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_output_0", "outputIndexes": +[ 3315 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3307 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5316", "outputIndexes": +[ 3316 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3309 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5319", "outputIndexes": +[ 3317 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3305, 3316, 3317, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5322", "outputIndexes": +[ 3318 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3318 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5323", "outputIndexes": +[ 3319 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 3319 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5324", "outputIndexes": +[ 3320 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3320, 3319 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5325", "outputIndexes": +[ 3321 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3304, 3321, 3307 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.35/self_attn/Gather_1_output_0", "outputIndexes": +[ 3322 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3322, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_1_output_0", "outputIndexes": +[ 3323 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3315, 3323, 30, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.35/self_attn/Concat_output_0", "outputIndexes": +[ 3324 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3303, 3324 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.35/self_attn/Reshape_output_0", "outputIndexes": +[ 3325 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3325 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4233407780, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.35/self_attn/q_norm/Mul_1_output_0", "outputIndexes": +[ 3327 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3327, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Mul_output_0", "outputIndexes": +[ 3328 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3327 ] +, "main_type": "NONE", "name": "/blocks.35/self_attn/Shape_2_output_0", "outputIndexes": +[ 3329 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3329 ] +, "main_type": "NONE", "name": "Shape5340", "outputIndexes": +[ 3330 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3329 ] +, "main_type": "NONE", "name": "Rank5342", "outputIndexes": +[ 3331 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3331, 3331 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5344", "outputIndexes": +[ 3332 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3332 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5345", "outputIndexes": +[ 3333 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3332, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5347", "outputIndexes": +[ 3334 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3334 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5348", "outputIndexes": +[ 3335 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3330, 3333, 3335, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5351", "outputIndexes": +[ 3336 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3336 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5352", "outputIndexes": +[ 3337 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3337 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5353", "outputIndexes": +[ 3338 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3338, 3337 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5354", "outputIndexes": +[ 3339 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3329, 3339, 3332 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.35/self_attn/Gather_2_output_0", "outputIndexes": +[ 3340 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3340, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Div_output_0", "outputIndexes": +[ 3341 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3341, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_7_output_0", "outputIndexes": +[ 3342 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3327, 3342, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.35/self_attn/Slice_1_output_0", "outputIndexes": +[ 3343 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3343 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.35/self_attn/Neg_output_0", "outputIndexes": +[ 3344 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3341, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_6_output_0", "outputIndexes": +[ 3345 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3327, 20, 3345, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.35/self_attn/Slice_output_0", "outputIndexes": +[ 3346 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3344, 3346 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.35/self_attn/Concat_3_output_0", "outputIndexes": +[ 3347 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3347, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Mul_1_output_0", "outputIndexes": +[ 3348 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3328, 3348 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Add_output_0", "outputIndexes": +[ 3349 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4378 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4233408804, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.35/self_attn/k_proj/Linear", "outputIndexes": +[ 4383 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4383 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4384 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4384 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3350 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3314, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_2_output_0", "outputIndexes": +[ 3351 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3322, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_3_output_0", "outputIndexes": +[ 3352 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3351, 3352, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.35/self_attn/Concat_1_output_0", "outputIndexes": +[ 3353 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3350, 3353 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.35/self_attn/Reshape_1_output_0", "outputIndexes": +[ 3354 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3354 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4236030266, 512, 512 ] +, "useRMSNorm": true } +, "name": "/blocks.35/self_attn/k_norm/Mul_1_output_0", "outputIndexes": +[ 3356 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3356, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Mul_2_output_0", "outputIndexes": +[ 3357 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3356 ] +, "main_type": "NONE", "name": "/blocks.35/self_attn/Shape_3_output_0", "outputIndexes": +[ 3358 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3358 ] +, "main_type": "NONE", "name": "Shape5382", "outputIndexes": +[ 3359 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3358 ] +, "main_type": "NONE", "name": "Rank5384", "outputIndexes": +[ 3360 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3360, 3360 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5386", "outputIndexes": +[ 3361 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3361 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5387", "outputIndexes": +[ 3362 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3361, 12 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5389", "outputIndexes": +[ 3363 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3363 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5390", "outputIndexes": +[ 3364 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3359, 3362, 3364, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5393", "outputIndexes": +[ 3365 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3365 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5394", "outputIndexes": +[ 3366 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 3366 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5395", "outputIndexes": +[ 3367 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3367, 3366 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MOD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5396", "outputIndexes": +[ 3368 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3358, 3368, 3361 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/blocks.35/self_attn/Gather_3_output_0", "outputIndexes": +[ 3369 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3369, 74 ] +, "main_type": "BinaryOp", "main": +{ "opType": "REALDIV", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Div_1_output_0", "outputIndexes": +[ 3370 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3370, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_9_output_0", "outputIndexes": +[ 3371 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3356, 3371, 77, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.35/self_attn/Slice_3_output_0", "outputIndexes": +[ 3372 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3372 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/blocks.35/self_attn/Neg_1_output_0", "outputIndexes": +[ 3373 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3370, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_8_output_0", "outputIndexes": +[ 3374 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3356, 20, 3374, 78, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/blocks.35/self_attn/Slice_2_output_0", "outputIndexes": +[ 3375 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3373, 3375 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/blocks.35/self_attn/Concat_4_output_0", "outputIndexes": +[ 3376 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3376, 90 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Mul_3_output_0", "outputIndexes": +[ 3377 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3357, 3377 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/self_attn/Add_1_output_0", "outputIndexes": +[ 3378 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4378 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 1024, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 65536, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4236031290, 2097174, 524288, 0, 0 ] + } +, "name": "/layers.35/self_attn/v_proj/Linear", "outputIndexes": +[ 4387 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4387 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4388 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4388 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3379 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3314, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_4_output_0", "outputIndexes": +[ 3380 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3322, 20 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/blocks.35/self_attn/Unsqueeze_5_output_0", "outputIndexes": +[ 3381 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3380, 3381, 96, 31 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/blocks.35/self_attn/Concat_2_output_0", "outputIndexes": +[ 3382 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3379, 3382 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/blocks.35/self_attn/Reshape_2_output_0", "outputIndexes": +[ 3383 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3349, 3378, 3383, 128 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "/layers.35/self_attn/FusedAttention", "outputIndexes": +[ 3384 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3384 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4389 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4389 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4390 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4390 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 262144, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4238652752, 8388630, 2097152, 0, 0 ] + } +, "name": "/layers.35/self_attn/o_proj/Linear", "outputIndexes": +[ 4391 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4391 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4392 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4392 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3385 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3301, 3385 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/Add_output_0", "outputIndexes": +[ 3386 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3386 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4249138534, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/blocks.35/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 3387 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3387 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4393 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4393 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4394 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4394 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4249171302, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.35/mlp/gate_proj/Linear", "outputIndexes": +[ 4395 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4395 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4396 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4396 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3388 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3388 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/blocks.35/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 3389 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4394 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 12288, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4280628604, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.35/mlp/up_proj/Linear", "outputIndexes": +[ 4399 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4399 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4400 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4400 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3390 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3389, 3390 ] +, "main_type": "BinaryOp", "main": +{ "opType": "MUL", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/mlp/Mul_output_0", "outputIndexes": +[ 3391 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3391 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4401 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4401 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4402 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4402 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 4096, "inputCount": 12288, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 786432, "has_scaleInt": false, "shapeInt32": false, "weightSize": 0 } +, "external": +[ 4312085906, 25165846, 6291456, 0, 0 ] + } +, "name": "/layers.35/mlp/down_proj/Linear", "outputIndexes": +[ 4403 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4403 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4404 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4404 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3392 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3386, 3392 ] +, "main_type": "BinaryOp", "main": +{ "opType": "ADD", "T": "DT_FLOAT", "activationType": 0 } +, "name": "/blocks.35/Add_1_output_0", "outputIndexes": +[ 3393 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3393, 1, 77, 15, 15 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_output_0", "outputIndexes": +[ 3394 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3394 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4343543208, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "hidden_states", "outputIndexes": +[ 3395 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3395 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/lm/lm_head/Linear/pre_reshape", "outputIndexes": +[ 4405 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4405 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/lm/lm_head/Linear/pre_convert", "outputIndexes": +[ 4406 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4406 ] +, "main_type": "Convolution2D", "main": +{ "common": +{ "padX": 0, "padY": 0, "kernelX": 1, "kernelY": 1, "strideX": 1, "strideY": 1, "dilateX": 1, "dilateY": 1, "padMode": "CAFFE", "group": 1, "outputCount": 151936, "inputCount": 4096, "relu": false, "relu6": false, "hasOutputShape": false } +, "quanParameter": +{ "type": 1, "useInt32": false, "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "aMaxOrBits": 4, "aMin": 1, "readType": 9723904, "has_scaleInt": false, "shapeInt32": true, "weightSize": 0 } +, "external": +[ 4343575976, 311164954, 77791232, 0, 0 ] + } +, "name": "/lm/lm_head/Linear", "outputIndexes": +[ 4407 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4407 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/lm/lm_head/Linear/post_convert", "outputIndexes": +[ 4408 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4408 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 151936 ] +, "dimType": "NCHW" } +, "name": "/lm/lm_head/Linear/post_reshape", "outputIndexes": +[ 3396 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } + ] +, "outputName": +[ "logits", "hidden_states" ] +, "preferForwardType": "CPU", "sourceType": "ONNX", "tensorName": +[ "position_ids", "logits_index", "input_ids", "/blocks.0/Constant_output_0", "/blocks.0/Reshape_output_0", "/blocks.0/input_layernorm/Mul_1_output_0", "/blocks.0/self_attn/q_proj/FakeLinear_output_0", "/blocks.0/self_attn/Shape_output_0", "Shape22", "Rank24", "BinaryOp26", "Unsqueeze27", "Const28", "BinaryOp29", "Unsqueeze30", "Unsqueeze32", "StridedSlice33", "Squeeze34", "BinaryOp36", "/blocks.0/self_attn/Gather_output_0", "/rotary/Constant_3_output_0", "/blocks.0/self_attn/Unsqueeze_output_0", "Unsqueeze46", "Unsqueeze49", "StridedSlice52", "Squeeze53", "BinaryOp54", "BinaryOp55", "/blocks.0/self_attn/Gather_1_output_0", "/blocks.0/self_attn/Unsqueeze_1_output_0", "/blocks.0/self_attn/Constant_2_output_0", "/blocks.0/self_attn/Constant_3_output_0", "/blocks.0/self_attn/Concat_output_0", "/blocks.0/self_attn/Reshape_output_0", "/blocks.0/self_attn/q_norm/Cast_output_0", "/blocks.0/self_attn/q_norm/Mul_1_output_0", "/rotary/Cast_output_0", "/rotary/Constant_output_0", "/rotary/Reshape_output_0", "/rotary/Constant_1_output_0", "/rotary/Mul_output_0", "/rotary/Cos_output_0", "/rotary/Unsqueeze_output_0", "/rotary/Sin_output_0", "/rotary/Unsqueeze_1_output_0", "/rotary/Concat_output_0", "/rotary/Concat_1_output_0", "/rotary/Constant_5_output_0", "/rotary/Unsqueeze_2_output_0", "/rotary/Unsqueeze_3_output_0", "Shape85", "Rank87", "BinaryOp89", "Unsqueeze90", "BinaryOp92", "Unsqueeze93", "StridedSlice96", "Squeeze97", "BinaryOp99", "/blocks.0/self_attn/Gather_2_output_0", "/blocks.0/self_attn/Mul_output_0", "/blocks.0/self_attn/Shape_2_output_0", "/blocks.0/self_attn/Constant_8_output_0", "Shape104", "Rank106", "BinaryOp108", "Unsqueeze109", "BinaryOp111", "Unsqueeze112", "StridedSlice115", "Squeeze116", "BinaryOp117", "BinaryOp118", "/blocks.0/self_attn/Gather_4_output_0", "/blocks.0/self_attn/Constant_9_output_0", "/blocks.0/self_attn/Div_output_0", "/blocks.0/self_attn/Unsqueeze_7_output_0", "/blocks.0/self_attn/Constant_16_output_0", "/blocks.0/self_attn/Constant_10_output_0", "/blocks.0/self_attn/Slice_1_output_0", "/blocks.0/self_attn/Neg_output_0", "/blocks.0/self_attn/Unsqueeze_6_output_0", "/blocks.0/self_attn/Slice_output_0", "/blocks.0/self_attn/Concat_3_output_0", "Unsqueeze136", "Unsqueeze139", "StridedSlice142", "Squeeze143", "BinaryOp144", "BinaryOp145", "/blocks.0/self_attn/Gather_3_output_0", "/blocks.0/self_attn/Mul_1_output_0", "/blocks.0/self_attn/Add_output_0", "/blocks.0/self_attn/k_proj/FakeLinear_output_0", "/blocks.0/self_attn/Unsqueeze_2_output_0", "/blocks.0/self_attn/Unsqueeze_3_output_0", "/blocks.0/self_attn/Constant_4_output_0", "/blocks.0/self_attn/Concat_1_output_0", "/blocks.0/self_attn/Reshape_1_output_0", "/blocks.0/self_attn/k_norm/Cast_output_0", "/blocks.0/self_attn/k_norm/Mul_1_output_0", "/blocks.0/self_attn/Mul_2_output_0", "/blocks.0/self_attn/Shape_3_output_0", "Shape166", "Rank168", "BinaryOp170", "Unsqueeze171", "BinaryOp173", "Unsqueeze174", "StridedSlice177", "Squeeze178", "BinaryOp179", "BinaryOp180", "/blocks.0/self_attn/Gather_5_output_0", "/blocks.0/self_attn/Div_1_output_0", "/blocks.0/self_attn/Unsqueeze_9_output_0", "/blocks.0/self_attn/Slice_3_output_0", "/blocks.0/self_attn/Neg_1_output_0", "/blocks.0/self_attn/Unsqueeze_8_output_0", "/blocks.0/self_attn/Slice_2_output_0", "/blocks.0/self_attn/Concat_4_output_0", "/blocks.0/self_attn/Mul_3_output_0", "/blocks.0/self_attn/Add_1_output_0", "/blocks.0/self_attn/v_proj/FakeLinear_output_0", "/blocks.0/self_attn/Unsqueeze_4_output_0", "/blocks.0/self_attn/Unsqueeze_5_output_0", "/blocks.0/self_attn/Concat_2_output_0", "/blocks.0/self_attn/Reshape_2_output_0", "attention_mask", "/blocks.0/self_attn/fused_attn/FusedAttention_output_0", "/blocks.0/self_attn/o_proj/FakeLinear_output_0", "/blocks.0/Add_output_0", "/blocks.0/post_attention_layernorm/Mul_1_output_0", "/blocks.0/mlp/gate_proj/FakeLinear_output_0", "/blocks.0/mlp/act_fn/Mul_output_0", "/blocks.0/mlp/up_proj/FakeLinear_output_0", "/blocks.0/mlp/Mul_output_0", "/blocks.0/mlp/down_proj/FakeLinear_output_0", "/blocks.0/Add_1_output_0", "/blocks.1/Reshape_output_0", "/blocks.1/input_layernorm/Mul_1_output_0", "/blocks.1/self_attn/q_proj/FakeLinear_output_0", "/blocks.1/self_attn/Shape_output_0", "Shape228", "Rank230", "BinaryOp232", "Unsqueeze233", "BinaryOp235", "Unsqueeze236", "StridedSlice239", "Squeeze240", "BinaryOp242", "/blocks.1/self_attn/Gather_output_0", "/blocks.1/self_attn/Unsqueeze_output_0", "Unsqueeze250", "Unsqueeze253", "StridedSlice256", "Squeeze257", "BinaryOp258", "BinaryOp259", "/blocks.1/self_attn/Gather_1_output_0", "/blocks.1/self_attn/Unsqueeze_1_output_0", "/blocks.1/self_attn/Concat_output_0", "/blocks.1/self_attn/Reshape_output_0", "/blocks.1/self_attn/q_norm/Cast_output_0", "/blocks.1/self_attn/q_norm/Mul_1_output_0", "/blocks.1/self_attn/Mul_output_0", "/blocks.1/self_attn/Shape_2_output_0", "Shape274", "Rank276", "BinaryOp278", "Unsqueeze279", "BinaryOp281", "Unsqueeze282", "StridedSlice285", "Squeeze286", "BinaryOp287", "BinaryOp288", "/blocks.1/self_attn/Gather_2_output_0", "/blocks.1/self_attn/Div_output_0", "/blocks.1/self_attn/Unsqueeze_7_output_0", "/blocks.1/self_attn/Slice_1_output_0", "/blocks.1/self_attn/Neg_output_0", "/blocks.1/self_attn/Unsqueeze_6_output_0", "/blocks.1/self_attn/Slice_output_0", "/blocks.1/self_attn/Concat_3_output_0", "/blocks.1/self_attn/Mul_1_output_0", "/blocks.1/self_attn/Add_output_0", "/blocks.1/self_attn/k_proj/FakeLinear_output_0", "/blocks.1/self_attn/Unsqueeze_2_output_0", "/blocks.1/self_attn/Unsqueeze_3_output_0", "/blocks.1/self_attn/Concat_1_output_0", "/blocks.1/self_attn/Reshape_1_output_0", "/blocks.1/self_attn/k_norm/Cast_output_0", "/blocks.1/self_attn/k_norm/Mul_1_output_0", "/blocks.1/self_attn/Mul_2_output_0", "/blocks.1/self_attn/Shape_3_output_0", "Shape316", "Rank318", "BinaryOp320", "Unsqueeze321", "BinaryOp323", "Unsqueeze324", "StridedSlice327", "Squeeze328", "BinaryOp329", "BinaryOp330", "/blocks.1/self_attn/Gather_3_output_0", "/blocks.1/self_attn/Div_1_output_0", "/blocks.1/self_attn/Unsqueeze_9_output_0", "/blocks.1/self_attn/Slice_3_output_0", "/blocks.1/self_attn/Neg_1_output_0", "/blocks.1/self_attn/Unsqueeze_8_output_0", "/blocks.1/self_attn/Slice_2_output_0", "/blocks.1/self_attn/Concat_4_output_0", "/blocks.1/self_attn/Mul_3_output_0", "/blocks.1/self_attn/Add_1_output_0", "/blocks.1/self_attn/v_proj/FakeLinear_output_0", "/blocks.1/self_attn/Unsqueeze_4_output_0", "/blocks.1/self_attn/Unsqueeze_5_output_0", "/blocks.1/self_attn/Concat_2_output_0", "/blocks.1/self_attn/Reshape_2_output_0", "/blocks.1/self_attn/fused_attn/FusedAttention_output_0", "/blocks.1/self_attn/o_proj/FakeLinear_output_0", "/blocks.1/Add_output_0", "/blocks.1/post_attention_layernorm/Mul_1_output_0", "/blocks.1/mlp/gate_proj/FakeLinear_output_0", "/blocks.1/mlp/act_fn/Mul_output_0", "/blocks.1/mlp/up_proj/FakeLinear_output_0", "/blocks.1/mlp/Mul_output_0", "/blocks.1/mlp/down_proj/FakeLinear_output_0", "/blocks.1/Add_1_output_0", "/blocks.2/Reshape_output_0", "/blocks.2/input_layernorm/Mul_1_output_0", "/blocks.2/self_attn/q_proj/FakeLinear_output_0", "/blocks.2/self_attn/Shape_output_0", "Shape377", "Rank379", "BinaryOp381", "Unsqueeze382", "BinaryOp384", "Unsqueeze385", "StridedSlice388", "Squeeze389", "BinaryOp391", "/blocks.2/self_attn/Gather_output_0", "/blocks.2/self_attn/Unsqueeze_output_0", "Unsqueeze399", "Unsqueeze402", "StridedSlice405", "Squeeze406", "BinaryOp407", "BinaryOp408", "/blocks.2/self_attn/Gather_1_output_0", "/blocks.2/self_attn/Unsqueeze_1_output_0", "/blocks.2/self_attn/Concat_output_0", "/blocks.2/self_attn/Reshape_output_0", "/blocks.2/self_attn/q_norm/Cast_output_0", "/blocks.2/self_attn/q_norm/Mul_1_output_0", "/blocks.2/self_attn/Mul_output_0", "/blocks.2/self_attn/Shape_2_output_0", "Shape423", "Rank425", "BinaryOp427", "Unsqueeze428", "BinaryOp430", "Unsqueeze431", "StridedSlice434", "Squeeze435", "BinaryOp436", "BinaryOp437", "/blocks.2/self_attn/Gather_2_output_0", "/blocks.2/self_attn/Div_output_0", "/blocks.2/self_attn/Unsqueeze_7_output_0", "/blocks.2/self_attn/Slice_1_output_0", "/blocks.2/self_attn/Neg_output_0", "/blocks.2/self_attn/Unsqueeze_6_output_0", "/blocks.2/self_attn/Slice_output_0", "/blocks.2/self_attn/Concat_3_output_0", "/blocks.2/self_attn/Mul_1_output_0", "/blocks.2/self_attn/Add_output_0", "/blocks.2/self_attn/k_proj/FakeLinear_output_0", "/blocks.2/self_attn/Unsqueeze_2_output_0", "/blocks.2/self_attn/Unsqueeze_3_output_0", "/blocks.2/self_attn/Concat_1_output_0", "/blocks.2/self_attn/Reshape_1_output_0", "/blocks.2/self_attn/k_norm/Cast_output_0", "/blocks.2/self_attn/k_norm/Mul_1_output_0", "/blocks.2/self_attn/Mul_2_output_0", "/blocks.2/self_attn/Shape_3_output_0", "Shape465", "Rank467", "BinaryOp469", "Unsqueeze470", "BinaryOp472", "Unsqueeze473", "StridedSlice476", "Squeeze477", "BinaryOp478", "BinaryOp479", "/blocks.2/self_attn/Gather_3_output_0", "/blocks.2/self_attn/Div_1_output_0", "/blocks.2/self_attn/Unsqueeze_9_output_0", "/blocks.2/self_attn/Slice_3_output_0", "/blocks.2/self_attn/Neg_1_output_0", "/blocks.2/self_attn/Unsqueeze_8_output_0", "/blocks.2/self_attn/Slice_2_output_0", "/blocks.2/self_attn/Concat_4_output_0", "/blocks.2/self_attn/Mul_3_output_0", "/blocks.2/self_attn/Add_1_output_0", "/blocks.2/self_attn/v_proj/FakeLinear_output_0", "/blocks.2/self_attn/Unsqueeze_4_output_0", "/blocks.2/self_attn/Unsqueeze_5_output_0", "/blocks.2/self_attn/Concat_2_output_0", "/blocks.2/self_attn/Reshape_2_output_0", "/blocks.2/self_attn/fused_attn/FusedAttention_output_0", "/blocks.2/self_attn/o_proj/FakeLinear_output_0", "/blocks.2/Add_output_0", "/blocks.2/post_attention_layernorm/Mul_1_output_0", "/blocks.2/mlp/gate_proj/FakeLinear_output_0", "/blocks.2/mlp/act_fn/Mul_output_0", "/blocks.2/mlp/up_proj/FakeLinear_output_0", "/blocks.2/mlp/Mul_output_0", "/blocks.2/mlp/down_proj/FakeLinear_output_0", "/blocks.2/Add_1_output_0", "/blocks.3/Reshape_output_0", "/blocks.3/input_layernorm/Mul_1_output_0", "/blocks.3/self_attn/q_proj/FakeLinear_output_0", "/blocks.3/self_attn/Shape_output_0", "Shape526", "Rank528", "BinaryOp530", "Unsqueeze531", "BinaryOp533", "Unsqueeze534", "StridedSlice537", "Squeeze538", "BinaryOp540", "/blocks.3/self_attn/Gather_output_0", "/blocks.3/self_attn/Unsqueeze_output_0", "Unsqueeze548", "Unsqueeze551", "StridedSlice554", "Squeeze555", "BinaryOp556", "BinaryOp557", "/blocks.3/self_attn/Gather_1_output_0", "/blocks.3/self_attn/Unsqueeze_1_output_0", "/blocks.3/self_attn/Concat_output_0", "/blocks.3/self_attn/Reshape_output_0", "/blocks.3/self_attn/q_norm/Cast_output_0", "/blocks.3/self_attn/q_norm/Mul_1_output_0", "/blocks.3/self_attn/Mul_output_0", "/blocks.3/self_attn/Shape_2_output_0", "Shape572", "Rank574", "BinaryOp576", "Unsqueeze577", "BinaryOp579", "Unsqueeze580", "StridedSlice583", "Squeeze584", "BinaryOp585", "BinaryOp586", "/blocks.3/self_attn/Gather_2_output_0", "/blocks.3/self_attn/Div_output_0", "/blocks.3/self_attn/Unsqueeze_7_output_0", "/blocks.3/self_attn/Slice_1_output_0", "/blocks.3/self_attn/Neg_output_0", "/blocks.3/self_attn/Unsqueeze_6_output_0", "/blocks.3/self_attn/Slice_output_0", "/blocks.3/self_attn/Concat_3_output_0", "/blocks.3/self_attn/Mul_1_output_0", "/blocks.3/self_attn/Add_output_0", "/blocks.3/self_attn/k_proj/FakeLinear_output_0", "/blocks.3/self_attn/Unsqueeze_2_output_0", "/blocks.3/self_attn/Unsqueeze_3_output_0", "/blocks.3/self_attn/Concat_1_output_0", "/blocks.3/self_attn/Reshape_1_output_0", "/blocks.3/self_attn/k_norm/Cast_output_0", "/blocks.3/self_attn/k_norm/Mul_1_output_0", "/blocks.3/self_attn/Mul_2_output_0", "/blocks.3/self_attn/Shape_3_output_0", "Shape614", "Rank616", "BinaryOp618", "Unsqueeze619", "BinaryOp621", "Unsqueeze622", "StridedSlice625", "Squeeze626", "BinaryOp627", "BinaryOp628", "/blocks.3/self_attn/Gather_3_output_0", "/blocks.3/self_attn/Div_1_output_0", "/blocks.3/self_attn/Unsqueeze_9_output_0", "/blocks.3/self_attn/Slice_3_output_0", "/blocks.3/self_attn/Neg_1_output_0", "/blocks.3/self_attn/Unsqueeze_8_output_0", "/blocks.3/self_attn/Slice_2_output_0", "/blocks.3/self_attn/Concat_4_output_0", "/blocks.3/self_attn/Mul_3_output_0", "/blocks.3/self_attn/Add_1_output_0", "/blocks.3/self_attn/v_proj/FakeLinear_output_0", "/blocks.3/self_attn/Unsqueeze_4_output_0", "/blocks.3/self_attn/Unsqueeze_5_output_0", "/blocks.3/self_attn/Concat_2_output_0", "/blocks.3/self_attn/Reshape_2_output_0", "/blocks.3/self_attn/fused_attn/FusedAttention_output_0", "/blocks.3/self_attn/o_proj/FakeLinear_output_0", "/blocks.3/Add_output_0", "/blocks.3/post_attention_layernorm/Mul_1_output_0", "/blocks.3/mlp/gate_proj/FakeLinear_output_0", "/blocks.3/mlp/act_fn/Mul_output_0", "/blocks.3/mlp/up_proj/FakeLinear_output_0", "/blocks.3/mlp/Mul_output_0", "/blocks.3/mlp/down_proj/FakeLinear_output_0", "/blocks.3/Add_1_output_0", "/blocks.4/Reshape_output_0", "/blocks.4/input_layernorm/Mul_1_output_0", "/blocks.4/self_attn/q_proj/FakeLinear_output_0", "/blocks.4/self_attn/Shape_output_0", "Shape675", "Rank677", "BinaryOp679", "Unsqueeze680", "BinaryOp682", "Unsqueeze683", "StridedSlice686", "Squeeze687", "BinaryOp689", "/blocks.4/self_attn/Gather_output_0", "/blocks.4/self_attn/Unsqueeze_output_0", "Unsqueeze697", "Unsqueeze700", "StridedSlice703", "Squeeze704", "BinaryOp705", "BinaryOp706", "/blocks.4/self_attn/Gather_1_output_0", "/blocks.4/self_attn/Unsqueeze_1_output_0", "/blocks.4/self_attn/Concat_output_0", "/blocks.4/self_attn/Reshape_output_0", "/blocks.4/self_attn/q_norm/Cast_output_0", "/blocks.4/self_attn/q_norm/Mul_1_output_0", "/blocks.4/self_attn/Mul_output_0", "/blocks.4/self_attn/Shape_2_output_0", "Shape721", "Rank723", "BinaryOp725", "Unsqueeze726", "BinaryOp728", "Unsqueeze729", "StridedSlice732", "Squeeze733", "BinaryOp734", "BinaryOp735", "/blocks.4/self_attn/Gather_2_output_0", "/blocks.4/self_attn/Div_output_0", "/blocks.4/self_attn/Unsqueeze_7_output_0", "/blocks.4/self_attn/Slice_1_output_0", "/blocks.4/self_attn/Neg_output_0", "/blocks.4/self_attn/Unsqueeze_6_output_0", "/blocks.4/self_attn/Slice_output_0", "/blocks.4/self_attn/Concat_3_output_0", "/blocks.4/self_attn/Mul_1_output_0", "/blocks.4/self_attn/Add_output_0", "/blocks.4/self_attn/k_proj/FakeLinear_output_0", "/blocks.4/self_attn/Unsqueeze_2_output_0", "/blocks.4/self_attn/Unsqueeze_3_output_0", "/blocks.4/self_attn/Concat_1_output_0", "/blocks.4/self_attn/Reshape_1_output_0", "/blocks.4/self_attn/k_norm/Cast_output_0", "/blocks.4/self_attn/k_norm/Mul_1_output_0", "/blocks.4/self_attn/Mul_2_output_0", "/blocks.4/self_attn/Shape_3_output_0", "Shape763", "Rank765", "BinaryOp767", "Unsqueeze768", "BinaryOp770", "Unsqueeze771", "StridedSlice774", "Squeeze775", "BinaryOp776", "BinaryOp777", "/blocks.4/self_attn/Gather_3_output_0", "/blocks.4/self_attn/Div_1_output_0", "/blocks.4/self_attn/Unsqueeze_9_output_0", "/blocks.4/self_attn/Slice_3_output_0", "/blocks.4/self_attn/Neg_1_output_0", "/blocks.4/self_attn/Unsqueeze_8_output_0", "/blocks.4/self_attn/Slice_2_output_0", "/blocks.4/self_attn/Concat_4_output_0", "/blocks.4/self_attn/Mul_3_output_0", "/blocks.4/self_attn/Add_1_output_0", "/blocks.4/self_attn/v_proj/FakeLinear_output_0", "/blocks.4/self_attn/Unsqueeze_4_output_0", "/blocks.4/self_attn/Unsqueeze_5_output_0", "/blocks.4/self_attn/Concat_2_output_0", "/blocks.4/self_attn/Reshape_2_output_0", "/blocks.4/self_attn/fused_attn/FusedAttention_output_0", "/blocks.4/self_attn/o_proj/FakeLinear_output_0", "/blocks.4/Add_output_0", "/blocks.4/post_attention_layernorm/Mul_1_output_0", "/blocks.4/mlp/gate_proj/FakeLinear_output_0", "/blocks.4/mlp/act_fn/Mul_output_0", "/blocks.4/mlp/up_proj/FakeLinear_output_0", "/blocks.4/mlp/Mul_output_0", "/blocks.4/mlp/down_proj/FakeLinear_output_0", "/blocks.4/Add_1_output_0", "/blocks.5/Reshape_output_0", "/blocks.5/input_layernorm/Mul_1_output_0", "/blocks.5/self_attn/q_proj/FakeLinear_output_0", "/blocks.5/self_attn/Shape_output_0", "Shape824", "Rank826", "BinaryOp828", "Unsqueeze829", "BinaryOp831", "Unsqueeze832", "StridedSlice835", "Squeeze836", "BinaryOp838", "/blocks.5/self_attn/Gather_output_0", "/blocks.5/self_attn/Unsqueeze_output_0", "Unsqueeze846", "Unsqueeze849", "StridedSlice852", "Squeeze853", "BinaryOp854", "BinaryOp855", "/blocks.5/self_attn/Gather_1_output_0", "/blocks.5/self_attn/Unsqueeze_1_output_0", "/blocks.5/self_attn/Concat_output_0", "/blocks.5/self_attn/Reshape_output_0", "/blocks.5/self_attn/q_norm/Cast_output_0", "/blocks.5/self_attn/q_norm/Mul_1_output_0", "/blocks.5/self_attn/Mul_output_0", "/blocks.5/self_attn/Shape_2_output_0", "Shape870", "Rank872", "BinaryOp874", "Unsqueeze875", "BinaryOp877", "Unsqueeze878", "StridedSlice881", "Squeeze882", "BinaryOp883", "BinaryOp884", "/blocks.5/self_attn/Gather_2_output_0", "/blocks.5/self_attn/Div_output_0", "/blocks.5/self_attn/Unsqueeze_7_output_0", "/blocks.5/self_attn/Slice_1_output_0", "/blocks.5/self_attn/Neg_output_0", "/blocks.5/self_attn/Unsqueeze_6_output_0", "/blocks.5/self_attn/Slice_output_0", "/blocks.5/self_attn/Concat_3_output_0", "/blocks.5/self_attn/Mul_1_output_0", "/blocks.5/self_attn/Add_output_0", "/blocks.5/self_attn/k_proj/FakeLinear_output_0", "/blocks.5/self_attn/Unsqueeze_2_output_0", "/blocks.5/self_attn/Unsqueeze_3_output_0", "/blocks.5/self_attn/Concat_1_output_0", "/blocks.5/self_attn/Reshape_1_output_0", "/blocks.5/self_attn/k_norm/Cast_output_0", "/blocks.5/self_attn/k_norm/Mul_1_output_0", "/blocks.5/self_attn/Mul_2_output_0", "/blocks.5/self_attn/Shape_3_output_0", "Shape912", "Rank914", "BinaryOp916", "Unsqueeze917", "BinaryOp919", "Unsqueeze920", "StridedSlice923", "Squeeze924", "BinaryOp925", "BinaryOp926", "/blocks.5/self_attn/Gather_3_output_0", "/blocks.5/self_attn/Div_1_output_0", "/blocks.5/self_attn/Unsqueeze_9_output_0", "/blocks.5/self_attn/Slice_3_output_0", "/blocks.5/self_attn/Neg_1_output_0", "/blocks.5/self_attn/Unsqueeze_8_output_0", "/blocks.5/self_attn/Slice_2_output_0", "/blocks.5/self_attn/Concat_4_output_0", "/blocks.5/self_attn/Mul_3_output_0", "/blocks.5/self_attn/Add_1_output_0", "/blocks.5/self_attn/v_proj/FakeLinear_output_0", "/blocks.5/self_attn/Unsqueeze_4_output_0", "/blocks.5/self_attn/Unsqueeze_5_output_0", "/blocks.5/self_attn/Concat_2_output_0", "/blocks.5/self_attn/Reshape_2_output_0", "/blocks.5/self_attn/fused_attn/FusedAttention_output_0", "/blocks.5/self_attn/o_proj/FakeLinear_output_0", "/blocks.5/Add_output_0", "/blocks.5/post_attention_layernorm/Mul_1_output_0", "/blocks.5/mlp/gate_proj/FakeLinear_output_0", "/blocks.5/mlp/act_fn/Mul_output_0", "/blocks.5/mlp/up_proj/FakeLinear_output_0", "/blocks.5/mlp/Mul_output_0", "/blocks.5/mlp/down_proj/FakeLinear_output_0", "/blocks.5/Add_1_output_0", "/blocks.6/Reshape_output_0", "/blocks.6/input_layernorm/Mul_1_output_0", "/blocks.6/self_attn/q_proj/FakeLinear_output_0", "/blocks.6/self_attn/Shape_output_0", "Shape973", "Rank975", "BinaryOp977", "Unsqueeze978", "BinaryOp980", "Unsqueeze981", "StridedSlice984", "Squeeze985", "BinaryOp987", "/blocks.6/self_attn/Gather_output_0", "/blocks.6/self_attn/Unsqueeze_output_0", "Unsqueeze995", "Unsqueeze998", "StridedSlice1001", "Squeeze1002", "BinaryOp1003", "BinaryOp1004", "/blocks.6/self_attn/Gather_1_output_0", "/blocks.6/self_attn/Unsqueeze_1_output_0", "/blocks.6/self_attn/Concat_output_0", "/blocks.6/self_attn/Reshape_output_0", "/blocks.6/self_attn/q_norm/Cast_output_0", "/blocks.6/self_attn/q_norm/Mul_1_output_0", "/blocks.6/self_attn/Mul_output_0", "/blocks.6/self_attn/Shape_2_output_0", "Shape1019", "Rank1021", "BinaryOp1023", "Unsqueeze1024", "BinaryOp1026", "Unsqueeze1027", "StridedSlice1030", "Squeeze1031", "BinaryOp1032", "BinaryOp1033", "/blocks.6/self_attn/Gather_2_output_0", "/blocks.6/self_attn/Div_output_0", "/blocks.6/self_attn/Unsqueeze_7_output_0", "/blocks.6/self_attn/Slice_1_output_0", "/blocks.6/self_attn/Neg_output_0", "/blocks.6/self_attn/Unsqueeze_6_output_0", "/blocks.6/self_attn/Slice_output_0", "/blocks.6/self_attn/Concat_3_output_0", "/blocks.6/self_attn/Mul_1_output_0", "/blocks.6/self_attn/Add_output_0", "/blocks.6/self_attn/k_proj/FakeLinear_output_0", "/blocks.6/self_attn/Unsqueeze_2_output_0", "/blocks.6/self_attn/Unsqueeze_3_output_0", "/blocks.6/self_attn/Concat_1_output_0", "/blocks.6/self_attn/Reshape_1_output_0", "/blocks.6/self_attn/k_norm/Cast_output_0", "/blocks.6/self_attn/k_norm/Mul_1_output_0", "/blocks.6/self_attn/Mul_2_output_0", "/blocks.6/self_attn/Shape_3_output_0", "Shape1061", "Rank1063", "BinaryOp1065", "Unsqueeze1066", "BinaryOp1068", "Unsqueeze1069", "StridedSlice1072", "Squeeze1073", "BinaryOp1074", "BinaryOp1075", "/blocks.6/self_attn/Gather_3_output_0", "/blocks.6/self_attn/Div_1_output_0", "/blocks.6/self_attn/Unsqueeze_9_output_0", "/blocks.6/self_attn/Slice_3_output_0", "/blocks.6/self_attn/Neg_1_output_0", "/blocks.6/self_attn/Unsqueeze_8_output_0", "/blocks.6/self_attn/Slice_2_output_0", "/blocks.6/self_attn/Concat_4_output_0", "/blocks.6/self_attn/Mul_3_output_0", "/blocks.6/self_attn/Add_1_output_0", "/blocks.6/self_attn/v_proj/FakeLinear_output_0", "/blocks.6/self_attn/Unsqueeze_4_output_0", "/blocks.6/self_attn/Unsqueeze_5_output_0", "/blocks.6/self_attn/Concat_2_output_0", "/blocks.6/self_attn/Reshape_2_output_0", "/blocks.6/self_attn/fused_attn/FusedAttention_output_0", "/blocks.6/self_attn/o_proj/FakeLinear_output_0", "/blocks.6/Add_output_0", "/blocks.6/post_attention_layernorm/Mul_1_output_0", "/blocks.6/mlp/gate_proj/FakeLinear_output_0", "/blocks.6/mlp/act_fn/Mul_output_0", "/blocks.6/mlp/up_proj/FakeLinear_output_0", "/blocks.6/mlp/Mul_output_0", "/blocks.6/mlp/down_proj/FakeLinear_output_0", "/blocks.6/Add_1_output_0", "/blocks.7/Reshape_output_0", "/blocks.7/input_layernorm/Mul_1_output_0", "/blocks.7/self_attn/q_proj/FakeLinear_output_0", "/blocks.7/self_attn/Shape_output_0", "Shape1122", "Rank1124", "BinaryOp1126", "Unsqueeze1127", "BinaryOp1129", "Unsqueeze1130", "StridedSlice1133", "Squeeze1134", "BinaryOp1136", "/blocks.7/self_attn/Gather_output_0", "/blocks.7/self_attn/Unsqueeze_output_0", "Unsqueeze1144", "Unsqueeze1147", "StridedSlice1150", "Squeeze1151", "BinaryOp1152", "BinaryOp1153", "/blocks.7/self_attn/Gather_1_output_0", "/blocks.7/self_attn/Unsqueeze_1_output_0", "/blocks.7/self_attn/Concat_output_0", "/blocks.7/self_attn/Reshape_output_0", "/blocks.7/self_attn/q_norm/Cast_output_0", "/blocks.7/self_attn/q_norm/Mul_1_output_0", "/blocks.7/self_attn/Mul_output_0", "/blocks.7/self_attn/Shape_2_output_0", "Shape1168", "Rank1170", "BinaryOp1172", "Unsqueeze1173", "BinaryOp1175", "Unsqueeze1176", "StridedSlice1179", "Squeeze1180", "BinaryOp1181", "BinaryOp1182", "/blocks.7/self_attn/Gather_2_output_0", "/blocks.7/self_attn/Div_output_0", "/blocks.7/self_attn/Unsqueeze_7_output_0", "/blocks.7/self_attn/Slice_1_output_0", "/blocks.7/self_attn/Neg_output_0", "/blocks.7/self_attn/Unsqueeze_6_output_0", "/blocks.7/self_attn/Slice_output_0", "/blocks.7/self_attn/Concat_3_output_0", "/blocks.7/self_attn/Mul_1_output_0", "/blocks.7/self_attn/Add_output_0", "/blocks.7/self_attn/k_proj/FakeLinear_output_0", "/blocks.7/self_attn/Unsqueeze_2_output_0", "/blocks.7/self_attn/Unsqueeze_3_output_0", "/blocks.7/self_attn/Concat_1_output_0", "/blocks.7/self_attn/Reshape_1_output_0", "/blocks.7/self_attn/k_norm/Cast_output_0", "/blocks.7/self_attn/k_norm/Mul_1_output_0", "/blocks.7/self_attn/Mul_2_output_0", "/blocks.7/self_attn/Shape_3_output_0", "Shape1210", "Rank1212", "BinaryOp1214", "Unsqueeze1215", "BinaryOp1217", "Unsqueeze1218", "StridedSlice1221", "Squeeze1222", "BinaryOp1223", "BinaryOp1224", "/blocks.7/self_attn/Gather_3_output_0", "/blocks.7/self_attn/Div_1_output_0", "/blocks.7/self_attn/Unsqueeze_9_output_0", "/blocks.7/self_attn/Slice_3_output_0", "/blocks.7/self_attn/Neg_1_output_0", "/blocks.7/self_attn/Unsqueeze_8_output_0", "/blocks.7/self_attn/Slice_2_output_0", "/blocks.7/self_attn/Concat_4_output_0", "/blocks.7/self_attn/Mul_3_output_0", "/blocks.7/self_attn/Add_1_output_0", "/blocks.7/self_attn/v_proj/FakeLinear_output_0", "/blocks.7/self_attn/Unsqueeze_4_output_0", "/blocks.7/self_attn/Unsqueeze_5_output_0", "/blocks.7/self_attn/Concat_2_output_0", "/blocks.7/self_attn/Reshape_2_output_0", "/blocks.7/self_attn/fused_attn/FusedAttention_output_0", "/blocks.7/self_attn/o_proj/FakeLinear_output_0", "/blocks.7/Add_output_0", "/blocks.7/post_attention_layernorm/Mul_1_output_0", "/blocks.7/mlp/gate_proj/FakeLinear_output_0", "/blocks.7/mlp/act_fn/Mul_output_0", "/blocks.7/mlp/up_proj/FakeLinear_output_0", "/blocks.7/mlp/Mul_output_0", "/blocks.7/mlp/down_proj/FakeLinear_output_0", "/blocks.7/Add_1_output_0", "/blocks.8/Reshape_output_0", "/blocks.8/input_layernorm/Mul_1_output_0", "/blocks.8/self_attn/q_proj/FakeLinear_output_0", "/blocks.8/self_attn/Shape_output_0", "Shape1271", "Rank1273", "BinaryOp1275", "Unsqueeze1276", "BinaryOp1278", "Unsqueeze1279", "StridedSlice1282", "Squeeze1283", "BinaryOp1285", "/blocks.8/self_attn/Gather_output_0", "/blocks.8/self_attn/Unsqueeze_output_0", "Unsqueeze1293", "Unsqueeze1296", "StridedSlice1299", "Squeeze1300", "BinaryOp1301", "BinaryOp1302", "/blocks.8/self_attn/Gather_1_output_0", "/blocks.8/self_attn/Unsqueeze_1_output_0", "/blocks.8/self_attn/Concat_output_0", "/blocks.8/self_attn/Reshape_output_0", "/blocks.8/self_attn/q_norm/Cast_output_0", "/blocks.8/self_attn/q_norm/Mul_1_output_0", "/blocks.8/self_attn/Mul_output_0", "/blocks.8/self_attn/Shape_2_output_0", "Shape1317", "Rank1319", "BinaryOp1321", "Unsqueeze1322", "BinaryOp1324", "Unsqueeze1325", "StridedSlice1328", "Squeeze1329", "BinaryOp1330", "BinaryOp1331", "/blocks.8/self_attn/Gather_2_output_0", "/blocks.8/self_attn/Div_output_0", "/blocks.8/self_attn/Unsqueeze_7_output_0", "/blocks.8/self_attn/Slice_1_output_0", "/blocks.8/self_attn/Neg_output_0", "/blocks.8/self_attn/Unsqueeze_6_output_0", "/blocks.8/self_attn/Slice_output_0", "/blocks.8/self_attn/Concat_3_output_0", "/blocks.8/self_attn/Mul_1_output_0", "/blocks.8/self_attn/Add_output_0", "/blocks.8/self_attn/k_proj/FakeLinear_output_0", "/blocks.8/self_attn/Unsqueeze_2_output_0", "/blocks.8/self_attn/Unsqueeze_3_output_0", "/blocks.8/self_attn/Concat_1_output_0", "/blocks.8/self_attn/Reshape_1_output_0", "/blocks.8/self_attn/k_norm/Cast_output_0", "/blocks.8/self_attn/k_norm/Mul_1_output_0", "/blocks.8/self_attn/Mul_2_output_0", "/blocks.8/self_attn/Shape_3_output_0", "Shape1359", "Rank1361", "BinaryOp1363", "Unsqueeze1364", "BinaryOp1366", "Unsqueeze1367", "StridedSlice1370", "Squeeze1371", "BinaryOp1372", "BinaryOp1373", "/blocks.8/self_attn/Gather_3_output_0", "/blocks.8/self_attn/Div_1_output_0", "/blocks.8/self_attn/Unsqueeze_9_output_0", "/blocks.8/self_attn/Slice_3_output_0", "/blocks.8/self_attn/Neg_1_output_0", "/blocks.8/self_attn/Unsqueeze_8_output_0", "/blocks.8/self_attn/Slice_2_output_0", "/blocks.8/self_attn/Concat_4_output_0", "/blocks.8/self_attn/Mul_3_output_0", "/blocks.8/self_attn/Add_1_output_0", "/blocks.8/self_attn/v_proj/FakeLinear_output_0", "/blocks.8/self_attn/Unsqueeze_4_output_0", "/blocks.8/self_attn/Unsqueeze_5_output_0", "/blocks.8/self_attn/Concat_2_output_0", "/blocks.8/self_attn/Reshape_2_output_0", "/blocks.8/self_attn/fused_attn/FusedAttention_output_0", "/blocks.8/self_attn/o_proj/FakeLinear_output_0", "/blocks.8/Add_output_0", "/blocks.8/post_attention_layernorm/Mul_1_output_0", "/blocks.8/mlp/gate_proj/FakeLinear_output_0", "/blocks.8/mlp/act_fn/Mul_output_0", "/blocks.8/mlp/up_proj/FakeLinear_output_0", "/blocks.8/mlp/Mul_output_0", "/blocks.8/mlp/down_proj/FakeLinear_output_0", "/blocks.8/Add_1_output_0", "/blocks.9/Reshape_output_0", "/blocks.9/input_layernorm/Mul_1_output_0", "/blocks.9/self_attn/q_proj/FakeLinear_output_0", "/blocks.9/self_attn/Shape_output_0", "Shape1420", "Rank1422", "BinaryOp1424", "Unsqueeze1425", "BinaryOp1427", "Unsqueeze1428", "StridedSlice1431", "Squeeze1432", "BinaryOp1434", "/blocks.9/self_attn/Gather_output_0", "/blocks.9/self_attn/Unsqueeze_output_0", "Unsqueeze1442", "Unsqueeze1445", "StridedSlice1448", "Squeeze1449", "BinaryOp1450", "BinaryOp1451", "/blocks.9/self_attn/Gather_1_output_0", "/blocks.9/self_attn/Unsqueeze_1_output_0", "/blocks.9/self_attn/Concat_output_0", "/blocks.9/self_attn/Reshape_output_0", "/blocks.9/self_attn/q_norm/Cast_output_0", "/blocks.9/self_attn/q_norm/Mul_1_output_0", "/blocks.9/self_attn/Mul_output_0", "/blocks.9/self_attn/Shape_2_output_0", "Shape1466", "Rank1468", "BinaryOp1470", "Unsqueeze1471", "BinaryOp1473", "Unsqueeze1474", "StridedSlice1477", "Squeeze1478", "BinaryOp1479", "BinaryOp1480", "/blocks.9/self_attn/Gather_2_output_0", "/blocks.9/self_attn/Div_output_0", "/blocks.9/self_attn/Unsqueeze_7_output_0", "/blocks.9/self_attn/Slice_1_output_0", "/blocks.9/self_attn/Neg_output_0", "/blocks.9/self_attn/Unsqueeze_6_output_0", "/blocks.9/self_attn/Slice_output_0", "/blocks.9/self_attn/Concat_3_output_0", "/blocks.9/self_attn/Mul_1_output_0", "/blocks.9/self_attn/Add_output_0", "/blocks.9/self_attn/k_proj/FakeLinear_output_0", "/blocks.9/self_attn/Unsqueeze_2_output_0", "/blocks.9/self_attn/Unsqueeze_3_output_0", "/blocks.9/self_attn/Concat_1_output_0", "/blocks.9/self_attn/Reshape_1_output_0", "/blocks.9/self_attn/k_norm/Cast_output_0", "/blocks.9/self_attn/k_norm/Mul_1_output_0", "/blocks.9/self_attn/Mul_2_output_0", "/blocks.9/self_attn/Shape_3_output_0", "Shape1508", "Rank1510", "BinaryOp1512", "Unsqueeze1513", "BinaryOp1515", "Unsqueeze1516", "StridedSlice1519", "Squeeze1520", "BinaryOp1521", "BinaryOp1522", "/blocks.9/self_attn/Gather_3_output_0", "/blocks.9/self_attn/Div_1_output_0", "/blocks.9/self_attn/Unsqueeze_9_output_0", "/blocks.9/self_attn/Slice_3_output_0", "/blocks.9/self_attn/Neg_1_output_0", "/blocks.9/self_attn/Unsqueeze_8_output_0", "/blocks.9/self_attn/Slice_2_output_0", "/blocks.9/self_attn/Concat_4_output_0", "/blocks.9/self_attn/Mul_3_output_0", "/blocks.9/self_attn/Add_1_output_0", "/blocks.9/self_attn/v_proj/FakeLinear_output_0", "/blocks.9/self_attn/Unsqueeze_4_output_0", "/blocks.9/self_attn/Unsqueeze_5_output_0", "/blocks.9/self_attn/Concat_2_output_0", "/blocks.9/self_attn/Reshape_2_output_0", "/blocks.9/self_attn/fused_attn/FusedAttention_output_0", "/blocks.9/self_attn/o_proj/FakeLinear_output_0", "/blocks.9/Add_output_0", "/blocks.9/post_attention_layernorm/Mul_1_output_0", "/blocks.9/mlp/gate_proj/FakeLinear_output_0", "/blocks.9/mlp/act_fn/Mul_output_0", "/blocks.9/mlp/up_proj/FakeLinear_output_0", "/blocks.9/mlp/Mul_output_0", "/blocks.9/mlp/down_proj/FakeLinear_output_0", "/blocks.9/Add_1_output_0", "/blocks.10/Reshape_output_0", "/blocks.10/input_layernorm/Mul_1_output_0", "/blocks.10/self_attn/q_proj/FakeLinear_output_0", "/blocks.10/self_attn/Shape_output_0", "Shape1569", "Rank1571", "BinaryOp1573", "Unsqueeze1574", "BinaryOp1576", "Unsqueeze1577", "StridedSlice1580", "Squeeze1581", "BinaryOp1583", "/blocks.10/self_attn/Gather_output_0", "/blocks.10/self_attn/Unsqueeze_output_0", "Unsqueeze1591", "Unsqueeze1594", "StridedSlice1597", "Squeeze1598", "BinaryOp1599", "BinaryOp1600", "/blocks.10/self_attn/Gather_1_output_0", "/blocks.10/self_attn/Unsqueeze_1_output_0", "/blocks.10/self_attn/Concat_output_0", "/blocks.10/self_attn/Reshape_output_0", "/blocks.10/self_attn/q_norm/Cast_output_0", "/blocks.10/self_attn/q_norm/Mul_1_output_0", "/blocks.10/self_attn/Mul_output_0", "/blocks.10/self_attn/Shape_2_output_0", "Shape1615", "Rank1617", "BinaryOp1619", "Unsqueeze1620", "BinaryOp1622", "Unsqueeze1623", "StridedSlice1626", "Squeeze1627", "BinaryOp1628", "BinaryOp1629", "/blocks.10/self_attn/Gather_2_output_0", "/blocks.10/self_attn/Div_output_0", "/blocks.10/self_attn/Unsqueeze_7_output_0", "/blocks.10/self_attn/Slice_1_output_0", "/blocks.10/self_attn/Neg_output_0", "/blocks.10/self_attn/Unsqueeze_6_output_0", "/blocks.10/self_attn/Slice_output_0", "/blocks.10/self_attn/Concat_3_output_0", "/blocks.10/self_attn/Mul_1_output_0", "/blocks.10/self_attn/Add_output_0", "/blocks.10/self_attn/k_proj/FakeLinear_output_0", "/blocks.10/self_attn/Unsqueeze_2_output_0", "/blocks.10/self_attn/Unsqueeze_3_output_0", "/blocks.10/self_attn/Concat_1_output_0", "/blocks.10/self_attn/Reshape_1_output_0", "/blocks.10/self_attn/k_norm/Cast_output_0", "/blocks.10/self_attn/k_norm/Mul_1_output_0", "/blocks.10/self_attn/Mul_2_output_0", "/blocks.10/self_attn/Shape_3_output_0", "Shape1657", "Rank1659", "BinaryOp1661", "Unsqueeze1662", "BinaryOp1664", "Unsqueeze1665", "StridedSlice1668", "Squeeze1669", "BinaryOp1670", "BinaryOp1671", "/blocks.10/self_attn/Gather_3_output_0", "/blocks.10/self_attn/Div_1_output_0", "/blocks.10/self_attn/Unsqueeze_9_output_0", "/blocks.10/self_attn/Slice_3_output_0", "/blocks.10/self_attn/Neg_1_output_0", "/blocks.10/self_attn/Unsqueeze_8_output_0", "/blocks.10/self_attn/Slice_2_output_0", "/blocks.10/self_attn/Concat_4_output_0", "/blocks.10/self_attn/Mul_3_output_0", "/blocks.10/self_attn/Add_1_output_0", "/blocks.10/self_attn/v_proj/FakeLinear_output_0", "/blocks.10/self_attn/Unsqueeze_4_output_0", "/blocks.10/self_attn/Unsqueeze_5_output_0", "/blocks.10/self_attn/Concat_2_output_0", "/blocks.10/self_attn/Reshape_2_output_0", "/blocks.10/self_attn/fused_attn/FusedAttention_output_0", "/blocks.10/self_attn/o_proj/FakeLinear_output_0", "/blocks.10/Add_output_0", "/blocks.10/post_attention_layernorm/Mul_1_output_0", "/blocks.10/mlp/gate_proj/FakeLinear_output_0", "/blocks.10/mlp/act_fn/Mul_output_0", "/blocks.10/mlp/up_proj/FakeLinear_output_0", "/blocks.10/mlp/Mul_output_0", "/blocks.10/mlp/down_proj/FakeLinear_output_0", "/blocks.10/Add_1_output_0", "/blocks.11/Reshape_output_0", "/blocks.11/input_layernorm/Mul_1_output_0", "/blocks.11/self_attn/q_proj/FakeLinear_output_0", "/blocks.11/self_attn/Shape_output_0", "Shape1718", "Rank1720", "BinaryOp1722", "Unsqueeze1723", "BinaryOp1725", "Unsqueeze1726", "StridedSlice1729", "Squeeze1730", "BinaryOp1732", "/blocks.11/self_attn/Gather_output_0", "/blocks.11/self_attn/Unsqueeze_output_0", "Unsqueeze1740", "Unsqueeze1743", "StridedSlice1746", "Squeeze1747", "BinaryOp1748", "BinaryOp1749", "/blocks.11/self_attn/Gather_1_output_0", "/blocks.11/self_attn/Unsqueeze_1_output_0", "/blocks.11/self_attn/Concat_output_0", "/blocks.11/self_attn/Reshape_output_0", "/blocks.11/self_attn/q_norm/Cast_output_0", "/blocks.11/self_attn/q_norm/Mul_1_output_0", "/blocks.11/self_attn/Mul_output_0", "/blocks.11/self_attn/Shape_2_output_0", "Shape1764", "Rank1766", "BinaryOp1768", "Unsqueeze1769", "BinaryOp1771", "Unsqueeze1772", "StridedSlice1775", "Squeeze1776", "BinaryOp1777", "BinaryOp1778", "/blocks.11/self_attn/Gather_2_output_0", "/blocks.11/self_attn/Div_output_0", "/blocks.11/self_attn/Unsqueeze_7_output_0", "/blocks.11/self_attn/Slice_1_output_0", "/blocks.11/self_attn/Neg_output_0", "/blocks.11/self_attn/Unsqueeze_6_output_0", "/blocks.11/self_attn/Slice_output_0", "/blocks.11/self_attn/Concat_3_output_0", "/blocks.11/self_attn/Mul_1_output_0", "/blocks.11/self_attn/Add_output_0", "/blocks.11/self_attn/k_proj/FakeLinear_output_0", "/blocks.11/self_attn/Unsqueeze_2_output_0", "/blocks.11/self_attn/Unsqueeze_3_output_0", "/blocks.11/self_attn/Concat_1_output_0", "/blocks.11/self_attn/Reshape_1_output_0", "/blocks.11/self_attn/k_norm/Cast_output_0", "/blocks.11/self_attn/k_norm/Mul_1_output_0", "/blocks.11/self_attn/Mul_2_output_0", "/blocks.11/self_attn/Shape_3_output_0", "Shape1806", "Rank1808", "BinaryOp1810", "Unsqueeze1811", "BinaryOp1813", "Unsqueeze1814", "StridedSlice1817", "Squeeze1818", "BinaryOp1819", "BinaryOp1820", "/blocks.11/self_attn/Gather_3_output_0", "/blocks.11/self_attn/Div_1_output_0", "/blocks.11/self_attn/Unsqueeze_9_output_0", "/blocks.11/self_attn/Slice_3_output_0", "/blocks.11/self_attn/Neg_1_output_0", "/blocks.11/self_attn/Unsqueeze_8_output_0", "/blocks.11/self_attn/Slice_2_output_0", "/blocks.11/self_attn/Concat_4_output_0", "/blocks.11/self_attn/Mul_3_output_0", "/blocks.11/self_attn/Add_1_output_0", "/blocks.11/self_attn/v_proj/FakeLinear_output_0", "/blocks.11/self_attn/Unsqueeze_4_output_0", "/blocks.11/self_attn/Unsqueeze_5_output_0", "/blocks.11/self_attn/Concat_2_output_0", "/blocks.11/self_attn/Reshape_2_output_0", "/blocks.11/self_attn/fused_attn/FusedAttention_output_0", "/blocks.11/self_attn/o_proj/FakeLinear_output_0", "/blocks.11/Add_output_0", "/blocks.11/post_attention_layernorm/Mul_1_output_0", "/blocks.11/mlp/gate_proj/FakeLinear_output_0", "/blocks.11/mlp/act_fn/Mul_output_0", "/blocks.11/mlp/up_proj/FakeLinear_output_0", "/blocks.11/mlp/Mul_output_0", "/blocks.11/mlp/down_proj/FakeLinear_output_0", "/blocks.11/Add_1_output_0", "/blocks.12/Reshape_output_0", "/blocks.12/input_layernorm/Mul_1_output_0", "/blocks.12/self_attn/q_proj/FakeLinear_output_0", "/blocks.12/self_attn/Shape_output_0", "Shape1867", "Rank1869", "BinaryOp1871", "Unsqueeze1872", "BinaryOp1874", "Unsqueeze1875", "StridedSlice1878", "Squeeze1879", "BinaryOp1881", "/blocks.12/self_attn/Gather_output_0", "/blocks.12/self_attn/Unsqueeze_output_0", "Unsqueeze1889", "Unsqueeze1892", "StridedSlice1895", "Squeeze1896", "BinaryOp1897", "BinaryOp1898", "/blocks.12/self_attn/Gather_1_output_0", "/blocks.12/self_attn/Unsqueeze_1_output_0", "/blocks.12/self_attn/Concat_output_0", "/blocks.12/self_attn/Reshape_output_0", "/blocks.12/self_attn/q_norm/Cast_output_0", "/blocks.12/self_attn/q_norm/Mul_1_output_0", "/blocks.12/self_attn/Mul_output_0", "/blocks.12/self_attn/Shape_2_output_0", "Shape1913", "Rank1915", "BinaryOp1917", "Unsqueeze1918", "BinaryOp1920", "Unsqueeze1921", "StridedSlice1924", "Squeeze1925", "BinaryOp1926", "BinaryOp1927", "/blocks.12/self_attn/Gather_2_output_0", "/blocks.12/self_attn/Div_output_0", "/blocks.12/self_attn/Unsqueeze_7_output_0", "/blocks.12/self_attn/Slice_1_output_0", "/blocks.12/self_attn/Neg_output_0", "/blocks.12/self_attn/Unsqueeze_6_output_0", "/blocks.12/self_attn/Slice_output_0", "/blocks.12/self_attn/Concat_3_output_0", "/blocks.12/self_attn/Mul_1_output_0", "/blocks.12/self_attn/Add_output_0", "/blocks.12/self_attn/k_proj/FakeLinear_output_0", "/blocks.12/self_attn/Unsqueeze_2_output_0", "/blocks.12/self_attn/Unsqueeze_3_output_0", "/blocks.12/self_attn/Concat_1_output_0", "/blocks.12/self_attn/Reshape_1_output_0", "/blocks.12/self_attn/k_norm/Cast_output_0", "/blocks.12/self_attn/k_norm/Mul_1_output_0", "/blocks.12/self_attn/Mul_2_output_0", "/blocks.12/self_attn/Shape_3_output_0", "Shape1955", "Rank1957", "BinaryOp1959", "Unsqueeze1960", "BinaryOp1962", "Unsqueeze1963", "StridedSlice1966", "Squeeze1967", "BinaryOp1968", "BinaryOp1969", "/blocks.12/self_attn/Gather_3_output_0", "/blocks.12/self_attn/Div_1_output_0", "/blocks.12/self_attn/Unsqueeze_9_output_0", "/blocks.12/self_attn/Slice_3_output_0", "/blocks.12/self_attn/Neg_1_output_0", "/blocks.12/self_attn/Unsqueeze_8_output_0", "/blocks.12/self_attn/Slice_2_output_0", "/blocks.12/self_attn/Concat_4_output_0", "/blocks.12/self_attn/Mul_3_output_0", "/blocks.12/self_attn/Add_1_output_0", "/blocks.12/self_attn/v_proj/FakeLinear_output_0", "/blocks.12/self_attn/Unsqueeze_4_output_0", "/blocks.12/self_attn/Unsqueeze_5_output_0", "/blocks.12/self_attn/Concat_2_output_0", "/blocks.12/self_attn/Reshape_2_output_0", "/blocks.12/self_attn/fused_attn/FusedAttention_output_0", "/blocks.12/self_attn/o_proj/FakeLinear_output_0", "/blocks.12/Add_output_0", "/blocks.12/post_attention_layernorm/Mul_1_output_0", "/blocks.12/mlp/gate_proj/FakeLinear_output_0", "/blocks.12/mlp/act_fn/Mul_output_0", "/blocks.12/mlp/up_proj/FakeLinear_output_0", "/blocks.12/mlp/Mul_output_0", "/blocks.12/mlp/down_proj/FakeLinear_output_0", "/blocks.12/Add_1_output_0", "/blocks.13/Reshape_output_0", "/blocks.13/input_layernorm/Mul_1_output_0", "/blocks.13/self_attn/q_proj/FakeLinear_output_0", "/blocks.13/self_attn/Shape_output_0", "Shape2016", "Rank2018", "BinaryOp2020", "Unsqueeze2021", "BinaryOp2023", "Unsqueeze2024", "StridedSlice2027", "Squeeze2028", "BinaryOp2030", "/blocks.13/self_attn/Gather_output_0", "/blocks.13/self_attn/Unsqueeze_output_0", "Unsqueeze2038", "Unsqueeze2041", "StridedSlice2044", "Squeeze2045", "BinaryOp2046", "BinaryOp2047", "/blocks.13/self_attn/Gather_1_output_0", "/blocks.13/self_attn/Unsqueeze_1_output_0", "/blocks.13/self_attn/Concat_output_0", "/blocks.13/self_attn/Reshape_output_0", "/blocks.13/self_attn/q_norm/Cast_output_0", "/blocks.13/self_attn/q_norm/Mul_1_output_0", "/blocks.13/self_attn/Mul_output_0", "/blocks.13/self_attn/Shape_2_output_0", "Shape2062", "Rank2064", "BinaryOp2066", "Unsqueeze2067", "BinaryOp2069", "Unsqueeze2070", "StridedSlice2073", "Squeeze2074", "BinaryOp2075", "BinaryOp2076", "/blocks.13/self_attn/Gather_2_output_0", "/blocks.13/self_attn/Div_output_0", "/blocks.13/self_attn/Unsqueeze_7_output_0", "/blocks.13/self_attn/Slice_1_output_0", "/blocks.13/self_attn/Neg_output_0", "/blocks.13/self_attn/Unsqueeze_6_output_0", "/blocks.13/self_attn/Slice_output_0", "/blocks.13/self_attn/Concat_3_output_0", "/blocks.13/self_attn/Mul_1_output_0", "/blocks.13/self_attn/Add_output_0", "/blocks.13/self_attn/k_proj/FakeLinear_output_0", "/blocks.13/self_attn/Unsqueeze_2_output_0", "/blocks.13/self_attn/Unsqueeze_3_output_0", "/blocks.13/self_attn/Concat_1_output_0", "/blocks.13/self_attn/Reshape_1_output_0", "/blocks.13/self_attn/k_norm/Cast_output_0", "/blocks.13/self_attn/k_norm/Mul_1_output_0", "/blocks.13/self_attn/Mul_2_output_0", "/blocks.13/self_attn/Shape_3_output_0", "Shape2104", "Rank2106", "BinaryOp2108", "Unsqueeze2109", "BinaryOp2111", "Unsqueeze2112", "StridedSlice2115", "Squeeze2116", "BinaryOp2117", "BinaryOp2118", "/blocks.13/self_attn/Gather_3_output_0", "/blocks.13/self_attn/Div_1_output_0", "/blocks.13/self_attn/Unsqueeze_9_output_0", "/blocks.13/self_attn/Slice_3_output_0", "/blocks.13/self_attn/Neg_1_output_0", "/blocks.13/self_attn/Unsqueeze_8_output_0", "/blocks.13/self_attn/Slice_2_output_0", "/blocks.13/self_attn/Concat_4_output_0", "/blocks.13/self_attn/Mul_3_output_0", "/blocks.13/self_attn/Add_1_output_0", "/blocks.13/self_attn/v_proj/FakeLinear_output_0", "/blocks.13/self_attn/Unsqueeze_4_output_0", "/blocks.13/self_attn/Unsqueeze_5_output_0", "/blocks.13/self_attn/Concat_2_output_0", "/blocks.13/self_attn/Reshape_2_output_0", "/blocks.13/self_attn/fused_attn/FusedAttention_output_0", "/blocks.13/self_attn/o_proj/FakeLinear_output_0", "/blocks.13/Add_output_0", "/blocks.13/post_attention_layernorm/Mul_1_output_0", "/blocks.13/mlp/gate_proj/FakeLinear_output_0", "/blocks.13/mlp/act_fn/Mul_output_0", "/blocks.13/mlp/up_proj/FakeLinear_output_0", "/blocks.13/mlp/Mul_output_0", "/blocks.13/mlp/down_proj/FakeLinear_output_0", "/blocks.13/Add_1_output_0", "/blocks.14/Reshape_output_0", "/blocks.14/input_layernorm/Mul_1_output_0", "/blocks.14/self_attn/q_proj/FakeLinear_output_0", "/blocks.14/self_attn/Shape_output_0", "Shape2165", "Rank2167", "BinaryOp2169", "Unsqueeze2170", "BinaryOp2172", "Unsqueeze2173", "StridedSlice2176", "Squeeze2177", "BinaryOp2179", "/blocks.14/self_attn/Gather_output_0", "/blocks.14/self_attn/Unsqueeze_output_0", "Unsqueeze2187", "Unsqueeze2190", "StridedSlice2193", "Squeeze2194", "BinaryOp2195", "BinaryOp2196", "/blocks.14/self_attn/Gather_1_output_0", "/blocks.14/self_attn/Unsqueeze_1_output_0", "/blocks.14/self_attn/Concat_output_0", "/blocks.14/self_attn/Reshape_output_0", "/blocks.14/self_attn/q_norm/Cast_output_0", "/blocks.14/self_attn/q_norm/Mul_1_output_0", "/blocks.14/self_attn/Mul_output_0", "/blocks.14/self_attn/Shape_2_output_0", "Shape2211", "Rank2213", "BinaryOp2215", "Unsqueeze2216", "BinaryOp2218", "Unsqueeze2219", "StridedSlice2222", "Squeeze2223", "BinaryOp2224", "BinaryOp2225", "/blocks.14/self_attn/Gather_2_output_0", "/blocks.14/self_attn/Div_output_0", "/blocks.14/self_attn/Unsqueeze_7_output_0", "/blocks.14/self_attn/Slice_1_output_0", "/blocks.14/self_attn/Neg_output_0", "/blocks.14/self_attn/Unsqueeze_6_output_0", "/blocks.14/self_attn/Slice_output_0", "/blocks.14/self_attn/Concat_3_output_0", "/blocks.14/self_attn/Mul_1_output_0", "/blocks.14/self_attn/Add_output_0", "/blocks.14/self_attn/k_proj/FakeLinear_output_0", "/blocks.14/self_attn/Unsqueeze_2_output_0", "/blocks.14/self_attn/Unsqueeze_3_output_0", "/blocks.14/self_attn/Concat_1_output_0", "/blocks.14/self_attn/Reshape_1_output_0", "/blocks.14/self_attn/k_norm/Cast_output_0", "/blocks.14/self_attn/k_norm/Mul_1_output_0", "/blocks.14/self_attn/Mul_2_output_0", "/blocks.14/self_attn/Shape_3_output_0", "Shape2253", "Rank2255", "BinaryOp2257", "Unsqueeze2258", "BinaryOp2260", "Unsqueeze2261", "StridedSlice2264", "Squeeze2265", "BinaryOp2266", "BinaryOp2267", "/blocks.14/self_attn/Gather_3_output_0", "/blocks.14/self_attn/Div_1_output_0", "/blocks.14/self_attn/Unsqueeze_9_output_0", "/blocks.14/self_attn/Slice_3_output_0", "/blocks.14/self_attn/Neg_1_output_0", "/blocks.14/self_attn/Unsqueeze_8_output_0", "/blocks.14/self_attn/Slice_2_output_0", "/blocks.14/self_attn/Concat_4_output_0", "/blocks.14/self_attn/Mul_3_output_0", "/blocks.14/self_attn/Add_1_output_0", "/blocks.14/self_attn/v_proj/FakeLinear_output_0", "/blocks.14/self_attn/Unsqueeze_4_output_0", "/blocks.14/self_attn/Unsqueeze_5_output_0", "/blocks.14/self_attn/Concat_2_output_0", "/blocks.14/self_attn/Reshape_2_output_0", "/blocks.14/self_attn/fused_attn/FusedAttention_output_0", "/blocks.14/self_attn/o_proj/FakeLinear_output_0", "/blocks.14/Add_output_0", "/blocks.14/post_attention_layernorm/Mul_1_output_0", "/blocks.14/mlp/gate_proj/FakeLinear_output_0", "/blocks.14/mlp/act_fn/Mul_output_0", "/blocks.14/mlp/up_proj/FakeLinear_output_0", "/blocks.14/mlp/Mul_output_0", "/blocks.14/mlp/down_proj/FakeLinear_output_0", "/blocks.14/Add_1_output_0", "/blocks.15/Reshape_output_0", "/blocks.15/input_layernorm/Mul_1_output_0", "/blocks.15/self_attn/q_proj/FakeLinear_output_0", "/blocks.15/self_attn/Shape_output_0", "Shape2314", "Rank2316", "BinaryOp2318", "Unsqueeze2319", "BinaryOp2321", "Unsqueeze2322", "StridedSlice2325", "Squeeze2326", "BinaryOp2328", "/blocks.15/self_attn/Gather_output_0", "/blocks.15/self_attn/Unsqueeze_output_0", "Unsqueeze2336", "Unsqueeze2339", "StridedSlice2342", "Squeeze2343", "BinaryOp2344", "BinaryOp2345", "/blocks.15/self_attn/Gather_1_output_0", "/blocks.15/self_attn/Unsqueeze_1_output_0", "/blocks.15/self_attn/Concat_output_0", "/blocks.15/self_attn/Reshape_output_0", "/blocks.15/self_attn/q_norm/Cast_output_0", "/blocks.15/self_attn/q_norm/Mul_1_output_0", "/blocks.15/self_attn/Mul_output_0", "/blocks.15/self_attn/Shape_2_output_0", "Shape2360", "Rank2362", "BinaryOp2364", "Unsqueeze2365", "BinaryOp2367", "Unsqueeze2368", "StridedSlice2371", "Squeeze2372", "BinaryOp2373", "BinaryOp2374", "/blocks.15/self_attn/Gather_2_output_0", "/blocks.15/self_attn/Div_output_0", "/blocks.15/self_attn/Unsqueeze_7_output_0", "/blocks.15/self_attn/Slice_1_output_0", "/blocks.15/self_attn/Neg_output_0", "/blocks.15/self_attn/Unsqueeze_6_output_0", "/blocks.15/self_attn/Slice_output_0", "/blocks.15/self_attn/Concat_3_output_0", "/blocks.15/self_attn/Mul_1_output_0", "/blocks.15/self_attn/Add_output_0", "/blocks.15/self_attn/k_proj/FakeLinear_output_0", "/blocks.15/self_attn/Unsqueeze_2_output_0", "/blocks.15/self_attn/Unsqueeze_3_output_0", "/blocks.15/self_attn/Concat_1_output_0", "/blocks.15/self_attn/Reshape_1_output_0", "/blocks.15/self_attn/k_norm/Cast_output_0", "/blocks.15/self_attn/k_norm/Mul_1_output_0", "/blocks.15/self_attn/Mul_2_output_0", "/blocks.15/self_attn/Shape_3_output_0", "Shape2402", "Rank2404", "BinaryOp2406", "Unsqueeze2407", "BinaryOp2409", "Unsqueeze2410", "StridedSlice2413", "Squeeze2414", "BinaryOp2415", "BinaryOp2416", "/blocks.15/self_attn/Gather_3_output_0", "/blocks.15/self_attn/Div_1_output_0", "/blocks.15/self_attn/Unsqueeze_9_output_0", "/blocks.15/self_attn/Slice_3_output_0", "/blocks.15/self_attn/Neg_1_output_0", "/blocks.15/self_attn/Unsqueeze_8_output_0", "/blocks.15/self_attn/Slice_2_output_0", "/blocks.15/self_attn/Concat_4_output_0", "/blocks.15/self_attn/Mul_3_output_0", "/blocks.15/self_attn/Add_1_output_0", "/blocks.15/self_attn/v_proj/FakeLinear_output_0", "/blocks.15/self_attn/Unsqueeze_4_output_0", "/blocks.15/self_attn/Unsqueeze_5_output_0", "/blocks.15/self_attn/Concat_2_output_0", "/blocks.15/self_attn/Reshape_2_output_0", "/blocks.15/self_attn/fused_attn/FusedAttention_output_0", "/blocks.15/self_attn/o_proj/FakeLinear_output_0", "/blocks.15/Add_output_0", "/blocks.15/post_attention_layernorm/Mul_1_output_0", "/blocks.15/mlp/gate_proj/FakeLinear_output_0", "/blocks.15/mlp/act_fn/Mul_output_0", "/blocks.15/mlp/up_proj/FakeLinear_output_0", "/blocks.15/mlp/Mul_output_0", "/blocks.15/mlp/down_proj/FakeLinear_output_0", "/blocks.15/Add_1_output_0", "/blocks.16/Reshape_output_0", "/blocks.16/input_layernorm/Mul_1_output_0", "/blocks.16/self_attn/q_proj/FakeLinear_output_0", "/blocks.16/self_attn/Shape_output_0", "Shape2463", "Rank2465", "BinaryOp2467", "Unsqueeze2468", "BinaryOp2470", "Unsqueeze2471", "StridedSlice2474", "Squeeze2475", "BinaryOp2477", "/blocks.16/self_attn/Gather_output_0", "/blocks.16/self_attn/Unsqueeze_output_0", "Unsqueeze2485", "Unsqueeze2488", "StridedSlice2491", "Squeeze2492", "BinaryOp2493", "BinaryOp2494", "/blocks.16/self_attn/Gather_1_output_0", "/blocks.16/self_attn/Unsqueeze_1_output_0", "/blocks.16/self_attn/Concat_output_0", "/blocks.16/self_attn/Reshape_output_0", "/blocks.16/self_attn/q_norm/Cast_output_0", "/blocks.16/self_attn/q_norm/Mul_1_output_0", "/blocks.16/self_attn/Mul_output_0", "/blocks.16/self_attn/Shape_2_output_0", "Shape2509", "Rank2511", "BinaryOp2513", "Unsqueeze2514", "BinaryOp2516", "Unsqueeze2517", "StridedSlice2520", "Squeeze2521", "BinaryOp2522", "BinaryOp2523", "/blocks.16/self_attn/Gather_2_output_0", "/blocks.16/self_attn/Div_output_0", "/blocks.16/self_attn/Unsqueeze_7_output_0", "/blocks.16/self_attn/Slice_1_output_0", "/blocks.16/self_attn/Neg_output_0", "/blocks.16/self_attn/Unsqueeze_6_output_0", "/blocks.16/self_attn/Slice_output_0", "/blocks.16/self_attn/Concat_3_output_0", "/blocks.16/self_attn/Mul_1_output_0", "/blocks.16/self_attn/Add_output_0", "/blocks.16/self_attn/k_proj/FakeLinear_output_0", "/blocks.16/self_attn/Unsqueeze_2_output_0", "/blocks.16/self_attn/Unsqueeze_3_output_0", "/blocks.16/self_attn/Concat_1_output_0", "/blocks.16/self_attn/Reshape_1_output_0", "/blocks.16/self_attn/k_norm/Cast_output_0", "/blocks.16/self_attn/k_norm/Mul_1_output_0", "/blocks.16/self_attn/Mul_2_output_0", "/blocks.16/self_attn/Shape_3_output_0", "Shape2551", "Rank2553", "BinaryOp2555", "Unsqueeze2556", "BinaryOp2558", "Unsqueeze2559", "StridedSlice2562", "Squeeze2563", "BinaryOp2564", "BinaryOp2565", "/blocks.16/self_attn/Gather_3_output_0", "/blocks.16/self_attn/Div_1_output_0", "/blocks.16/self_attn/Unsqueeze_9_output_0", "/blocks.16/self_attn/Slice_3_output_0", "/blocks.16/self_attn/Neg_1_output_0", "/blocks.16/self_attn/Unsqueeze_8_output_0", "/blocks.16/self_attn/Slice_2_output_0", "/blocks.16/self_attn/Concat_4_output_0", "/blocks.16/self_attn/Mul_3_output_0", "/blocks.16/self_attn/Add_1_output_0", "/blocks.16/self_attn/v_proj/FakeLinear_output_0", "/blocks.16/self_attn/Unsqueeze_4_output_0", "/blocks.16/self_attn/Unsqueeze_5_output_0", "/blocks.16/self_attn/Concat_2_output_0", "/blocks.16/self_attn/Reshape_2_output_0", "/blocks.16/self_attn/fused_attn/FusedAttention_output_0", "/blocks.16/self_attn/o_proj/FakeLinear_output_0", "/blocks.16/Add_output_0", "/blocks.16/post_attention_layernorm/Mul_1_output_0", "/blocks.16/mlp/gate_proj/FakeLinear_output_0", "/blocks.16/mlp/act_fn/Mul_output_0", "/blocks.16/mlp/up_proj/FakeLinear_output_0", "/blocks.16/mlp/Mul_output_0", "/blocks.16/mlp/down_proj/FakeLinear_output_0", "/blocks.16/Add_1_output_0", "/blocks.17/Reshape_output_0", "/blocks.17/input_layernorm/Mul_1_output_0", "/blocks.17/self_attn/q_proj/FakeLinear_output_0", "/blocks.17/self_attn/Shape_output_0", "Shape2612", "Rank2614", "BinaryOp2616", "Unsqueeze2617", "BinaryOp2619", "Unsqueeze2620", "StridedSlice2623", "Squeeze2624", "BinaryOp2626", "/blocks.17/self_attn/Gather_output_0", "/blocks.17/self_attn/Unsqueeze_output_0", "Unsqueeze2634", "Unsqueeze2637", "StridedSlice2640", "Squeeze2641", "BinaryOp2642", "BinaryOp2643", "/blocks.17/self_attn/Gather_1_output_0", "/blocks.17/self_attn/Unsqueeze_1_output_0", "/blocks.17/self_attn/Concat_output_0", "/blocks.17/self_attn/Reshape_output_0", "/blocks.17/self_attn/q_norm/Cast_output_0", "/blocks.17/self_attn/q_norm/Mul_1_output_0", "/blocks.17/self_attn/Mul_output_0", "/blocks.17/self_attn/Shape_2_output_0", "Shape2658", "Rank2660", "BinaryOp2662", "Unsqueeze2663", "BinaryOp2665", "Unsqueeze2666", "StridedSlice2669", "Squeeze2670", "BinaryOp2671", "BinaryOp2672", "/blocks.17/self_attn/Gather_2_output_0", "/blocks.17/self_attn/Div_output_0", "/blocks.17/self_attn/Unsqueeze_7_output_0", "/blocks.17/self_attn/Slice_1_output_0", "/blocks.17/self_attn/Neg_output_0", "/blocks.17/self_attn/Unsqueeze_6_output_0", "/blocks.17/self_attn/Slice_output_0", "/blocks.17/self_attn/Concat_3_output_0", "/blocks.17/self_attn/Mul_1_output_0", "/blocks.17/self_attn/Add_output_0", "/blocks.17/self_attn/k_proj/FakeLinear_output_0", "/blocks.17/self_attn/Unsqueeze_2_output_0", "/blocks.17/self_attn/Unsqueeze_3_output_0", "/blocks.17/self_attn/Concat_1_output_0", "/blocks.17/self_attn/Reshape_1_output_0", "/blocks.17/self_attn/k_norm/Cast_output_0", "/blocks.17/self_attn/k_norm/Mul_1_output_0", "/blocks.17/self_attn/Mul_2_output_0", "/blocks.17/self_attn/Shape_3_output_0", "Shape2700", "Rank2702", "BinaryOp2704", "Unsqueeze2705", "BinaryOp2707", "Unsqueeze2708", "StridedSlice2711", "Squeeze2712", "BinaryOp2713", "BinaryOp2714", "/blocks.17/self_attn/Gather_3_output_0", "/blocks.17/self_attn/Div_1_output_0", "/blocks.17/self_attn/Unsqueeze_9_output_0", "/blocks.17/self_attn/Slice_3_output_0", "/blocks.17/self_attn/Neg_1_output_0", "/blocks.17/self_attn/Unsqueeze_8_output_0", "/blocks.17/self_attn/Slice_2_output_0", "/blocks.17/self_attn/Concat_4_output_0", "/blocks.17/self_attn/Mul_3_output_0", "/blocks.17/self_attn/Add_1_output_0", "/blocks.17/self_attn/v_proj/FakeLinear_output_0", "/blocks.17/self_attn/Unsqueeze_4_output_0", "/blocks.17/self_attn/Unsqueeze_5_output_0", "/blocks.17/self_attn/Concat_2_output_0", "/blocks.17/self_attn/Reshape_2_output_0", "/blocks.17/self_attn/fused_attn/FusedAttention_output_0", "/blocks.17/self_attn/o_proj/FakeLinear_output_0", "/blocks.17/Add_output_0", "/blocks.17/post_attention_layernorm/Mul_1_output_0", "/blocks.17/mlp/gate_proj/FakeLinear_output_0", "/blocks.17/mlp/act_fn/Mul_output_0", "/blocks.17/mlp/up_proj/FakeLinear_output_0", "/blocks.17/mlp/Mul_output_0", "/blocks.17/mlp/down_proj/FakeLinear_output_0", "/blocks.17/Add_1_output_0", "/blocks.18/Reshape_output_0", "/blocks.18/input_layernorm/Mul_1_output_0", "/blocks.18/self_attn/q_proj/FakeLinear_output_0", "/blocks.18/self_attn/Shape_output_0", "Shape2761", "Rank2763", "BinaryOp2765", "Unsqueeze2766", "BinaryOp2768", "Unsqueeze2769", "StridedSlice2772", "Squeeze2773", "BinaryOp2775", "/blocks.18/self_attn/Gather_output_0", "/blocks.18/self_attn/Unsqueeze_output_0", "Unsqueeze2783", "Unsqueeze2786", "StridedSlice2789", "Squeeze2790", "BinaryOp2791", "BinaryOp2792", "/blocks.18/self_attn/Gather_1_output_0", "/blocks.18/self_attn/Unsqueeze_1_output_0", "/blocks.18/self_attn/Concat_output_0", "/blocks.18/self_attn/Reshape_output_0", "/blocks.18/self_attn/q_norm/Cast_output_0", "/blocks.18/self_attn/q_norm/Mul_1_output_0", "/blocks.18/self_attn/Mul_output_0", "/blocks.18/self_attn/Shape_2_output_0", "Shape2807", "Rank2809", "BinaryOp2811", "Unsqueeze2812", "BinaryOp2814", "Unsqueeze2815", "StridedSlice2818", "Squeeze2819", "BinaryOp2820", "BinaryOp2821", "/blocks.18/self_attn/Gather_2_output_0", "/blocks.18/self_attn/Div_output_0", "/blocks.18/self_attn/Unsqueeze_7_output_0", "/blocks.18/self_attn/Slice_1_output_0", "/blocks.18/self_attn/Neg_output_0", "/blocks.18/self_attn/Unsqueeze_6_output_0", "/blocks.18/self_attn/Slice_output_0", "/blocks.18/self_attn/Concat_3_output_0", "/blocks.18/self_attn/Mul_1_output_0", "/blocks.18/self_attn/Add_output_0", "/blocks.18/self_attn/k_proj/FakeLinear_output_0", "/blocks.18/self_attn/Unsqueeze_2_output_0", "/blocks.18/self_attn/Unsqueeze_3_output_0", "/blocks.18/self_attn/Concat_1_output_0", "/blocks.18/self_attn/Reshape_1_output_0", "/blocks.18/self_attn/k_norm/Cast_output_0", "/blocks.18/self_attn/k_norm/Mul_1_output_0", "/blocks.18/self_attn/Mul_2_output_0", "/blocks.18/self_attn/Shape_3_output_0", "Shape2849", "Rank2851", "BinaryOp2853", "Unsqueeze2854", "BinaryOp2856", "Unsqueeze2857", "StridedSlice2860", "Squeeze2861", "BinaryOp2862", "BinaryOp2863", "/blocks.18/self_attn/Gather_3_output_0", "/blocks.18/self_attn/Div_1_output_0", "/blocks.18/self_attn/Unsqueeze_9_output_0", "/blocks.18/self_attn/Slice_3_output_0", "/blocks.18/self_attn/Neg_1_output_0", "/blocks.18/self_attn/Unsqueeze_8_output_0", "/blocks.18/self_attn/Slice_2_output_0", "/blocks.18/self_attn/Concat_4_output_0", "/blocks.18/self_attn/Mul_3_output_0", "/blocks.18/self_attn/Add_1_output_0", "/blocks.18/self_attn/v_proj/FakeLinear_output_0", "/blocks.18/self_attn/Unsqueeze_4_output_0", "/blocks.18/self_attn/Unsqueeze_5_output_0", "/blocks.18/self_attn/Concat_2_output_0", "/blocks.18/self_attn/Reshape_2_output_0", "/blocks.18/self_attn/fused_attn/FusedAttention_output_0", "/blocks.18/self_attn/o_proj/FakeLinear_output_0", "/blocks.18/Add_output_0", "/blocks.18/post_attention_layernorm/Mul_1_output_0", "/blocks.18/mlp/gate_proj/FakeLinear_output_0", "/blocks.18/mlp/act_fn/Mul_output_0", "/blocks.18/mlp/up_proj/FakeLinear_output_0", "/blocks.18/mlp/Mul_output_0", "/blocks.18/mlp/down_proj/FakeLinear_output_0", "/blocks.18/Add_1_output_0", "/blocks.19/Reshape_output_0", "/blocks.19/input_layernorm/Mul_1_output_0", "/blocks.19/self_attn/q_proj/FakeLinear_output_0", "/blocks.19/self_attn/Shape_output_0", "Shape2910", "Rank2912", "BinaryOp2914", "Unsqueeze2915", "BinaryOp2917", "Unsqueeze2918", "StridedSlice2921", "Squeeze2922", "BinaryOp2924", "/blocks.19/self_attn/Gather_output_0", "/blocks.19/self_attn/Unsqueeze_output_0", "Unsqueeze2932", "Unsqueeze2935", "StridedSlice2938", "Squeeze2939", "BinaryOp2940", "BinaryOp2941", "/blocks.19/self_attn/Gather_1_output_0", "/blocks.19/self_attn/Unsqueeze_1_output_0", "/blocks.19/self_attn/Concat_output_0", "/blocks.19/self_attn/Reshape_output_0", "/blocks.19/self_attn/q_norm/Cast_output_0", "/blocks.19/self_attn/q_norm/Mul_1_output_0", "/blocks.19/self_attn/Mul_output_0", "/blocks.19/self_attn/Shape_2_output_0", "Shape2956", "Rank2958", "BinaryOp2960", "Unsqueeze2961", "BinaryOp2963", "Unsqueeze2964", "StridedSlice2967", "Squeeze2968", "BinaryOp2969", "BinaryOp2970", "/blocks.19/self_attn/Gather_2_output_0", "/blocks.19/self_attn/Div_output_0", "/blocks.19/self_attn/Unsqueeze_7_output_0", "/blocks.19/self_attn/Slice_1_output_0", "/blocks.19/self_attn/Neg_output_0", "/blocks.19/self_attn/Unsqueeze_6_output_0", "/blocks.19/self_attn/Slice_output_0", "/blocks.19/self_attn/Concat_3_output_0", "/blocks.19/self_attn/Mul_1_output_0", "/blocks.19/self_attn/Add_output_0", "/blocks.19/self_attn/k_proj/FakeLinear_output_0", "/blocks.19/self_attn/Unsqueeze_2_output_0", "/blocks.19/self_attn/Unsqueeze_3_output_0", "/blocks.19/self_attn/Concat_1_output_0", "/blocks.19/self_attn/Reshape_1_output_0", "/blocks.19/self_attn/k_norm/Cast_output_0", "/blocks.19/self_attn/k_norm/Mul_1_output_0", "/blocks.19/self_attn/Mul_2_output_0", "/blocks.19/self_attn/Shape_3_output_0", "Shape2998", "Rank3000", "BinaryOp3002", "Unsqueeze3003", "BinaryOp3005", "Unsqueeze3006", "StridedSlice3009", "Squeeze3010", "BinaryOp3011", "BinaryOp3012", "/blocks.19/self_attn/Gather_3_output_0", "/blocks.19/self_attn/Div_1_output_0", "/blocks.19/self_attn/Unsqueeze_9_output_0", "/blocks.19/self_attn/Slice_3_output_0", "/blocks.19/self_attn/Neg_1_output_0", "/blocks.19/self_attn/Unsqueeze_8_output_0", "/blocks.19/self_attn/Slice_2_output_0", "/blocks.19/self_attn/Concat_4_output_0", "/blocks.19/self_attn/Mul_3_output_0", "/blocks.19/self_attn/Add_1_output_0", "/blocks.19/self_attn/v_proj/FakeLinear_output_0", "/blocks.19/self_attn/Unsqueeze_4_output_0", "/blocks.19/self_attn/Unsqueeze_5_output_0", "/blocks.19/self_attn/Concat_2_output_0", "/blocks.19/self_attn/Reshape_2_output_0", "/blocks.19/self_attn/fused_attn/FusedAttention_output_0", "/blocks.19/self_attn/o_proj/FakeLinear_output_0", "/blocks.19/Add_output_0", "/blocks.19/post_attention_layernorm/Mul_1_output_0", "/blocks.19/mlp/gate_proj/FakeLinear_output_0", "/blocks.19/mlp/act_fn/Mul_output_0", "/blocks.19/mlp/up_proj/FakeLinear_output_0", "/blocks.19/mlp/Mul_output_0", "/blocks.19/mlp/down_proj/FakeLinear_output_0", "/blocks.19/Add_1_output_0", "/blocks.20/Reshape_output_0", "/blocks.20/input_layernorm/Mul_1_output_0", "/blocks.20/self_attn/q_proj/FakeLinear_output_0", "/blocks.20/self_attn/Shape_output_0", "Shape3059", "Rank3061", "BinaryOp3063", "Unsqueeze3064", "BinaryOp3066", "Unsqueeze3067", "StridedSlice3070", "Squeeze3071", "BinaryOp3073", "/blocks.20/self_attn/Gather_output_0", "/blocks.20/self_attn/Unsqueeze_output_0", "Unsqueeze3081", "Unsqueeze3084", "StridedSlice3087", "Squeeze3088", "BinaryOp3089", "BinaryOp3090", "/blocks.20/self_attn/Gather_1_output_0", "/blocks.20/self_attn/Unsqueeze_1_output_0", "/blocks.20/self_attn/Concat_output_0", "/blocks.20/self_attn/Reshape_output_0", "/blocks.20/self_attn/q_norm/Cast_output_0", "/blocks.20/self_attn/q_norm/Mul_1_output_0", "/blocks.20/self_attn/Mul_output_0", "/blocks.20/self_attn/Shape_2_output_0", "Shape3105", "Rank3107", "BinaryOp3109", "Unsqueeze3110", "BinaryOp3112", "Unsqueeze3113", "StridedSlice3116", "Squeeze3117", "BinaryOp3118", "BinaryOp3119", "/blocks.20/self_attn/Gather_2_output_0", "/blocks.20/self_attn/Div_output_0", "/blocks.20/self_attn/Unsqueeze_7_output_0", "/blocks.20/self_attn/Slice_1_output_0", "/blocks.20/self_attn/Neg_output_0", "/blocks.20/self_attn/Unsqueeze_6_output_0", "/blocks.20/self_attn/Slice_output_0", "/blocks.20/self_attn/Concat_3_output_0", "/blocks.20/self_attn/Mul_1_output_0", "/blocks.20/self_attn/Add_output_0", "/blocks.20/self_attn/k_proj/FakeLinear_output_0", "/blocks.20/self_attn/Unsqueeze_2_output_0", "/blocks.20/self_attn/Unsqueeze_3_output_0", "/blocks.20/self_attn/Concat_1_output_0", "/blocks.20/self_attn/Reshape_1_output_0", "/blocks.20/self_attn/k_norm/Cast_output_0", "/blocks.20/self_attn/k_norm/Mul_1_output_0", "/blocks.20/self_attn/Mul_2_output_0", "/blocks.20/self_attn/Shape_3_output_0", "Shape3147", "Rank3149", "BinaryOp3151", "Unsqueeze3152", "BinaryOp3154", "Unsqueeze3155", "StridedSlice3158", "Squeeze3159", "BinaryOp3160", "BinaryOp3161", "/blocks.20/self_attn/Gather_3_output_0", "/blocks.20/self_attn/Div_1_output_0", "/blocks.20/self_attn/Unsqueeze_9_output_0", "/blocks.20/self_attn/Slice_3_output_0", "/blocks.20/self_attn/Neg_1_output_0", "/blocks.20/self_attn/Unsqueeze_8_output_0", "/blocks.20/self_attn/Slice_2_output_0", "/blocks.20/self_attn/Concat_4_output_0", "/blocks.20/self_attn/Mul_3_output_0", "/blocks.20/self_attn/Add_1_output_0", "/blocks.20/self_attn/v_proj/FakeLinear_output_0", "/blocks.20/self_attn/Unsqueeze_4_output_0", "/blocks.20/self_attn/Unsqueeze_5_output_0", "/blocks.20/self_attn/Concat_2_output_0", "/blocks.20/self_attn/Reshape_2_output_0", "/blocks.20/self_attn/fused_attn/FusedAttention_output_0", "/blocks.20/self_attn/o_proj/FakeLinear_output_0", "/blocks.20/Add_output_0", "/blocks.20/post_attention_layernorm/Mul_1_output_0", "/blocks.20/mlp/gate_proj/FakeLinear_output_0", "/blocks.20/mlp/act_fn/Mul_output_0", "/blocks.20/mlp/up_proj/FakeLinear_output_0", "/blocks.20/mlp/Mul_output_0", "/blocks.20/mlp/down_proj/FakeLinear_output_0", "/blocks.20/Add_1_output_0", "/blocks.21/Reshape_output_0", "/blocks.21/input_layernorm/Mul_1_output_0", "/blocks.21/self_attn/q_proj/FakeLinear_output_0", "/blocks.21/self_attn/Shape_output_0", "Shape3208", "Rank3210", "BinaryOp3212", "Unsqueeze3213", "BinaryOp3215", "Unsqueeze3216", "StridedSlice3219", "Squeeze3220", "BinaryOp3222", "/blocks.21/self_attn/Gather_output_0", "/blocks.21/self_attn/Unsqueeze_output_0", "Unsqueeze3230", "Unsqueeze3233", "StridedSlice3236", "Squeeze3237", "BinaryOp3238", "BinaryOp3239", "/blocks.21/self_attn/Gather_1_output_0", "/blocks.21/self_attn/Unsqueeze_1_output_0", "/blocks.21/self_attn/Concat_output_0", "/blocks.21/self_attn/Reshape_output_0", "/blocks.21/self_attn/q_norm/Cast_output_0", "/blocks.21/self_attn/q_norm/Mul_1_output_0", "/blocks.21/self_attn/Mul_output_0", "/blocks.21/self_attn/Shape_2_output_0", "Shape3254", "Rank3256", "BinaryOp3258", "Unsqueeze3259", "BinaryOp3261", "Unsqueeze3262", "StridedSlice3265", "Squeeze3266", "BinaryOp3267", "BinaryOp3268", "/blocks.21/self_attn/Gather_2_output_0", "/blocks.21/self_attn/Div_output_0", "/blocks.21/self_attn/Unsqueeze_7_output_0", "/blocks.21/self_attn/Slice_1_output_0", "/blocks.21/self_attn/Neg_output_0", "/blocks.21/self_attn/Unsqueeze_6_output_0", "/blocks.21/self_attn/Slice_output_0", "/blocks.21/self_attn/Concat_3_output_0", "/blocks.21/self_attn/Mul_1_output_0", "/blocks.21/self_attn/Add_output_0", "/blocks.21/self_attn/k_proj/FakeLinear_output_0", "/blocks.21/self_attn/Unsqueeze_2_output_0", "/blocks.21/self_attn/Unsqueeze_3_output_0", "/blocks.21/self_attn/Concat_1_output_0", "/blocks.21/self_attn/Reshape_1_output_0", "/blocks.21/self_attn/k_norm/Cast_output_0", "/blocks.21/self_attn/k_norm/Mul_1_output_0", "/blocks.21/self_attn/Mul_2_output_0", "/blocks.21/self_attn/Shape_3_output_0", "Shape3296", "Rank3298", "BinaryOp3300", "Unsqueeze3301", "BinaryOp3303", "Unsqueeze3304", "StridedSlice3307", "Squeeze3308", "BinaryOp3309", "BinaryOp3310", "/blocks.21/self_attn/Gather_3_output_0", "/blocks.21/self_attn/Div_1_output_0", "/blocks.21/self_attn/Unsqueeze_9_output_0", "/blocks.21/self_attn/Slice_3_output_0", "/blocks.21/self_attn/Neg_1_output_0", "/blocks.21/self_attn/Unsqueeze_8_output_0", "/blocks.21/self_attn/Slice_2_output_0", "/blocks.21/self_attn/Concat_4_output_0", "/blocks.21/self_attn/Mul_3_output_0", "/blocks.21/self_attn/Add_1_output_0", "/blocks.21/self_attn/v_proj/FakeLinear_output_0", "/blocks.21/self_attn/Unsqueeze_4_output_0", "/blocks.21/self_attn/Unsqueeze_5_output_0", "/blocks.21/self_attn/Concat_2_output_0", "/blocks.21/self_attn/Reshape_2_output_0", "/blocks.21/self_attn/fused_attn/FusedAttention_output_0", "/blocks.21/self_attn/o_proj/FakeLinear_output_0", "/blocks.21/Add_output_0", "/blocks.21/post_attention_layernorm/Mul_1_output_0", "/blocks.21/mlp/gate_proj/FakeLinear_output_0", "/blocks.21/mlp/act_fn/Mul_output_0", "/blocks.21/mlp/up_proj/FakeLinear_output_0", "/blocks.21/mlp/Mul_output_0", "/blocks.21/mlp/down_proj/FakeLinear_output_0", "/blocks.21/Add_1_output_0", "/blocks.22/Reshape_output_0", "/blocks.22/input_layernorm/Mul_1_output_0", "/blocks.22/self_attn/q_proj/FakeLinear_output_0", "/blocks.22/self_attn/Shape_output_0", "Shape3357", "Rank3359", "BinaryOp3361", "Unsqueeze3362", "BinaryOp3364", "Unsqueeze3365", "StridedSlice3368", "Squeeze3369", "BinaryOp3371", "/blocks.22/self_attn/Gather_output_0", "/blocks.22/self_attn/Unsqueeze_output_0", "Unsqueeze3379", "Unsqueeze3382", "StridedSlice3385", "Squeeze3386", "BinaryOp3387", "BinaryOp3388", "/blocks.22/self_attn/Gather_1_output_0", "/blocks.22/self_attn/Unsqueeze_1_output_0", "/blocks.22/self_attn/Concat_output_0", "/blocks.22/self_attn/Reshape_output_0", "/blocks.22/self_attn/q_norm/Cast_output_0", "/blocks.22/self_attn/q_norm/Mul_1_output_0", "/blocks.22/self_attn/Mul_output_0", "/blocks.22/self_attn/Shape_2_output_0", "Shape3403", "Rank3405", "BinaryOp3407", "Unsqueeze3408", "BinaryOp3410", "Unsqueeze3411", "StridedSlice3414", "Squeeze3415", "BinaryOp3416", "BinaryOp3417", "/blocks.22/self_attn/Gather_2_output_0", "/blocks.22/self_attn/Div_output_0", "/blocks.22/self_attn/Unsqueeze_7_output_0", "/blocks.22/self_attn/Slice_1_output_0", "/blocks.22/self_attn/Neg_output_0", "/blocks.22/self_attn/Unsqueeze_6_output_0", "/blocks.22/self_attn/Slice_output_0", "/blocks.22/self_attn/Concat_3_output_0", "/blocks.22/self_attn/Mul_1_output_0", "/blocks.22/self_attn/Add_output_0", "/blocks.22/self_attn/k_proj/FakeLinear_output_0", "/blocks.22/self_attn/Unsqueeze_2_output_0", "/blocks.22/self_attn/Unsqueeze_3_output_0", "/blocks.22/self_attn/Concat_1_output_0", "/blocks.22/self_attn/Reshape_1_output_0", "/blocks.22/self_attn/k_norm/Cast_output_0", "/blocks.22/self_attn/k_norm/Mul_1_output_0", "/blocks.22/self_attn/Mul_2_output_0", "/blocks.22/self_attn/Shape_3_output_0", "Shape3445", "Rank3447", "BinaryOp3449", "Unsqueeze3450", "BinaryOp3452", "Unsqueeze3453", "StridedSlice3456", "Squeeze3457", "BinaryOp3458", "BinaryOp3459", "/blocks.22/self_attn/Gather_3_output_0", "/blocks.22/self_attn/Div_1_output_0", "/blocks.22/self_attn/Unsqueeze_9_output_0", "/blocks.22/self_attn/Slice_3_output_0", "/blocks.22/self_attn/Neg_1_output_0", "/blocks.22/self_attn/Unsqueeze_8_output_0", "/blocks.22/self_attn/Slice_2_output_0", "/blocks.22/self_attn/Concat_4_output_0", "/blocks.22/self_attn/Mul_3_output_0", "/blocks.22/self_attn/Add_1_output_0", "/blocks.22/self_attn/v_proj/FakeLinear_output_0", "/blocks.22/self_attn/Unsqueeze_4_output_0", "/blocks.22/self_attn/Unsqueeze_5_output_0", "/blocks.22/self_attn/Concat_2_output_0", "/blocks.22/self_attn/Reshape_2_output_0", "/blocks.22/self_attn/fused_attn/FusedAttention_output_0", "/blocks.22/self_attn/o_proj/FakeLinear_output_0", "/blocks.22/Add_output_0", "/blocks.22/post_attention_layernorm/Mul_1_output_0", "/blocks.22/mlp/gate_proj/FakeLinear_output_0", "/blocks.22/mlp/act_fn/Mul_output_0", "/blocks.22/mlp/up_proj/FakeLinear_output_0", "/blocks.22/mlp/Mul_output_0", "/blocks.22/mlp/down_proj/FakeLinear_output_0", "/blocks.22/Add_1_output_0", "/blocks.23/Reshape_output_0", "/blocks.23/input_layernorm/Mul_1_output_0", "/blocks.23/self_attn/q_proj/FakeLinear_output_0", "/blocks.23/self_attn/Shape_output_0", "Shape3506", "Rank3508", "BinaryOp3510", "Unsqueeze3511", "BinaryOp3513", "Unsqueeze3514", "StridedSlice3517", "Squeeze3518", "BinaryOp3520", "/blocks.23/self_attn/Gather_output_0", "/blocks.23/self_attn/Unsqueeze_output_0", "Unsqueeze3528", "Unsqueeze3531", "StridedSlice3534", "Squeeze3535", "BinaryOp3536", "BinaryOp3537", "/blocks.23/self_attn/Gather_1_output_0", "/blocks.23/self_attn/Unsqueeze_1_output_0", "/blocks.23/self_attn/Concat_output_0", "/blocks.23/self_attn/Reshape_output_0", "/blocks.23/self_attn/q_norm/Cast_output_0", "/blocks.23/self_attn/q_norm/Mul_1_output_0", "/blocks.23/self_attn/Mul_output_0", "/blocks.23/self_attn/Shape_2_output_0", "Shape3552", "Rank3554", "BinaryOp3556", "Unsqueeze3557", "BinaryOp3559", "Unsqueeze3560", "StridedSlice3563", "Squeeze3564", "BinaryOp3565", "BinaryOp3566", "/blocks.23/self_attn/Gather_2_output_0", "/blocks.23/self_attn/Div_output_0", "/blocks.23/self_attn/Unsqueeze_7_output_0", "/blocks.23/self_attn/Slice_1_output_0", "/blocks.23/self_attn/Neg_output_0", "/blocks.23/self_attn/Unsqueeze_6_output_0", "/blocks.23/self_attn/Slice_output_0", "/blocks.23/self_attn/Concat_3_output_0", "/blocks.23/self_attn/Mul_1_output_0", "/blocks.23/self_attn/Add_output_0", "/blocks.23/self_attn/k_proj/FakeLinear_output_0", "/blocks.23/self_attn/Unsqueeze_2_output_0", "/blocks.23/self_attn/Unsqueeze_3_output_0", "/blocks.23/self_attn/Concat_1_output_0", "/blocks.23/self_attn/Reshape_1_output_0", "/blocks.23/self_attn/k_norm/Cast_output_0", "/blocks.23/self_attn/k_norm/Mul_1_output_0", "/blocks.23/self_attn/Mul_2_output_0", "/blocks.23/self_attn/Shape_3_output_0", "Shape3594", "Rank3596", "BinaryOp3598", "Unsqueeze3599", "BinaryOp3601", "Unsqueeze3602", "StridedSlice3605", "Squeeze3606", "BinaryOp3607", "BinaryOp3608", "/blocks.23/self_attn/Gather_3_output_0", "/blocks.23/self_attn/Div_1_output_0", "/blocks.23/self_attn/Unsqueeze_9_output_0", "/blocks.23/self_attn/Slice_3_output_0", "/blocks.23/self_attn/Neg_1_output_0", "/blocks.23/self_attn/Unsqueeze_8_output_0", "/blocks.23/self_attn/Slice_2_output_0", "/blocks.23/self_attn/Concat_4_output_0", "/blocks.23/self_attn/Mul_3_output_0", "/blocks.23/self_attn/Add_1_output_0", "/blocks.23/self_attn/v_proj/FakeLinear_output_0", "/blocks.23/self_attn/Unsqueeze_4_output_0", "/blocks.23/self_attn/Unsqueeze_5_output_0", "/blocks.23/self_attn/Concat_2_output_0", "/blocks.23/self_attn/Reshape_2_output_0", "/blocks.23/self_attn/fused_attn/FusedAttention_output_0", "/blocks.23/self_attn/o_proj/FakeLinear_output_0", "/blocks.23/Add_output_0", "/blocks.23/post_attention_layernorm/Mul_1_output_0", "/blocks.23/mlp/gate_proj/FakeLinear_output_0", "/blocks.23/mlp/act_fn/Mul_output_0", "/blocks.23/mlp/up_proj/FakeLinear_output_0", "/blocks.23/mlp/Mul_output_0", "/blocks.23/mlp/down_proj/FakeLinear_output_0", "/blocks.23/Add_1_output_0", "/blocks.24/Reshape_output_0", "/blocks.24/input_layernorm/Mul_1_output_0", "/blocks.24/self_attn/q_proj/FakeLinear_output_0", "/blocks.24/self_attn/Shape_output_0", "Shape3655", "Rank3657", "BinaryOp3659", "Unsqueeze3660", "BinaryOp3662", "Unsqueeze3663", "StridedSlice3666", "Squeeze3667", "BinaryOp3669", "/blocks.24/self_attn/Gather_output_0", "/blocks.24/self_attn/Unsqueeze_output_0", "Unsqueeze3677", "Unsqueeze3680", "StridedSlice3683", "Squeeze3684", "BinaryOp3685", "BinaryOp3686", "/blocks.24/self_attn/Gather_1_output_0", "/blocks.24/self_attn/Unsqueeze_1_output_0", "/blocks.24/self_attn/Concat_output_0", "/blocks.24/self_attn/Reshape_output_0", "/blocks.24/self_attn/q_norm/Cast_output_0", "/blocks.24/self_attn/q_norm/Mul_1_output_0", "/blocks.24/self_attn/Mul_output_0", "/blocks.24/self_attn/Shape_2_output_0", "Shape3701", "Rank3703", "BinaryOp3705", "Unsqueeze3706", "BinaryOp3708", "Unsqueeze3709", "StridedSlice3712", "Squeeze3713", "BinaryOp3714", "BinaryOp3715", "/blocks.24/self_attn/Gather_2_output_0", "/blocks.24/self_attn/Div_output_0", "/blocks.24/self_attn/Unsqueeze_7_output_0", "/blocks.24/self_attn/Slice_1_output_0", "/blocks.24/self_attn/Neg_output_0", "/blocks.24/self_attn/Unsqueeze_6_output_0", "/blocks.24/self_attn/Slice_output_0", "/blocks.24/self_attn/Concat_3_output_0", "/blocks.24/self_attn/Mul_1_output_0", "/blocks.24/self_attn/Add_output_0", "/blocks.24/self_attn/k_proj/FakeLinear_output_0", "/blocks.24/self_attn/Unsqueeze_2_output_0", "/blocks.24/self_attn/Unsqueeze_3_output_0", "/blocks.24/self_attn/Concat_1_output_0", "/blocks.24/self_attn/Reshape_1_output_0", "/blocks.24/self_attn/k_norm/Cast_output_0", "/blocks.24/self_attn/k_norm/Mul_1_output_0", "/blocks.24/self_attn/Mul_2_output_0", "/blocks.24/self_attn/Shape_3_output_0", "Shape3743", "Rank3745", "BinaryOp3747", "Unsqueeze3748", "BinaryOp3750", "Unsqueeze3751", "StridedSlice3754", "Squeeze3755", "BinaryOp3756", "BinaryOp3757", "/blocks.24/self_attn/Gather_3_output_0", "/blocks.24/self_attn/Div_1_output_0", "/blocks.24/self_attn/Unsqueeze_9_output_0", "/blocks.24/self_attn/Slice_3_output_0", "/blocks.24/self_attn/Neg_1_output_0", "/blocks.24/self_attn/Unsqueeze_8_output_0", "/blocks.24/self_attn/Slice_2_output_0", "/blocks.24/self_attn/Concat_4_output_0", "/blocks.24/self_attn/Mul_3_output_0", "/blocks.24/self_attn/Add_1_output_0", "/blocks.24/self_attn/v_proj/FakeLinear_output_0", "/blocks.24/self_attn/Unsqueeze_4_output_0", "/blocks.24/self_attn/Unsqueeze_5_output_0", "/blocks.24/self_attn/Concat_2_output_0", "/blocks.24/self_attn/Reshape_2_output_0", "/blocks.24/self_attn/fused_attn/FusedAttention_output_0", "/blocks.24/self_attn/o_proj/FakeLinear_output_0", "/blocks.24/Add_output_0", "/blocks.24/post_attention_layernorm/Mul_1_output_0", "/blocks.24/mlp/gate_proj/FakeLinear_output_0", "/blocks.24/mlp/act_fn/Mul_output_0", "/blocks.24/mlp/up_proj/FakeLinear_output_0", "/blocks.24/mlp/Mul_output_0", "/blocks.24/mlp/down_proj/FakeLinear_output_0", "/blocks.24/Add_1_output_0", "/blocks.25/Reshape_output_0", "/blocks.25/input_layernorm/Mul_1_output_0", "/blocks.25/self_attn/q_proj/FakeLinear_output_0", "/blocks.25/self_attn/Shape_output_0", "Shape3804", "Rank3806", "BinaryOp3808", "Unsqueeze3809", "BinaryOp3811", "Unsqueeze3812", "StridedSlice3815", "Squeeze3816", "BinaryOp3818", "/blocks.25/self_attn/Gather_output_0", "/blocks.25/self_attn/Unsqueeze_output_0", "Unsqueeze3826", "Unsqueeze3829", "StridedSlice3832", "Squeeze3833", "BinaryOp3834", "BinaryOp3835", "/blocks.25/self_attn/Gather_1_output_0", "/blocks.25/self_attn/Unsqueeze_1_output_0", "/blocks.25/self_attn/Concat_output_0", "/blocks.25/self_attn/Reshape_output_0", "/blocks.25/self_attn/q_norm/Cast_output_0", "/blocks.25/self_attn/q_norm/Mul_1_output_0", "/blocks.25/self_attn/Mul_output_0", "/blocks.25/self_attn/Shape_2_output_0", "Shape3850", "Rank3852", "BinaryOp3854", "Unsqueeze3855", "BinaryOp3857", "Unsqueeze3858", "StridedSlice3861", "Squeeze3862", "BinaryOp3863", "BinaryOp3864", "/blocks.25/self_attn/Gather_2_output_0", "/blocks.25/self_attn/Div_output_0", "/blocks.25/self_attn/Unsqueeze_7_output_0", "/blocks.25/self_attn/Slice_1_output_0", "/blocks.25/self_attn/Neg_output_0", "/blocks.25/self_attn/Unsqueeze_6_output_0", "/blocks.25/self_attn/Slice_output_0", "/blocks.25/self_attn/Concat_3_output_0", "/blocks.25/self_attn/Mul_1_output_0", "/blocks.25/self_attn/Add_output_0", "/blocks.25/self_attn/k_proj/FakeLinear_output_0", "/blocks.25/self_attn/Unsqueeze_2_output_0", "/blocks.25/self_attn/Unsqueeze_3_output_0", "/blocks.25/self_attn/Concat_1_output_0", "/blocks.25/self_attn/Reshape_1_output_0", "/blocks.25/self_attn/k_norm/Cast_output_0", "/blocks.25/self_attn/k_norm/Mul_1_output_0", "/blocks.25/self_attn/Mul_2_output_0", "/blocks.25/self_attn/Shape_3_output_0", "Shape3892", "Rank3894", "BinaryOp3896", "Unsqueeze3897", "BinaryOp3899", "Unsqueeze3900", "StridedSlice3903", "Squeeze3904", "BinaryOp3905", "BinaryOp3906", "/blocks.25/self_attn/Gather_3_output_0", "/blocks.25/self_attn/Div_1_output_0", "/blocks.25/self_attn/Unsqueeze_9_output_0", "/blocks.25/self_attn/Slice_3_output_0", "/blocks.25/self_attn/Neg_1_output_0", "/blocks.25/self_attn/Unsqueeze_8_output_0", "/blocks.25/self_attn/Slice_2_output_0", "/blocks.25/self_attn/Concat_4_output_0", "/blocks.25/self_attn/Mul_3_output_0", "/blocks.25/self_attn/Add_1_output_0", "/blocks.25/self_attn/v_proj/FakeLinear_output_0", "/blocks.25/self_attn/Unsqueeze_4_output_0", "/blocks.25/self_attn/Unsqueeze_5_output_0", "/blocks.25/self_attn/Concat_2_output_0", "/blocks.25/self_attn/Reshape_2_output_0", "/blocks.25/self_attn/fused_attn/FusedAttention_output_0", "/blocks.25/self_attn/o_proj/FakeLinear_output_0", "/blocks.25/Add_output_0", "/blocks.25/post_attention_layernorm/Mul_1_output_0", "/blocks.25/mlp/gate_proj/FakeLinear_output_0", "/blocks.25/mlp/act_fn/Mul_output_0", "/blocks.25/mlp/up_proj/FakeLinear_output_0", "/blocks.25/mlp/Mul_output_0", "/blocks.25/mlp/down_proj/FakeLinear_output_0", "/blocks.25/Add_1_output_0", "/blocks.26/Reshape_output_0", "/blocks.26/input_layernorm/Mul_1_output_0", "/blocks.26/self_attn/q_proj/FakeLinear_output_0", "/blocks.26/self_attn/Shape_output_0", "Shape3953", "Rank3955", "BinaryOp3957", "Unsqueeze3958", "BinaryOp3960", "Unsqueeze3961", "StridedSlice3964", "Squeeze3965", "BinaryOp3967", "/blocks.26/self_attn/Gather_output_0", "/blocks.26/self_attn/Unsqueeze_output_0", "Unsqueeze3975", "Unsqueeze3978", "StridedSlice3981", "Squeeze3982", "BinaryOp3983", "BinaryOp3984", "/blocks.26/self_attn/Gather_1_output_0", "/blocks.26/self_attn/Unsqueeze_1_output_0", "/blocks.26/self_attn/Concat_output_0", "/blocks.26/self_attn/Reshape_output_0", "/blocks.26/self_attn/q_norm/Cast_output_0", "/blocks.26/self_attn/q_norm/Mul_1_output_0", "/blocks.26/self_attn/Mul_output_0", "/blocks.26/self_attn/Shape_2_output_0", "Shape3999", "Rank4001", "BinaryOp4003", "Unsqueeze4004", "BinaryOp4006", "Unsqueeze4007", "StridedSlice4010", "Squeeze4011", "BinaryOp4012", "BinaryOp4013", "/blocks.26/self_attn/Gather_2_output_0", "/blocks.26/self_attn/Div_output_0", "/blocks.26/self_attn/Unsqueeze_7_output_0", "/blocks.26/self_attn/Slice_1_output_0", "/blocks.26/self_attn/Neg_output_0", "/blocks.26/self_attn/Unsqueeze_6_output_0", "/blocks.26/self_attn/Slice_output_0", "/blocks.26/self_attn/Concat_3_output_0", "/blocks.26/self_attn/Mul_1_output_0", "/blocks.26/self_attn/Add_output_0", "/blocks.26/self_attn/k_proj/FakeLinear_output_0", "/blocks.26/self_attn/Unsqueeze_2_output_0", "/blocks.26/self_attn/Unsqueeze_3_output_0", "/blocks.26/self_attn/Concat_1_output_0", "/blocks.26/self_attn/Reshape_1_output_0", "/blocks.26/self_attn/k_norm/Cast_output_0", "/blocks.26/self_attn/k_norm/Mul_1_output_0", "/blocks.26/self_attn/Mul_2_output_0", "/blocks.26/self_attn/Shape_3_output_0", "Shape4041", "Rank4043", "BinaryOp4045", "Unsqueeze4046", "BinaryOp4048", "Unsqueeze4049", "StridedSlice4052", "Squeeze4053", "BinaryOp4054", "BinaryOp4055", "/blocks.26/self_attn/Gather_3_output_0", "/blocks.26/self_attn/Div_1_output_0", "/blocks.26/self_attn/Unsqueeze_9_output_0", "/blocks.26/self_attn/Slice_3_output_0", "/blocks.26/self_attn/Neg_1_output_0", "/blocks.26/self_attn/Unsqueeze_8_output_0", "/blocks.26/self_attn/Slice_2_output_0", "/blocks.26/self_attn/Concat_4_output_0", "/blocks.26/self_attn/Mul_3_output_0", "/blocks.26/self_attn/Add_1_output_0", "/blocks.26/self_attn/v_proj/FakeLinear_output_0", "/blocks.26/self_attn/Unsqueeze_4_output_0", "/blocks.26/self_attn/Unsqueeze_5_output_0", "/blocks.26/self_attn/Concat_2_output_0", "/blocks.26/self_attn/Reshape_2_output_0", "/blocks.26/self_attn/fused_attn/FusedAttention_output_0", "/blocks.26/self_attn/o_proj/FakeLinear_output_0", "/blocks.26/Add_output_0", "/blocks.26/post_attention_layernorm/Mul_1_output_0", "/blocks.26/mlp/gate_proj/FakeLinear_output_0", "/blocks.26/mlp/act_fn/Mul_output_0", "/blocks.26/mlp/up_proj/FakeLinear_output_0", "/blocks.26/mlp/Mul_output_0", "/blocks.26/mlp/down_proj/FakeLinear_output_0", "/blocks.26/Add_1_output_0", "/blocks.27/Reshape_output_0", "/blocks.27/input_layernorm/Mul_1_output_0", "/blocks.27/self_attn/q_proj/FakeLinear_output_0", "/blocks.27/self_attn/Shape_output_0", "Shape4102", "Rank4104", "BinaryOp4106", "Unsqueeze4107", "BinaryOp4109", "Unsqueeze4110", "StridedSlice4113", "Squeeze4114", "BinaryOp4116", "/blocks.27/self_attn/Gather_output_0", "/blocks.27/self_attn/Unsqueeze_output_0", "Unsqueeze4124", "Unsqueeze4127", "StridedSlice4130", "Squeeze4131", "BinaryOp4132", "BinaryOp4133", "/blocks.27/self_attn/Gather_1_output_0", "/blocks.27/self_attn/Unsqueeze_1_output_0", "/blocks.27/self_attn/Concat_output_0", "/blocks.27/self_attn/Reshape_output_0", "/blocks.27/self_attn/q_norm/Cast_output_0", "/blocks.27/self_attn/q_norm/Mul_1_output_0", "/blocks.27/self_attn/Mul_output_0", "/blocks.27/self_attn/Shape_2_output_0", "Shape4148", "Rank4150", "BinaryOp4152", "Unsqueeze4153", "BinaryOp4155", "Unsqueeze4156", "StridedSlice4159", "Squeeze4160", "BinaryOp4161", "BinaryOp4162", "/blocks.27/self_attn/Gather_2_output_0", "/blocks.27/self_attn/Div_output_0", "/blocks.27/self_attn/Unsqueeze_7_output_0", "/blocks.27/self_attn/Slice_1_output_0", "/blocks.27/self_attn/Neg_output_0", "/blocks.27/self_attn/Unsqueeze_6_output_0", "/blocks.27/self_attn/Slice_output_0", "/blocks.27/self_attn/Concat_3_output_0", "/blocks.27/self_attn/Mul_1_output_0", "/blocks.27/self_attn/Add_output_0", "/blocks.27/self_attn/k_proj/FakeLinear_output_0", "/blocks.27/self_attn/Unsqueeze_2_output_0", "/blocks.27/self_attn/Unsqueeze_3_output_0", "/blocks.27/self_attn/Concat_1_output_0", "/blocks.27/self_attn/Reshape_1_output_0", "/blocks.27/self_attn/k_norm/Cast_output_0", "/blocks.27/self_attn/k_norm/Mul_1_output_0", "/blocks.27/self_attn/Mul_2_output_0", "/blocks.27/self_attn/Shape_3_output_0", "Shape4190", "Rank4192", "BinaryOp4194", "Unsqueeze4195", "BinaryOp4197", "Unsqueeze4198", "StridedSlice4201", "Squeeze4202", "BinaryOp4203", "BinaryOp4204", "/blocks.27/self_attn/Gather_3_output_0", "/blocks.27/self_attn/Div_1_output_0", "/blocks.27/self_attn/Unsqueeze_9_output_0", "/blocks.27/self_attn/Slice_3_output_0", "/blocks.27/self_attn/Neg_1_output_0", "/blocks.27/self_attn/Unsqueeze_8_output_0", "/blocks.27/self_attn/Slice_2_output_0", "/blocks.27/self_attn/Concat_4_output_0", "/blocks.27/self_attn/Mul_3_output_0", "/blocks.27/self_attn/Add_1_output_0", "/blocks.27/self_attn/v_proj/FakeLinear_output_0", "/blocks.27/self_attn/Unsqueeze_4_output_0", "/blocks.27/self_attn/Unsqueeze_5_output_0", "/blocks.27/self_attn/Concat_2_output_0", "/blocks.27/self_attn/Reshape_2_output_0", "/blocks.27/self_attn/fused_attn/FusedAttention_output_0", "/blocks.27/self_attn/o_proj/FakeLinear_output_0", "/blocks.27/Add_output_0", "/blocks.27/post_attention_layernorm/Mul_1_output_0", "/blocks.27/mlp/gate_proj/FakeLinear_output_0", "/blocks.27/mlp/act_fn/Mul_output_0", "/blocks.27/mlp/up_proj/FakeLinear_output_0", "/blocks.27/mlp/Mul_output_0", "/blocks.27/mlp/down_proj/FakeLinear_output_0", "/blocks.27/Add_1_output_0", "/blocks.28/Reshape_output_0", "/blocks.28/input_layernorm/Mul_1_output_0", "/blocks.28/self_attn/q_proj/FakeLinear_output_0", "/blocks.28/self_attn/Shape_output_0", "Shape4251", "Rank4253", "BinaryOp4255", "Unsqueeze4256", "BinaryOp4258", "Unsqueeze4259", "StridedSlice4262", "Squeeze4263", "BinaryOp4265", "/blocks.28/self_attn/Gather_output_0", "/blocks.28/self_attn/Unsqueeze_output_0", "Unsqueeze4273", "Unsqueeze4276", "StridedSlice4279", "Squeeze4280", "BinaryOp4281", "BinaryOp4282", "/blocks.28/self_attn/Gather_1_output_0", "/blocks.28/self_attn/Unsqueeze_1_output_0", "/blocks.28/self_attn/Concat_output_0", "/blocks.28/self_attn/Reshape_output_0", "/blocks.28/self_attn/q_norm/Cast_output_0", "/blocks.28/self_attn/q_norm/Mul_1_output_0", "/blocks.28/self_attn/Mul_output_0", "/blocks.28/self_attn/Shape_2_output_0", "Shape4297", "Rank4299", "BinaryOp4301", "Unsqueeze4302", "BinaryOp4304", "Unsqueeze4305", "StridedSlice4308", "Squeeze4309", "BinaryOp4310", "BinaryOp4311", "/blocks.28/self_attn/Gather_2_output_0", "/blocks.28/self_attn/Div_output_0", "/blocks.28/self_attn/Unsqueeze_7_output_0", "/blocks.28/self_attn/Slice_1_output_0", "/blocks.28/self_attn/Neg_output_0", "/blocks.28/self_attn/Unsqueeze_6_output_0", "/blocks.28/self_attn/Slice_output_0", "/blocks.28/self_attn/Concat_3_output_0", "/blocks.28/self_attn/Mul_1_output_0", "/blocks.28/self_attn/Add_output_0", "/blocks.28/self_attn/k_proj/FakeLinear_output_0", "/blocks.28/self_attn/Unsqueeze_2_output_0", "/blocks.28/self_attn/Unsqueeze_3_output_0", "/blocks.28/self_attn/Concat_1_output_0", "/blocks.28/self_attn/Reshape_1_output_0", "/blocks.28/self_attn/k_norm/Cast_output_0", "/blocks.28/self_attn/k_norm/Mul_1_output_0", "/blocks.28/self_attn/Mul_2_output_0", "/blocks.28/self_attn/Shape_3_output_0", "Shape4339", "Rank4341", "BinaryOp4343", "Unsqueeze4344", "BinaryOp4346", "Unsqueeze4347", "StridedSlice4350", "Squeeze4351", "BinaryOp4352", "BinaryOp4353", "/blocks.28/self_attn/Gather_3_output_0", "/blocks.28/self_attn/Div_1_output_0", "/blocks.28/self_attn/Unsqueeze_9_output_0", "/blocks.28/self_attn/Slice_3_output_0", "/blocks.28/self_attn/Neg_1_output_0", "/blocks.28/self_attn/Unsqueeze_8_output_0", "/blocks.28/self_attn/Slice_2_output_0", "/blocks.28/self_attn/Concat_4_output_0", "/blocks.28/self_attn/Mul_3_output_0", "/blocks.28/self_attn/Add_1_output_0", "/blocks.28/self_attn/v_proj/FakeLinear_output_0", "/blocks.28/self_attn/Unsqueeze_4_output_0", "/blocks.28/self_attn/Unsqueeze_5_output_0", "/blocks.28/self_attn/Concat_2_output_0", "/blocks.28/self_attn/Reshape_2_output_0", "/blocks.28/self_attn/fused_attn/FusedAttention_output_0", "/blocks.28/self_attn/o_proj/FakeLinear_output_0", "/blocks.28/Add_output_0", "/blocks.28/post_attention_layernorm/Mul_1_output_0", "/blocks.28/mlp/gate_proj/FakeLinear_output_0", "/blocks.28/mlp/act_fn/Mul_output_0", "/blocks.28/mlp/up_proj/FakeLinear_output_0", "/blocks.28/mlp/Mul_output_0", "/blocks.28/mlp/down_proj/FakeLinear_output_0", "/blocks.28/Add_1_output_0", "/blocks.29/Reshape_output_0", "/blocks.29/input_layernorm/Mul_1_output_0", "/blocks.29/self_attn/q_proj/FakeLinear_output_0", "/blocks.29/self_attn/Shape_output_0", "Shape4400", "Rank4402", "BinaryOp4404", "Unsqueeze4405", "BinaryOp4407", "Unsqueeze4408", "StridedSlice4411", "Squeeze4412", "BinaryOp4414", "/blocks.29/self_attn/Gather_output_0", "/blocks.29/self_attn/Unsqueeze_output_0", "Unsqueeze4422", "Unsqueeze4425", "StridedSlice4428", "Squeeze4429", "BinaryOp4430", "BinaryOp4431", "/blocks.29/self_attn/Gather_1_output_0", "/blocks.29/self_attn/Unsqueeze_1_output_0", "/blocks.29/self_attn/Concat_output_0", "/blocks.29/self_attn/Reshape_output_0", "/blocks.29/self_attn/q_norm/Cast_output_0", "/blocks.29/self_attn/q_norm/Mul_1_output_0", "/blocks.29/self_attn/Mul_output_0", "/blocks.29/self_attn/Shape_2_output_0", "Shape4446", "Rank4448", "BinaryOp4450", "Unsqueeze4451", "BinaryOp4453", "Unsqueeze4454", "StridedSlice4457", "Squeeze4458", "BinaryOp4459", "BinaryOp4460", "/blocks.29/self_attn/Gather_2_output_0", "/blocks.29/self_attn/Div_output_0", "/blocks.29/self_attn/Unsqueeze_7_output_0", "/blocks.29/self_attn/Slice_1_output_0", "/blocks.29/self_attn/Neg_output_0", "/blocks.29/self_attn/Unsqueeze_6_output_0", "/blocks.29/self_attn/Slice_output_0", "/blocks.29/self_attn/Concat_3_output_0", "/blocks.29/self_attn/Mul_1_output_0", "/blocks.29/self_attn/Add_output_0", "/blocks.29/self_attn/k_proj/FakeLinear_output_0", "/blocks.29/self_attn/Unsqueeze_2_output_0", "/blocks.29/self_attn/Unsqueeze_3_output_0", "/blocks.29/self_attn/Concat_1_output_0", "/blocks.29/self_attn/Reshape_1_output_0", "/blocks.29/self_attn/k_norm/Cast_output_0", "/blocks.29/self_attn/k_norm/Mul_1_output_0", "/blocks.29/self_attn/Mul_2_output_0", "/blocks.29/self_attn/Shape_3_output_0", "Shape4488", "Rank4490", "BinaryOp4492", "Unsqueeze4493", "BinaryOp4495", "Unsqueeze4496", "StridedSlice4499", "Squeeze4500", "BinaryOp4501", "BinaryOp4502", "/blocks.29/self_attn/Gather_3_output_0", "/blocks.29/self_attn/Div_1_output_0", "/blocks.29/self_attn/Unsqueeze_9_output_0", "/blocks.29/self_attn/Slice_3_output_0", "/blocks.29/self_attn/Neg_1_output_0", "/blocks.29/self_attn/Unsqueeze_8_output_0", "/blocks.29/self_attn/Slice_2_output_0", "/blocks.29/self_attn/Concat_4_output_0", "/blocks.29/self_attn/Mul_3_output_0", "/blocks.29/self_attn/Add_1_output_0", "/blocks.29/self_attn/v_proj/FakeLinear_output_0", "/blocks.29/self_attn/Unsqueeze_4_output_0", "/blocks.29/self_attn/Unsqueeze_5_output_0", "/blocks.29/self_attn/Concat_2_output_0", "/blocks.29/self_attn/Reshape_2_output_0", "/blocks.29/self_attn/fused_attn/FusedAttention_output_0", "/blocks.29/self_attn/o_proj/FakeLinear_output_0", "/blocks.29/Add_output_0", "/blocks.29/post_attention_layernorm/Mul_1_output_0", "/blocks.29/mlp/gate_proj/FakeLinear_output_0", "/blocks.29/mlp/act_fn/Mul_output_0", "/blocks.29/mlp/up_proj/FakeLinear_output_0", "/blocks.29/mlp/Mul_output_0", "/blocks.29/mlp/down_proj/FakeLinear_output_0", "/blocks.29/Add_1_output_0", "/blocks.30/Reshape_output_0", "/blocks.30/input_layernorm/Mul_1_output_0", "/blocks.30/self_attn/q_proj/FakeLinear_output_0", "/blocks.30/self_attn/Shape_output_0", "Shape4549", "Rank4551", "BinaryOp4553", "Unsqueeze4554", "BinaryOp4556", "Unsqueeze4557", "StridedSlice4560", "Squeeze4561", "BinaryOp4563", "/blocks.30/self_attn/Gather_output_0", "/blocks.30/self_attn/Unsqueeze_output_0", "Unsqueeze4571", "Unsqueeze4574", "StridedSlice4577", "Squeeze4578", "BinaryOp4579", "BinaryOp4580", "/blocks.30/self_attn/Gather_1_output_0", "/blocks.30/self_attn/Unsqueeze_1_output_0", "/blocks.30/self_attn/Concat_output_0", "/blocks.30/self_attn/Reshape_output_0", "/blocks.30/self_attn/q_norm/Cast_output_0", "/blocks.30/self_attn/q_norm/Mul_1_output_0", "/blocks.30/self_attn/Mul_output_0", "/blocks.30/self_attn/Shape_2_output_0", "Shape4595", "Rank4597", "BinaryOp4599", "Unsqueeze4600", "BinaryOp4602", "Unsqueeze4603", "StridedSlice4606", "Squeeze4607", "BinaryOp4608", "BinaryOp4609", "/blocks.30/self_attn/Gather_2_output_0", "/blocks.30/self_attn/Div_output_0", "/blocks.30/self_attn/Unsqueeze_7_output_0", "/blocks.30/self_attn/Slice_1_output_0", "/blocks.30/self_attn/Neg_output_0", "/blocks.30/self_attn/Unsqueeze_6_output_0", "/blocks.30/self_attn/Slice_output_0", "/blocks.30/self_attn/Concat_3_output_0", "/blocks.30/self_attn/Mul_1_output_0", "/blocks.30/self_attn/Add_output_0", "/blocks.30/self_attn/k_proj/FakeLinear_output_0", "/blocks.30/self_attn/Unsqueeze_2_output_0", "/blocks.30/self_attn/Unsqueeze_3_output_0", "/blocks.30/self_attn/Concat_1_output_0", "/blocks.30/self_attn/Reshape_1_output_0", "/blocks.30/self_attn/k_norm/Cast_output_0", "/blocks.30/self_attn/k_norm/Mul_1_output_0", "/blocks.30/self_attn/Mul_2_output_0", "/blocks.30/self_attn/Shape_3_output_0", "Shape4637", "Rank4639", "BinaryOp4641", "Unsqueeze4642", "BinaryOp4644", "Unsqueeze4645", "StridedSlice4648", "Squeeze4649", "BinaryOp4650", "BinaryOp4651", "/blocks.30/self_attn/Gather_3_output_0", "/blocks.30/self_attn/Div_1_output_0", "/blocks.30/self_attn/Unsqueeze_9_output_0", "/blocks.30/self_attn/Slice_3_output_0", "/blocks.30/self_attn/Neg_1_output_0", "/blocks.30/self_attn/Unsqueeze_8_output_0", "/blocks.30/self_attn/Slice_2_output_0", "/blocks.30/self_attn/Concat_4_output_0", "/blocks.30/self_attn/Mul_3_output_0", "/blocks.30/self_attn/Add_1_output_0", "/blocks.30/self_attn/v_proj/FakeLinear_output_0", "/blocks.30/self_attn/Unsqueeze_4_output_0", "/blocks.30/self_attn/Unsqueeze_5_output_0", "/blocks.30/self_attn/Concat_2_output_0", "/blocks.30/self_attn/Reshape_2_output_0", "/blocks.30/self_attn/fused_attn/FusedAttention_output_0", "/blocks.30/self_attn/o_proj/FakeLinear_output_0", "/blocks.30/Add_output_0", "/blocks.30/post_attention_layernorm/Mul_1_output_0", "/blocks.30/mlp/gate_proj/FakeLinear_output_0", "/blocks.30/mlp/act_fn/Mul_output_0", "/blocks.30/mlp/up_proj/FakeLinear_output_0", "/blocks.30/mlp/Mul_output_0", "/blocks.30/mlp/down_proj/FakeLinear_output_0", "/blocks.30/Add_1_output_0", "/blocks.31/Reshape_output_0", "/blocks.31/input_layernorm/Mul_1_output_0", "/blocks.31/self_attn/q_proj/FakeLinear_output_0", "/blocks.31/self_attn/Shape_output_0", "Shape4698", "Rank4700", "BinaryOp4702", "Unsqueeze4703", "BinaryOp4705", "Unsqueeze4706", "StridedSlice4709", "Squeeze4710", "BinaryOp4712", "/blocks.31/self_attn/Gather_output_0", "/blocks.31/self_attn/Unsqueeze_output_0", "Unsqueeze4720", "Unsqueeze4723", "StridedSlice4726", "Squeeze4727", "BinaryOp4728", "BinaryOp4729", "/blocks.31/self_attn/Gather_1_output_0", "/blocks.31/self_attn/Unsqueeze_1_output_0", "/blocks.31/self_attn/Concat_output_0", "/blocks.31/self_attn/Reshape_output_0", "/blocks.31/self_attn/q_norm/Cast_output_0", "/blocks.31/self_attn/q_norm/Mul_1_output_0", "/blocks.31/self_attn/Mul_output_0", "/blocks.31/self_attn/Shape_2_output_0", "Shape4744", "Rank4746", "BinaryOp4748", "Unsqueeze4749", "BinaryOp4751", "Unsqueeze4752", "StridedSlice4755", "Squeeze4756", "BinaryOp4757", "BinaryOp4758", "/blocks.31/self_attn/Gather_2_output_0", "/blocks.31/self_attn/Div_output_0", "/blocks.31/self_attn/Unsqueeze_7_output_0", "/blocks.31/self_attn/Slice_1_output_0", "/blocks.31/self_attn/Neg_output_0", "/blocks.31/self_attn/Unsqueeze_6_output_0", "/blocks.31/self_attn/Slice_output_0", "/blocks.31/self_attn/Concat_3_output_0", "/blocks.31/self_attn/Mul_1_output_0", "/blocks.31/self_attn/Add_output_0", "/blocks.31/self_attn/k_proj/FakeLinear_output_0", "/blocks.31/self_attn/Unsqueeze_2_output_0", "/blocks.31/self_attn/Unsqueeze_3_output_0", "/blocks.31/self_attn/Concat_1_output_0", "/blocks.31/self_attn/Reshape_1_output_0", "/blocks.31/self_attn/k_norm/Cast_output_0", "/blocks.31/self_attn/k_norm/Mul_1_output_0", "/blocks.31/self_attn/Mul_2_output_0", "/blocks.31/self_attn/Shape_3_output_0", "Shape4786", "Rank4788", "BinaryOp4790", "Unsqueeze4791", "BinaryOp4793", "Unsqueeze4794", "StridedSlice4797", "Squeeze4798", "BinaryOp4799", "BinaryOp4800", "/blocks.31/self_attn/Gather_3_output_0", "/blocks.31/self_attn/Div_1_output_0", "/blocks.31/self_attn/Unsqueeze_9_output_0", "/blocks.31/self_attn/Slice_3_output_0", "/blocks.31/self_attn/Neg_1_output_0", "/blocks.31/self_attn/Unsqueeze_8_output_0", "/blocks.31/self_attn/Slice_2_output_0", "/blocks.31/self_attn/Concat_4_output_0", "/blocks.31/self_attn/Mul_3_output_0", "/blocks.31/self_attn/Add_1_output_0", "/blocks.31/self_attn/v_proj/FakeLinear_output_0", "/blocks.31/self_attn/Unsqueeze_4_output_0", "/blocks.31/self_attn/Unsqueeze_5_output_0", "/blocks.31/self_attn/Concat_2_output_0", "/blocks.31/self_attn/Reshape_2_output_0", "/blocks.31/self_attn/fused_attn/FusedAttention_output_0", "/blocks.31/self_attn/o_proj/FakeLinear_output_0", "/blocks.31/Add_output_0", "/blocks.31/post_attention_layernorm/Mul_1_output_0", "/blocks.31/mlp/gate_proj/FakeLinear_output_0", "/blocks.31/mlp/act_fn/Mul_output_0", "/blocks.31/mlp/up_proj/FakeLinear_output_0", "/blocks.31/mlp/Mul_output_0", "/blocks.31/mlp/down_proj/FakeLinear_output_0", "/blocks.31/Add_1_output_0", "/blocks.32/Reshape_output_0", "/blocks.32/input_layernorm/Mul_1_output_0", "/blocks.32/self_attn/q_proj/FakeLinear_output_0", "/blocks.32/self_attn/Shape_output_0", "Shape4847", "Rank4849", "BinaryOp4851", "Unsqueeze4852", "BinaryOp4854", "Unsqueeze4855", "StridedSlice4858", "Squeeze4859", "BinaryOp4861", "/blocks.32/self_attn/Gather_output_0", "/blocks.32/self_attn/Unsqueeze_output_0", "Unsqueeze4869", "Unsqueeze4872", "StridedSlice4875", "Squeeze4876", "BinaryOp4877", "BinaryOp4878", "/blocks.32/self_attn/Gather_1_output_0", "/blocks.32/self_attn/Unsqueeze_1_output_0", "/blocks.32/self_attn/Concat_output_0", "/blocks.32/self_attn/Reshape_output_0", "/blocks.32/self_attn/q_norm/Cast_output_0", "/blocks.32/self_attn/q_norm/Mul_1_output_0", "/blocks.32/self_attn/Mul_output_0", "/blocks.32/self_attn/Shape_2_output_0", "Shape4893", "Rank4895", "BinaryOp4897", "Unsqueeze4898", "BinaryOp4900", "Unsqueeze4901", "StridedSlice4904", "Squeeze4905", "BinaryOp4906", "BinaryOp4907", "/blocks.32/self_attn/Gather_2_output_0", "/blocks.32/self_attn/Div_output_0", "/blocks.32/self_attn/Unsqueeze_7_output_0", "/blocks.32/self_attn/Slice_1_output_0", "/blocks.32/self_attn/Neg_output_0", "/blocks.32/self_attn/Unsqueeze_6_output_0", "/blocks.32/self_attn/Slice_output_0", "/blocks.32/self_attn/Concat_3_output_0", "/blocks.32/self_attn/Mul_1_output_0", "/blocks.32/self_attn/Add_output_0", "/blocks.32/self_attn/k_proj/FakeLinear_output_0", "/blocks.32/self_attn/Unsqueeze_2_output_0", "/blocks.32/self_attn/Unsqueeze_3_output_0", "/blocks.32/self_attn/Concat_1_output_0", "/blocks.32/self_attn/Reshape_1_output_0", "/blocks.32/self_attn/k_norm/Cast_output_0", "/blocks.32/self_attn/k_norm/Mul_1_output_0", "/blocks.32/self_attn/Mul_2_output_0", "/blocks.32/self_attn/Shape_3_output_0", "Shape4935", "Rank4937", "BinaryOp4939", "Unsqueeze4940", "BinaryOp4942", "Unsqueeze4943", "StridedSlice4946", "Squeeze4947", "BinaryOp4948", "BinaryOp4949", "/blocks.32/self_attn/Gather_3_output_0", "/blocks.32/self_attn/Div_1_output_0", "/blocks.32/self_attn/Unsqueeze_9_output_0", "/blocks.32/self_attn/Slice_3_output_0", "/blocks.32/self_attn/Neg_1_output_0", "/blocks.32/self_attn/Unsqueeze_8_output_0", "/blocks.32/self_attn/Slice_2_output_0", "/blocks.32/self_attn/Concat_4_output_0", "/blocks.32/self_attn/Mul_3_output_0", "/blocks.32/self_attn/Add_1_output_0", "/blocks.32/self_attn/v_proj/FakeLinear_output_0", "/blocks.32/self_attn/Unsqueeze_4_output_0", "/blocks.32/self_attn/Unsqueeze_5_output_0", "/blocks.32/self_attn/Concat_2_output_0", "/blocks.32/self_attn/Reshape_2_output_0", "/blocks.32/self_attn/fused_attn/FusedAttention_output_0", "/blocks.32/self_attn/o_proj/FakeLinear_output_0", "/blocks.32/Add_output_0", "/blocks.32/post_attention_layernorm/Mul_1_output_0", "/blocks.32/mlp/gate_proj/FakeLinear_output_0", "/blocks.32/mlp/act_fn/Mul_output_0", "/blocks.32/mlp/up_proj/FakeLinear_output_0", "/blocks.32/mlp/Mul_output_0", "/blocks.32/mlp/down_proj/FakeLinear_output_0", "/blocks.32/Add_1_output_0", "/blocks.33/Reshape_output_0", "/blocks.33/input_layernorm/Mul_1_output_0", "/blocks.33/self_attn/q_proj/FakeLinear_output_0", "/blocks.33/self_attn/Shape_output_0", "Shape4996", "Rank4998", "BinaryOp5000", "Unsqueeze5001", "BinaryOp5003", "Unsqueeze5004", "StridedSlice5007", "Squeeze5008", "BinaryOp5010", "/blocks.33/self_attn/Gather_output_0", "/blocks.33/self_attn/Unsqueeze_output_0", "Unsqueeze5018", "Unsqueeze5021", "StridedSlice5024", "Squeeze5025", "BinaryOp5026", "BinaryOp5027", "/blocks.33/self_attn/Gather_1_output_0", "/blocks.33/self_attn/Unsqueeze_1_output_0", "/blocks.33/self_attn/Concat_output_0", "/blocks.33/self_attn/Reshape_output_0", "/blocks.33/self_attn/q_norm/Cast_output_0", "/blocks.33/self_attn/q_norm/Mul_1_output_0", "/blocks.33/self_attn/Mul_output_0", "/blocks.33/self_attn/Shape_2_output_0", "Shape5042", "Rank5044", "BinaryOp5046", "Unsqueeze5047", "BinaryOp5049", "Unsqueeze5050", "StridedSlice5053", "Squeeze5054", "BinaryOp5055", "BinaryOp5056", "/blocks.33/self_attn/Gather_2_output_0", "/blocks.33/self_attn/Div_output_0", "/blocks.33/self_attn/Unsqueeze_7_output_0", "/blocks.33/self_attn/Slice_1_output_0", "/blocks.33/self_attn/Neg_output_0", "/blocks.33/self_attn/Unsqueeze_6_output_0", "/blocks.33/self_attn/Slice_output_0", "/blocks.33/self_attn/Concat_3_output_0", "/blocks.33/self_attn/Mul_1_output_0", "/blocks.33/self_attn/Add_output_0", "/blocks.33/self_attn/k_proj/FakeLinear_output_0", "/blocks.33/self_attn/Unsqueeze_2_output_0", "/blocks.33/self_attn/Unsqueeze_3_output_0", "/blocks.33/self_attn/Concat_1_output_0", "/blocks.33/self_attn/Reshape_1_output_0", "/blocks.33/self_attn/k_norm/Cast_output_0", "/blocks.33/self_attn/k_norm/Mul_1_output_0", "/blocks.33/self_attn/Mul_2_output_0", "/blocks.33/self_attn/Shape_3_output_0", "Shape5084", "Rank5086", "BinaryOp5088", "Unsqueeze5089", "BinaryOp5091", "Unsqueeze5092", "StridedSlice5095", "Squeeze5096", "BinaryOp5097", "BinaryOp5098", "/blocks.33/self_attn/Gather_3_output_0", "/blocks.33/self_attn/Div_1_output_0", "/blocks.33/self_attn/Unsqueeze_9_output_0", "/blocks.33/self_attn/Slice_3_output_0", "/blocks.33/self_attn/Neg_1_output_0", "/blocks.33/self_attn/Unsqueeze_8_output_0", "/blocks.33/self_attn/Slice_2_output_0", "/blocks.33/self_attn/Concat_4_output_0", "/blocks.33/self_attn/Mul_3_output_0", "/blocks.33/self_attn/Add_1_output_0", "/blocks.33/self_attn/v_proj/FakeLinear_output_0", "/blocks.33/self_attn/Unsqueeze_4_output_0", "/blocks.33/self_attn/Unsqueeze_5_output_0", "/blocks.33/self_attn/Concat_2_output_0", "/blocks.33/self_attn/Reshape_2_output_0", "/blocks.33/self_attn/fused_attn/FusedAttention_output_0", "/blocks.33/self_attn/o_proj/FakeLinear_output_0", "/blocks.33/Add_output_0", "/blocks.33/post_attention_layernorm/Mul_1_output_0", "/blocks.33/mlp/gate_proj/FakeLinear_output_0", "/blocks.33/mlp/act_fn/Mul_output_0", "/blocks.33/mlp/up_proj/FakeLinear_output_0", "/blocks.33/mlp/Mul_output_0", "/blocks.33/mlp/down_proj/FakeLinear_output_0", "/blocks.33/Add_1_output_0", "/blocks.34/Reshape_output_0", "/blocks.34/input_layernorm/Mul_1_output_0", "/blocks.34/self_attn/q_proj/FakeLinear_output_0", "/blocks.34/self_attn/Shape_output_0", "Shape5145", "Rank5147", "BinaryOp5149", "Unsqueeze5150", "BinaryOp5152", "Unsqueeze5153", "StridedSlice5156", "Squeeze5157", "BinaryOp5159", "/blocks.34/self_attn/Gather_output_0", "/blocks.34/self_attn/Unsqueeze_output_0", "Unsqueeze5167", "Unsqueeze5170", "StridedSlice5173", "Squeeze5174", "BinaryOp5175", "BinaryOp5176", "/blocks.34/self_attn/Gather_1_output_0", "/blocks.34/self_attn/Unsqueeze_1_output_0", "/blocks.34/self_attn/Concat_output_0", "/blocks.34/self_attn/Reshape_output_0", "/blocks.34/self_attn/q_norm/Cast_output_0", "/blocks.34/self_attn/q_norm/Mul_1_output_0", "/blocks.34/self_attn/Mul_output_0", "/blocks.34/self_attn/Shape_2_output_0", "Shape5191", "Rank5193", "BinaryOp5195", "Unsqueeze5196", "BinaryOp5198", "Unsqueeze5199", "StridedSlice5202", "Squeeze5203", "BinaryOp5204", "BinaryOp5205", "/blocks.34/self_attn/Gather_2_output_0", "/blocks.34/self_attn/Div_output_0", "/blocks.34/self_attn/Unsqueeze_7_output_0", "/blocks.34/self_attn/Slice_1_output_0", "/blocks.34/self_attn/Neg_output_0", "/blocks.34/self_attn/Unsqueeze_6_output_0", "/blocks.34/self_attn/Slice_output_0", "/blocks.34/self_attn/Concat_3_output_0", "/blocks.34/self_attn/Mul_1_output_0", "/blocks.34/self_attn/Add_output_0", "/blocks.34/self_attn/k_proj/FakeLinear_output_0", "/blocks.34/self_attn/Unsqueeze_2_output_0", "/blocks.34/self_attn/Unsqueeze_3_output_0", "/blocks.34/self_attn/Concat_1_output_0", "/blocks.34/self_attn/Reshape_1_output_0", "/blocks.34/self_attn/k_norm/Cast_output_0", "/blocks.34/self_attn/k_norm/Mul_1_output_0", "/blocks.34/self_attn/Mul_2_output_0", "/blocks.34/self_attn/Shape_3_output_0", "Shape5233", "Rank5235", "BinaryOp5237", "Unsqueeze5238", "BinaryOp5240", "Unsqueeze5241", "StridedSlice5244", "Squeeze5245", "BinaryOp5246", "BinaryOp5247", "/blocks.34/self_attn/Gather_3_output_0", "/blocks.34/self_attn/Div_1_output_0", "/blocks.34/self_attn/Unsqueeze_9_output_0", "/blocks.34/self_attn/Slice_3_output_0", "/blocks.34/self_attn/Neg_1_output_0", "/blocks.34/self_attn/Unsqueeze_8_output_0", "/blocks.34/self_attn/Slice_2_output_0", "/blocks.34/self_attn/Concat_4_output_0", "/blocks.34/self_attn/Mul_3_output_0", "/blocks.34/self_attn/Add_1_output_0", "/blocks.34/self_attn/v_proj/FakeLinear_output_0", "/blocks.34/self_attn/Unsqueeze_4_output_0", "/blocks.34/self_attn/Unsqueeze_5_output_0", "/blocks.34/self_attn/Concat_2_output_0", "/blocks.34/self_attn/Reshape_2_output_0", "/blocks.34/self_attn/fused_attn/FusedAttention_output_0", "/blocks.34/self_attn/o_proj/FakeLinear_output_0", "/blocks.34/Add_output_0", "/blocks.34/post_attention_layernorm/Mul_1_output_0", "/blocks.34/mlp/gate_proj/FakeLinear_output_0", "/blocks.34/mlp/act_fn/Mul_output_0", "/blocks.34/mlp/up_proj/FakeLinear_output_0", "/blocks.34/mlp/Mul_output_0", "/blocks.34/mlp/down_proj/FakeLinear_output_0", "/blocks.34/Add_1_output_0", "/blocks.35/Reshape_output_0", "/blocks.35/input_layernorm/Mul_1_output_0", "/blocks.35/self_attn/q_proj/FakeLinear_output_0", "/blocks.35/self_attn/Shape_output_0", "Shape5294", "Rank5296", "BinaryOp5298", "Unsqueeze5299", "BinaryOp5301", "Unsqueeze5302", "StridedSlice5305", "Squeeze5306", "BinaryOp5308", "/blocks.35/self_attn/Gather_output_0", "/blocks.35/self_attn/Unsqueeze_output_0", "Unsqueeze5316", "Unsqueeze5319", "StridedSlice5322", "Squeeze5323", "BinaryOp5324", "BinaryOp5325", "/blocks.35/self_attn/Gather_1_output_0", "/blocks.35/self_attn/Unsqueeze_1_output_0", "/blocks.35/self_attn/Concat_output_0", "/blocks.35/self_attn/Reshape_output_0", "/blocks.35/self_attn/q_norm/Cast_output_0", "/blocks.35/self_attn/q_norm/Mul_1_output_0", "/blocks.35/self_attn/Mul_output_0", "/blocks.35/self_attn/Shape_2_output_0", "Shape5340", "Rank5342", "BinaryOp5344", "Unsqueeze5345", "BinaryOp5347", "Unsqueeze5348", "StridedSlice5351", "Squeeze5352", "BinaryOp5353", "BinaryOp5354", "/blocks.35/self_attn/Gather_2_output_0", "/blocks.35/self_attn/Div_output_0", "/blocks.35/self_attn/Unsqueeze_7_output_0", "/blocks.35/self_attn/Slice_1_output_0", "/blocks.35/self_attn/Neg_output_0", "/blocks.35/self_attn/Unsqueeze_6_output_0", "/blocks.35/self_attn/Slice_output_0", "/blocks.35/self_attn/Concat_3_output_0", "/blocks.35/self_attn/Mul_1_output_0", "/blocks.35/self_attn/Add_output_0", "/blocks.35/self_attn/k_proj/FakeLinear_output_0", "/blocks.35/self_attn/Unsqueeze_2_output_0", "/blocks.35/self_attn/Unsqueeze_3_output_0", "/blocks.35/self_attn/Concat_1_output_0", "/blocks.35/self_attn/Reshape_1_output_0", "/blocks.35/self_attn/k_norm/Cast_output_0", "/blocks.35/self_attn/k_norm/Mul_1_output_0", "/blocks.35/self_attn/Mul_2_output_0", "/blocks.35/self_attn/Shape_3_output_0", "Shape5382", "Rank5384", "BinaryOp5386", "Unsqueeze5387", "BinaryOp5389", "Unsqueeze5390", "StridedSlice5393", "Squeeze5394", "BinaryOp5395", "BinaryOp5396", "/blocks.35/self_attn/Gather_3_output_0", "/blocks.35/self_attn/Div_1_output_0", "/blocks.35/self_attn/Unsqueeze_9_output_0", "/blocks.35/self_attn/Slice_3_output_0", "/blocks.35/self_attn/Neg_1_output_0", "/blocks.35/self_attn/Unsqueeze_8_output_0", "/blocks.35/self_attn/Slice_2_output_0", "/blocks.35/self_attn/Concat_4_output_0", "/blocks.35/self_attn/Mul_3_output_0", "/blocks.35/self_attn/Add_1_output_0", "/blocks.35/self_attn/v_proj/FakeLinear_output_0", "/blocks.35/self_attn/Unsqueeze_4_output_0", "/blocks.35/self_attn/Unsqueeze_5_output_0", "/blocks.35/self_attn/Concat_2_output_0", "/blocks.35/self_attn/Reshape_2_output_0", "/blocks.35/self_attn/fused_attn/FusedAttention_output_0", "/blocks.35/self_attn/o_proj/FakeLinear_output_0", "/blocks.35/Add_output_0", "/blocks.35/post_attention_layernorm/Mul_1_output_0", "/blocks.35/mlp/gate_proj/FakeLinear_output_0", "/blocks.35/mlp/act_fn/Mul_output_0", "/blocks.35/mlp/up_proj/FakeLinear_output_0", "/blocks.35/mlp/Mul_output_0", "/blocks.35/mlp/down_proj/FakeLinear_output_0", "/blocks.35/Add_1_output_0", "/Slice_output_0", "hidden_states", "logits", "/layers.0/self_attn/q_proj/Linear/pre_reshape", "/layers.0/self_attn/q_proj/Linear/pre_convert", "/layers.0/self_attn/q_proj/Linear", "/layers.0/self_attn/q_proj/Linear/post_convert", "/layers.0/self_attn/k_proj/Linear/pre_reshape", "/layers.0/self_attn/k_proj/Linear/pre_convert", "/layers.0/self_attn/k_proj/Linear", "/layers.0/self_attn/k_proj/Linear/post_convert", "/layers.0/self_attn/v_proj/Linear/pre_reshape", "/layers.0/self_attn/v_proj/Linear/pre_convert", "/layers.0/self_attn/v_proj/Linear", "/layers.0/self_attn/v_proj/Linear/post_convert", "/layers.0/self_attn/o_proj/Linear/pre_reshape", "/layers.0/self_attn/o_proj/Linear/pre_convert", "/layers.0/self_attn/o_proj/Linear", "/layers.0/self_attn/o_proj/Linear/post_convert", "/layers.0/mlp/gate_proj/Linear/pre_reshape", "/layers.0/mlp/gate_proj/Linear/pre_convert", "/layers.0/mlp/gate_proj/Linear", "/layers.0/mlp/gate_proj/Linear/post_convert", "/layers.0/mlp/up_proj/Linear/pre_reshape", "/layers.0/mlp/up_proj/Linear/pre_convert", "/layers.0/mlp/up_proj/Linear", "/layers.0/mlp/up_proj/Linear/post_convert", "/layers.0/mlp/down_proj/Linear/pre_reshape", "/layers.0/mlp/down_proj/Linear/pre_convert", "/layers.0/mlp/down_proj/Linear", "/layers.0/mlp/down_proj/Linear/post_convert", "/layers.1/self_attn/q_proj/Linear/pre_reshape", "/layers.1/self_attn/q_proj/Linear/pre_convert", "/layers.1/self_attn/q_proj/Linear", "/layers.1/self_attn/q_proj/Linear/post_convert", "/layers.1/self_attn/k_proj/Linear/pre_reshape", "/layers.1/self_attn/k_proj/Linear/pre_convert", "/layers.1/self_attn/k_proj/Linear", "/layers.1/self_attn/k_proj/Linear/post_convert", "/layers.1/self_attn/v_proj/Linear/pre_reshape", "/layers.1/self_attn/v_proj/Linear/pre_convert", "/layers.1/self_attn/v_proj/Linear", "/layers.1/self_attn/v_proj/Linear/post_convert", "/layers.1/self_attn/o_proj/Linear/pre_reshape", "/layers.1/self_attn/o_proj/Linear/pre_convert", "/layers.1/self_attn/o_proj/Linear", "/layers.1/self_attn/o_proj/Linear/post_convert", "/layers.1/mlp/gate_proj/Linear/pre_reshape", "/layers.1/mlp/gate_proj/Linear/pre_convert", "/layers.1/mlp/gate_proj/Linear", "/layers.1/mlp/gate_proj/Linear/post_convert", "/layers.1/mlp/up_proj/Linear/pre_reshape", "/layers.1/mlp/up_proj/Linear/pre_convert", "/layers.1/mlp/up_proj/Linear", "/layers.1/mlp/up_proj/Linear/post_convert", "/layers.1/mlp/down_proj/Linear/pre_reshape", "/layers.1/mlp/down_proj/Linear/pre_convert", "/layers.1/mlp/down_proj/Linear", "/layers.1/mlp/down_proj/Linear/post_convert", "/layers.2/self_attn/q_proj/Linear/pre_reshape", "/layers.2/self_attn/q_proj/Linear/pre_convert", "/layers.2/self_attn/q_proj/Linear", "/layers.2/self_attn/q_proj/Linear/post_convert", "/layers.2/self_attn/k_proj/Linear/pre_reshape", "/layers.2/self_attn/k_proj/Linear/pre_convert", "/layers.2/self_attn/k_proj/Linear", "/layers.2/self_attn/k_proj/Linear/post_convert", "/layers.2/self_attn/v_proj/Linear/pre_reshape", "/layers.2/self_attn/v_proj/Linear/pre_convert", "/layers.2/self_attn/v_proj/Linear", "/layers.2/self_attn/v_proj/Linear/post_convert", "/layers.2/self_attn/o_proj/Linear/pre_reshape", "/layers.2/self_attn/o_proj/Linear/pre_convert", "/layers.2/self_attn/o_proj/Linear", "/layers.2/self_attn/o_proj/Linear/post_convert", "/layers.2/mlp/gate_proj/Linear/pre_reshape", "/layers.2/mlp/gate_proj/Linear/pre_convert", "/layers.2/mlp/gate_proj/Linear", "/layers.2/mlp/gate_proj/Linear/post_convert", "/layers.2/mlp/up_proj/Linear/pre_reshape", "/layers.2/mlp/up_proj/Linear/pre_convert", "/layers.2/mlp/up_proj/Linear", "/layers.2/mlp/up_proj/Linear/post_convert", "/layers.2/mlp/down_proj/Linear/pre_reshape", "/layers.2/mlp/down_proj/Linear/pre_convert", "/layers.2/mlp/down_proj/Linear", "/layers.2/mlp/down_proj/Linear/post_convert", "/layers.3/self_attn/q_proj/Linear/pre_reshape", "/layers.3/self_attn/q_proj/Linear/pre_convert", "/layers.3/self_attn/q_proj/Linear", "/layers.3/self_attn/q_proj/Linear/post_convert", "/layers.3/self_attn/k_proj/Linear/pre_reshape", "/layers.3/self_attn/k_proj/Linear/pre_convert", "/layers.3/self_attn/k_proj/Linear", "/layers.3/self_attn/k_proj/Linear/post_convert", "/layers.3/self_attn/v_proj/Linear/pre_reshape", "/layers.3/self_attn/v_proj/Linear/pre_convert", "/layers.3/self_attn/v_proj/Linear", "/layers.3/self_attn/v_proj/Linear/post_convert", "/layers.3/self_attn/o_proj/Linear/pre_reshape", "/layers.3/self_attn/o_proj/Linear/pre_convert", "/layers.3/self_attn/o_proj/Linear", "/layers.3/self_attn/o_proj/Linear/post_convert", "/layers.3/mlp/gate_proj/Linear/pre_reshape", "/layers.3/mlp/gate_proj/Linear/pre_convert", "/layers.3/mlp/gate_proj/Linear", "/layers.3/mlp/gate_proj/Linear/post_convert", "/layers.3/mlp/up_proj/Linear/pre_reshape", "/layers.3/mlp/up_proj/Linear/pre_convert", "/layers.3/mlp/up_proj/Linear", "/layers.3/mlp/up_proj/Linear/post_convert", "/layers.3/mlp/down_proj/Linear/pre_reshape", "/layers.3/mlp/down_proj/Linear/pre_convert", "/layers.3/mlp/down_proj/Linear", "/layers.3/mlp/down_proj/Linear/post_convert", "/layers.4/self_attn/q_proj/Linear/pre_reshape", "/layers.4/self_attn/q_proj/Linear/pre_convert", "/layers.4/self_attn/q_proj/Linear", "/layers.4/self_attn/q_proj/Linear/post_convert", "/layers.4/self_attn/k_proj/Linear/pre_reshape", "/layers.4/self_attn/k_proj/Linear/pre_convert", "/layers.4/self_attn/k_proj/Linear", "/layers.4/self_attn/k_proj/Linear/post_convert", "/layers.4/self_attn/v_proj/Linear/pre_reshape", "/layers.4/self_attn/v_proj/Linear/pre_convert", "/layers.4/self_attn/v_proj/Linear", "/layers.4/self_attn/v_proj/Linear/post_convert", "/layers.4/self_attn/o_proj/Linear/pre_reshape", "/layers.4/self_attn/o_proj/Linear/pre_convert", "/layers.4/self_attn/o_proj/Linear", "/layers.4/self_attn/o_proj/Linear/post_convert", "/layers.4/mlp/gate_proj/Linear/pre_reshape", "/layers.4/mlp/gate_proj/Linear/pre_convert", "/layers.4/mlp/gate_proj/Linear", "/layers.4/mlp/gate_proj/Linear/post_convert", "/layers.4/mlp/up_proj/Linear/pre_reshape", "/layers.4/mlp/up_proj/Linear/pre_convert", "/layers.4/mlp/up_proj/Linear", "/layers.4/mlp/up_proj/Linear/post_convert", "/layers.4/mlp/down_proj/Linear/pre_reshape", "/layers.4/mlp/down_proj/Linear/pre_convert", "/layers.4/mlp/down_proj/Linear", "/layers.4/mlp/down_proj/Linear/post_convert", "/layers.5/self_attn/q_proj/Linear/pre_reshape", "/layers.5/self_attn/q_proj/Linear/pre_convert", "/layers.5/self_attn/q_proj/Linear", "/layers.5/self_attn/q_proj/Linear/post_convert", "/layers.5/self_attn/k_proj/Linear/pre_reshape", "/layers.5/self_attn/k_proj/Linear/pre_convert", "/layers.5/self_attn/k_proj/Linear", "/layers.5/self_attn/k_proj/Linear/post_convert", "/layers.5/self_attn/v_proj/Linear/pre_reshape", "/layers.5/self_attn/v_proj/Linear/pre_convert", "/layers.5/self_attn/v_proj/Linear", "/layers.5/self_attn/v_proj/Linear/post_convert", "/layers.5/self_attn/o_proj/Linear/pre_reshape", "/layers.5/self_attn/o_proj/Linear/pre_convert", "/layers.5/self_attn/o_proj/Linear", "/layers.5/self_attn/o_proj/Linear/post_convert", "/layers.5/mlp/gate_proj/Linear/pre_reshape", "/layers.5/mlp/gate_proj/Linear/pre_convert", "/layers.5/mlp/gate_proj/Linear", "/layers.5/mlp/gate_proj/Linear/post_convert", "/layers.5/mlp/up_proj/Linear/pre_reshape", "/layers.5/mlp/up_proj/Linear/pre_convert", "/layers.5/mlp/up_proj/Linear", "/layers.5/mlp/up_proj/Linear/post_convert", "/layers.5/mlp/down_proj/Linear/pre_reshape", "/layers.5/mlp/down_proj/Linear/pre_convert", "/layers.5/mlp/down_proj/Linear", "/layers.5/mlp/down_proj/Linear/post_convert", "/layers.6/self_attn/q_proj/Linear/pre_reshape", "/layers.6/self_attn/q_proj/Linear/pre_convert", "/layers.6/self_attn/q_proj/Linear", "/layers.6/self_attn/q_proj/Linear/post_convert", "/layers.6/self_attn/k_proj/Linear/pre_reshape", "/layers.6/self_attn/k_proj/Linear/pre_convert", "/layers.6/self_attn/k_proj/Linear", "/layers.6/self_attn/k_proj/Linear/post_convert", "/layers.6/self_attn/v_proj/Linear/pre_reshape", "/layers.6/self_attn/v_proj/Linear/pre_convert", "/layers.6/self_attn/v_proj/Linear", "/layers.6/self_attn/v_proj/Linear/post_convert", "/layers.6/self_attn/o_proj/Linear/pre_reshape", "/layers.6/self_attn/o_proj/Linear/pre_convert", "/layers.6/self_attn/o_proj/Linear", "/layers.6/self_attn/o_proj/Linear/post_convert", "/layers.6/mlp/gate_proj/Linear/pre_reshape", "/layers.6/mlp/gate_proj/Linear/pre_convert", "/layers.6/mlp/gate_proj/Linear", "/layers.6/mlp/gate_proj/Linear/post_convert", "/layers.6/mlp/up_proj/Linear/pre_reshape", "/layers.6/mlp/up_proj/Linear/pre_convert", "/layers.6/mlp/up_proj/Linear", "/layers.6/mlp/up_proj/Linear/post_convert", "/layers.6/mlp/down_proj/Linear/pre_reshape", "/layers.6/mlp/down_proj/Linear/pre_convert", "/layers.6/mlp/down_proj/Linear", "/layers.6/mlp/down_proj/Linear/post_convert", "/layers.7/self_attn/q_proj/Linear/pre_reshape", "/layers.7/self_attn/q_proj/Linear/pre_convert", "/layers.7/self_attn/q_proj/Linear", "/layers.7/self_attn/q_proj/Linear/post_convert", "/layers.7/self_attn/k_proj/Linear/pre_reshape", "/layers.7/self_attn/k_proj/Linear/pre_convert", "/layers.7/self_attn/k_proj/Linear", "/layers.7/self_attn/k_proj/Linear/post_convert", "/layers.7/self_attn/v_proj/Linear/pre_reshape", "/layers.7/self_attn/v_proj/Linear/pre_convert", "/layers.7/self_attn/v_proj/Linear", "/layers.7/self_attn/v_proj/Linear/post_convert", "/layers.7/self_attn/o_proj/Linear/pre_reshape", "/layers.7/self_attn/o_proj/Linear/pre_convert", "/layers.7/self_attn/o_proj/Linear", "/layers.7/self_attn/o_proj/Linear/post_convert", "/layers.7/mlp/gate_proj/Linear/pre_reshape", "/layers.7/mlp/gate_proj/Linear/pre_convert", "/layers.7/mlp/gate_proj/Linear", "/layers.7/mlp/gate_proj/Linear/post_convert", "/layers.7/mlp/up_proj/Linear/pre_reshape", "/layers.7/mlp/up_proj/Linear/pre_convert", "/layers.7/mlp/up_proj/Linear", "/layers.7/mlp/up_proj/Linear/post_convert", "/layers.7/mlp/down_proj/Linear/pre_reshape", "/layers.7/mlp/down_proj/Linear/pre_convert", "/layers.7/mlp/down_proj/Linear", "/layers.7/mlp/down_proj/Linear/post_convert", "/layers.8/self_attn/q_proj/Linear/pre_reshape", "/layers.8/self_attn/q_proj/Linear/pre_convert", "/layers.8/self_attn/q_proj/Linear", "/layers.8/self_attn/q_proj/Linear/post_convert", "/layers.8/self_attn/k_proj/Linear/pre_reshape", "/layers.8/self_attn/k_proj/Linear/pre_convert", "/layers.8/self_attn/k_proj/Linear", "/layers.8/self_attn/k_proj/Linear/post_convert", "/layers.8/self_attn/v_proj/Linear/pre_reshape", "/layers.8/self_attn/v_proj/Linear/pre_convert", "/layers.8/self_attn/v_proj/Linear", "/layers.8/self_attn/v_proj/Linear/post_convert", "/layers.8/self_attn/o_proj/Linear/pre_reshape", "/layers.8/self_attn/o_proj/Linear/pre_convert", "/layers.8/self_attn/o_proj/Linear", "/layers.8/self_attn/o_proj/Linear/post_convert", "/layers.8/mlp/gate_proj/Linear/pre_reshape", "/layers.8/mlp/gate_proj/Linear/pre_convert", "/layers.8/mlp/gate_proj/Linear", "/layers.8/mlp/gate_proj/Linear/post_convert", "/layers.8/mlp/up_proj/Linear/pre_reshape", "/layers.8/mlp/up_proj/Linear/pre_convert", "/layers.8/mlp/up_proj/Linear", "/layers.8/mlp/up_proj/Linear/post_convert", "/layers.8/mlp/down_proj/Linear/pre_reshape", "/layers.8/mlp/down_proj/Linear/pre_convert", "/layers.8/mlp/down_proj/Linear", "/layers.8/mlp/down_proj/Linear/post_convert", "/layers.9/self_attn/q_proj/Linear/pre_reshape", "/layers.9/self_attn/q_proj/Linear/pre_convert", "/layers.9/self_attn/q_proj/Linear", "/layers.9/self_attn/q_proj/Linear/post_convert", "/layers.9/self_attn/k_proj/Linear/pre_reshape", "/layers.9/self_attn/k_proj/Linear/pre_convert", "/layers.9/self_attn/k_proj/Linear", "/layers.9/self_attn/k_proj/Linear/post_convert", "/layers.9/self_attn/v_proj/Linear/pre_reshape", "/layers.9/self_attn/v_proj/Linear/pre_convert", "/layers.9/self_attn/v_proj/Linear", "/layers.9/self_attn/v_proj/Linear/post_convert", "/layers.9/self_attn/o_proj/Linear/pre_reshape", "/layers.9/self_attn/o_proj/Linear/pre_convert", "/layers.9/self_attn/o_proj/Linear", "/layers.9/self_attn/o_proj/Linear/post_convert", "/layers.9/mlp/gate_proj/Linear/pre_reshape", "/layers.9/mlp/gate_proj/Linear/pre_convert", "/layers.9/mlp/gate_proj/Linear", "/layers.9/mlp/gate_proj/Linear/post_convert", "/layers.9/mlp/up_proj/Linear/pre_reshape", "/layers.9/mlp/up_proj/Linear/pre_convert", "/layers.9/mlp/up_proj/Linear", "/layers.9/mlp/up_proj/Linear/post_convert", "/layers.9/mlp/down_proj/Linear/pre_reshape", "/layers.9/mlp/down_proj/Linear/pre_convert", "/layers.9/mlp/down_proj/Linear", "/layers.9/mlp/down_proj/Linear/post_convert", "/layers.10/self_attn/q_proj/Linear/pre_reshape", "/layers.10/self_attn/q_proj/Linear/pre_convert", "/layers.10/self_attn/q_proj/Linear", "/layers.10/self_attn/q_proj/Linear/post_convert", "/layers.10/self_attn/k_proj/Linear/pre_reshape", "/layers.10/self_attn/k_proj/Linear/pre_convert", "/layers.10/self_attn/k_proj/Linear", "/layers.10/self_attn/k_proj/Linear/post_convert", "/layers.10/self_attn/v_proj/Linear/pre_reshape", "/layers.10/self_attn/v_proj/Linear/pre_convert", "/layers.10/self_attn/v_proj/Linear", "/layers.10/self_attn/v_proj/Linear/post_convert", "/layers.10/self_attn/o_proj/Linear/pre_reshape", "/layers.10/self_attn/o_proj/Linear/pre_convert", "/layers.10/self_attn/o_proj/Linear", "/layers.10/self_attn/o_proj/Linear/post_convert", "/layers.10/mlp/gate_proj/Linear/pre_reshape", "/layers.10/mlp/gate_proj/Linear/pre_convert", "/layers.10/mlp/gate_proj/Linear", "/layers.10/mlp/gate_proj/Linear/post_convert", "/layers.10/mlp/up_proj/Linear/pre_reshape", "/layers.10/mlp/up_proj/Linear/pre_convert", "/layers.10/mlp/up_proj/Linear", "/layers.10/mlp/up_proj/Linear/post_convert", "/layers.10/mlp/down_proj/Linear/pre_reshape", "/layers.10/mlp/down_proj/Linear/pre_convert", "/layers.10/mlp/down_proj/Linear", "/layers.10/mlp/down_proj/Linear/post_convert", "/layers.11/self_attn/q_proj/Linear/pre_reshape", "/layers.11/self_attn/q_proj/Linear/pre_convert", "/layers.11/self_attn/q_proj/Linear", "/layers.11/self_attn/q_proj/Linear/post_convert", "/layers.11/self_attn/k_proj/Linear/pre_reshape", "/layers.11/self_attn/k_proj/Linear/pre_convert", "/layers.11/self_attn/k_proj/Linear", "/layers.11/self_attn/k_proj/Linear/post_convert", "/layers.11/self_attn/v_proj/Linear/pre_reshape", "/layers.11/self_attn/v_proj/Linear/pre_convert", "/layers.11/self_attn/v_proj/Linear", "/layers.11/self_attn/v_proj/Linear/post_convert", "/layers.11/self_attn/o_proj/Linear/pre_reshape", "/layers.11/self_attn/o_proj/Linear/pre_convert", "/layers.11/self_attn/o_proj/Linear", "/layers.11/self_attn/o_proj/Linear/post_convert", "/layers.11/mlp/gate_proj/Linear/pre_reshape", "/layers.11/mlp/gate_proj/Linear/pre_convert", "/layers.11/mlp/gate_proj/Linear", "/layers.11/mlp/gate_proj/Linear/post_convert", "/layers.11/mlp/up_proj/Linear/pre_reshape", "/layers.11/mlp/up_proj/Linear/pre_convert", "/layers.11/mlp/up_proj/Linear", "/layers.11/mlp/up_proj/Linear/post_convert", "/layers.11/mlp/down_proj/Linear/pre_reshape", "/layers.11/mlp/down_proj/Linear/pre_convert", "/layers.11/mlp/down_proj/Linear", "/layers.11/mlp/down_proj/Linear/post_convert", "/layers.12/self_attn/q_proj/Linear/pre_reshape", "/layers.12/self_attn/q_proj/Linear/pre_convert", "/layers.12/self_attn/q_proj/Linear", "/layers.12/self_attn/q_proj/Linear/post_convert", "/layers.12/self_attn/k_proj/Linear/pre_reshape", "/layers.12/self_attn/k_proj/Linear/pre_convert", "/layers.12/self_attn/k_proj/Linear", "/layers.12/self_attn/k_proj/Linear/post_convert", "/layers.12/self_attn/v_proj/Linear/pre_reshape", "/layers.12/self_attn/v_proj/Linear/pre_convert", "/layers.12/self_attn/v_proj/Linear", "/layers.12/self_attn/v_proj/Linear/post_convert", "/layers.12/self_attn/o_proj/Linear/pre_reshape", "/layers.12/self_attn/o_proj/Linear/pre_convert", "/layers.12/self_attn/o_proj/Linear", "/layers.12/self_attn/o_proj/Linear/post_convert", "/layers.12/mlp/gate_proj/Linear/pre_reshape", "/layers.12/mlp/gate_proj/Linear/pre_convert", "/layers.12/mlp/gate_proj/Linear", "/layers.12/mlp/gate_proj/Linear/post_convert", "/layers.12/mlp/up_proj/Linear/pre_reshape", "/layers.12/mlp/up_proj/Linear/pre_convert", "/layers.12/mlp/up_proj/Linear", "/layers.12/mlp/up_proj/Linear/post_convert", "/layers.12/mlp/down_proj/Linear/pre_reshape", "/layers.12/mlp/down_proj/Linear/pre_convert", "/layers.12/mlp/down_proj/Linear", "/layers.12/mlp/down_proj/Linear/post_convert", "/layers.13/self_attn/q_proj/Linear/pre_reshape", "/layers.13/self_attn/q_proj/Linear/pre_convert", "/layers.13/self_attn/q_proj/Linear", "/layers.13/self_attn/q_proj/Linear/post_convert", "/layers.13/self_attn/k_proj/Linear/pre_reshape", "/layers.13/self_attn/k_proj/Linear/pre_convert", "/layers.13/self_attn/k_proj/Linear", "/layers.13/self_attn/k_proj/Linear/post_convert", "/layers.13/self_attn/v_proj/Linear/pre_reshape", "/layers.13/self_attn/v_proj/Linear/pre_convert", "/layers.13/self_attn/v_proj/Linear", "/layers.13/self_attn/v_proj/Linear/post_convert", "/layers.13/self_attn/o_proj/Linear/pre_reshape", "/layers.13/self_attn/o_proj/Linear/pre_convert", "/layers.13/self_attn/o_proj/Linear", "/layers.13/self_attn/o_proj/Linear/post_convert", "/layers.13/mlp/gate_proj/Linear/pre_reshape", "/layers.13/mlp/gate_proj/Linear/pre_convert", "/layers.13/mlp/gate_proj/Linear", "/layers.13/mlp/gate_proj/Linear/post_convert", "/layers.13/mlp/up_proj/Linear/pre_reshape", "/layers.13/mlp/up_proj/Linear/pre_convert", "/layers.13/mlp/up_proj/Linear", "/layers.13/mlp/up_proj/Linear/post_convert", "/layers.13/mlp/down_proj/Linear/pre_reshape", "/layers.13/mlp/down_proj/Linear/pre_convert", "/layers.13/mlp/down_proj/Linear", "/layers.13/mlp/down_proj/Linear/post_convert", "/layers.14/self_attn/q_proj/Linear/pre_reshape", "/layers.14/self_attn/q_proj/Linear/pre_convert", "/layers.14/self_attn/q_proj/Linear", "/layers.14/self_attn/q_proj/Linear/post_convert", "/layers.14/self_attn/k_proj/Linear/pre_reshape", "/layers.14/self_attn/k_proj/Linear/pre_convert", "/layers.14/self_attn/k_proj/Linear", "/layers.14/self_attn/k_proj/Linear/post_convert", "/layers.14/self_attn/v_proj/Linear/pre_reshape", "/layers.14/self_attn/v_proj/Linear/pre_convert", "/layers.14/self_attn/v_proj/Linear", "/layers.14/self_attn/v_proj/Linear/post_convert", "/layers.14/self_attn/o_proj/Linear/pre_reshape", "/layers.14/self_attn/o_proj/Linear/pre_convert", "/layers.14/self_attn/o_proj/Linear", "/layers.14/self_attn/o_proj/Linear/post_convert", "/layers.14/mlp/gate_proj/Linear/pre_reshape", "/layers.14/mlp/gate_proj/Linear/pre_convert", "/layers.14/mlp/gate_proj/Linear", "/layers.14/mlp/gate_proj/Linear/post_convert", "/layers.14/mlp/up_proj/Linear/pre_reshape", "/layers.14/mlp/up_proj/Linear/pre_convert", "/layers.14/mlp/up_proj/Linear", "/layers.14/mlp/up_proj/Linear/post_convert", "/layers.14/mlp/down_proj/Linear/pre_reshape", "/layers.14/mlp/down_proj/Linear/pre_convert", "/layers.14/mlp/down_proj/Linear", "/layers.14/mlp/down_proj/Linear/post_convert", "/layers.15/self_attn/q_proj/Linear/pre_reshape", "/layers.15/self_attn/q_proj/Linear/pre_convert", "/layers.15/self_attn/q_proj/Linear", "/layers.15/self_attn/q_proj/Linear/post_convert", "/layers.15/self_attn/k_proj/Linear/pre_reshape", "/layers.15/self_attn/k_proj/Linear/pre_convert", "/layers.15/self_attn/k_proj/Linear", "/layers.15/self_attn/k_proj/Linear/post_convert", "/layers.15/self_attn/v_proj/Linear/pre_reshape", "/layers.15/self_attn/v_proj/Linear/pre_convert", "/layers.15/self_attn/v_proj/Linear", "/layers.15/self_attn/v_proj/Linear/post_convert", "/layers.15/self_attn/o_proj/Linear/pre_reshape", "/layers.15/self_attn/o_proj/Linear/pre_convert", "/layers.15/self_attn/o_proj/Linear", "/layers.15/self_attn/o_proj/Linear/post_convert", "/layers.15/mlp/gate_proj/Linear/pre_reshape", "/layers.15/mlp/gate_proj/Linear/pre_convert", "/layers.15/mlp/gate_proj/Linear", "/layers.15/mlp/gate_proj/Linear/post_convert", "/layers.15/mlp/up_proj/Linear/pre_reshape", "/layers.15/mlp/up_proj/Linear/pre_convert", "/layers.15/mlp/up_proj/Linear", "/layers.15/mlp/up_proj/Linear/post_convert", "/layers.15/mlp/down_proj/Linear/pre_reshape", "/layers.15/mlp/down_proj/Linear/pre_convert", "/layers.15/mlp/down_proj/Linear", "/layers.15/mlp/down_proj/Linear/post_convert", "/layers.16/self_attn/q_proj/Linear/pre_reshape", "/layers.16/self_attn/q_proj/Linear/pre_convert", "/layers.16/self_attn/q_proj/Linear", "/layers.16/self_attn/q_proj/Linear/post_convert", "/layers.16/self_attn/k_proj/Linear/pre_reshape", "/layers.16/self_attn/k_proj/Linear/pre_convert", "/layers.16/self_attn/k_proj/Linear", "/layers.16/self_attn/k_proj/Linear/post_convert", "/layers.16/self_attn/v_proj/Linear/pre_reshape", "/layers.16/self_attn/v_proj/Linear/pre_convert", "/layers.16/self_attn/v_proj/Linear", "/layers.16/self_attn/v_proj/Linear/post_convert", "/layers.16/self_attn/o_proj/Linear/pre_reshape", "/layers.16/self_attn/o_proj/Linear/pre_convert", "/layers.16/self_attn/o_proj/Linear", "/layers.16/self_attn/o_proj/Linear/post_convert", "/layers.16/mlp/gate_proj/Linear/pre_reshape", "/layers.16/mlp/gate_proj/Linear/pre_convert", "/layers.16/mlp/gate_proj/Linear", "/layers.16/mlp/gate_proj/Linear/post_convert", "/layers.16/mlp/up_proj/Linear/pre_reshape", "/layers.16/mlp/up_proj/Linear/pre_convert", "/layers.16/mlp/up_proj/Linear", "/layers.16/mlp/up_proj/Linear/post_convert", "/layers.16/mlp/down_proj/Linear/pre_reshape", "/layers.16/mlp/down_proj/Linear/pre_convert", "/layers.16/mlp/down_proj/Linear", "/layers.16/mlp/down_proj/Linear/post_convert", "/layers.17/self_attn/q_proj/Linear/pre_reshape", "/layers.17/self_attn/q_proj/Linear/pre_convert", "/layers.17/self_attn/q_proj/Linear", "/layers.17/self_attn/q_proj/Linear/post_convert", "/layers.17/self_attn/k_proj/Linear/pre_reshape", "/layers.17/self_attn/k_proj/Linear/pre_convert", "/layers.17/self_attn/k_proj/Linear", "/layers.17/self_attn/k_proj/Linear/post_convert", "/layers.17/self_attn/v_proj/Linear/pre_reshape", "/layers.17/self_attn/v_proj/Linear/pre_convert", "/layers.17/self_attn/v_proj/Linear", "/layers.17/self_attn/v_proj/Linear/post_convert", "/layers.17/self_attn/o_proj/Linear/pre_reshape", "/layers.17/self_attn/o_proj/Linear/pre_convert", "/layers.17/self_attn/o_proj/Linear", "/layers.17/self_attn/o_proj/Linear/post_convert", "/layers.17/mlp/gate_proj/Linear/pre_reshape", "/layers.17/mlp/gate_proj/Linear/pre_convert", "/layers.17/mlp/gate_proj/Linear", "/layers.17/mlp/gate_proj/Linear/post_convert", "/layers.17/mlp/up_proj/Linear/pre_reshape", "/layers.17/mlp/up_proj/Linear/pre_convert", "/layers.17/mlp/up_proj/Linear", "/layers.17/mlp/up_proj/Linear/post_convert", "/layers.17/mlp/down_proj/Linear/pre_reshape", "/layers.17/mlp/down_proj/Linear/pre_convert", "/layers.17/mlp/down_proj/Linear", "/layers.17/mlp/down_proj/Linear/post_convert", "/layers.18/self_attn/q_proj/Linear/pre_reshape", "/layers.18/self_attn/q_proj/Linear/pre_convert", "/layers.18/self_attn/q_proj/Linear", "/layers.18/self_attn/q_proj/Linear/post_convert", "/layers.18/self_attn/k_proj/Linear/pre_reshape", "/layers.18/self_attn/k_proj/Linear/pre_convert", "/layers.18/self_attn/k_proj/Linear", "/layers.18/self_attn/k_proj/Linear/post_convert", "/layers.18/self_attn/v_proj/Linear/pre_reshape", "/layers.18/self_attn/v_proj/Linear/pre_convert", "/layers.18/self_attn/v_proj/Linear", "/layers.18/self_attn/v_proj/Linear/post_convert", "/layers.18/self_attn/o_proj/Linear/pre_reshape", "/layers.18/self_attn/o_proj/Linear/pre_convert", "/layers.18/self_attn/o_proj/Linear", "/layers.18/self_attn/o_proj/Linear/post_convert", "/layers.18/mlp/gate_proj/Linear/pre_reshape", "/layers.18/mlp/gate_proj/Linear/pre_convert", "/layers.18/mlp/gate_proj/Linear", "/layers.18/mlp/gate_proj/Linear/post_convert", "/layers.18/mlp/up_proj/Linear/pre_reshape", "/layers.18/mlp/up_proj/Linear/pre_convert", "/layers.18/mlp/up_proj/Linear", "/layers.18/mlp/up_proj/Linear/post_convert", "/layers.18/mlp/down_proj/Linear/pre_reshape", "/layers.18/mlp/down_proj/Linear/pre_convert", "/layers.18/mlp/down_proj/Linear", "/layers.18/mlp/down_proj/Linear/post_convert", "/layers.19/self_attn/q_proj/Linear/pre_reshape", "/layers.19/self_attn/q_proj/Linear/pre_convert", "/layers.19/self_attn/q_proj/Linear", "/layers.19/self_attn/q_proj/Linear/post_convert", "/layers.19/self_attn/k_proj/Linear/pre_reshape", "/layers.19/self_attn/k_proj/Linear/pre_convert", "/layers.19/self_attn/k_proj/Linear", "/layers.19/self_attn/k_proj/Linear/post_convert", "/layers.19/self_attn/v_proj/Linear/pre_reshape", "/layers.19/self_attn/v_proj/Linear/pre_convert", "/layers.19/self_attn/v_proj/Linear", "/layers.19/self_attn/v_proj/Linear/post_convert", "/layers.19/self_attn/o_proj/Linear/pre_reshape", "/layers.19/self_attn/o_proj/Linear/pre_convert", "/layers.19/self_attn/o_proj/Linear", "/layers.19/self_attn/o_proj/Linear/post_convert", "/layers.19/mlp/gate_proj/Linear/pre_reshape", "/layers.19/mlp/gate_proj/Linear/pre_convert", "/layers.19/mlp/gate_proj/Linear", "/layers.19/mlp/gate_proj/Linear/post_convert", "/layers.19/mlp/up_proj/Linear/pre_reshape", "/layers.19/mlp/up_proj/Linear/pre_convert", "/layers.19/mlp/up_proj/Linear", "/layers.19/mlp/up_proj/Linear/post_convert", "/layers.19/mlp/down_proj/Linear/pre_reshape", "/layers.19/mlp/down_proj/Linear/pre_convert", "/layers.19/mlp/down_proj/Linear", "/layers.19/mlp/down_proj/Linear/post_convert", "/layers.20/self_attn/q_proj/Linear/pre_reshape", "/layers.20/self_attn/q_proj/Linear/pre_convert", "/layers.20/self_attn/q_proj/Linear", "/layers.20/self_attn/q_proj/Linear/post_convert", "/layers.20/self_attn/k_proj/Linear/pre_reshape", "/layers.20/self_attn/k_proj/Linear/pre_convert", "/layers.20/self_attn/k_proj/Linear", "/layers.20/self_attn/k_proj/Linear/post_convert", "/layers.20/self_attn/v_proj/Linear/pre_reshape", "/layers.20/self_attn/v_proj/Linear/pre_convert", "/layers.20/self_attn/v_proj/Linear", "/layers.20/self_attn/v_proj/Linear/post_convert", "/layers.20/self_attn/o_proj/Linear/pre_reshape", "/layers.20/self_attn/o_proj/Linear/pre_convert", "/layers.20/self_attn/o_proj/Linear", "/layers.20/self_attn/o_proj/Linear/post_convert", "/layers.20/mlp/gate_proj/Linear/pre_reshape", "/layers.20/mlp/gate_proj/Linear/pre_convert", "/layers.20/mlp/gate_proj/Linear", "/layers.20/mlp/gate_proj/Linear/post_convert", "/layers.20/mlp/up_proj/Linear/pre_reshape", "/layers.20/mlp/up_proj/Linear/pre_convert", "/layers.20/mlp/up_proj/Linear", "/layers.20/mlp/up_proj/Linear/post_convert", "/layers.20/mlp/down_proj/Linear/pre_reshape", "/layers.20/mlp/down_proj/Linear/pre_convert", "/layers.20/mlp/down_proj/Linear", "/layers.20/mlp/down_proj/Linear/post_convert", "/layers.21/self_attn/q_proj/Linear/pre_reshape", "/layers.21/self_attn/q_proj/Linear/pre_convert", "/layers.21/self_attn/q_proj/Linear", "/layers.21/self_attn/q_proj/Linear/post_convert", "/layers.21/self_attn/k_proj/Linear/pre_reshape", "/layers.21/self_attn/k_proj/Linear/pre_convert", "/layers.21/self_attn/k_proj/Linear", "/layers.21/self_attn/k_proj/Linear/post_convert", "/layers.21/self_attn/v_proj/Linear/pre_reshape", "/layers.21/self_attn/v_proj/Linear/pre_convert", "/layers.21/self_attn/v_proj/Linear", "/layers.21/self_attn/v_proj/Linear/post_convert", "/layers.21/self_attn/o_proj/Linear/pre_reshape", "/layers.21/self_attn/o_proj/Linear/pre_convert", "/layers.21/self_attn/o_proj/Linear", "/layers.21/self_attn/o_proj/Linear/post_convert", "/layers.21/mlp/gate_proj/Linear/pre_reshape", "/layers.21/mlp/gate_proj/Linear/pre_convert", "/layers.21/mlp/gate_proj/Linear", "/layers.21/mlp/gate_proj/Linear/post_convert", "/layers.21/mlp/up_proj/Linear/pre_reshape", "/layers.21/mlp/up_proj/Linear/pre_convert", "/layers.21/mlp/up_proj/Linear", "/layers.21/mlp/up_proj/Linear/post_convert", "/layers.21/mlp/down_proj/Linear/pre_reshape", "/layers.21/mlp/down_proj/Linear/pre_convert", "/layers.21/mlp/down_proj/Linear", "/layers.21/mlp/down_proj/Linear/post_convert", "/layers.22/self_attn/q_proj/Linear/pre_reshape", "/layers.22/self_attn/q_proj/Linear/pre_convert", "/layers.22/self_attn/q_proj/Linear", "/layers.22/self_attn/q_proj/Linear/post_convert", "/layers.22/self_attn/k_proj/Linear/pre_reshape", "/layers.22/self_attn/k_proj/Linear/pre_convert", "/layers.22/self_attn/k_proj/Linear", "/layers.22/self_attn/k_proj/Linear/post_convert", "/layers.22/self_attn/v_proj/Linear/pre_reshape", "/layers.22/self_attn/v_proj/Linear/pre_convert", "/layers.22/self_attn/v_proj/Linear", "/layers.22/self_attn/v_proj/Linear/post_convert", "/layers.22/self_attn/o_proj/Linear/pre_reshape", "/layers.22/self_attn/o_proj/Linear/pre_convert", "/layers.22/self_attn/o_proj/Linear", "/layers.22/self_attn/o_proj/Linear/post_convert", "/layers.22/mlp/gate_proj/Linear/pre_reshape", "/layers.22/mlp/gate_proj/Linear/pre_convert", "/layers.22/mlp/gate_proj/Linear", "/layers.22/mlp/gate_proj/Linear/post_convert", "/layers.22/mlp/up_proj/Linear/pre_reshape", "/layers.22/mlp/up_proj/Linear/pre_convert", "/layers.22/mlp/up_proj/Linear", "/layers.22/mlp/up_proj/Linear/post_convert", "/layers.22/mlp/down_proj/Linear/pre_reshape", "/layers.22/mlp/down_proj/Linear/pre_convert", "/layers.22/mlp/down_proj/Linear", "/layers.22/mlp/down_proj/Linear/post_convert", "/layers.23/self_attn/q_proj/Linear/pre_reshape", "/layers.23/self_attn/q_proj/Linear/pre_convert", "/layers.23/self_attn/q_proj/Linear", "/layers.23/self_attn/q_proj/Linear/post_convert", "/layers.23/self_attn/k_proj/Linear/pre_reshape", "/layers.23/self_attn/k_proj/Linear/pre_convert", "/layers.23/self_attn/k_proj/Linear", "/layers.23/self_attn/k_proj/Linear/post_convert", "/layers.23/self_attn/v_proj/Linear/pre_reshape", "/layers.23/self_attn/v_proj/Linear/pre_convert", "/layers.23/self_attn/v_proj/Linear", "/layers.23/self_attn/v_proj/Linear/post_convert", "/layers.23/self_attn/o_proj/Linear/pre_reshape", "/layers.23/self_attn/o_proj/Linear/pre_convert", "/layers.23/self_attn/o_proj/Linear", "/layers.23/self_attn/o_proj/Linear/post_convert", "/layers.23/mlp/gate_proj/Linear/pre_reshape", "/layers.23/mlp/gate_proj/Linear/pre_convert", "/layers.23/mlp/gate_proj/Linear", "/layers.23/mlp/gate_proj/Linear/post_convert", "/layers.23/mlp/up_proj/Linear/pre_reshape", "/layers.23/mlp/up_proj/Linear/pre_convert", "/layers.23/mlp/up_proj/Linear", "/layers.23/mlp/up_proj/Linear/post_convert", "/layers.23/mlp/down_proj/Linear/pre_reshape", "/layers.23/mlp/down_proj/Linear/pre_convert", "/layers.23/mlp/down_proj/Linear", "/layers.23/mlp/down_proj/Linear/post_convert", "/layers.24/self_attn/q_proj/Linear/pre_reshape", "/layers.24/self_attn/q_proj/Linear/pre_convert", "/layers.24/self_attn/q_proj/Linear", "/layers.24/self_attn/q_proj/Linear/post_convert", "/layers.24/self_attn/k_proj/Linear/pre_reshape", "/layers.24/self_attn/k_proj/Linear/pre_convert", "/layers.24/self_attn/k_proj/Linear", "/layers.24/self_attn/k_proj/Linear/post_convert", "/layers.24/self_attn/v_proj/Linear/pre_reshape", "/layers.24/self_attn/v_proj/Linear/pre_convert", "/layers.24/self_attn/v_proj/Linear", "/layers.24/self_attn/v_proj/Linear/post_convert", "/layers.24/self_attn/o_proj/Linear/pre_reshape", "/layers.24/self_attn/o_proj/Linear/pre_convert", "/layers.24/self_attn/o_proj/Linear", "/layers.24/self_attn/o_proj/Linear/post_convert", "/layers.24/mlp/gate_proj/Linear/pre_reshape", "/layers.24/mlp/gate_proj/Linear/pre_convert", "/layers.24/mlp/gate_proj/Linear", "/layers.24/mlp/gate_proj/Linear/post_convert", "/layers.24/mlp/up_proj/Linear/pre_reshape", "/layers.24/mlp/up_proj/Linear/pre_convert", "/layers.24/mlp/up_proj/Linear", "/layers.24/mlp/up_proj/Linear/post_convert", "/layers.24/mlp/down_proj/Linear/pre_reshape", "/layers.24/mlp/down_proj/Linear/pre_convert", "/layers.24/mlp/down_proj/Linear", "/layers.24/mlp/down_proj/Linear/post_convert", "/layers.25/self_attn/q_proj/Linear/pre_reshape", "/layers.25/self_attn/q_proj/Linear/pre_convert", "/layers.25/self_attn/q_proj/Linear", "/layers.25/self_attn/q_proj/Linear/post_convert", "/layers.25/self_attn/k_proj/Linear/pre_reshape", "/layers.25/self_attn/k_proj/Linear/pre_convert", "/layers.25/self_attn/k_proj/Linear", "/layers.25/self_attn/k_proj/Linear/post_convert", "/layers.25/self_attn/v_proj/Linear/pre_reshape", "/layers.25/self_attn/v_proj/Linear/pre_convert", "/layers.25/self_attn/v_proj/Linear", "/layers.25/self_attn/v_proj/Linear/post_convert", "/layers.25/self_attn/o_proj/Linear/pre_reshape", "/layers.25/self_attn/o_proj/Linear/pre_convert", "/layers.25/self_attn/o_proj/Linear", "/layers.25/self_attn/o_proj/Linear/post_convert", "/layers.25/mlp/gate_proj/Linear/pre_reshape", "/layers.25/mlp/gate_proj/Linear/pre_convert", "/layers.25/mlp/gate_proj/Linear", "/layers.25/mlp/gate_proj/Linear/post_convert", "/layers.25/mlp/up_proj/Linear/pre_reshape", "/layers.25/mlp/up_proj/Linear/pre_convert", "/layers.25/mlp/up_proj/Linear", "/layers.25/mlp/up_proj/Linear/post_convert", "/layers.25/mlp/down_proj/Linear/pre_reshape", "/layers.25/mlp/down_proj/Linear/pre_convert", "/layers.25/mlp/down_proj/Linear", "/layers.25/mlp/down_proj/Linear/post_convert", "/layers.26/self_attn/q_proj/Linear/pre_reshape", "/layers.26/self_attn/q_proj/Linear/pre_convert", "/layers.26/self_attn/q_proj/Linear", "/layers.26/self_attn/q_proj/Linear/post_convert", "/layers.26/self_attn/k_proj/Linear/pre_reshape", "/layers.26/self_attn/k_proj/Linear/pre_convert", "/layers.26/self_attn/k_proj/Linear", "/layers.26/self_attn/k_proj/Linear/post_convert", "/layers.26/self_attn/v_proj/Linear/pre_reshape", "/layers.26/self_attn/v_proj/Linear/pre_convert", "/layers.26/self_attn/v_proj/Linear", "/layers.26/self_attn/v_proj/Linear/post_convert", "/layers.26/self_attn/o_proj/Linear/pre_reshape", "/layers.26/self_attn/o_proj/Linear/pre_convert", "/layers.26/self_attn/o_proj/Linear", "/layers.26/self_attn/o_proj/Linear/post_convert", "/layers.26/mlp/gate_proj/Linear/pre_reshape", "/layers.26/mlp/gate_proj/Linear/pre_convert", "/layers.26/mlp/gate_proj/Linear", "/layers.26/mlp/gate_proj/Linear/post_convert", "/layers.26/mlp/up_proj/Linear/pre_reshape", "/layers.26/mlp/up_proj/Linear/pre_convert", "/layers.26/mlp/up_proj/Linear", "/layers.26/mlp/up_proj/Linear/post_convert", "/layers.26/mlp/down_proj/Linear/pre_reshape", "/layers.26/mlp/down_proj/Linear/pre_convert", "/layers.26/mlp/down_proj/Linear", "/layers.26/mlp/down_proj/Linear/post_convert", "/layers.27/self_attn/q_proj/Linear/pre_reshape", "/layers.27/self_attn/q_proj/Linear/pre_convert", "/layers.27/self_attn/q_proj/Linear", "/layers.27/self_attn/q_proj/Linear/post_convert", "/layers.27/self_attn/k_proj/Linear/pre_reshape", "/layers.27/self_attn/k_proj/Linear/pre_convert", "/layers.27/self_attn/k_proj/Linear", "/layers.27/self_attn/k_proj/Linear/post_convert", "/layers.27/self_attn/v_proj/Linear/pre_reshape", "/layers.27/self_attn/v_proj/Linear/pre_convert", "/layers.27/self_attn/v_proj/Linear", "/layers.27/self_attn/v_proj/Linear/post_convert", "/layers.27/self_attn/o_proj/Linear/pre_reshape", "/layers.27/self_attn/o_proj/Linear/pre_convert", "/layers.27/self_attn/o_proj/Linear", "/layers.27/self_attn/o_proj/Linear/post_convert", "/layers.27/mlp/gate_proj/Linear/pre_reshape", "/layers.27/mlp/gate_proj/Linear/pre_convert", "/layers.27/mlp/gate_proj/Linear", "/layers.27/mlp/gate_proj/Linear/post_convert", "/layers.27/mlp/up_proj/Linear/pre_reshape", "/layers.27/mlp/up_proj/Linear/pre_convert", "/layers.27/mlp/up_proj/Linear", "/layers.27/mlp/up_proj/Linear/post_convert", "/layers.27/mlp/down_proj/Linear/pre_reshape", "/layers.27/mlp/down_proj/Linear/pre_convert", "/layers.27/mlp/down_proj/Linear", "/layers.27/mlp/down_proj/Linear/post_convert", "/layers.28/self_attn/q_proj/Linear/pre_reshape", "/layers.28/self_attn/q_proj/Linear/pre_convert", "/layers.28/self_attn/q_proj/Linear", "/layers.28/self_attn/q_proj/Linear/post_convert", "/layers.28/self_attn/k_proj/Linear/pre_reshape", "/layers.28/self_attn/k_proj/Linear/pre_convert", "/layers.28/self_attn/k_proj/Linear", "/layers.28/self_attn/k_proj/Linear/post_convert", "/layers.28/self_attn/v_proj/Linear/pre_reshape", "/layers.28/self_attn/v_proj/Linear/pre_convert", "/layers.28/self_attn/v_proj/Linear", "/layers.28/self_attn/v_proj/Linear/post_convert", "/layers.28/self_attn/o_proj/Linear/pre_reshape", "/layers.28/self_attn/o_proj/Linear/pre_convert", "/layers.28/self_attn/o_proj/Linear", "/layers.28/self_attn/o_proj/Linear/post_convert", "/layers.28/mlp/gate_proj/Linear/pre_reshape", "/layers.28/mlp/gate_proj/Linear/pre_convert", "/layers.28/mlp/gate_proj/Linear", "/layers.28/mlp/gate_proj/Linear/post_convert", "/layers.28/mlp/up_proj/Linear/pre_reshape", "/layers.28/mlp/up_proj/Linear/pre_convert", "/layers.28/mlp/up_proj/Linear", "/layers.28/mlp/up_proj/Linear/post_convert", "/layers.28/mlp/down_proj/Linear/pre_reshape", "/layers.28/mlp/down_proj/Linear/pre_convert", "/layers.28/mlp/down_proj/Linear", "/layers.28/mlp/down_proj/Linear/post_convert", "/layers.29/self_attn/q_proj/Linear/pre_reshape", "/layers.29/self_attn/q_proj/Linear/pre_convert", "/layers.29/self_attn/q_proj/Linear", "/layers.29/self_attn/q_proj/Linear/post_convert", "/layers.29/self_attn/k_proj/Linear/pre_reshape", "/layers.29/self_attn/k_proj/Linear/pre_convert", "/layers.29/self_attn/k_proj/Linear", "/layers.29/self_attn/k_proj/Linear/post_convert", "/layers.29/self_attn/v_proj/Linear/pre_reshape", "/layers.29/self_attn/v_proj/Linear/pre_convert", "/layers.29/self_attn/v_proj/Linear", "/layers.29/self_attn/v_proj/Linear/post_convert", "/layers.29/self_attn/o_proj/Linear/pre_reshape", "/layers.29/self_attn/o_proj/Linear/pre_convert", "/layers.29/self_attn/o_proj/Linear", "/layers.29/self_attn/o_proj/Linear/post_convert", "/layers.29/mlp/gate_proj/Linear/pre_reshape", "/layers.29/mlp/gate_proj/Linear/pre_convert", "/layers.29/mlp/gate_proj/Linear", "/layers.29/mlp/gate_proj/Linear/post_convert", "/layers.29/mlp/up_proj/Linear/pre_reshape", "/layers.29/mlp/up_proj/Linear/pre_convert", "/layers.29/mlp/up_proj/Linear", "/layers.29/mlp/up_proj/Linear/post_convert", "/layers.29/mlp/down_proj/Linear/pre_reshape", "/layers.29/mlp/down_proj/Linear/pre_convert", "/layers.29/mlp/down_proj/Linear", "/layers.29/mlp/down_proj/Linear/post_convert", "/layers.30/self_attn/q_proj/Linear/pre_reshape", "/layers.30/self_attn/q_proj/Linear/pre_convert", "/layers.30/self_attn/q_proj/Linear", "/layers.30/self_attn/q_proj/Linear/post_convert", "/layers.30/self_attn/k_proj/Linear/pre_reshape", "/layers.30/self_attn/k_proj/Linear/pre_convert", "/layers.30/self_attn/k_proj/Linear", "/layers.30/self_attn/k_proj/Linear/post_convert", "/layers.30/self_attn/v_proj/Linear/pre_reshape", "/layers.30/self_attn/v_proj/Linear/pre_convert", "/layers.30/self_attn/v_proj/Linear", "/layers.30/self_attn/v_proj/Linear/post_convert", "/layers.30/self_attn/o_proj/Linear/pre_reshape", "/layers.30/self_attn/o_proj/Linear/pre_convert", "/layers.30/self_attn/o_proj/Linear", "/layers.30/self_attn/o_proj/Linear/post_convert", "/layers.30/mlp/gate_proj/Linear/pre_reshape", "/layers.30/mlp/gate_proj/Linear/pre_convert", "/layers.30/mlp/gate_proj/Linear", "/layers.30/mlp/gate_proj/Linear/post_convert", "/layers.30/mlp/up_proj/Linear/pre_reshape", "/layers.30/mlp/up_proj/Linear/pre_convert", "/layers.30/mlp/up_proj/Linear", "/layers.30/mlp/up_proj/Linear/post_convert", "/layers.30/mlp/down_proj/Linear/pre_reshape", "/layers.30/mlp/down_proj/Linear/pre_convert", "/layers.30/mlp/down_proj/Linear", "/layers.30/mlp/down_proj/Linear/post_convert", "/layers.31/self_attn/q_proj/Linear/pre_reshape", "/layers.31/self_attn/q_proj/Linear/pre_convert", "/layers.31/self_attn/q_proj/Linear", "/layers.31/self_attn/q_proj/Linear/post_convert", "/layers.31/self_attn/k_proj/Linear/pre_reshape", "/layers.31/self_attn/k_proj/Linear/pre_convert", "/layers.31/self_attn/k_proj/Linear", "/layers.31/self_attn/k_proj/Linear/post_convert", "/layers.31/self_attn/v_proj/Linear/pre_reshape", "/layers.31/self_attn/v_proj/Linear/pre_convert", "/layers.31/self_attn/v_proj/Linear", "/layers.31/self_attn/v_proj/Linear/post_convert", "/layers.31/self_attn/o_proj/Linear/pre_reshape", "/layers.31/self_attn/o_proj/Linear/pre_convert", "/layers.31/self_attn/o_proj/Linear", "/layers.31/self_attn/o_proj/Linear/post_convert", "/layers.31/mlp/gate_proj/Linear/pre_reshape", "/layers.31/mlp/gate_proj/Linear/pre_convert", "/layers.31/mlp/gate_proj/Linear", "/layers.31/mlp/gate_proj/Linear/post_convert", "/layers.31/mlp/up_proj/Linear/pre_reshape", "/layers.31/mlp/up_proj/Linear/pre_convert", "/layers.31/mlp/up_proj/Linear", "/layers.31/mlp/up_proj/Linear/post_convert", "/layers.31/mlp/down_proj/Linear/pre_reshape", "/layers.31/mlp/down_proj/Linear/pre_convert", "/layers.31/mlp/down_proj/Linear", "/layers.31/mlp/down_proj/Linear/post_convert", "/layers.32/self_attn/q_proj/Linear/pre_reshape", "/layers.32/self_attn/q_proj/Linear/pre_convert", "/layers.32/self_attn/q_proj/Linear", "/layers.32/self_attn/q_proj/Linear/post_convert", "/layers.32/self_attn/k_proj/Linear/pre_reshape", "/layers.32/self_attn/k_proj/Linear/pre_convert", "/layers.32/self_attn/k_proj/Linear", "/layers.32/self_attn/k_proj/Linear/post_convert", "/layers.32/self_attn/v_proj/Linear/pre_reshape", "/layers.32/self_attn/v_proj/Linear/pre_convert", "/layers.32/self_attn/v_proj/Linear", "/layers.32/self_attn/v_proj/Linear/post_convert", "/layers.32/self_attn/o_proj/Linear/pre_reshape", "/layers.32/self_attn/o_proj/Linear/pre_convert", "/layers.32/self_attn/o_proj/Linear", "/layers.32/self_attn/o_proj/Linear/post_convert", "/layers.32/mlp/gate_proj/Linear/pre_reshape", "/layers.32/mlp/gate_proj/Linear/pre_convert", "/layers.32/mlp/gate_proj/Linear", "/layers.32/mlp/gate_proj/Linear/post_convert", "/layers.32/mlp/up_proj/Linear/pre_reshape", "/layers.32/mlp/up_proj/Linear/pre_convert", "/layers.32/mlp/up_proj/Linear", "/layers.32/mlp/up_proj/Linear/post_convert", "/layers.32/mlp/down_proj/Linear/pre_reshape", "/layers.32/mlp/down_proj/Linear/pre_convert", "/layers.32/mlp/down_proj/Linear", "/layers.32/mlp/down_proj/Linear/post_convert", "/layers.33/self_attn/q_proj/Linear/pre_reshape", "/layers.33/self_attn/q_proj/Linear/pre_convert", "/layers.33/self_attn/q_proj/Linear", "/layers.33/self_attn/q_proj/Linear/post_convert", "/layers.33/self_attn/k_proj/Linear/pre_reshape", "/layers.33/self_attn/k_proj/Linear/pre_convert", "/layers.33/self_attn/k_proj/Linear", "/layers.33/self_attn/k_proj/Linear/post_convert", "/layers.33/self_attn/v_proj/Linear/pre_reshape", "/layers.33/self_attn/v_proj/Linear/pre_convert", "/layers.33/self_attn/v_proj/Linear", "/layers.33/self_attn/v_proj/Linear/post_convert", "/layers.33/self_attn/o_proj/Linear/pre_reshape", "/layers.33/self_attn/o_proj/Linear/pre_convert", "/layers.33/self_attn/o_proj/Linear", "/layers.33/self_attn/o_proj/Linear/post_convert", "/layers.33/mlp/gate_proj/Linear/pre_reshape", "/layers.33/mlp/gate_proj/Linear/pre_convert", "/layers.33/mlp/gate_proj/Linear", "/layers.33/mlp/gate_proj/Linear/post_convert", "/layers.33/mlp/up_proj/Linear/pre_reshape", "/layers.33/mlp/up_proj/Linear/pre_convert", "/layers.33/mlp/up_proj/Linear", "/layers.33/mlp/up_proj/Linear/post_convert", "/layers.33/mlp/down_proj/Linear/pre_reshape", "/layers.33/mlp/down_proj/Linear/pre_convert", "/layers.33/mlp/down_proj/Linear", "/layers.33/mlp/down_proj/Linear/post_convert", "/layers.34/self_attn/q_proj/Linear/pre_reshape", "/layers.34/self_attn/q_proj/Linear/pre_convert", "/layers.34/self_attn/q_proj/Linear", "/layers.34/self_attn/q_proj/Linear/post_convert", "/layers.34/self_attn/k_proj/Linear/pre_reshape", "/layers.34/self_attn/k_proj/Linear/pre_convert", "/layers.34/self_attn/k_proj/Linear", "/layers.34/self_attn/k_proj/Linear/post_convert", "/layers.34/self_attn/v_proj/Linear/pre_reshape", "/layers.34/self_attn/v_proj/Linear/pre_convert", "/layers.34/self_attn/v_proj/Linear", "/layers.34/self_attn/v_proj/Linear/post_convert", "/layers.34/self_attn/o_proj/Linear/pre_reshape", "/layers.34/self_attn/o_proj/Linear/pre_convert", "/layers.34/self_attn/o_proj/Linear", "/layers.34/self_attn/o_proj/Linear/post_convert", "/layers.34/mlp/gate_proj/Linear/pre_reshape", "/layers.34/mlp/gate_proj/Linear/pre_convert", "/layers.34/mlp/gate_proj/Linear", "/layers.34/mlp/gate_proj/Linear/post_convert", "/layers.34/mlp/up_proj/Linear/pre_reshape", "/layers.34/mlp/up_proj/Linear/pre_convert", "/layers.34/mlp/up_proj/Linear", "/layers.34/mlp/up_proj/Linear/post_convert", "/layers.34/mlp/down_proj/Linear/pre_reshape", "/layers.34/mlp/down_proj/Linear/pre_convert", "/layers.34/mlp/down_proj/Linear", "/layers.34/mlp/down_proj/Linear/post_convert", "/layers.35/self_attn/q_proj/Linear/pre_reshape", "/layers.35/self_attn/q_proj/Linear/pre_convert", "/layers.35/self_attn/q_proj/Linear", "/layers.35/self_attn/q_proj/Linear/post_convert", "/layers.35/self_attn/k_proj/Linear/pre_reshape", "/layers.35/self_attn/k_proj/Linear/pre_convert", "/layers.35/self_attn/k_proj/Linear", "/layers.35/self_attn/k_proj/Linear/post_convert", "/layers.35/self_attn/v_proj/Linear/pre_reshape", "/layers.35/self_attn/v_proj/Linear/pre_convert", "/layers.35/self_attn/v_proj/Linear", "/layers.35/self_attn/v_proj/Linear/post_convert", "/layers.35/self_attn/o_proj/Linear/pre_reshape", "/layers.35/self_attn/o_proj/Linear/pre_convert", "/layers.35/self_attn/o_proj/Linear", "/layers.35/self_attn/o_proj/Linear/post_convert", "/layers.35/mlp/gate_proj/Linear/pre_reshape", "/layers.35/mlp/gate_proj/Linear/pre_convert", "/layers.35/mlp/gate_proj/Linear", "/layers.35/mlp/gate_proj/Linear/post_convert", "/layers.35/mlp/up_proj/Linear/pre_reshape", "/layers.35/mlp/up_proj/Linear/pre_convert", "/layers.35/mlp/up_proj/Linear", "/layers.35/mlp/up_proj/Linear/post_convert", "/layers.35/mlp/down_proj/Linear/pre_reshape", "/layers.35/mlp/down_proj/Linear/pre_convert", "/layers.35/mlp/down_proj/Linear", "/layers.35/mlp/down_proj/Linear/post_convert", "/lm/lm_head/Linear/pre_reshape", "/lm/lm_head/Linear/pre_convert", "/lm/lm_head/Linear", "/lm/lm_head/Linear/post_convert" ] +, "tensorNumber": 0, "usage": "INFERENCE", "mnn_uuid": "6a43b9bc-0aa0-4ce2-b1c2-3e185776650d" }