diff --git "a/llm.mnn.json" "b/llm.mnn.json" new file mode 100644--- /dev/null +++ "b/llm.mnn.json" @@ -0,0 +1,35844 @@ + +{ "bizCode": "MNNTest", "extraInfo": +{ "version": "3.2.4" } +, "oplists": +[ +{ "main_type": "Input", "main": +{ "dims": +[ 1, 1, 4096 ] +, "dtype": "DT_FLOAT", "dformat": "NCHW" } +, "name": "input_ids", "outputIndexes": +[ 4 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Input", "main": +{ "dims": +[ 1, 1, -1, -1 ] +, "dtype": "DT_FLOAT", "dformat": "NCHW" } +, "name": "attention_mask", "outputIndexes": +[ 130 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Input", "main": +{ "dims": +[ 1, -1 ] +, "dtype": "DT_INT32", "dformat": "NCHW" } +, "name": "position_ids", "outputIndexes": +[ 2 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Input", "main": +{ "dims": +[ 36, 2, 1, -1, 8, 128 ] +, "dtype": "DT_FLOAT", "dformat": "NCHW" } +, "name": "past_key_values", "outputIndexes": +[ 0 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Input", "main": +{ "dims": +[ 1 ] +, "dtype": "DT_INT32", "dformat": "NCHW" } +, "name": "logits_index", "outputIndexes": +[ 3 ] +, "type": "Input", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 0 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1 ] +, "dimType": "NCHW" } +, "name": "presents", "outputIndexes": +[ 1 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 3 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1, -1, 4096 ] + } +, "name": "/Constant_output_0", "outputIndexes": +[ 5 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_output_0", "outputIndexes": +[ 6 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 6 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 0, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm/Mul_1_output_0", "outputIndexes": +[ 7 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 7 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3399 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3399 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3400 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3400 ] +, "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": +[ 3401 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3401 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3402 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3402 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 8 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 7 ] +, "main_type": "NONE", "name": "/Shape_output_0", "outputIndexes": +[ 9 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 9 ] +, "main_type": "NONE", "name": "Shape52", "outputIndexes": +[ 10 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 9 ] +, "main_type": "NONE", "name": "Rank54", "outputIndexes": +[ 11 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 11, 11 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp56", "outputIndexes": +[ 12 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze57", "outputIndexes": +[ 13 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1 ] + } +, "name": "Const9", "outputIndexes": +[ 14 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp59", "outputIndexes": +[ 15 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 15 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze60", "outputIndexes": +[ 16 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1 ] + } +, "name": "Unsqueeze62", "outputIndexes": +[ 17 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 10, 13, 16, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice63", "outputIndexes": +[ 18 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 18 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze64", "outputIndexes": +[ 19 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 19, 19 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp66", "outputIndexes": +[ 20 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 9, 20, 12 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_1_output_0", "outputIndexes": +[ 21 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 0 ] + } +, "name": "/rotary/Constant_4_output_0", "outputIndexes": +[ 22 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 21, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_output_0", "outputIndexes": +[ 23 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 12 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze76", "outputIndexes": +[ 24 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 15 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze79", "outputIndexes": +[ 25 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 10, 24, 25, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice82", "outputIndexes": +[ 26 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 26 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze83", "outputIndexes": +[ 27 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 27 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp84", "outputIndexes": +[ 28 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 28, 27 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp85", "outputIndexes": +[ 29 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 9, 29, 12 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_2_output_0", "outputIndexes": +[ 30 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 30, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1_output_0", "outputIndexes": +[ 31 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 32 ] + } +, "name": "/Constant_4_output_0", "outputIndexes": +[ 32 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 128 ] + } +, "name": "/Constant_5_output_0", "outputIndexes": +[ 33 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 23, 31, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_output_0", "outputIndexes": +[ 34 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 8, 34 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_1_output_0", "outputIndexes": +[ 35 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 35 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 10518550, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm/Mul_1_output_0", "outputIndexes": +[ 37 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2 ] +, "main_type": "CastParam", "main": +{ "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" } +, "name": "/rotary/Cast_output_0", "outputIndexes": +[ 38 ] +, "type": "Cast", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 2 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ -1, 1 ] + } +, "name": "/rotary/Constant_1_output_0", "outputIndexes": +[ 39 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 38, 39 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/rotary/Reshape_output_0", "outputIndexes": +[ 40 ] +, "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_2_output_0", "outputIndexes": +[ 41 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 40, 41 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/rotary/Mul_output_0", "outputIndexes": +[ 42 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 42 ] +, "main_type": "UnaryOp", "main": +{ "opType": "COS", "T": "DT_FLOAT" } +, "name": "/rotary/Cos_output_0", "outputIndexes": +[ 43 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 43, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_output_0", "outputIndexes": +[ 44 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 42 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SIN", "T": "DT_FLOAT" } +, "name": "/rotary/Sin_output_0", "outputIndexes": +[ 45 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 45, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_1_output_0", "outputIndexes": +[ 46 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 44, 46 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/rotary/Concat_output_0", "outputIndexes": +[ 47 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 47, 47 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/rotary/Concat_1_output_0", "outputIndexes": +[ 48 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 2 ] + } +, "name": "/rotary/Constant_7_output_0", "outputIndexes": +[ 49 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 48, 49 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_2_output_0", "outputIndexes": +[ 50 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 50, 17 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/rotary/Unsqueeze_3_output_0", "outputIndexes": +[ 51 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 51 ] +, "main_type": "NONE", "name": "Shape116", "outputIndexes": +[ 52 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 51 ] +, "main_type": "NONE", "name": "Rank118", "outputIndexes": +[ 53 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 53, 53 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp120", "outputIndexes": +[ 54 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 54 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze121", "outputIndexes": +[ 55 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 54, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp123", "outputIndexes": +[ 56 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 56 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze124", "outputIndexes": +[ 57 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 52, 55, 57, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice127", "outputIndexes": +[ 58 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 58 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze128", "outputIndexes": +[ 59 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 59, 59 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp130", "outputIndexes": +[ 60 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 51, 60, 54 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_3_output_0", "outputIndexes": +[ 61 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 37, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_output_0", "outputIndexes": +[ 62 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 37 ] +, "main_type": "NONE", "name": "/Shape_2_output_0", "outputIndexes": +[ 63 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "main_type": "Blob", "main": +{ "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 3 ] + } +, "name": "/Constant_11_output_0", "outputIndexes": +[ 64 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 63 ] +, "main_type": "NONE", "name": "Shape258", "outputIndexes": +[ 65 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 63 ] +, "main_type": "NONE", "name": "Rank260", "outputIndexes": +[ 66 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 66, 66 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp262", "outputIndexes": +[ 67 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 67 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze263", "outputIndexes": +[ 68 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 67, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp265", "outputIndexes": +[ 69 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 69 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze266", "outputIndexes": +[ 70 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 65, 68, 70, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice269", "outputIndexes": +[ 71 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 71 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze270", "outputIndexes": +[ 72 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 72 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp271", "outputIndexes": +[ 73 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 73, 72 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp272", "outputIndexes": +[ 74 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 63, 74, 67 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_5_output_0", "outputIndexes": +[ 75 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 2 ] + } +, "name": "/rotary/Constant_6_output_0", "outputIndexes": +[ 76 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 75, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_output_0", "outputIndexes": +[ 77 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 77, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_7_output_0", "outputIndexes": +[ 78 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 1073741824 ] + } +, "name": "/Constant_20_output_0", "outputIndexes": +[ 79 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 3 ] + } +, "name": "/Constant_14_output_0", "outputIndexes": +[ 80 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 37, 78, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_1_output_0", "outputIndexes": +[ 81 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 81 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_output_0", "outputIndexes": +[ 82 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 77, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_6_output_0", "outputIndexes": +[ 83 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 37, 22, 83, 80, 17 ] +, "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": +[ 84 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 82, 84 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_3_output_0", "outputIndexes": +[ 85 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 54 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze167", "outputIndexes": +[ 86 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 56 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze170", "outputIndexes": +[ 87 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 52, 86, 87, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice173", "outputIndexes": +[ 88 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 88 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze174", "outputIndexes": +[ 89 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 89 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp175", "outputIndexes": +[ 90 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 90, 89 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp176", "outputIndexes": +[ 91 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 51, 91, 54 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_4_output_0", "outputIndexes": +[ 92 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 85, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_1_output_0", "outputIndexes": +[ 93 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 62, 93 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_output_0", "outputIndexes": +[ 94 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3400 ] +, "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": +[ 3405 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3405 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3406 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3406 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 95 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 21, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_2_output_0", "outputIndexes": +[ 96 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 30, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_3_output_0", "outputIndexes": +[ 97 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "main_type": "Blob", "main": +{ "dims": +[ 1 ] +, "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": +[ 8 ] + } +, "name": "/Constant_7_output_0", "outputIndexes": +[ 98 ] +, "type": "Const", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 96, 97, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_1_output_0", "outputIndexes": +[ 99 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 95, 99 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_2_output_0", "outputIndexes": +[ 100 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 100 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 13141036, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm/Mul_1_output_0", "outputIndexes": +[ 102 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 102, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_2_output_0", "outputIndexes": +[ 103 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 102 ] +, "main_type": "NONE", "name": "/Shape_3_output_0", "outputIndexes": +[ 104 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 104 ] +, "main_type": "NONE", "name": "Shape135", "outputIndexes": +[ 105 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 104 ] +, "main_type": "NONE", "name": "Rank137", "outputIndexes": +[ 106 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 106, 106 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp139", "outputIndexes": +[ 107 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 107 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze140", "outputIndexes": +[ 108 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 107, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp142", "outputIndexes": +[ 109 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 109 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze143", "outputIndexes": +[ 110 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 105, 108, 110, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice146", "outputIndexes": +[ 111 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 111 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze147", "outputIndexes": +[ 112 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 112 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp148", "outputIndexes": +[ 113 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 113, 112 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp149", "outputIndexes": +[ 114 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 104, 114, 107 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_6_output_0", "outputIndexes": +[ 115 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 115, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_1_output_0", "outputIndexes": +[ 116 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 116, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_9_output_0", "outputIndexes": +[ 117 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 102, 117, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_3_output_0", "outputIndexes": +[ 118 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 118 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_1_output_0", "outputIndexes": +[ 119 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 116, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_8_output_0", "outputIndexes": +[ 120 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 102, 22, 120, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_2_output_0", "outputIndexes": +[ 121 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 119, 121 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_4_output_0", "outputIndexes": +[ 122 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 122, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_3_output_0", "outputIndexes": +[ 123 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 103, 123 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_1_output_0", "outputIndexes": +[ 124 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3400 ] +, "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": +[ 3409 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3409 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3410 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3410 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 125 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 21, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_4_output_0", "outputIndexes": +[ 126 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 30, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_5_output_0", "outputIndexes": +[ 127 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 126, 127, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_2_output_0", "outputIndexes": +[ 128 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 125, 128 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_3_output_0", "outputIndexes": +[ 129 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 94, 124, 129, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_8_output_0", "outputIndexes": +[ 131 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 131 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3411 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3411 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3412 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3412 ] +, "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": +[ 3413 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3413 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3414 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3414 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.0/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 132 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 6, 132 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_3_output_0", "outputIndexes": +[ 133 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 133 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 26249304, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm/Mul_1_output_0", "outputIndexes": +[ 134 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 134 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3415 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3415 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3416 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3416 ] +, "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": +[ 3417 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3417 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3418 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3418 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 135 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 135 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn/Mul_output_0", "outputIndexes": +[ 136 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3416 ] +, "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": +[ 3421 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3421 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3422 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3422 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 137 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 136, 137 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp/Mul_output_0", "outputIndexes": +[ 138 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 138 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3423 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3423 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.0/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3424 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3424 ] +, "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": +[ 3425 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3425 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.0/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3426 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3426 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.0/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 139 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 133, 139 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_4_output_0", "outputIndexes": +[ 140 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 140, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_9_output_0", "outputIndexes": +[ 141 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 141 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 120653978, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_1/Mul_1_output_0", "outputIndexes": +[ 142 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 142 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3427 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3427 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3428 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3428 ] +, "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": +[ 3429 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3429 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3430 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3430 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 143 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 142 ] +, "main_type": "NONE", "name": "/Shape_14_output_0", "outputIndexes": +[ 144 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 144 ] +, "main_type": "NONE", "name": "Shape503", "outputIndexes": +[ 145 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 144 ] +, "main_type": "NONE", "name": "Rank505", "outputIndexes": +[ 146 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 146, 146 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp507", "outputIndexes": +[ 147 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 147 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze508", "outputIndexes": +[ 148 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 147, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp510", "outputIndexes": +[ 149 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 149 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze511", "outputIndexes": +[ 150 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 145, 148, 150, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice514", "outputIndexes": +[ 151 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 151 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze515", "outputIndexes": +[ 152 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 152, 152 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp517", "outputIndexes": +[ 153 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 144, 153, 147 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_18_output_0", "outputIndexes": +[ 154 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 154, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_32_output_0", "outputIndexes": +[ 155 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 147 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze525", "outputIndexes": +[ 156 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 149 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze528", "outputIndexes": +[ 157 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 145, 156, 157, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice531", "outputIndexes": +[ 158 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 158 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze532", "outputIndexes": +[ 159 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 159 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp533", "outputIndexes": +[ 160 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 160, 159 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp534", "outputIndexes": +[ 161 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 144, 161, 147 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_19_output_0", "outputIndexes": +[ 162 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 162, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_33_output_0", "outputIndexes": +[ 163 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 155, 163, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_13_output_0", "outputIndexes": +[ 164 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 143, 164 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_10_output_0", "outputIndexes": +[ 165 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 165 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 131172528, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_1/Mul_1_output_0", "outputIndexes": +[ 167 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 167, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_8_output_0", "outputIndexes": +[ 168 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 167 ] +, "main_type": "NONE", "name": "/Shape_16_output_0", "outputIndexes": +[ 169 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 169 ] +, "main_type": "NONE", "name": "Shape652", "outputIndexes": +[ 170 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 169 ] +, "main_type": "NONE", "name": "Rank654", "outputIndexes": +[ 171 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 171, 171 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp656", "outputIndexes": +[ 172 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 172 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze657", "outputIndexes": +[ 173 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 172, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp659", "outputIndexes": +[ 174 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 174 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze660", "outputIndexes": +[ 175 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 170, 173, 175, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice663", "outputIndexes": +[ 176 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 176 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze664", "outputIndexes": +[ 177 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 177 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp665", "outputIndexes": +[ 178 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 178, 177 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp666", "outputIndexes": +[ 179 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 169, 179, 172 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_20_output_0", "outputIndexes": +[ 180 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 180, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_3_output_0", "outputIndexes": +[ 181 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 181, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_39_output_0", "outputIndexes": +[ 182 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 167, 182, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_5_output_0", "outputIndexes": +[ 183 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 183 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_2_output_0", "outputIndexes": +[ 184 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 181, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_38_output_0", "outputIndexes": +[ 185 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 167, 22, 185, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_4_output_0", "outputIndexes": +[ 186 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 184, 186 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_16_output_0", "outputIndexes": +[ 187 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 187, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_9_output_0", "outputIndexes": +[ 188 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 168, 188 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_5_output_0", "outputIndexes": +[ 189 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3428 ] +, "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": +[ 3433 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3433 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3434 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3434 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 190 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 154, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_34_output_0", "outputIndexes": +[ 191 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 162, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_35_output_0", "outputIndexes": +[ 192 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 191, 192, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_14_output_0", "outputIndexes": +[ 193 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 190, 193 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_11_output_0", "outputIndexes": +[ 194 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 194 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 133795014, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_1/Mul_1_output_0", "outputIndexes": +[ 196 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 196, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_10_output_0", "outputIndexes": +[ 197 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 196 ] +, "main_type": "NONE", "name": "/Shape_17_output_0", "outputIndexes": +[ 198 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 198 ] +, "main_type": "NONE", "name": "Shape549", "outputIndexes": +[ 199 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 198 ] +, "main_type": "NONE", "name": "Rank551", "outputIndexes": +[ 200 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 200, 200 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp553", "outputIndexes": +[ 201 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 201 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze554", "outputIndexes": +[ 202 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 201, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp556", "outputIndexes": +[ 203 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 203 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze557", "outputIndexes": +[ 204 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 199, 202, 204, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice560", "outputIndexes": +[ 205 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 205 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze561", "outputIndexes": +[ 206 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 206 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp562", "outputIndexes": +[ 207 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 207, 206 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp563", "outputIndexes": +[ 208 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 198, 208, 201 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_21_output_0", "outputIndexes": +[ 209 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 209, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_4_output_0", "outputIndexes": +[ 210 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 210, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_41_output_0", "outputIndexes": +[ 211 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 196, 211, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_7_output_0", "outputIndexes": +[ 212 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 212 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_3_output_0", "outputIndexes": +[ 213 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 210, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_40_output_0", "outputIndexes": +[ 214 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 196, 22, 214, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_6_output_0", "outputIndexes": +[ 215 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 213, 215 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_17_output_0", "outputIndexes": +[ 216 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 216, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_11_output_0", "outputIndexes": +[ 217 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 197, 217 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_6_output_0", "outputIndexes": +[ 218 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3428 ] +, "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": +[ 3437 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3437 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3438 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3438 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 219 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 154, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_36_output_0", "outputIndexes": +[ 220 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 162, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_37_output_0", "outputIndexes": +[ 221 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 220, 221, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_15_output_0", "outputIndexes": +[ 222 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 219, 222 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_12_output_0", "outputIndexes": +[ 223 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 189, 218, 223, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_17_output_0", "outputIndexes": +[ 224 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 224 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3439 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3439 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3440 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3440 ] +, "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": +[ 3441 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3441 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3442 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3442 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.1/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 225 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 141, 225 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_8_output_0", "outputIndexes": +[ 226 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 226 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 146903282, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_1/Mul_1_output_0", "outputIndexes": +[ 227 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 227 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3443 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3443 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3444 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3444 ] +, "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": +[ 3445 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3445 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3446 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3446 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 228 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 228 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_1/Mul_output_0", "outputIndexes": +[ 229 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3444 ] +, "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": +[ 3449 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3449 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3450 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3450 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 230 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 229, 230 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_1/Mul_output_0", "outputIndexes": +[ 231 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 231 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3451 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3451 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.1/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3452 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3452 ] +, "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": +[ 3453 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3453 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.1/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3454 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3454 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.1/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 232 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 226, 232 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_9_output_0", "outputIndexes": +[ 233 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 233, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_18_output_0", "outputIndexes": +[ 234 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 234 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 241307956, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_2/Mul_1_output_0", "outputIndexes": +[ 235 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 235 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3455 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3455 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3456 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3456 ] +, "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": +[ 3457 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3457 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3458 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3458 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 236 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 235 ] +, "main_type": "NONE", "name": "/Shape_28_output_0", "outputIndexes": +[ 237 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 237 ] +, "main_type": "NONE", "name": "Shape891", "outputIndexes": +[ 238 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 237 ] +, "main_type": "NONE", "name": "Rank893", "outputIndexes": +[ 239 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 239, 239 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp895", "outputIndexes": +[ 240 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 240 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze896", "outputIndexes": +[ 241 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 240, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp898", "outputIndexes": +[ 242 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 242 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze899", "outputIndexes": +[ 243 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 238, 241, 243, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice902", "outputIndexes": +[ 244 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 244 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze903", "outputIndexes": +[ 245 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 245, 245 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp905", "outputIndexes": +[ 246 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 237, 246, 240 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_33_output_0", "outputIndexes": +[ 247 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 247, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_64_output_0", "outputIndexes": +[ 248 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 240 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze913", "outputIndexes": +[ 249 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 242 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze916", "outputIndexes": +[ 250 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 238, 249, 250, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice919", "outputIndexes": +[ 251 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 251 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze920", "outputIndexes": +[ 252 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 252 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp921", "outputIndexes": +[ 253 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 253, 252 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp922", "outputIndexes": +[ 254 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 237, 254, 240 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_34_output_0", "outputIndexes": +[ 255 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 255, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_65_output_0", "outputIndexes": +[ 256 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 248, 256, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_26_output_0", "outputIndexes": +[ 257 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 236, 257 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_19_output_0", "outputIndexes": +[ 258 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 258 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 251826506, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_2/Mul_1_output_0", "outputIndexes": +[ 260 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 260, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_16_output_0", "outputIndexes": +[ 261 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 260 ] +, "main_type": "NONE", "name": "/Shape_30_output_0", "outputIndexes": +[ 262 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 262 ] +, "main_type": "NONE", "name": "Shape1040", "outputIndexes": +[ 263 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 262 ] +, "main_type": "NONE", "name": "Rank1042", "outputIndexes": +[ 264 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 264, 264 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1044", "outputIndexes": +[ 265 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 265 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1045", "outputIndexes": +[ 266 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 265, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1047", "outputIndexes": +[ 267 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 267 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1048", "outputIndexes": +[ 268 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 263, 266, 268, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1051", "outputIndexes": +[ 269 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 269 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1052", "outputIndexes": +[ 270 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 270 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1053", "outputIndexes": +[ 271 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 271, 270 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1054", "outputIndexes": +[ 272 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 262, 272, 265 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_35_output_0", "outputIndexes": +[ 273 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 273, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_6_output_0", "outputIndexes": +[ 274 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 274, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_71_output_0", "outputIndexes": +[ 275 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 260, 275, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_9_output_0", "outputIndexes": +[ 276 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 276 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_4_output_0", "outputIndexes": +[ 277 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 274, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_70_output_0", "outputIndexes": +[ 278 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 260, 22, 278, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_8_output_0", "outputIndexes": +[ 279 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 277, 279 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_29_output_0", "outputIndexes": +[ 280 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 280, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_17_output_0", "outputIndexes": +[ 281 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 261, 281 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_10_output_0", "outputIndexes": +[ 282 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3456 ] +, "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": +[ 3461 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3461 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3462 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3462 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 283 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 247, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_66_output_0", "outputIndexes": +[ 284 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 255, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_67_output_0", "outputIndexes": +[ 285 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 284, 285, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_27_output_0", "outputIndexes": +[ 286 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 283, 286 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_20_output_0", "outputIndexes": +[ 287 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 287 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 254448992, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_2/Mul_1_output_0", "outputIndexes": +[ 289 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 289, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_18_output_0", "outputIndexes": +[ 290 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 289 ] +, "main_type": "NONE", "name": "/Shape_31_output_0", "outputIndexes": +[ 291 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 291 ] +, "main_type": "NONE", "name": "Shape937", "outputIndexes": +[ 292 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 291 ] +, "main_type": "NONE", "name": "Rank939", "outputIndexes": +[ 293 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 293, 293 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp941", "outputIndexes": +[ 294 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 294 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze942", "outputIndexes": +[ 295 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 294, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp944", "outputIndexes": +[ 296 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 296 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze945", "outputIndexes": +[ 297 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 292, 295, 297, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice948", "outputIndexes": +[ 298 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 298 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze949", "outputIndexes": +[ 299 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 299 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp950", "outputIndexes": +[ 300 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 300, 299 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp951", "outputIndexes": +[ 301 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 291, 301, 294 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_36_output_0", "outputIndexes": +[ 302 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 302, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_7_output_0", "outputIndexes": +[ 303 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 303, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_73_output_0", "outputIndexes": +[ 304 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 289, 304, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_11_output_0", "outputIndexes": +[ 305 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 305 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_5_output_0", "outputIndexes": +[ 306 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 303, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_72_output_0", "outputIndexes": +[ 307 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 289, 22, 307, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_10_output_0", "outputIndexes": +[ 308 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 306, 308 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_30_output_0", "outputIndexes": +[ 309 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 309, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_19_output_0", "outputIndexes": +[ 310 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 290, 310 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_11_output_0", "outputIndexes": +[ 311 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3456 ] +, "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": +[ 3465 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3465 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3466 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3466 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 312 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 247, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_68_output_0", "outputIndexes": +[ 313 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 255, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_69_output_0", "outputIndexes": +[ 314 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 313, 314, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_28_output_0", "outputIndexes": +[ 315 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 312, 315 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_21_output_0", "outputIndexes": +[ 316 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 282, 311, 316, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_26_output_0", "outputIndexes": +[ 317 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 317 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3467 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3467 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3468 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3468 ] +, "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": +[ 3469 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3469 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3470 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3470 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.2/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 318 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 234, 318 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_13_output_0", "outputIndexes": +[ 319 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 319 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 267557260, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_2/Mul_1_output_0", "outputIndexes": +[ 320 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 320 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3471 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3471 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3472 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3472 ] +, "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": +[ 3473 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3473 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3474 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3474 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 321 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 321 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_2/Mul_output_0", "outputIndexes": +[ 322 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3472 ] +, "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": +[ 3477 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3477 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3478 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3478 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 323 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 322, 323 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_2/Mul_output_0", "outputIndexes": +[ 324 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 324 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3479 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3479 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.2/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3480 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3480 ] +, "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": +[ 3481 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3481 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.2/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3482 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3482 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.2/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 325 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 319, 325 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_14_output_0", "outputIndexes": +[ 326 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 326, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_27_output_0", "outputIndexes": +[ 327 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 327 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 361961934, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_3/Mul_1_output_0", "outputIndexes": +[ 328 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 328 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3483 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3483 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3484 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3484 ] +, "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": +[ 3485 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3485 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3486 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3486 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 329 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 328 ] +, "main_type": "NONE", "name": "/Shape_42_output_0", "outputIndexes": +[ 330 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 330 ] +, "main_type": "NONE", "name": "Shape1279", "outputIndexes": +[ 331 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 330 ] +, "main_type": "NONE", "name": "Rank1281", "outputIndexes": +[ 332 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 332, 332 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1283", "outputIndexes": +[ 333 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 333 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1284", "outputIndexes": +[ 334 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 333, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1286", "outputIndexes": +[ 335 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 335 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1287", "outputIndexes": +[ 336 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 331, 334, 336, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1290", "outputIndexes": +[ 337 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 337 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1291", "outputIndexes": +[ 338 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 338, 338 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1293", "outputIndexes": +[ 339 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 330, 339, 333 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_48_output_0", "outputIndexes": +[ 340 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 340, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_96_output_0", "outputIndexes": +[ 341 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 333 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1301", "outputIndexes": +[ 342 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 335 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1304", "outputIndexes": +[ 343 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 331, 342, 343, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1307", "outputIndexes": +[ 344 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 344 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1308", "outputIndexes": +[ 345 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 345 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1309", "outputIndexes": +[ 346 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 346, 345 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1310", "outputIndexes": +[ 347 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 330, 347, 333 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_49_output_0", "outputIndexes": +[ 348 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 348, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_97_output_0", "outputIndexes": +[ 349 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 341, 349, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_39_output_0", "outputIndexes": +[ 350 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 329, 350 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_28_output_0", "outputIndexes": +[ 351 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 351 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 372480484, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_3/Mul_1_output_0", "outputIndexes": +[ 353 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 353, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_24_output_0", "outputIndexes": +[ 354 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 353 ] +, "main_type": "NONE", "name": "/Shape_44_output_0", "outputIndexes": +[ 355 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 355 ] +, "main_type": "NONE", "name": "Shape1428", "outputIndexes": +[ 356 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 355 ] +, "main_type": "NONE", "name": "Rank1430", "outputIndexes": +[ 357 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 357, 357 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1432", "outputIndexes": +[ 358 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 358 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1433", "outputIndexes": +[ 359 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 358, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1435", "outputIndexes": +[ 360 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 360 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1436", "outputIndexes": +[ 361 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 356, 359, 361, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1439", "outputIndexes": +[ 362 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 362 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1440", "outputIndexes": +[ 363 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 363 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1441", "outputIndexes": +[ 364 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 364, 363 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1442", "outputIndexes": +[ 365 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 355, 365, 358 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_50_output_0", "outputIndexes": +[ 366 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 366, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_9_output_0", "outputIndexes": +[ 367 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 367, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_103_output_0", "outputIndexes": +[ 368 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 353, 368, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_13_output_0", "outputIndexes": +[ 369 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 369 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_6_output_0", "outputIndexes": +[ 370 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 367, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_102_output_0", "outputIndexes": +[ 371 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 353, 22, 371, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_12_output_0", "outputIndexes": +[ 372 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 370, 372 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_42_output_0", "outputIndexes": +[ 373 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 373, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_25_output_0", "outputIndexes": +[ 374 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 354, 374 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_15_output_0", "outputIndexes": +[ 375 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3484 ] +, "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": +[ 3489 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3489 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3490 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3490 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 376 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 340, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_98_output_0", "outputIndexes": +[ 377 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 348, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_99_output_0", "outputIndexes": +[ 378 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 377, 378, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_40_output_0", "outputIndexes": +[ 379 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 376, 379 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_29_output_0", "outputIndexes": +[ 380 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 380 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 375102970, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_3/Mul_1_output_0", "outputIndexes": +[ 382 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 382, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_26_output_0", "outputIndexes": +[ 383 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 382 ] +, "main_type": "NONE", "name": "/Shape_45_output_0", "outputIndexes": +[ 384 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 384 ] +, "main_type": "NONE", "name": "Shape1325", "outputIndexes": +[ 385 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 384 ] +, "main_type": "NONE", "name": "Rank1327", "outputIndexes": +[ 386 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 386, 386 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1329", "outputIndexes": +[ 387 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 387 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1330", "outputIndexes": +[ 388 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 387, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1332", "outputIndexes": +[ 389 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 389 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1333", "outputIndexes": +[ 390 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 385, 388, 390, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1336", "outputIndexes": +[ 391 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 391 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1337", "outputIndexes": +[ 392 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 392 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1338", "outputIndexes": +[ 393 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 393, 392 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1339", "outputIndexes": +[ 394 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 384, 394, 387 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_51_output_0", "outputIndexes": +[ 395 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 395, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_10_output_0", "outputIndexes": +[ 396 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 396, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_105_output_0", "outputIndexes": +[ 397 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 382, 397, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_15_output_0", "outputIndexes": +[ 398 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 398 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_7_output_0", "outputIndexes": +[ 399 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 396, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_104_output_0", "outputIndexes": +[ 400 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 382, 22, 400, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_14_output_0", "outputIndexes": +[ 401 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 399, 401 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_43_output_0", "outputIndexes": +[ 402 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 402, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_27_output_0", "outputIndexes": +[ 403 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 383, 403 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_16_output_0", "outputIndexes": +[ 404 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3484 ] +, "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": +[ 3493 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3493 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3494 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3494 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 405 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 340, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_100_output_0", "outputIndexes": +[ 406 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 348, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_101_output_0", "outputIndexes": +[ 407 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 406, 407, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_41_output_0", "outputIndexes": +[ 408 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 405, 408 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_30_output_0", "outputIndexes": +[ 409 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 375, 404, 409, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_35_output_0", "outputIndexes": +[ 410 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 410 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3495 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3495 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3496 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3496 ] +, "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": +[ 3497 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3497 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3498 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3498 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.3/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 411 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 327, 411 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_18_output_0", "outputIndexes": +[ 412 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 412 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 388211238, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_3/Mul_1_output_0", "outputIndexes": +[ 413 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 413 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3499 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3499 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3500 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3500 ] +, "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": +[ 3501 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3501 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3502 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3502 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 414 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 414 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_3/Mul_output_0", "outputIndexes": +[ 415 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3500 ] +, "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": +[ 3505 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3505 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3506 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3506 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 416 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 415, 416 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_3/Mul_output_0", "outputIndexes": +[ 417 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 417 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3507 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3507 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.3/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3508 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3508 ] +, "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": +[ 3509 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3509 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.3/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3510 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3510 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.3/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 418 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 412, 418 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_19_output_0", "outputIndexes": +[ 419 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 419, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_36_output_0", "outputIndexes": +[ 420 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 420 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 482615912, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_4/Mul_1_output_0", "outputIndexes": +[ 421 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 421 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3511 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3511 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3512 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3512 ] +, "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": +[ 3513 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3513 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3514 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3514 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 422 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 421 ] +, "main_type": "NONE", "name": "/Shape_56_output_0", "outputIndexes": +[ 423 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 423 ] +, "main_type": "NONE", "name": "Shape1667", "outputIndexes": +[ 424 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 423 ] +, "main_type": "NONE", "name": "Rank1669", "outputIndexes": +[ 425 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 425, 425 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1671", "outputIndexes": +[ 426 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 426 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1672", "outputIndexes": +[ 427 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 426, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1674", "outputIndexes": +[ 428 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 428 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1675", "outputIndexes": +[ 429 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 424, 427, 429, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1678", "outputIndexes": +[ 430 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 430 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1679", "outputIndexes": +[ 431 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 431, 431 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1681", "outputIndexes": +[ 432 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 423, 432, 426 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_63_output_0", "outputIndexes": +[ 433 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 433, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_128_output_0", "outputIndexes": +[ 434 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 426 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1689", "outputIndexes": +[ 435 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 428 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1692", "outputIndexes": +[ 436 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 424, 435, 436, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1695", "outputIndexes": +[ 437 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 437 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1696", "outputIndexes": +[ 438 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 438 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1697", "outputIndexes": +[ 439 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 439, 438 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1698", "outputIndexes": +[ 440 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 423, 440, 426 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_64_output_0", "outputIndexes": +[ 441 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 441, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_129_output_0", "outputIndexes": +[ 442 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 434, 442, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_52_output_0", "outputIndexes": +[ 443 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 422, 443 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_37_output_0", "outputIndexes": +[ 444 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 444 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 493134462, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_4/Mul_1_output_0", "outputIndexes": +[ 446 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 446, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_32_output_0", "outputIndexes": +[ 447 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 446 ] +, "main_type": "NONE", "name": "/Shape_58_output_0", "outputIndexes": +[ 448 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 448 ] +, "main_type": "NONE", "name": "Shape1817", "outputIndexes": +[ 449 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 448 ] +, "main_type": "NONE", "name": "Rank1819", "outputIndexes": +[ 450 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 450, 450 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1821", "outputIndexes": +[ 451 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 451 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1822", "outputIndexes": +[ 452 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 451, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1824", "outputIndexes": +[ 453 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 453 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1825", "outputIndexes": +[ 454 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 449, 452, 454, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1828", "outputIndexes": +[ 455 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 455 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1829", "outputIndexes": +[ 456 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 456 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1830", "outputIndexes": +[ 457 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 457, 456 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1831", "outputIndexes": +[ 458 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 448, 458, 451 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_65_output_0", "outputIndexes": +[ 459 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 459, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_12_output_0", "outputIndexes": +[ 460 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 460, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_135_output_0", "outputIndexes": +[ 461 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 446, 461, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_17_output_0", "outputIndexes": +[ 462 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 462 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_8_output_0", "outputIndexes": +[ 463 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 460, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_134_output_0", "outputIndexes": +[ 464 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 446, 22, 464, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_16_output_0", "outputIndexes": +[ 465 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 463, 465 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_55_output_0", "outputIndexes": +[ 466 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 466, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_33_output_0", "outputIndexes": +[ 467 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 447, 467 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_20_output_0", "outputIndexes": +[ 468 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3512 ] +, "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": +[ 3517 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3517 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3518 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3518 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 469 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 433, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_130_output_0", "outputIndexes": +[ 470 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 441, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_131_output_0", "outputIndexes": +[ 471 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 470, 471, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_53_output_0", "outputIndexes": +[ 472 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 469, 472 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_38_output_0", "outputIndexes": +[ 473 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 473 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 495756948, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_4/Mul_1_output_0", "outputIndexes": +[ 475 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 475, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_34_output_0", "outputIndexes": +[ 476 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 475 ] +, "main_type": "NONE", "name": "/Shape_59_output_0", "outputIndexes": +[ 477 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 477 ] +, "main_type": "NONE", "name": "Shape1713", "outputIndexes": +[ 478 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 477 ] +, "main_type": "NONE", "name": "Rank1715", "outputIndexes": +[ 479 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 479, 479 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1717", "outputIndexes": +[ 480 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 480 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1718", "outputIndexes": +[ 481 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 480, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1720", "outputIndexes": +[ 482 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 482 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze1721", "outputIndexes": +[ 483 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 478, 481, 483, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice1724", "outputIndexes": +[ 484 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 484 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze1725", "outputIndexes": +[ 485 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 485 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1726", "outputIndexes": +[ 486 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 486, 485 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp1727", "outputIndexes": +[ 487 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 477, 487, 480 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_66_output_0", "outputIndexes": +[ 488 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 488, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_13_output_0", "outputIndexes": +[ 489 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 489, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_137_output_0", "outputIndexes": +[ 490 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 475, 490, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_19_output_0", "outputIndexes": +[ 491 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 491 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_9_output_0", "outputIndexes": +[ 492 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 489, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_136_output_0", "outputIndexes": +[ 493 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 475, 22, 493, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_18_output_0", "outputIndexes": +[ 494 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 492, 494 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_56_output_0", "outputIndexes": +[ 495 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 495, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_35_output_0", "outputIndexes": +[ 496 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 476, 496 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_21_output_0", "outputIndexes": +[ 497 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3512 ] +, "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": +[ 3521 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3521 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3522 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3522 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 498 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 433, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_132_output_0", "outputIndexes": +[ 499 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 441, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_133_output_0", "outputIndexes": +[ 500 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 499, 500, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_54_output_0", "outputIndexes": +[ 501 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 498, 501 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_39_output_0", "outputIndexes": +[ 502 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 468, 497, 502, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_44_output_0", "outputIndexes": +[ 503 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 503 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3523 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3523 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3524 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3524 ] +, "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": +[ 3525 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3525 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3526 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3526 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.4/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 504 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 420, 504 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_23_output_0", "outputIndexes": +[ 505 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 505 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 508865216, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_4/Mul_1_output_0", "outputIndexes": +[ 506 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 506 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3527 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3527 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3528 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3528 ] +, "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": +[ 3529 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3529 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3530 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3530 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 507 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 507 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_4/Mul_output_0", "outputIndexes": +[ 508 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3528 ] +, "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": +[ 3533 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3533 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3534 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3534 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 509 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 508, 509 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_4/Mul_output_0", "outputIndexes": +[ 510 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 510 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3535 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3535 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.4/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3536 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3536 ] +, "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": +[ 3537 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3537 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.4/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3538 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3538 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.4/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 511 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 505, 511 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_24_output_0", "outputIndexes": +[ 512 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 512, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_45_output_0", "outputIndexes": +[ 513 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 513 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 603269890, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_5/Mul_1_output_0", "outputIndexes": +[ 514 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 514 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3539 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3539 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3540 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3540 ] +, "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": +[ 3541 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3541 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3542 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3542 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 515 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 514 ] +, "main_type": "NONE", "name": "/Shape_70_output_0", "outputIndexes": +[ 516 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 516 ] +, "main_type": "NONE", "name": "Shape2056", "outputIndexes": +[ 517 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 516 ] +, "main_type": "NONE", "name": "Rank2058", "outputIndexes": +[ 518 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 518, 518 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2060", "outputIndexes": +[ 519 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 519 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2061", "outputIndexes": +[ 520 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 519, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2063", "outputIndexes": +[ 521 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 521 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2064", "outputIndexes": +[ 522 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 517, 520, 522, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2067", "outputIndexes": +[ 523 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 523 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2068", "outputIndexes": +[ 524 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 524, 524 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2070", "outputIndexes": +[ 525 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 516, 525, 519 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_78_output_0", "outputIndexes": +[ 526 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 526, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_160_output_0", "outputIndexes": +[ 527 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 519 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2078", "outputIndexes": +[ 528 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 521 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2081", "outputIndexes": +[ 529 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 517, 528, 529, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2084", "outputIndexes": +[ 530 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 530 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2085", "outputIndexes": +[ 531 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 531 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2086", "outputIndexes": +[ 532 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 532, 531 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2087", "outputIndexes": +[ 533 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 516, 533, 519 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_79_output_0", "outputIndexes": +[ 534 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 534, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_161_output_0", "outputIndexes": +[ 535 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 527, 535, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_65_output_0", "outputIndexes": +[ 536 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 515, 536 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_46_output_0", "outputIndexes": +[ 537 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 537 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 613788440, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_5/Mul_1_output_0", "outputIndexes": +[ 539 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 539, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_40_output_0", "outputIndexes": +[ 540 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 539 ] +, "main_type": "NONE", "name": "/Shape_72_output_0", "outputIndexes": +[ 541 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 541 ] +, "main_type": "NONE", "name": "Shape2206", "outputIndexes": +[ 542 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 541 ] +, "main_type": "NONE", "name": "Rank2208", "outputIndexes": +[ 543 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 543, 543 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2210", "outputIndexes": +[ 544 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 544 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2211", "outputIndexes": +[ 545 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 544, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2213", "outputIndexes": +[ 546 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 546 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2214", "outputIndexes": +[ 547 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 542, 545, 547, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2217", "outputIndexes": +[ 548 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 548 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2218", "outputIndexes": +[ 549 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 549 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2219", "outputIndexes": +[ 550 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 550, 549 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2220", "outputIndexes": +[ 551 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 541, 551, 544 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_80_output_0", "outputIndexes": +[ 552 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 552, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_15_output_0", "outputIndexes": +[ 553 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 553, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_167_output_0", "outputIndexes": +[ 554 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 539, 554, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_21_output_0", "outputIndexes": +[ 555 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 555 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_10_output_0", "outputIndexes": +[ 556 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 553, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_166_output_0", "outputIndexes": +[ 557 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 539, 22, 557, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_20_output_0", "outputIndexes": +[ 558 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 556, 558 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_68_output_0", "outputIndexes": +[ 559 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 559, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_41_output_0", "outputIndexes": +[ 560 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 540, 560 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_25_output_0", "outputIndexes": +[ 561 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3540 ] +, "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": +[ 3545 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3545 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3546 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3546 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 562 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 526, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_162_output_0", "outputIndexes": +[ 563 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 534, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_163_output_0", "outputIndexes": +[ 564 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 563, 564, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_66_output_0", "outputIndexes": +[ 565 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 562, 565 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_47_output_0", "outputIndexes": +[ 566 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 566 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 616410926, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_5/Mul_1_output_0", "outputIndexes": +[ 568 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 568, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_42_output_0", "outputIndexes": +[ 569 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 568 ] +, "main_type": "NONE", "name": "/Shape_73_output_0", "outputIndexes": +[ 570 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 570 ] +, "main_type": "NONE", "name": "Shape2102", "outputIndexes": +[ 571 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 570 ] +, "main_type": "NONE", "name": "Rank2104", "outputIndexes": +[ 572 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 572, 572 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2106", "outputIndexes": +[ 573 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 573 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2107", "outputIndexes": +[ 574 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 573, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2109", "outputIndexes": +[ 575 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 575 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2110", "outputIndexes": +[ 576 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 571, 574, 576, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2113", "outputIndexes": +[ 577 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 577 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2114", "outputIndexes": +[ 578 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 578 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2115", "outputIndexes": +[ 579 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 579, 578 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2116", "outputIndexes": +[ 580 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 570, 580, 573 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_81_output_0", "outputIndexes": +[ 581 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 581, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_16_output_0", "outputIndexes": +[ 582 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 582, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_169_output_0", "outputIndexes": +[ 583 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 568, 583, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_23_output_0", "outputIndexes": +[ 584 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 584 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_11_output_0", "outputIndexes": +[ 585 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 582, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_168_output_0", "outputIndexes": +[ 586 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 568, 22, 586, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_22_output_0", "outputIndexes": +[ 587 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 585, 587 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_69_output_0", "outputIndexes": +[ 588 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 588, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_43_output_0", "outputIndexes": +[ 589 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 569, 589 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_26_output_0", "outputIndexes": +[ 590 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3540 ] +, "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": +[ 3549 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3549 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3550 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3550 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 591 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 526, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_164_output_0", "outputIndexes": +[ 592 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 534, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_165_output_0", "outputIndexes": +[ 593 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 592, 593, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_67_output_0", "outputIndexes": +[ 594 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 591, 594 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_48_output_0", "outputIndexes": +[ 595 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 561, 590, 595, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_53_output_0", "outputIndexes": +[ 596 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 596 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3551 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3551 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3552 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3552 ] +, "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": +[ 3553 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3553 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3554 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3554 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.5/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 597 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 513, 597 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_28_output_0", "outputIndexes": +[ 598 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 598 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 629519194, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_5/Mul_1_output_0", "outputIndexes": +[ 599 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 599 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3555 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3555 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3556 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3556 ] +, "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": +[ 3557 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3557 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3558 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3558 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 600 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 600 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_5/Mul_output_0", "outputIndexes": +[ 601 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3556 ] +, "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": +[ 3561 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3561 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3562 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3562 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 602 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 601, 602 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_5/Mul_output_0", "outputIndexes": +[ 603 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 603 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3563 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3563 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.5/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3564 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3564 ] +, "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": +[ 3565 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3565 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.5/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3566 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3566 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.5/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 604 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 598, 604 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_29_output_0", "outputIndexes": +[ 605 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 605, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_54_output_0", "outputIndexes": +[ 606 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 606 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 723923868, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_6/Mul_1_output_0", "outputIndexes": +[ 607 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 607 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3567 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3567 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3568 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3568 ] +, "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": +[ 3569 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3569 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3570 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3570 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 608 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 607 ] +, "main_type": "NONE", "name": "/Shape_84_output_0", "outputIndexes": +[ 609 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 609 ] +, "main_type": "NONE", "name": "Shape2445", "outputIndexes": +[ 610 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 609 ] +, "main_type": "NONE", "name": "Rank2447", "outputIndexes": +[ 611 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 611, 611 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2449", "outputIndexes": +[ 612 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 612 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2450", "outputIndexes": +[ 613 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 612, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2452", "outputIndexes": +[ 614 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 614 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2453", "outputIndexes": +[ 615 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 610, 613, 615, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2456", "outputIndexes": +[ 616 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 616 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2457", "outputIndexes": +[ 617 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 617, 617 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2459", "outputIndexes": +[ 618 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 609, 618, 612 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_93_output_0", "outputIndexes": +[ 619 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 619, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_192_output_0", "outputIndexes": +[ 620 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 612 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2467", "outputIndexes": +[ 621 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 614 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2470", "outputIndexes": +[ 622 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 610, 621, 622, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2473", "outputIndexes": +[ 623 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 623 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2474", "outputIndexes": +[ 624 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 624 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2475", "outputIndexes": +[ 625 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 625, 624 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2476", "outputIndexes": +[ 626 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 609, 626, 612 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_94_output_0", "outputIndexes": +[ 627 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 627, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_193_output_0", "outputIndexes": +[ 628 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 620, 628, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_78_output_0", "outputIndexes": +[ 629 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 608, 629 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_55_output_0", "outputIndexes": +[ 630 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 630 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 734442418, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_6/Mul_1_output_0", "outputIndexes": +[ 632 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 632, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_48_output_0", "outputIndexes": +[ 633 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 632 ] +, "main_type": "NONE", "name": "/Shape_86_output_0", "outputIndexes": +[ 634 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 634 ] +, "main_type": "NONE", "name": "Shape2595", "outputIndexes": +[ 635 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 634 ] +, "main_type": "NONE", "name": "Rank2597", "outputIndexes": +[ 636 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 636, 636 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2599", "outputIndexes": +[ 637 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 637 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2600", "outputIndexes": +[ 638 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 637, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2602", "outputIndexes": +[ 639 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 639 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2603", "outputIndexes": +[ 640 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 635, 638, 640, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2606", "outputIndexes": +[ 641 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 641 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2607", "outputIndexes": +[ 642 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 642 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2608", "outputIndexes": +[ 643 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 643, 642 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2609", "outputIndexes": +[ 644 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 634, 644, 637 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_95_output_0", "outputIndexes": +[ 645 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 645, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_18_output_0", "outputIndexes": +[ 646 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 646, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_199_output_0", "outputIndexes": +[ 647 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 632, 647, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_25_output_0", "outputIndexes": +[ 648 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 648 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_12_output_0", "outputIndexes": +[ 649 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 646, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_198_output_0", "outputIndexes": +[ 650 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 632, 22, 650, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_24_output_0", "outputIndexes": +[ 651 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 649, 651 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_81_output_0", "outputIndexes": +[ 652 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 652, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_49_output_0", "outputIndexes": +[ 653 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 633, 653 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_30_output_0", "outputIndexes": +[ 654 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3568 ] +, "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": +[ 3573 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3573 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3574 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3574 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 655 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 619, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_194_output_0", "outputIndexes": +[ 656 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 627, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_195_output_0", "outputIndexes": +[ 657 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 656, 657, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_79_output_0", "outputIndexes": +[ 658 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 655, 658 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_56_output_0", "outputIndexes": +[ 659 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 659 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 737064904, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_6/Mul_1_output_0", "outputIndexes": +[ 661 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 661, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_50_output_0", "outputIndexes": +[ 662 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 661 ] +, "main_type": "NONE", "name": "/Shape_87_output_0", "outputIndexes": +[ 663 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 663 ] +, "main_type": "NONE", "name": "Shape2491", "outputIndexes": +[ 664 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 663 ] +, "main_type": "NONE", "name": "Rank2493", "outputIndexes": +[ 665 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 665, 665 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2495", "outputIndexes": +[ 666 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 666 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2496", "outputIndexes": +[ 667 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 666, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2498", "outputIndexes": +[ 668 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 668 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2499", "outputIndexes": +[ 669 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 664, 667, 669, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2502", "outputIndexes": +[ 670 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 670 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2503", "outputIndexes": +[ 671 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 671 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2504", "outputIndexes": +[ 672 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 672, 671 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2505", "outputIndexes": +[ 673 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 663, 673, 666 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_96_output_0", "outputIndexes": +[ 674 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 674, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_19_output_0", "outputIndexes": +[ 675 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 675, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_201_output_0", "outputIndexes": +[ 676 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 661, 676, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_27_output_0", "outputIndexes": +[ 677 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 677 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_13_output_0", "outputIndexes": +[ 678 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 675, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_200_output_0", "outputIndexes": +[ 679 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 661, 22, 679, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_26_output_0", "outputIndexes": +[ 680 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 678, 680 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_82_output_0", "outputIndexes": +[ 681 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 681, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_51_output_0", "outputIndexes": +[ 682 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 662, 682 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_31_output_0", "outputIndexes": +[ 683 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3568 ] +, "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": +[ 3577 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3577 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3578 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3578 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 684 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 619, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_196_output_0", "outputIndexes": +[ 685 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 627, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_197_output_0", "outputIndexes": +[ 686 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 685, 686, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_80_output_0", "outputIndexes": +[ 687 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 684, 687 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_57_output_0", "outputIndexes": +[ 688 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 654, 683, 688, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_62_output_0", "outputIndexes": +[ 689 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 689 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3579 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3579 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3580 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3580 ] +, "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": +[ 3581 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3581 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3582 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3582 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.6/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 690 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 606, 690 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_33_output_0", "outputIndexes": +[ 691 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 691 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 750173172, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_6/Mul_1_output_0", "outputIndexes": +[ 692 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 692 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3583 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3583 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3584 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3584 ] +, "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": +[ 3585 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3585 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3586 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3586 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 693 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 693 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_6/Mul_output_0", "outputIndexes": +[ 694 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3584 ] +, "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": +[ 3589 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3589 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3590 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3590 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 695 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 694, 695 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_6/Mul_output_0", "outputIndexes": +[ 696 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 696 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3591 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3591 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.6/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3592 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3592 ] +, "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": +[ 3593 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3593 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.6/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3594 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3594 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.6/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 697 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 691, 697 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_34_output_0", "outputIndexes": +[ 698 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 698, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_63_output_0", "outputIndexes": +[ 699 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 699 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 844577846, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_7/Mul_1_output_0", "outputIndexes": +[ 700 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 700 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3595 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3595 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3596 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3596 ] +, "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": +[ 3597 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3597 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3598 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3598 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 701 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 700 ] +, "main_type": "NONE", "name": "/Shape_98_output_0", "outputIndexes": +[ 702 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 702 ] +, "main_type": "NONE", "name": "Shape2834", "outputIndexes": +[ 703 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 702 ] +, "main_type": "NONE", "name": "Rank2836", "outputIndexes": +[ 704 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 704, 704 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2838", "outputIndexes": +[ 705 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 705 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2839", "outputIndexes": +[ 706 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 705, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2841", "outputIndexes": +[ 707 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 707 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2842", "outputIndexes": +[ 708 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 703, 706, 708, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2845", "outputIndexes": +[ 709 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 709 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2846", "outputIndexes": +[ 710 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 710, 710 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2848", "outputIndexes": +[ 711 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 702, 711, 705 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_108_output_0", "outputIndexes": +[ 712 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 712, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_224_output_0", "outputIndexes": +[ 713 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 705 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2856", "outputIndexes": +[ 714 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 707 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2859", "outputIndexes": +[ 715 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 703, 714, 715, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2862", "outputIndexes": +[ 716 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 716 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2863", "outputIndexes": +[ 717 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 717 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2864", "outputIndexes": +[ 718 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 718, 717 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2865", "outputIndexes": +[ 719 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 702, 719, 705 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_109_output_0", "outputIndexes": +[ 720 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 720, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_225_output_0", "outputIndexes": +[ 721 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 713, 721, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_91_output_0", "outputIndexes": +[ 722 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 701, 722 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_64_output_0", "outputIndexes": +[ 723 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 723 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 855096396, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_7/Mul_1_output_0", "outputIndexes": +[ 725 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 725, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_56_output_0", "outputIndexes": +[ 726 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 725 ] +, "main_type": "NONE", "name": "/Shape_100_output_0", "outputIndexes": +[ 727 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 727 ] +, "main_type": "NONE", "name": "Shape2984", "outputIndexes": +[ 728 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 727 ] +, "main_type": "NONE", "name": "Rank2986", "outputIndexes": +[ 729 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 729, 729 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2988", "outputIndexes": +[ 730 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 730 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2989", "outputIndexes": +[ 731 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 730, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2991", "outputIndexes": +[ 732 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 732 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2992", "outputIndexes": +[ 733 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 728, 731, 733, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2995", "outputIndexes": +[ 734 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 734 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2996", "outputIndexes": +[ 735 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 735 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2997", "outputIndexes": +[ 736 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 736, 735 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2998", "outputIndexes": +[ 737 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 727, 737, 730 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_110_output_0", "outputIndexes": +[ 738 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 738, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_21_output_0", "outputIndexes": +[ 739 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 739, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_231_output_0", "outputIndexes": +[ 740 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 725, 740, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_29_output_0", "outputIndexes": +[ 741 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 741 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_14_output_0", "outputIndexes": +[ 742 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 739, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_230_output_0", "outputIndexes": +[ 743 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 725, 22, 743, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_28_output_0", "outputIndexes": +[ 744 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 742, 744 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_94_output_0", "outputIndexes": +[ 745 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 745, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_57_output_0", "outputIndexes": +[ 746 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 726, 746 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_35_output_0", "outputIndexes": +[ 747 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3596 ] +, "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": +[ 3601 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3601 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3602 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3602 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 748 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 712, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_226_output_0", "outputIndexes": +[ 749 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 720, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_227_output_0", "outputIndexes": +[ 750 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 749, 750, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_92_output_0", "outputIndexes": +[ 751 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 748, 751 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_65_output_0", "outputIndexes": +[ 752 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 752 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 857718882, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_7/Mul_1_output_0", "outputIndexes": +[ 754 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 754, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_58_output_0", "outputIndexes": +[ 755 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 754 ] +, "main_type": "NONE", "name": "/Shape_101_output_0", "outputIndexes": +[ 756 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 756 ] +, "main_type": "NONE", "name": "Shape2880", "outputIndexes": +[ 757 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 756 ] +, "main_type": "NONE", "name": "Rank2882", "outputIndexes": +[ 758 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 758, 758 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2884", "outputIndexes": +[ 759 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 759 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2885", "outputIndexes": +[ 760 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 759, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2887", "outputIndexes": +[ 761 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 761 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze2888", "outputIndexes": +[ 762 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 757, 760, 762, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice2891", "outputIndexes": +[ 763 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 763 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze2892", "outputIndexes": +[ 764 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 764 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2893", "outputIndexes": +[ 765 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 765, 764 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp2894", "outputIndexes": +[ 766 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 756, 766, 759 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_111_output_0", "outputIndexes": +[ 767 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 767, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_22_output_0", "outputIndexes": +[ 768 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 768, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_233_output_0", "outputIndexes": +[ 769 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 754, 769, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_31_output_0", "outputIndexes": +[ 770 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 770 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_15_output_0", "outputIndexes": +[ 771 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 768, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_232_output_0", "outputIndexes": +[ 772 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 754, 22, 772, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_30_output_0", "outputIndexes": +[ 773 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 771, 773 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_95_output_0", "outputIndexes": +[ 774 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 774, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_59_output_0", "outputIndexes": +[ 775 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 755, 775 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_36_output_0", "outputIndexes": +[ 776 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3596 ] +, "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": +[ 3605 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3605 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3606 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3606 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 777 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 712, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_228_output_0", "outputIndexes": +[ 778 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 720, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_229_output_0", "outputIndexes": +[ 779 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 778, 779, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_93_output_0", "outputIndexes": +[ 780 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 777, 780 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_66_output_0", "outputIndexes": +[ 781 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 747, 776, 781, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_71_output_0", "outputIndexes": +[ 782 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 782 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3607 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3607 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3608 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3608 ] +, "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": +[ 3609 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3609 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3610 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3610 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.7/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 783 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 699, 783 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_38_output_0", "outputIndexes": +[ 784 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 784 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 870827150, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_7/Mul_1_output_0", "outputIndexes": +[ 785 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 785 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3611 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3611 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3612 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3612 ] +, "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": +[ 3613 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3613 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3614 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3614 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 786 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 786 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_7/Mul_output_0", "outputIndexes": +[ 787 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3612 ] +, "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": +[ 3617 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3617 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3618 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3618 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 788 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 787, 788 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_7/Mul_output_0", "outputIndexes": +[ 789 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 789 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3619 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3619 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.7/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3620 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3620 ] +, "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": +[ 3621 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3621 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.7/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3622 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3622 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.7/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 790 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 784, 790 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_39_output_0", "outputIndexes": +[ 791 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 791, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_72_output_0", "outputIndexes": +[ 792 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 792 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 965231824, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_8/Mul_1_output_0", "outputIndexes": +[ 793 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 793 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3623 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3623 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3624 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3624 ] +, "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": +[ 3625 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3625 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3626 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3626 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 794 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 793 ] +, "main_type": "NONE", "name": "/Shape_112_output_0", "outputIndexes": +[ 795 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 795 ] +, "main_type": "NONE", "name": "Shape3223", "outputIndexes": +[ 796 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 795 ] +, "main_type": "NONE", "name": "Rank3225", "outputIndexes": +[ 797 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 797, 797 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3227", "outputIndexes": +[ 798 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 798 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3228", "outputIndexes": +[ 799 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 798, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3230", "outputIndexes": +[ 800 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 800 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3231", "outputIndexes": +[ 801 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 796, 799, 801, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3234", "outputIndexes": +[ 802 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 802 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3235", "outputIndexes": +[ 803 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 803, 803 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3237", "outputIndexes": +[ 804 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 795, 804, 798 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_123_output_0", "outputIndexes": +[ 805 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 805, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_256_output_0", "outputIndexes": +[ 806 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 798 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3245", "outputIndexes": +[ 807 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 800 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3248", "outputIndexes": +[ 808 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 796, 807, 808, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3251", "outputIndexes": +[ 809 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 809 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3252", "outputIndexes": +[ 810 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 810 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3253", "outputIndexes": +[ 811 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 811, 810 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3254", "outputIndexes": +[ 812 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 795, 812, 798 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_124_output_0", "outputIndexes": +[ 813 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 813, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_257_output_0", "outputIndexes": +[ 814 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 806, 814, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_104_output_0", "outputIndexes": +[ 815 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 794, 815 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_73_output_0", "outputIndexes": +[ 816 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 816 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 975750374, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_8/Mul_1_output_0", "outputIndexes": +[ 818 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 818, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_64_output_0", "outputIndexes": +[ 819 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 818 ] +, "main_type": "NONE", "name": "/Shape_114_output_0", "outputIndexes": +[ 820 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 820 ] +, "main_type": "NONE", "name": "Shape3373", "outputIndexes": +[ 821 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 820 ] +, "main_type": "NONE", "name": "Rank3375", "outputIndexes": +[ 822 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 822, 822 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3377", "outputIndexes": +[ 823 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 823 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3378", "outputIndexes": +[ 824 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 823, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3380", "outputIndexes": +[ 825 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 825 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3381", "outputIndexes": +[ 826 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 821, 824, 826, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3384", "outputIndexes": +[ 827 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 827 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3385", "outputIndexes": +[ 828 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 828 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3386", "outputIndexes": +[ 829 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 829, 828 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3387", "outputIndexes": +[ 830 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 820, 830, 823 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_125_output_0", "outputIndexes": +[ 831 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 831, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_24_output_0", "outputIndexes": +[ 832 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 832, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_263_output_0", "outputIndexes": +[ 833 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 818, 833, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_33_output_0", "outputIndexes": +[ 834 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 834 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_16_output_0", "outputIndexes": +[ 835 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 832, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_262_output_0", "outputIndexes": +[ 836 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 818, 22, 836, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_32_output_0", "outputIndexes": +[ 837 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 835, 837 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_107_output_0", "outputIndexes": +[ 838 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 838, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_65_output_0", "outputIndexes": +[ 839 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 819, 839 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_40_output_0", "outputIndexes": +[ 840 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3624 ] +, "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": +[ 3629 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3629 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3630 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3630 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 841 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 805, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_258_output_0", "outputIndexes": +[ 842 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 813, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_259_output_0", "outputIndexes": +[ 843 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 842, 843, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_105_output_0", "outputIndexes": +[ 844 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 841, 844 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_74_output_0", "outputIndexes": +[ 845 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 845 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 978372860, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_8/Mul_1_output_0", "outputIndexes": +[ 847 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 847, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_66_output_0", "outputIndexes": +[ 848 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 847 ] +, "main_type": "NONE", "name": "/Shape_115_output_0", "outputIndexes": +[ 849 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 849 ] +, "main_type": "NONE", "name": "Shape3269", "outputIndexes": +[ 850 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 849 ] +, "main_type": "NONE", "name": "Rank3271", "outputIndexes": +[ 851 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 851, 851 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3273", "outputIndexes": +[ 852 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 852 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3274", "outputIndexes": +[ 853 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 852, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3276", "outputIndexes": +[ 854 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 854 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3277", "outputIndexes": +[ 855 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 850, 853, 855, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3280", "outputIndexes": +[ 856 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 856 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3281", "outputIndexes": +[ 857 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 857 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3282", "outputIndexes": +[ 858 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 858, 857 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3283", "outputIndexes": +[ 859 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 849, 859, 852 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_126_output_0", "outputIndexes": +[ 860 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 860, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_25_output_0", "outputIndexes": +[ 861 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 861, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_265_output_0", "outputIndexes": +[ 862 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 847, 862, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_35_output_0", "outputIndexes": +[ 863 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 863 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_17_output_0", "outputIndexes": +[ 864 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 861, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_264_output_0", "outputIndexes": +[ 865 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 847, 22, 865, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_34_output_0", "outputIndexes": +[ 866 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 864, 866 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_108_output_0", "outputIndexes": +[ 867 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 867, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_67_output_0", "outputIndexes": +[ 868 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 848, 868 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_41_output_0", "outputIndexes": +[ 869 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3624 ] +, "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": +[ 3633 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3633 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3634 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3634 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 870 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 805, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_260_output_0", "outputIndexes": +[ 871 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 813, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_261_output_0", "outputIndexes": +[ 872 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 871, 872, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_106_output_0", "outputIndexes": +[ 873 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 870, 873 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_75_output_0", "outputIndexes": +[ 874 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 840, 869, 874, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_80_output_0", "outputIndexes": +[ 875 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 875 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3635 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3635 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3636 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3636 ] +, "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": +[ 3637 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3637 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3638 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3638 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.8/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 876 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 792, 876 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_43_output_0", "outputIndexes": +[ 877 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 877 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 991481128, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_8/Mul_1_output_0", "outputIndexes": +[ 878 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 878 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3639 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3639 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3640 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3640 ] +, "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": +[ 3641 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3641 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3642 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3642 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 879 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 879 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_8/Mul_output_0", "outputIndexes": +[ 880 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3640 ] +, "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": +[ 3645 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3645 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3646 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3646 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 881 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 880, 881 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_8/Mul_output_0", "outputIndexes": +[ 882 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 882 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3647 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3647 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.8/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3648 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3648 ] +, "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": +[ 3649 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3649 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.8/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3650 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3650 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.8/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 883 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 877, 883 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_44_output_0", "outputIndexes": +[ 884 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 884, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_81_output_0", "outputIndexes": +[ 885 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 885 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1085885802, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_9/Mul_1_output_0", "outputIndexes": +[ 886 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 886 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3651 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3651 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3652 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3652 ] +, "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": +[ 3653 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3653 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3654 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3654 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 887 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 886 ] +, "main_type": "NONE", "name": "/Shape_126_output_0", "outputIndexes": +[ 888 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 888 ] +, "main_type": "NONE", "name": "Shape3612", "outputIndexes": +[ 889 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 888 ] +, "main_type": "NONE", "name": "Rank3614", "outputIndexes": +[ 890 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 890, 890 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3616", "outputIndexes": +[ 891 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 891 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3617", "outputIndexes": +[ 892 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 891, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3619", "outputIndexes": +[ 893 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 893 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3620", "outputIndexes": +[ 894 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 889, 892, 894, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3623", "outputIndexes": +[ 895 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 895 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3624", "outputIndexes": +[ 896 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 896, 896 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3626", "outputIndexes": +[ 897 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 888, 897, 891 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_138_output_0", "outputIndexes": +[ 898 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 898, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_288_output_0", "outputIndexes": +[ 899 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 891 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3634", "outputIndexes": +[ 900 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 893 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3637", "outputIndexes": +[ 901 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 889, 900, 901, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3640", "outputIndexes": +[ 902 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 902 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3641", "outputIndexes": +[ 903 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 903 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3642", "outputIndexes": +[ 904 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 904, 903 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3643", "outputIndexes": +[ 905 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 888, 905, 891 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_139_output_0", "outputIndexes": +[ 906 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 906, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_289_output_0", "outputIndexes": +[ 907 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 899, 907, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_117_output_0", "outputIndexes": +[ 908 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 887, 908 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_82_output_0", "outputIndexes": +[ 909 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 909 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1096404352, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_9/Mul_1_output_0", "outputIndexes": +[ 911 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 911, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_72_output_0", "outputIndexes": +[ 912 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 911 ] +, "main_type": "NONE", "name": "/Shape_128_output_0", "outputIndexes": +[ 913 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 913 ] +, "main_type": "NONE", "name": "Shape3762", "outputIndexes": +[ 914 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 913 ] +, "main_type": "NONE", "name": "Rank3764", "outputIndexes": +[ 915 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 915, 915 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3766", "outputIndexes": +[ 916 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 916 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3767", "outputIndexes": +[ 917 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 916, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3769", "outputIndexes": +[ 918 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 918 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3770", "outputIndexes": +[ 919 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 914, 917, 919, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3773", "outputIndexes": +[ 920 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 920 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3774", "outputIndexes": +[ 921 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 921 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3775", "outputIndexes": +[ 922 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 922, 921 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3776", "outputIndexes": +[ 923 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 913, 923, 916 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_140_output_0", "outputIndexes": +[ 924 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 924, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_27_output_0", "outputIndexes": +[ 925 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 925, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_295_output_0", "outputIndexes": +[ 926 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 911, 926, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_37_output_0", "outputIndexes": +[ 927 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 927 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_18_output_0", "outputIndexes": +[ 928 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 925, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_294_output_0", "outputIndexes": +[ 929 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 911, 22, 929, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_36_output_0", "outputIndexes": +[ 930 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 928, 930 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_120_output_0", "outputIndexes": +[ 931 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 931, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_73_output_0", "outputIndexes": +[ 932 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 912, 932 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_45_output_0", "outputIndexes": +[ 933 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3652 ] +, "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": +[ 3657 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3657 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3658 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3658 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 934 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 898, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_290_output_0", "outputIndexes": +[ 935 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 906, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_291_output_0", "outputIndexes": +[ 936 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 935, 936, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_118_output_0", "outputIndexes": +[ 937 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 934, 937 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_83_output_0", "outputIndexes": +[ 938 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 938 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1099026838, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_9/Mul_1_output_0", "outputIndexes": +[ 940 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 940, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_74_output_0", "outputIndexes": +[ 941 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 940 ] +, "main_type": "NONE", "name": "/Shape_129_output_0", "outputIndexes": +[ 942 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 942 ] +, "main_type": "NONE", "name": "Shape3658", "outputIndexes": +[ 943 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 942 ] +, "main_type": "NONE", "name": "Rank3660", "outputIndexes": +[ 944 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 944, 944 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3662", "outputIndexes": +[ 945 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 945 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3663", "outputIndexes": +[ 946 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 945, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3665", "outputIndexes": +[ 947 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 947 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze3666", "outputIndexes": +[ 948 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 943, 946, 948, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice3669", "outputIndexes": +[ 949 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 949 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze3670", "outputIndexes": +[ 950 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 950 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3671", "outputIndexes": +[ 951 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 951, 950 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp3672", "outputIndexes": +[ 952 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 942, 952, 945 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_141_output_0", "outputIndexes": +[ 953 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 953, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_28_output_0", "outputIndexes": +[ 954 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 954, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_297_output_0", "outputIndexes": +[ 955 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 940, 955, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_39_output_0", "outputIndexes": +[ 956 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 956 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_19_output_0", "outputIndexes": +[ 957 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 954, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_296_output_0", "outputIndexes": +[ 958 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 940, 22, 958, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_38_output_0", "outputIndexes": +[ 959 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 957, 959 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_121_output_0", "outputIndexes": +[ 960 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 960, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_75_output_0", "outputIndexes": +[ 961 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 941, 961 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_46_output_0", "outputIndexes": +[ 962 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3652 ] +, "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": +[ 3661 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3661 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3662 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3662 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 963 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 898, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_292_output_0", "outputIndexes": +[ 964 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 906, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_293_output_0", "outputIndexes": +[ 965 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 964, 965, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_119_output_0", "outputIndexes": +[ 966 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 963, 966 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_84_output_0", "outputIndexes": +[ 967 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 933, 962, 967, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_89_output_0", "outputIndexes": +[ 968 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 968 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3663 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3663 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3664 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3664 ] +, "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": +[ 3665 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3665 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3666 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3666 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.9/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 969 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 885, 969 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_48_output_0", "outputIndexes": +[ 970 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 970 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1112135106, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_9/Mul_1_output_0", "outputIndexes": +[ 971 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 971 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3667 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3667 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3668 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3668 ] +, "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": +[ 3669 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3669 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3670 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3670 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 972 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 972 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_9/Mul_output_0", "outputIndexes": +[ 973 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3668 ] +, "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": +[ 3673 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3673 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3674 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3674 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 974 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 973, 974 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_9/Mul_output_0", "outputIndexes": +[ 975 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 975 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3675 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3675 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.9/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3676 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3676 ] +, "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": +[ 3677 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3677 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.9/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3678 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3678 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.9/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 976 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 970, 976 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_49_output_0", "outputIndexes": +[ 977 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 977, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_90_output_0", "outputIndexes": +[ 978 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 978 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1206539780, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_10/Mul_1_output_0", "outputIndexes": +[ 979 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 979 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3679 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3679 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3680 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3680 ] +, "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": +[ 3681 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3681 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3682 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3682 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 980 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 979 ] +, "main_type": "NONE", "name": "/Shape_140_output_0", "outputIndexes": +[ 981 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 981 ] +, "main_type": "NONE", "name": "Shape4001", "outputIndexes": +[ 982 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 981 ] +, "main_type": "NONE", "name": "Rank4003", "outputIndexes": +[ 983 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 983, 983 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4005", "outputIndexes": +[ 984 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 984 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4006", "outputIndexes": +[ 985 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 984, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4008", "outputIndexes": +[ 986 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 986 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4009", "outputIndexes": +[ 987 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 982, 985, 987, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4012", "outputIndexes": +[ 988 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 988 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4013", "outputIndexes": +[ 989 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 989, 989 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4015", "outputIndexes": +[ 990 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 981, 990, 984 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_153_output_0", "outputIndexes": +[ 991 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 991, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_320_output_0", "outputIndexes": +[ 992 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 984 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4023", "outputIndexes": +[ 993 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 986 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4026", "outputIndexes": +[ 994 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 982, 993, 994, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4029", "outputIndexes": +[ 995 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 995 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4030", "outputIndexes": +[ 996 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 996 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4031", "outputIndexes": +[ 997 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 997, 996 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4032", "outputIndexes": +[ 998 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 981, 998, 984 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_154_output_0", "outputIndexes": +[ 999 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 999, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_321_output_0", "outputIndexes": +[ 1000 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 992, 1000, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_130_output_0", "outputIndexes": +[ 1001 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 980, 1001 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_91_output_0", "outputIndexes": +[ 1002 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1002 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1217058330, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_10/Mul_1_output_0", "outputIndexes": +[ 1004 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1004, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_80_output_0", "outputIndexes": +[ 1005 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1004 ] +, "main_type": "NONE", "name": "/Shape_142_output_0", "outputIndexes": +[ 1006 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1006 ] +, "main_type": "NONE", "name": "Shape4151", "outputIndexes": +[ 1007 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1006 ] +, "main_type": "NONE", "name": "Rank4153", "outputIndexes": +[ 1008 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1008, 1008 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4155", "outputIndexes": +[ 1009 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1009 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4156", "outputIndexes": +[ 1010 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1009, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4158", "outputIndexes": +[ 1011 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1011 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4159", "outputIndexes": +[ 1012 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1007, 1010, 1012, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4162", "outputIndexes": +[ 1013 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1013 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4163", "outputIndexes": +[ 1014 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1014 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4164", "outputIndexes": +[ 1015 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1015, 1014 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4165", "outputIndexes": +[ 1016 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1006, 1016, 1009 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_155_output_0", "outputIndexes": +[ 1017 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1017, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_30_output_0", "outputIndexes": +[ 1018 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1018, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_327_output_0", "outputIndexes": +[ 1019 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1004, 1019, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_41_output_0", "outputIndexes": +[ 1020 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1020 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_20_output_0", "outputIndexes": +[ 1021 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1018, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_326_output_0", "outputIndexes": +[ 1022 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1004, 22, 1022, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_40_output_0", "outputIndexes": +[ 1023 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1021, 1023 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_133_output_0", "outputIndexes": +[ 1024 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1024, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_81_output_0", "outputIndexes": +[ 1025 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1005, 1025 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_50_output_0", "outputIndexes": +[ 1026 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3680 ] +, "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": +[ 3685 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3685 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3686 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3686 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1027 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 991, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_322_output_0", "outputIndexes": +[ 1028 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 999, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_323_output_0", "outputIndexes": +[ 1029 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1028, 1029, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_131_output_0", "outputIndexes": +[ 1030 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1027, 1030 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_92_output_0", "outputIndexes": +[ 1031 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1031 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1219680816, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_10/Mul_1_output_0", "outputIndexes": +[ 1033 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1033, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_82_output_0", "outputIndexes": +[ 1034 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1033 ] +, "main_type": "NONE", "name": "/Shape_143_output_0", "outputIndexes": +[ 1035 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1035 ] +, "main_type": "NONE", "name": "Shape4047", "outputIndexes": +[ 1036 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1035 ] +, "main_type": "NONE", "name": "Rank4049", "outputIndexes": +[ 1037 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1037, 1037 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4051", "outputIndexes": +[ 1038 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1038 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4052", "outputIndexes": +[ 1039 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1038, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4054", "outputIndexes": +[ 1040 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1040 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4055", "outputIndexes": +[ 1041 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1036, 1039, 1041, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4058", "outputIndexes": +[ 1042 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1042 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4059", "outputIndexes": +[ 1043 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1043 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4060", "outputIndexes": +[ 1044 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1044, 1043 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4061", "outputIndexes": +[ 1045 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1035, 1045, 1038 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_156_output_0", "outputIndexes": +[ 1046 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1046, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_31_output_0", "outputIndexes": +[ 1047 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1047, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_329_output_0", "outputIndexes": +[ 1048 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1033, 1048, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_43_output_0", "outputIndexes": +[ 1049 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1049 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_21_output_0", "outputIndexes": +[ 1050 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1047, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_328_output_0", "outputIndexes": +[ 1051 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1033, 22, 1051, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_42_output_0", "outputIndexes": +[ 1052 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1050, 1052 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_134_output_0", "outputIndexes": +[ 1053 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1053, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_83_output_0", "outputIndexes": +[ 1054 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1034, 1054 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_51_output_0", "outputIndexes": +[ 1055 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3680 ] +, "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": +[ 3689 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3689 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3690 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3690 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1056 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 991, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_324_output_0", "outputIndexes": +[ 1057 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 999, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_325_output_0", "outputIndexes": +[ 1058 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1057, 1058, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_132_output_0", "outputIndexes": +[ 1059 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1056, 1059 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_93_output_0", "outputIndexes": +[ 1060 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1026, 1055, 1060, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_98_output_0", "outputIndexes": +[ 1061 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1061 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3691 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3691 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3692 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3692 ] +, "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": +[ 3693 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3693 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3694 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3694 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.10/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1062 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 978, 1062 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_53_output_0", "outputIndexes": +[ 1063 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1063 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1232789084, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_10/Mul_1_output_0", "outputIndexes": +[ 1064 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1064 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3695 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3695 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3696 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3696 ] +, "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": +[ 3697 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3697 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3698 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3698 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1065 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1065 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_10/Mul_output_0", "outputIndexes": +[ 1066 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3696 ] +, "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": +[ 3701 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3701 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3702 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3702 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1067 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1066, 1067 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_10/Mul_output_0", "outputIndexes": +[ 1068 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1068 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3703 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3703 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.10/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3704 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3704 ] +, "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": +[ 3705 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3705 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.10/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3706 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3706 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.10/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1069 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1063, 1069 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_54_output_0", "outputIndexes": +[ 1070 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1070, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_99_output_0", "outputIndexes": +[ 1071 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1071 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1327193758, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_11/Mul_1_output_0", "outputIndexes": +[ 1072 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1072 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3707 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3707 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3708 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3708 ] +, "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": +[ 3709 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3709 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3710 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3710 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1073 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1072 ] +, "main_type": "NONE", "name": "/Shape_154_output_0", "outputIndexes": +[ 1074 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1074 ] +, "main_type": "NONE", "name": "Shape4390", "outputIndexes": +[ 1075 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1074 ] +, "main_type": "NONE", "name": "Rank4392", "outputIndexes": +[ 1076 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1076, 1076 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4394", "outputIndexes": +[ 1077 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1077 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4395", "outputIndexes": +[ 1078 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1077, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4397", "outputIndexes": +[ 1079 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1079 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4398", "outputIndexes": +[ 1080 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1075, 1078, 1080, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4401", "outputIndexes": +[ 1081 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1081 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4402", "outputIndexes": +[ 1082 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1082, 1082 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4404", "outputIndexes": +[ 1083 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1074, 1083, 1077 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_168_output_0", "outputIndexes": +[ 1084 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1084, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_352_output_0", "outputIndexes": +[ 1085 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1077 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4412", "outputIndexes": +[ 1086 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1079 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4415", "outputIndexes": +[ 1087 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1075, 1086, 1087, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4418", "outputIndexes": +[ 1088 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1088 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4419", "outputIndexes": +[ 1089 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1089 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4420", "outputIndexes": +[ 1090 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1090, 1089 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4421", "outputIndexes": +[ 1091 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1074, 1091, 1077 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_169_output_0", "outputIndexes": +[ 1092 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1092, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_353_output_0", "outputIndexes": +[ 1093 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1085, 1093, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_143_output_0", "outputIndexes": +[ 1094 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1073, 1094 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_100_output_0", "outputIndexes": +[ 1095 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1095 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1337712308, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_11/Mul_1_output_0", "outputIndexes": +[ 1097 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1097, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_88_output_0", "outputIndexes": +[ 1098 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1097 ] +, "main_type": "NONE", "name": "/Shape_156_output_0", "outputIndexes": +[ 1099 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1099 ] +, "main_type": "NONE", "name": "Shape4540", "outputIndexes": +[ 1100 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1099 ] +, "main_type": "NONE", "name": "Rank4542", "outputIndexes": +[ 1101 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1101, 1101 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4544", "outputIndexes": +[ 1102 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1102 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4545", "outputIndexes": +[ 1103 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1102, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4547", "outputIndexes": +[ 1104 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1104 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4548", "outputIndexes": +[ 1105 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1100, 1103, 1105, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4551", "outputIndexes": +[ 1106 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1106 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4552", "outputIndexes": +[ 1107 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1107 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4553", "outputIndexes": +[ 1108 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1108, 1107 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4554", "outputIndexes": +[ 1109 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1099, 1109, 1102 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_170_output_0", "outputIndexes": +[ 1110 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1110, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_33_output_0", "outputIndexes": +[ 1111 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1111, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_359_output_0", "outputIndexes": +[ 1112 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1097, 1112, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_45_output_0", "outputIndexes": +[ 1113 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1113 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_22_output_0", "outputIndexes": +[ 1114 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1111, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_358_output_0", "outputIndexes": +[ 1115 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1097, 22, 1115, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_44_output_0", "outputIndexes": +[ 1116 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1114, 1116 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_146_output_0", "outputIndexes": +[ 1117 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1117, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_89_output_0", "outputIndexes": +[ 1118 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1098, 1118 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_55_output_0", "outputIndexes": +[ 1119 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3708 ] +, "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": +[ 3713 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3713 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3714 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3714 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1120 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1084, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_354_output_0", "outputIndexes": +[ 1121 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1092, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_355_output_0", "outputIndexes": +[ 1122 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1121, 1122, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_144_output_0", "outputIndexes": +[ 1123 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1120, 1123 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_101_output_0", "outputIndexes": +[ 1124 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1124 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1340334794, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_11/Mul_1_output_0", "outputIndexes": +[ 1126 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1126, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_90_output_0", "outputIndexes": +[ 1127 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1126 ] +, "main_type": "NONE", "name": "/Shape_157_output_0", "outputIndexes": +[ 1128 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1128 ] +, "main_type": "NONE", "name": "Shape4436", "outputIndexes": +[ 1129 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1128 ] +, "main_type": "NONE", "name": "Rank4438", "outputIndexes": +[ 1130 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1130, 1130 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4440", "outputIndexes": +[ 1131 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1131 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4441", "outputIndexes": +[ 1132 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1131, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4443", "outputIndexes": +[ 1133 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1133 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4444", "outputIndexes": +[ 1134 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1129, 1132, 1134, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4447", "outputIndexes": +[ 1135 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1135 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4448", "outputIndexes": +[ 1136 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1136 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4449", "outputIndexes": +[ 1137 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1137, 1136 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4450", "outputIndexes": +[ 1138 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1128, 1138, 1131 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_171_output_0", "outputIndexes": +[ 1139 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1139, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_34_output_0", "outputIndexes": +[ 1140 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1140, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_361_output_0", "outputIndexes": +[ 1141 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1126, 1141, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_47_output_0", "outputIndexes": +[ 1142 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1142 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_23_output_0", "outputIndexes": +[ 1143 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1140, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_360_output_0", "outputIndexes": +[ 1144 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1126, 22, 1144, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_46_output_0", "outputIndexes": +[ 1145 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1143, 1145 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_147_output_0", "outputIndexes": +[ 1146 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1146, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_91_output_0", "outputIndexes": +[ 1147 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1127, 1147 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_56_output_0", "outputIndexes": +[ 1148 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3708 ] +, "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": +[ 3717 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3717 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3718 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3718 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1149 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1084, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_356_output_0", "outputIndexes": +[ 1150 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1092, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_357_output_0", "outputIndexes": +[ 1151 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1150, 1151, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_145_output_0", "outputIndexes": +[ 1152 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1149, 1152 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_102_output_0", "outputIndexes": +[ 1153 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1119, 1148, 1153, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_107_output_0", "outputIndexes": +[ 1154 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1154 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3719 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3719 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3720 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3720 ] +, "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": +[ 3721 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3721 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3722 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3722 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.11/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1155 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1071, 1155 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_58_output_0", "outputIndexes": +[ 1156 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1156 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1353443062, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_11/Mul_1_output_0", "outputIndexes": +[ 1157 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1157 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3723 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3723 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3724 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3724 ] +, "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": +[ 3725 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3725 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3726 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3726 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1158 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1158 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_11/Mul_output_0", "outputIndexes": +[ 1159 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3724 ] +, "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": +[ 3729 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3729 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3730 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3730 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1160 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1159, 1160 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_11/Mul_output_0", "outputIndexes": +[ 1161 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1161 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3731 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3731 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.11/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3732 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3732 ] +, "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": +[ 3733 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3733 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.11/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3734 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3734 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.11/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1162 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1156, 1162 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_59_output_0", "outputIndexes": +[ 1163 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1163, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_108_output_0", "outputIndexes": +[ 1164 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1164 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1447847736, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_12/Mul_1_output_0", "outputIndexes": +[ 1165 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1165 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3735 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3735 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3736 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3736 ] +, "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": +[ 3737 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3737 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3738 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3738 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1166 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1165 ] +, "main_type": "NONE", "name": "/Shape_168_output_0", "outputIndexes": +[ 1167 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1167 ] +, "main_type": "NONE", "name": "Shape4779", "outputIndexes": +[ 1168 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1167 ] +, "main_type": "NONE", "name": "Rank4781", "outputIndexes": +[ 1169 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1169, 1169 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4783", "outputIndexes": +[ 1170 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1170 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4784", "outputIndexes": +[ 1171 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1170, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4786", "outputIndexes": +[ 1172 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1172 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4787", "outputIndexes": +[ 1173 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1168, 1171, 1173, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4790", "outputIndexes": +[ 1174 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1174 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4791", "outputIndexes": +[ 1175 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1175, 1175 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4793", "outputIndexes": +[ 1176 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1167, 1176, 1170 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_183_output_0", "outputIndexes": +[ 1177 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1177, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_384_output_0", "outputIndexes": +[ 1178 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1170 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4801", "outputIndexes": +[ 1179 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1172 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4804", "outputIndexes": +[ 1180 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1168, 1179, 1180, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4807", "outputIndexes": +[ 1181 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1181 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4808", "outputIndexes": +[ 1182 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1182 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4809", "outputIndexes": +[ 1183 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1183, 1182 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4810", "outputIndexes": +[ 1184 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1167, 1184, 1170 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_184_output_0", "outputIndexes": +[ 1185 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1185, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_385_output_0", "outputIndexes": +[ 1186 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1178, 1186, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_156_output_0", "outputIndexes": +[ 1187 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1166, 1187 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_109_output_0", "outputIndexes": +[ 1188 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1188 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1458366286, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_12/Mul_1_output_0", "outputIndexes": +[ 1190 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1190, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_96_output_0", "outputIndexes": +[ 1191 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1190 ] +, "main_type": "NONE", "name": "/Shape_170_output_0", "outputIndexes": +[ 1192 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1192 ] +, "main_type": "NONE", "name": "Shape4929", "outputIndexes": +[ 1193 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1192 ] +, "main_type": "NONE", "name": "Rank4931", "outputIndexes": +[ 1194 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1194, 1194 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4933", "outputIndexes": +[ 1195 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1195 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4934", "outputIndexes": +[ 1196 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1195, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4936", "outputIndexes": +[ 1197 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1197 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4937", "outputIndexes": +[ 1198 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1193, 1196, 1198, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4940", "outputIndexes": +[ 1199 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1199 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4941", "outputIndexes": +[ 1200 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1200 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4942", "outputIndexes": +[ 1201 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1201, 1200 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4943", "outputIndexes": +[ 1202 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1192, 1202, 1195 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_185_output_0", "outputIndexes": +[ 1203 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1203, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_36_output_0", "outputIndexes": +[ 1204 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1204, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_391_output_0", "outputIndexes": +[ 1205 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1190, 1205, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_49_output_0", "outputIndexes": +[ 1206 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1206 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_24_output_0", "outputIndexes": +[ 1207 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1204, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_390_output_0", "outputIndexes": +[ 1208 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1190, 22, 1208, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_48_output_0", "outputIndexes": +[ 1209 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1207, 1209 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_159_output_0", "outputIndexes": +[ 1210 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1210, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_97_output_0", "outputIndexes": +[ 1211 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1191, 1211 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_60_output_0", "outputIndexes": +[ 1212 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3736 ] +, "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": +[ 3741 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3741 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3742 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3742 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1213 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1177, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_386_output_0", "outputIndexes": +[ 1214 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1185, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_387_output_0", "outputIndexes": +[ 1215 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1214, 1215, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_157_output_0", "outputIndexes": +[ 1216 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1213, 1216 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_110_output_0", "outputIndexes": +[ 1217 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1217 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1460988772, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_12/Mul_1_output_0", "outputIndexes": +[ 1219 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1219, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_98_output_0", "outputIndexes": +[ 1220 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1219 ] +, "main_type": "NONE", "name": "/Shape_171_output_0", "outputIndexes": +[ 1221 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1221 ] +, "main_type": "NONE", "name": "Shape4825", "outputIndexes": +[ 1222 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1221 ] +, "main_type": "NONE", "name": "Rank4827", "outputIndexes": +[ 1223 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1223, 1223 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4829", "outputIndexes": +[ 1224 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1224 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4830", "outputIndexes": +[ 1225 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1224, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4832", "outputIndexes": +[ 1226 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1226 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze4833", "outputIndexes": +[ 1227 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1222, 1225, 1227, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice4836", "outputIndexes": +[ 1228 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1228 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze4837", "outputIndexes": +[ 1229 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1229 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4838", "outputIndexes": +[ 1230 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1230, 1229 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp4839", "outputIndexes": +[ 1231 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1221, 1231, 1224 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_186_output_0", "outputIndexes": +[ 1232 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1232, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_37_output_0", "outputIndexes": +[ 1233 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1233, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_393_output_0", "outputIndexes": +[ 1234 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1219, 1234, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_51_output_0", "outputIndexes": +[ 1235 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1235 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_25_output_0", "outputIndexes": +[ 1236 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1233, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_392_output_0", "outputIndexes": +[ 1237 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1219, 22, 1237, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_50_output_0", "outputIndexes": +[ 1238 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1236, 1238 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_160_output_0", "outputIndexes": +[ 1239 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1239, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_99_output_0", "outputIndexes": +[ 1240 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1220, 1240 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_61_output_0", "outputIndexes": +[ 1241 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3736 ] +, "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": +[ 3745 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3745 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3746 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3746 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1242 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1177, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_388_output_0", "outputIndexes": +[ 1243 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1185, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_389_output_0", "outputIndexes": +[ 1244 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1243, 1244, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_158_output_0", "outputIndexes": +[ 1245 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1242, 1245 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_111_output_0", "outputIndexes": +[ 1246 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1212, 1241, 1246, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_116_output_0", "outputIndexes": +[ 1247 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1247 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3747 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3747 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3748 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3748 ] +, "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": +[ 3749 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3749 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3750 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3750 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.12/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1248 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1164, 1248 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_63_output_0", "outputIndexes": +[ 1249 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1249 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1474097040, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_12/Mul_1_output_0", "outputIndexes": +[ 1250 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1250 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3751 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3751 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3752 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3752 ] +, "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": +[ 3753 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3753 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3754 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3754 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1251 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1251 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_12/Mul_output_0", "outputIndexes": +[ 1252 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3752 ] +, "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": +[ 3757 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3757 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3758 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3758 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1253 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1252, 1253 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_12/Mul_output_0", "outputIndexes": +[ 1254 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1254 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3759 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3759 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.12/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3760 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3760 ] +, "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": +[ 3761 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3761 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.12/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3762 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3762 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.12/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1255 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1249, 1255 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_64_output_0", "outputIndexes": +[ 1256 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1256, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_117_output_0", "outputIndexes": +[ 1257 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1257 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1568501714, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_13/Mul_1_output_0", "outputIndexes": +[ 1258 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1258 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3763 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3763 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3764 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3764 ] +, "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": +[ 3765 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3765 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3766 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3766 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1259 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1258 ] +, "main_type": "NONE", "name": "/Shape_182_output_0", "outputIndexes": +[ 1260 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1260 ] +, "main_type": "NONE", "name": "Shape5168", "outputIndexes": +[ 1261 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1260 ] +, "main_type": "NONE", "name": "Rank5170", "outputIndexes": +[ 1262 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1262, 1262 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5172", "outputIndexes": +[ 1263 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1263 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5173", "outputIndexes": +[ 1264 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1263, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5175", "outputIndexes": +[ 1265 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1265 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5176", "outputIndexes": +[ 1266 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1261, 1264, 1266, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5179", "outputIndexes": +[ 1267 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1267 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5180", "outputIndexes": +[ 1268 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1268, 1268 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5182", "outputIndexes": +[ 1269 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1260, 1269, 1263 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_198_output_0", "outputIndexes": +[ 1270 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1270, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_416_output_0", "outputIndexes": +[ 1271 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1263 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5190", "outputIndexes": +[ 1272 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1265 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5193", "outputIndexes": +[ 1273 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1261, 1272, 1273, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5196", "outputIndexes": +[ 1274 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1274 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5197", "outputIndexes": +[ 1275 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1275 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5198", "outputIndexes": +[ 1276 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1276, 1275 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5199", "outputIndexes": +[ 1277 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1260, 1277, 1263 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_199_output_0", "outputIndexes": +[ 1278 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1278, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_417_output_0", "outputIndexes": +[ 1279 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1271, 1279, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_169_output_0", "outputIndexes": +[ 1280 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1259, 1280 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_118_output_0", "outputIndexes": +[ 1281 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1281 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1579020264, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_13/Mul_1_output_0", "outputIndexes": +[ 1283 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1283, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_104_output_0", "outputIndexes": +[ 1284 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1283 ] +, "main_type": "NONE", "name": "/Shape_184_output_0", "outputIndexes": +[ 1285 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1285 ] +, "main_type": "NONE", "name": "Shape5318", "outputIndexes": +[ 1286 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1285 ] +, "main_type": "NONE", "name": "Rank5320", "outputIndexes": +[ 1287 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1287, 1287 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5322", "outputIndexes": +[ 1288 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1288 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5323", "outputIndexes": +[ 1289 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1288, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5325", "outputIndexes": +[ 1290 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1290 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5326", "outputIndexes": +[ 1291 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1286, 1289, 1291, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5329", "outputIndexes": +[ 1292 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1292 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5330", "outputIndexes": +[ 1293 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1293 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5331", "outputIndexes": +[ 1294 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1294, 1293 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5332", "outputIndexes": +[ 1295 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1285, 1295, 1288 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_200_output_0", "outputIndexes": +[ 1296 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1296, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_39_output_0", "outputIndexes": +[ 1297 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1297, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_423_output_0", "outputIndexes": +[ 1298 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1283, 1298, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_53_output_0", "outputIndexes": +[ 1299 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1299 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_26_output_0", "outputIndexes": +[ 1300 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1297, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_422_output_0", "outputIndexes": +[ 1301 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1283, 22, 1301, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_52_output_0", "outputIndexes": +[ 1302 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1300, 1302 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_172_output_0", "outputIndexes": +[ 1303 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1303, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_105_output_0", "outputIndexes": +[ 1304 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1284, 1304 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_65_output_0", "outputIndexes": +[ 1305 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3764 ] +, "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": +[ 3769 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3769 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3770 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3770 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1306 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1270, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_418_output_0", "outputIndexes": +[ 1307 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1278, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_419_output_0", "outputIndexes": +[ 1308 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1307, 1308, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_170_output_0", "outputIndexes": +[ 1309 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1306, 1309 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_119_output_0", "outputIndexes": +[ 1310 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1310 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1581642750, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_13/Mul_1_output_0", "outputIndexes": +[ 1312 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1312, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_106_output_0", "outputIndexes": +[ 1313 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1312 ] +, "main_type": "NONE", "name": "/Shape_185_output_0", "outputIndexes": +[ 1314 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1314 ] +, "main_type": "NONE", "name": "Shape5214", "outputIndexes": +[ 1315 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1314 ] +, "main_type": "NONE", "name": "Rank5216", "outputIndexes": +[ 1316 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1316, 1316 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5218", "outputIndexes": +[ 1317 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1317 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5219", "outputIndexes": +[ 1318 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1317, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5221", "outputIndexes": +[ 1319 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1319 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5222", "outputIndexes": +[ 1320 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1315, 1318, 1320, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5225", "outputIndexes": +[ 1321 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1321 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5226", "outputIndexes": +[ 1322 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1322 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5227", "outputIndexes": +[ 1323 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1323, 1322 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5228", "outputIndexes": +[ 1324 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1314, 1324, 1317 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_201_output_0", "outputIndexes": +[ 1325 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1325, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_40_output_0", "outputIndexes": +[ 1326 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1326, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_425_output_0", "outputIndexes": +[ 1327 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1312, 1327, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_55_output_0", "outputIndexes": +[ 1328 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1328 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_27_output_0", "outputIndexes": +[ 1329 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1326, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_424_output_0", "outputIndexes": +[ 1330 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1312, 22, 1330, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_54_output_0", "outputIndexes": +[ 1331 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1329, 1331 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_173_output_0", "outputIndexes": +[ 1332 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1332, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_107_output_0", "outputIndexes": +[ 1333 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1313, 1333 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_66_output_0", "outputIndexes": +[ 1334 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3764 ] +, "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": +[ 3773 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3773 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3774 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3774 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1335 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1270, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_420_output_0", "outputIndexes": +[ 1336 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1278, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_421_output_0", "outputIndexes": +[ 1337 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1336, 1337, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_171_output_0", "outputIndexes": +[ 1338 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1335, 1338 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_120_output_0", "outputIndexes": +[ 1339 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1305, 1334, 1339, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_125_output_0", "outputIndexes": +[ 1340 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1340 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3775 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3775 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3776 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3776 ] +, "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": +[ 3777 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3777 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3778 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3778 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.13/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1341 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1257, 1341 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_68_output_0", "outputIndexes": +[ 1342 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1342 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1594751018, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_13/Mul_1_output_0", "outputIndexes": +[ 1343 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1343 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3779 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3779 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3780 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3780 ] +, "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": +[ 3781 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3781 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3782 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3782 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1344 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1344 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_13/Mul_output_0", "outputIndexes": +[ 1345 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3780 ] +, "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": +[ 3785 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3785 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3786 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3786 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1346 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1345, 1346 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_13/Mul_output_0", "outputIndexes": +[ 1347 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1347 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3787 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3787 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.13/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3788 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3788 ] +, "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": +[ 3789 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3789 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.13/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3790 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3790 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.13/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1348 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1342, 1348 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_69_output_0", "outputIndexes": +[ 1349 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1349, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_126_output_0", "outputIndexes": +[ 1350 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1350 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1689155692, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_14/Mul_1_output_0", "outputIndexes": +[ 1351 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1351 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3791 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3791 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3792 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3792 ] +, "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": +[ 3793 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3793 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3794 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3794 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1352 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1351 ] +, "main_type": "NONE", "name": "/Shape_196_output_0", "outputIndexes": +[ 1353 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1353 ] +, "main_type": "NONE", "name": "Shape5557", "outputIndexes": +[ 1354 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1353 ] +, "main_type": "NONE", "name": "Rank5559", "outputIndexes": +[ 1355 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1355, 1355 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5561", "outputIndexes": +[ 1356 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1356 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5562", "outputIndexes": +[ 1357 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1356, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5564", "outputIndexes": +[ 1358 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1358 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5565", "outputIndexes": +[ 1359 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1354, 1357, 1359, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5568", "outputIndexes": +[ 1360 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1360 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5569", "outputIndexes": +[ 1361 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1361, 1361 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5571", "outputIndexes": +[ 1362 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1353, 1362, 1356 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_213_output_0", "outputIndexes": +[ 1363 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1363, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_448_output_0", "outputIndexes": +[ 1364 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1356 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5579", "outputIndexes": +[ 1365 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1358 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5582", "outputIndexes": +[ 1366 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1354, 1365, 1366, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5585", "outputIndexes": +[ 1367 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1367 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5586", "outputIndexes": +[ 1368 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1368 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5587", "outputIndexes": +[ 1369 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1369, 1368 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5588", "outputIndexes": +[ 1370 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1353, 1370, 1356 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_214_output_0", "outputIndexes": +[ 1371 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1371, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_449_output_0", "outputIndexes": +[ 1372 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1364, 1372, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_182_output_0", "outputIndexes": +[ 1373 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1352, 1373 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_127_output_0", "outputIndexes": +[ 1374 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1374 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1699674242, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_14/Mul_1_output_0", "outputIndexes": +[ 1376 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1376, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_112_output_0", "outputIndexes": +[ 1377 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1376 ] +, "main_type": "NONE", "name": "/Shape_198_output_0", "outputIndexes": +[ 1378 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1378 ] +, "main_type": "NONE", "name": "Shape5707", "outputIndexes": +[ 1379 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1378 ] +, "main_type": "NONE", "name": "Rank5709", "outputIndexes": +[ 1380 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1380, 1380 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5711", "outputIndexes": +[ 1381 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1381 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5712", "outputIndexes": +[ 1382 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1381, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5714", "outputIndexes": +[ 1383 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1383 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5715", "outputIndexes": +[ 1384 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1379, 1382, 1384, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5718", "outputIndexes": +[ 1385 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1385 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5719", "outputIndexes": +[ 1386 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1386 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5720", "outputIndexes": +[ 1387 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1387, 1386 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5721", "outputIndexes": +[ 1388 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1378, 1388, 1381 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_215_output_0", "outputIndexes": +[ 1389 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1389, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_42_output_0", "outputIndexes": +[ 1390 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1390, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_455_output_0", "outputIndexes": +[ 1391 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1376, 1391, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_57_output_0", "outputIndexes": +[ 1392 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1392 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_28_output_0", "outputIndexes": +[ 1393 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1390, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_454_output_0", "outputIndexes": +[ 1394 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1376, 22, 1394, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_56_output_0", "outputIndexes": +[ 1395 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1393, 1395 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_185_output_0", "outputIndexes": +[ 1396 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1396, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_113_output_0", "outputIndexes": +[ 1397 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1377, 1397 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_70_output_0", "outputIndexes": +[ 1398 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3792 ] +, "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": +[ 3797 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3797 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3798 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3798 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1399 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1363, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_450_output_0", "outputIndexes": +[ 1400 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1371, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_451_output_0", "outputIndexes": +[ 1401 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1400, 1401, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_183_output_0", "outputIndexes": +[ 1402 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1399, 1402 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_128_output_0", "outputIndexes": +[ 1403 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1403 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1702296728, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_14/Mul_1_output_0", "outputIndexes": +[ 1405 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1405, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_114_output_0", "outputIndexes": +[ 1406 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1405 ] +, "main_type": "NONE", "name": "/Shape_199_output_0", "outputIndexes": +[ 1407 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1407 ] +, "main_type": "NONE", "name": "Shape5603", "outputIndexes": +[ 1408 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1407 ] +, "main_type": "NONE", "name": "Rank5605", "outputIndexes": +[ 1409 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1409, 1409 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5607", "outputIndexes": +[ 1410 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1410 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5608", "outputIndexes": +[ 1411 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1410, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5610", "outputIndexes": +[ 1412 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1412 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5611", "outputIndexes": +[ 1413 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1408, 1411, 1413, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5614", "outputIndexes": +[ 1414 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1414 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5615", "outputIndexes": +[ 1415 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1415 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5616", "outputIndexes": +[ 1416 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1416, 1415 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5617", "outputIndexes": +[ 1417 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1407, 1417, 1410 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_216_output_0", "outputIndexes": +[ 1418 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1418, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_43_output_0", "outputIndexes": +[ 1419 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1419, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_457_output_0", "outputIndexes": +[ 1420 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1405, 1420, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_59_output_0", "outputIndexes": +[ 1421 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1421 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_29_output_0", "outputIndexes": +[ 1422 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1419, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_456_output_0", "outputIndexes": +[ 1423 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1405, 22, 1423, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_58_output_0", "outputIndexes": +[ 1424 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1422, 1424 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_186_output_0", "outputIndexes": +[ 1425 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1425, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_115_output_0", "outputIndexes": +[ 1426 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1406, 1426 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_71_output_0", "outputIndexes": +[ 1427 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3792 ] +, "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": +[ 3801 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3801 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3802 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3802 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1428 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1363, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_452_output_0", "outputIndexes": +[ 1429 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1371, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_453_output_0", "outputIndexes": +[ 1430 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1429, 1430, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_184_output_0", "outputIndexes": +[ 1431 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1428, 1431 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_129_output_0", "outputIndexes": +[ 1432 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1398, 1427, 1432, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_134_output_0", "outputIndexes": +[ 1433 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1433 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3803 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3803 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3804 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3804 ] +, "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": +[ 3805 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3805 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3806 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3806 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.14/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1434 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1350, 1434 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_73_output_0", "outputIndexes": +[ 1435 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1435 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1715404996, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_14/Mul_1_output_0", "outputIndexes": +[ 1436 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1436 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3807 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3807 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3808 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3808 ] +, "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": +[ 3809 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3809 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3810 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3810 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1437 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1437 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_14/Mul_output_0", "outputIndexes": +[ 1438 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3808 ] +, "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": +[ 3813 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3813 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3814 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3814 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1439 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1438, 1439 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_14/Mul_output_0", "outputIndexes": +[ 1440 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1440 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3815 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3815 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.14/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3816 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3816 ] +, "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": +[ 3817 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3817 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.14/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3818 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3818 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.14/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1441 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1435, 1441 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_74_output_0", "outputIndexes": +[ 1442 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1442, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_135_output_0", "outputIndexes": +[ 1443 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1443 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1809809670, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_15/Mul_1_output_0", "outputIndexes": +[ 1444 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1444 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3819 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3819 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3820 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3820 ] +, "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": +[ 3821 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3821 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3822 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3822 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1445 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1444 ] +, "main_type": "NONE", "name": "/Shape_210_output_0", "outputIndexes": +[ 1446 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1446 ] +, "main_type": "NONE", "name": "Shape5946", "outputIndexes": +[ 1447 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1446 ] +, "main_type": "NONE", "name": "Rank5948", "outputIndexes": +[ 1448 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1448, 1448 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5950", "outputIndexes": +[ 1449 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1449 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5951", "outputIndexes": +[ 1450 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1449, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5953", "outputIndexes": +[ 1451 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1451 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5954", "outputIndexes": +[ 1452 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1447, 1450, 1452, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5957", "outputIndexes": +[ 1453 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1453 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5958", "outputIndexes": +[ 1454 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1454, 1454 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5960", "outputIndexes": +[ 1455 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1446, 1455, 1449 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_228_output_0", "outputIndexes": +[ 1456 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1456, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_480_output_0", "outputIndexes": +[ 1457 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1449 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5968", "outputIndexes": +[ 1458 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1451 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5971", "outputIndexes": +[ 1459 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1447, 1458, 1459, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice5974", "outputIndexes": +[ 1460 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1460 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze5975", "outputIndexes": +[ 1461 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1461 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5976", "outputIndexes": +[ 1462 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1462, 1461 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5977", "outputIndexes": +[ 1463 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1446, 1463, 1449 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_229_output_0", "outputIndexes": +[ 1464 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1464, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_481_output_0", "outputIndexes": +[ 1465 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1457, 1465, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_195_output_0", "outputIndexes": +[ 1466 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1445, 1466 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_136_output_0", "outputIndexes": +[ 1467 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1467 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1820328220, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_15/Mul_1_output_0", "outputIndexes": +[ 1469 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1469, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_120_output_0", "outputIndexes": +[ 1470 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1469 ] +, "main_type": "NONE", "name": "/Shape_212_output_0", "outputIndexes": +[ 1471 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1471 ] +, "main_type": "NONE", "name": "Shape6096", "outputIndexes": +[ 1472 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1471 ] +, "main_type": "NONE", "name": "Rank6098", "outputIndexes": +[ 1473 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1473, 1473 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6100", "outputIndexes": +[ 1474 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1474 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6101", "outputIndexes": +[ 1475 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1474, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6103", "outputIndexes": +[ 1476 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1476 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6104", "outputIndexes": +[ 1477 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1472, 1475, 1477, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6107", "outputIndexes": +[ 1478 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1478 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6108", "outputIndexes": +[ 1479 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1479 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6109", "outputIndexes": +[ 1480 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1480, 1479 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6110", "outputIndexes": +[ 1481 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1471, 1481, 1474 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_230_output_0", "outputIndexes": +[ 1482 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1482, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_45_output_0", "outputIndexes": +[ 1483 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1483, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_487_output_0", "outputIndexes": +[ 1484 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1469, 1484, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_61_output_0", "outputIndexes": +[ 1485 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1485 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_30_output_0", "outputIndexes": +[ 1486 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1483, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_486_output_0", "outputIndexes": +[ 1487 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1469, 22, 1487, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_60_output_0", "outputIndexes": +[ 1488 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1486, 1488 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_198_output_0", "outputIndexes": +[ 1489 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1489, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_121_output_0", "outputIndexes": +[ 1490 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1470, 1490 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_75_output_0", "outputIndexes": +[ 1491 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3820 ] +, "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": +[ 3825 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3825 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3826 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3826 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1492 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1456, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_482_output_0", "outputIndexes": +[ 1493 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1464, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_483_output_0", "outputIndexes": +[ 1494 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1493, 1494, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_196_output_0", "outputIndexes": +[ 1495 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1492, 1495 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_137_output_0", "outputIndexes": +[ 1496 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1496 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1822950706, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_15/Mul_1_output_0", "outputIndexes": +[ 1498 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1498, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_122_output_0", "outputIndexes": +[ 1499 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1498 ] +, "main_type": "NONE", "name": "/Shape_213_output_0", "outputIndexes": +[ 1500 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1500 ] +, "main_type": "NONE", "name": "Shape5992", "outputIndexes": +[ 1501 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1500 ] +, "main_type": "NONE", "name": "Rank5994", "outputIndexes": +[ 1502 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1502, 1502 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5996", "outputIndexes": +[ 1503 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1503 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze5997", "outputIndexes": +[ 1504 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1503, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp5999", "outputIndexes": +[ 1505 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1505 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6000", "outputIndexes": +[ 1506 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1501, 1504, 1506, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6003", "outputIndexes": +[ 1507 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1507 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6004", "outputIndexes": +[ 1508 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1508 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6005", "outputIndexes": +[ 1509 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1509, 1508 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6006", "outputIndexes": +[ 1510 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1500, 1510, 1503 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_231_output_0", "outputIndexes": +[ 1511 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1511, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_46_output_0", "outputIndexes": +[ 1512 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1512, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_489_output_0", "outputIndexes": +[ 1513 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1498, 1513, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_63_output_0", "outputIndexes": +[ 1514 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1514 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_31_output_0", "outputIndexes": +[ 1515 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1512, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_488_output_0", "outputIndexes": +[ 1516 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1498, 22, 1516, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_62_output_0", "outputIndexes": +[ 1517 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1515, 1517 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_199_output_0", "outputIndexes": +[ 1518 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1518, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_123_output_0", "outputIndexes": +[ 1519 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1499, 1519 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_76_output_0", "outputIndexes": +[ 1520 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3820 ] +, "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": +[ 3829 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3829 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3830 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3830 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1521 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1456, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_484_output_0", "outputIndexes": +[ 1522 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1464, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_485_output_0", "outputIndexes": +[ 1523 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1522, 1523, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_197_output_0", "outputIndexes": +[ 1524 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1521, 1524 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_138_output_0", "outputIndexes": +[ 1525 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1491, 1520, 1525, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_143_output_0", "outputIndexes": +[ 1526 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1526 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3831 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3831 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3832 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3832 ] +, "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": +[ 3833 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3833 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3834 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3834 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.15/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1527 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1443, 1527 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_78_output_0", "outputIndexes": +[ 1528 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1528 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1836058974, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_15/Mul_1_output_0", "outputIndexes": +[ 1529 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1529 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3835 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3835 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3836 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3836 ] +, "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": +[ 3837 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3837 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3838 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3838 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1530 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1530 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_15/Mul_output_0", "outputIndexes": +[ 1531 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3836 ] +, "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": +[ 3841 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3841 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3842 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3842 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1532 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1531, 1532 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_15/Mul_output_0", "outputIndexes": +[ 1533 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1533 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3843 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3843 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.15/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3844 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3844 ] +, "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": +[ 3845 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3845 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.15/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3846 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3846 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.15/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1534 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1528, 1534 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_79_output_0", "outputIndexes": +[ 1535 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1535, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_144_output_0", "outputIndexes": +[ 1536 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1536 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1930463648, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_16/Mul_1_output_0", "outputIndexes": +[ 1537 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1537 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3847 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3847 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3848 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3848 ] +, "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": +[ 3849 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3849 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3850 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3850 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1538 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1537 ] +, "main_type": "NONE", "name": "/Shape_224_output_0", "outputIndexes": +[ 1539 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1539 ] +, "main_type": "NONE", "name": "Shape6335", "outputIndexes": +[ 1540 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1539 ] +, "main_type": "NONE", "name": "Rank6337", "outputIndexes": +[ 1541 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1541, 1541 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6339", "outputIndexes": +[ 1542 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1542 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6340", "outputIndexes": +[ 1543 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1542, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6342", "outputIndexes": +[ 1544 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1544 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6343", "outputIndexes": +[ 1545 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1540, 1543, 1545, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6346", "outputIndexes": +[ 1546 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1546 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6347", "outputIndexes": +[ 1547 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1547, 1547 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6349", "outputIndexes": +[ 1548 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1539, 1548, 1542 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_243_output_0", "outputIndexes": +[ 1549 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1549, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_512_output_0", "outputIndexes": +[ 1550 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1542 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6357", "outputIndexes": +[ 1551 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1544 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6360", "outputIndexes": +[ 1552 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1540, 1551, 1552, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6363", "outputIndexes": +[ 1553 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1553 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6364", "outputIndexes": +[ 1554 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1554 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6365", "outputIndexes": +[ 1555 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1555, 1554 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6366", "outputIndexes": +[ 1556 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1539, 1556, 1542 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_244_output_0", "outputIndexes": +[ 1557 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1557, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_513_output_0", "outputIndexes": +[ 1558 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1550, 1558, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_208_output_0", "outputIndexes": +[ 1559 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1538, 1559 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_145_output_0", "outputIndexes": +[ 1560 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1560 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1940982198, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_16/Mul_1_output_0", "outputIndexes": +[ 1562 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1562, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_128_output_0", "outputIndexes": +[ 1563 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1562 ] +, "main_type": "NONE", "name": "/Shape_226_output_0", "outputIndexes": +[ 1564 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1564 ] +, "main_type": "NONE", "name": "Shape6485", "outputIndexes": +[ 1565 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1564 ] +, "main_type": "NONE", "name": "Rank6487", "outputIndexes": +[ 1566 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1566, 1566 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6489", "outputIndexes": +[ 1567 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1567 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6490", "outputIndexes": +[ 1568 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1567, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6492", "outputIndexes": +[ 1569 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1569 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6493", "outputIndexes": +[ 1570 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1565, 1568, 1570, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6496", "outputIndexes": +[ 1571 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1571 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6497", "outputIndexes": +[ 1572 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1572 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6498", "outputIndexes": +[ 1573 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1573, 1572 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6499", "outputIndexes": +[ 1574 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1564, 1574, 1567 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_245_output_0", "outputIndexes": +[ 1575 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1575, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_48_output_0", "outputIndexes": +[ 1576 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1576, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_519_output_0", "outputIndexes": +[ 1577 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1562, 1577, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_65_output_0", "outputIndexes": +[ 1578 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1578 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_32_output_0", "outputIndexes": +[ 1579 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1576, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_518_output_0", "outputIndexes": +[ 1580 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1562, 22, 1580, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_64_output_0", "outputIndexes": +[ 1581 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1579, 1581 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_211_output_0", "outputIndexes": +[ 1582 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1582, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_129_output_0", "outputIndexes": +[ 1583 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1563, 1583 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_80_output_0", "outputIndexes": +[ 1584 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3848 ] +, "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": +[ 3853 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3853 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3854 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3854 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1585 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1549, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_514_output_0", "outputIndexes": +[ 1586 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1557, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_515_output_0", "outputIndexes": +[ 1587 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1586, 1587, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_209_output_0", "outputIndexes": +[ 1588 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1585, 1588 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_146_output_0", "outputIndexes": +[ 1589 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1589 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1943604684, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_16/Mul_1_output_0", "outputIndexes": +[ 1591 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1591, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_130_output_0", "outputIndexes": +[ 1592 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1591 ] +, "main_type": "NONE", "name": "/Shape_227_output_0", "outputIndexes": +[ 1593 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1593 ] +, "main_type": "NONE", "name": "Shape6381", "outputIndexes": +[ 1594 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1593 ] +, "main_type": "NONE", "name": "Rank6383", "outputIndexes": +[ 1595 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1595, 1595 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6385", "outputIndexes": +[ 1596 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1596 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6386", "outputIndexes": +[ 1597 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1596, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6388", "outputIndexes": +[ 1598 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1598 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6389", "outputIndexes": +[ 1599 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1594, 1597, 1599, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6392", "outputIndexes": +[ 1600 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1600 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6393", "outputIndexes": +[ 1601 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1601 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6394", "outputIndexes": +[ 1602 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1602, 1601 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6395", "outputIndexes": +[ 1603 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1593, 1603, 1596 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_246_output_0", "outputIndexes": +[ 1604 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1604, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_49_output_0", "outputIndexes": +[ 1605 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1605, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_521_output_0", "outputIndexes": +[ 1606 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1591, 1606, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_67_output_0", "outputIndexes": +[ 1607 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1607 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_33_output_0", "outputIndexes": +[ 1608 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1605, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_520_output_0", "outputIndexes": +[ 1609 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1591, 22, 1609, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_66_output_0", "outputIndexes": +[ 1610 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1608, 1610 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_212_output_0", "outputIndexes": +[ 1611 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1611, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_131_output_0", "outputIndexes": +[ 1612 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1592, 1612 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_81_output_0", "outputIndexes": +[ 1613 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3848 ] +, "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": +[ 3857 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3857 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3858 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3858 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1614 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1549, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_516_output_0", "outputIndexes": +[ 1615 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1557, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_517_output_0", "outputIndexes": +[ 1616 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1615, 1616, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_210_output_0", "outputIndexes": +[ 1617 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1614, 1617 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_147_output_0", "outputIndexes": +[ 1618 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1584, 1613, 1618, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_152_output_0", "outputIndexes": +[ 1619 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1619 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3859 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3859 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3860 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3860 ] +, "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": +[ 3861 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3861 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3862 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3862 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.16/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1620 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1536, 1620 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_83_output_0", "outputIndexes": +[ 1621 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1621 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 1956712952, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_16/Mul_1_output_0", "outputIndexes": +[ 1622 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1622 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3863 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3863 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3864 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3864 ] +, "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": +[ 3865 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3865 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3866 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3866 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1623 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1623 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_16/Mul_output_0", "outputIndexes": +[ 1624 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3864 ] +, "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": +[ 3869 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3869 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3870 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3870 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1625 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1624, 1625 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_16/Mul_output_0", "outputIndexes": +[ 1626 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1626 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3871 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3871 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.16/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3872 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3872 ] +, "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": +[ 3873 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3873 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.16/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3874 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3874 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.16/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1627 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1621, 1627 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_84_output_0", "outputIndexes": +[ 1628 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1628, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_153_output_0", "outputIndexes": +[ 1629 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1629 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2051117626, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_17/Mul_1_output_0", "outputIndexes": +[ 1630 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1630 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3875 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3875 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3876 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3876 ] +, "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": +[ 3877 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3877 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3878 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3878 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1631 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1630 ] +, "main_type": "NONE", "name": "/Shape_238_output_0", "outputIndexes": +[ 1632 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1632 ] +, "main_type": "NONE", "name": "Shape6724", "outputIndexes": +[ 1633 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1632 ] +, "main_type": "NONE", "name": "Rank6726", "outputIndexes": +[ 1634 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1634, 1634 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6728", "outputIndexes": +[ 1635 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1635 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6729", "outputIndexes": +[ 1636 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1635, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6731", "outputIndexes": +[ 1637 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1637 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6732", "outputIndexes": +[ 1638 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1633, 1636, 1638, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6735", "outputIndexes": +[ 1639 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1639 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6736", "outputIndexes": +[ 1640 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1640, 1640 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6738", "outputIndexes": +[ 1641 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1632, 1641, 1635 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_258_output_0", "outputIndexes": +[ 1642 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1642, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_544_output_0", "outputIndexes": +[ 1643 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1635 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6746", "outputIndexes": +[ 1644 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1637 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6749", "outputIndexes": +[ 1645 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1633, 1644, 1645, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6752", "outputIndexes": +[ 1646 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1646 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6753", "outputIndexes": +[ 1647 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1647 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6754", "outputIndexes": +[ 1648 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1648, 1647 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6755", "outputIndexes": +[ 1649 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1632, 1649, 1635 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_259_output_0", "outputIndexes": +[ 1650 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1650, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_545_output_0", "outputIndexes": +[ 1651 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1643, 1651, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_221_output_0", "outputIndexes": +[ 1652 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1631, 1652 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_154_output_0", "outputIndexes": +[ 1653 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1653 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2061636176, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_17/Mul_1_output_0", "outputIndexes": +[ 1655 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1655, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_136_output_0", "outputIndexes": +[ 1656 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1655 ] +, "main_type": "NONE", "name": "/Shape_240_output_0", "outputIndexes": +[ 1657 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1657 ] +, "main_type": "NONE", "name": "Shape6874", "outputIndexes": +[ 1658 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1657 ] +, "main_type": "NONE", "name": "Rank6876", "outputIndexes": +[ 1659 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1659, 1659 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6878", "outputIndexes": +[ 1660 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1660 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6879", "outputIndexes": +[ 1661 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1660, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6881", "outputIndexes": +[ 1662 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1662 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6882", "outputIndexes": +[ 1663 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1658, 1661, 1663, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6885", "outputIndexes": +[ 1664 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1664 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6886", "outputIndexes": +[ 1665 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1665 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6887", "outputIndexes": +[ 1666 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1666, 1665 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6888", "outputIndexes": +[ 1667 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1657, 1667, 1660 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_260_output_0", "outputIndexes": +[ 1668 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1668, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_51_output_0", "outputIndexes": +[ 1669 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1669, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_551_output_0", "outputIndexes": +[ 1670 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1655, 1670, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_69_output_0", "outputIndexes": +[ 1671 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1671 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_34_output_0", "outputIndexes": +[ 1672 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1669, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_550_output_0", "outputIndexes": +[ 1673 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1655, 22, 1673, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_68_output_0", "outputIndexes": +[ 1674 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1672, 1674 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_224_output_0", "outputIndexes": +[ 1675 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1675, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_137_output_0", "outputIndexes": +[ 1676 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1656, 1676 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_85_output_0", "outputIndexes": +[ 1677 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3876 ] +, "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": +[ 3881 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3881 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3882 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3882 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1678 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1642, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_546_output_0", "outputIndexes": +[ 1679 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1650, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_547_output_0", "outputIndexes": +[ 1680 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1679, 1680, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_222_output_0", "outputIndexes": +[ 1681 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1678, 1681 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_155_output_0", "outputIndexes": +[ 1682 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1682 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2064258662, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_17/Mul_1_output_0", "outputIndexes": +[ 1684 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1684, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_138_output_0", "outputIndexes": +[ 1685 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1684 ] +, "main_type": "NONE", "name": "/Shape_241_output_0", "outputIndexes": +[ 1686 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1686 ] +, "main_type": "NONE", "name": "Shape6770", "outputIndexes": +[ 1687 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1686 ] +, "main_type": "NONE", "name": "Rank6772", "outputIndexes": +[ 1688 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1688, 1688 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6774", "outputIndexes": +[ 1689 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1689 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6775", "outputIndexes": +[ 1690 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1689, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6777", "outputIndexes": +[ 1691 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1691 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze6778", "outputIndexes": +[ 1692 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1687, 1690, 1692, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice6781", "outputIndexes": +[ 1693 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1693 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze6782", "outputIndexes": +[ 1694 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1694 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6783", "outputIndexes": +[ 1695 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1695, 1694 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp6784", "outputIndexes": +[ 1696 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1686, 1696, 1689 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_261_output_0", "outputIndexes": +[ 1697 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1697, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_52_output_0", "outputIndexes": +[ 1698 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1698, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_553_output_0", "outputIndexes": +[ 1699 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1684, 1699, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_71_output_0", "outputIndexes": +[ 1700 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1700 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_35_output_0", "outputIndexes": +[ 1701 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1698, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_552_output_0", "outputIndexes": +[ 1702 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1684, 22, 1702, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_70_output_0", "outputIndexes": +[ 1703 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1701, 1703 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_225_output_0", "outputIndexes": +[ 1704 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1704, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_139_output_0", "outputIndexes": +[ 1705 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1685, 1705 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_86_output_0", "outputIndexes": +[ 1706 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3876 ] +, "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": +[ 3885 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3885 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3886 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3886 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1707 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1642, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_548_output_0", "outputIndexes": +[ 1708 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1650, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_549_output_0", "outputIndexes": +[ 1709 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1708, 1709, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_223_output_0", "outputIndexes": +[ 1710 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1707, 1710 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_156_output_0", "outputIndexes": +[ 1711 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1677, 1706, 1711, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_161_output_0", "outputIndexes": +[ 1712 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1712 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3887 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3887 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3888 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3888 ] +, "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": +[ 3889 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3889 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3890 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3890 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.17/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1713 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1629, 1713 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_88_output_0", "outputIndexes": +[ 1714 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1714 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2077366930, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_17/Mul_1_output_0", "outputIndexes": +[ 1715 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1715 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3891 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3891 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3892 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3892 ] +, "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": +[ 3893 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3893 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3894 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3894 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1716 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1716 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_17/Mul_output_0", "outputIndexes": +[ 1717 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3892 ] +, "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": +[ 3897 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3897 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3898 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3898 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1718 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1717, 1718 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_17/Mul_output_0", "outputIndexes": +[ 1719 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1719 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3899 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3899 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.17/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3900 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3900 ] +, "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": +[ 3901 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3901 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.17/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3902 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3902 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.17/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1720 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1714, 1720 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_89_output_0", "outputIndexes": +[ 1721 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1721, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_162_output_0", "outputIndexes": +[ 1722 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1722 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2171771604, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_18/Mul_1_output_0", "outputIndexes": +[ 1723 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1723 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3903 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3903 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3904 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3904 ] +, "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": +[ 3905 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3905 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3906 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3906 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1724 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1723 ] +, "main_type": "NONE", "name": "/Shape_252_output_0", "outputIndexes": +[ 1725 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1725 ] +, "main_type": "NONE", "name": "Shape7113", "outputIndexes": +[ 1726 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1725 ] +, "main_type": "NONE", "name": "Rank7115", "outputIndexes": +[ 1727 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1727, 1727 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7117", "outputIndexes": +[ 1728 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1728 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7118", "outputIndexes": +[ 1729 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1728, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7120", "outputIndexes": +[ 1730 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1730 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7121", "outputIndexes": +[ 1731 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1726, 1729, 1731, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7124", "outputIndexes": +[ 1732 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1732 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7125", "outputIndexes": +[ 1733 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1733, 1733 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7127", "outputIndexes": +[ 1734 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1725, 1734, 1728 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_273_output_0", "outputIndexes": +[ 1735 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1735, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_576_output_0", "outputIndexes": +[ 1736 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1728 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7135", "outputIndexes": +[ 1737 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1730 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7138", "outputIndexes": +[ 1738 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1726, 1737, 1738, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7141", "outputIndexes": +[ 1739 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1739 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7142", "outputIndexes": +[ 1740 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1740 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7143", "outputIndexes": +[ 1741 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1741, 1740 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7144", "outputIndexes": +[ 1742 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1725, 1742, 1728 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_274_output_0", "outputIndexes": +[ 1743 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1743, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_577_output_0", "outputIndexes": +[ 1744 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1736, 1744, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_234_output_0", "outputIndexes": +[ 1745 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1724, 1745 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_163_output_0", "outputIndexes": +[ 1746 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1746 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2182290154, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_18/Mul_1_output_0", "outputIndexes": +[ 1748 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1748, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_144_output_0", "outputIndexes": +[ 1749 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1748 ] +, "main_type": "NONE", "name": "/Shape_254_output_0", "outputIndexes": +[ 1750 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1750 ] +, "main_type": "NONE", "name": "Shape7263", "outputIndexes": +[ 1751 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1750 ] +, "main_type": "NONE", "name": "Rank7265", "outputIndexes": +[ 1752 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1752, 1752 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7267", "outputIndexes": +[ 1753 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1753 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7268", "outputIndexes": +[ 1754 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1753, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7270", "outputIndexes": +[ 1755 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1755 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7271", "outputIndexes": +[ 1756 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1751, 1754, 1756, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7274", "outputIndexes": +[ 1757 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1757 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7275", "outputIndexes": +[ 1758 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1758 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7276", "outputIndexes": +[ 1759 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1759, 1758 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7277", "outputIndexes": +[ 1760 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1750, 1760, 1753 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_275_output_0", "outputIndexes": +[ 1761 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1761, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_54_output_0", "outputIndexes": +[ 1762 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1762, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_583_output_0", "outputIndexes": +[ 1763 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1748, 1763, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_73_output_0", "outputIndexes": +[ 1764 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1764 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_36_output_0", "outputIndexes": +[ 1765 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1762, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_582_output_0", "outputIndexes": +[ 1766 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1748, 22, 1766, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_72_output_0", "outputIndexes": +[ 1767 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1765, 1767 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_237_output_0", "outputIndexes": +[ 1768 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1768, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_145_output_0", "outputIndexes": +[ 1769 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1749, 1769 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_90_output_0", "outputIndexes": +[ 1770 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3904 ] +, "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": +[ 3909 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3909 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3910 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3910 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1771 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1735, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_578_output_0", "outputIndexes": +[ 1772 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1743, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_579_output_0", "outputIndexes": +[ 1773 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1772, 1773, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_235_output_0", "outputIndexes": +[ 1774 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1771, 1774 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_164_output_0", "outputIndexes": +[ 1775 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1775 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2184912640, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_18/Mul_1_output_0", "outputIndexes": +[ 1777 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1777, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_146_output_0", "outputIndexes": +[ 1778 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1777 ] +, "main_type": "NONE", "name": "/Shape_255_output_0", "outputIndexes": +[ 1779 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1779 ] +, "main_type": "NONE", "name": "Shape7159", "outputIndexes": +[ 1780 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1779 ] +, "main_type": "NONE", "name": "Rank7161", "outputIndexes": +[ 1781 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1781, 1781 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7163", "outputIndexes": +[ 1782 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1782 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7164", "outputIndexes": +[ 1783 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1782, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7166", "outputIndexes": +[ 1784 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1784 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7167", "outputIndexes": +[ 1785 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1780, 1783, 1785, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7170", "outputIndexes": +[ 1786 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1786 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7171", "outputIndexes": +[ 1787 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1787 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7172", "outputIndexes": +[ 1788 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1788, 1787 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7173", "outputIndexes": +[ 1789 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1779, 1789, 1782 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_276_output_0", "outputIndexes": +[ 1790 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1790, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_55_output_0", "outputIndexes": +[ 1791 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1791, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_585_output_0", "outputIndexes": +[ 1792 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1777, 1792, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_75_output_0", "outputIndexes": +[ 1793 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1793 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_37_output_0", "outputIndexes": +[ 1794 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1791, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_584_output_0", "outputIndexes": +[ 1795 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1777, 22, 1795, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_74_output_0", "outputIndexes": +[ 1796 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1794, 1796 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_238_output_0", "outputIndexes": +[ 1797 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1797, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_147_output_0", "outputIndexes": +[ 1798 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1778, 1798 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_91_output_0", "outputIndexes": +[ 1799 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3904 ] +, "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": +[ 3913 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3913 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3914 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3914 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1800 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1735, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_580_output_0", "outputIndexes": +[ 1801 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1743, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_581_output_0", "outputIndexes": +[ 1802 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1801, 1802, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_236_output_0", "outputIndexes": +[ 1803 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1800, 1803 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_165_output_0", "outputIndexes": +[ 1804 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1770, 1799, 1804, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_170_output_0", "outputIndexes": +[ 1805 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1805 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3915 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3915 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3916 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3916 ] +, "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": +[ 3917 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3917 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3918 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3918 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.18/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1806 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1722, 1806 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_93_output_0", "outputIndexes": +[ 1807 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1807 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2198020908, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_18/Mul_1_output_0", "outputIndexes": +[ 1808 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1808 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3919 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3919 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3920 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3920 ] +, "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": +[ 3921 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3921 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3922 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3922 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1809 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1809 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_18/Mul_output_0", "outputIndexes": +[ 1810 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3920 ] +, "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": +[ 3925 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3925 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3926 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3926 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1811 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1810, 1811 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_18/Mul_output_0", "outputIndexes": +[ 1812 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1812 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3927 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3927 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.18/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3928 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3928 ] +, "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": +[ 3929 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3929 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.18/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3930 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3930 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.18/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1813 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1807, 1813 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_94_output_0", "outputIndexes": +[ 1814 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1814, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_171_output_0", "outputIndexes": +[ 1815 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1815 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2292425582, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_19/Mul_1_output_0", "outputIndexes": +[ 1816 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1816 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3931 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3931 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3932 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3932 ] +, "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": +[ 3933 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3933 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3934 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3934 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1817 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1816 ] +, "main_type": "NONE", "name": "/Shape_266_output_0", "outputIndexes": +[ 1818 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1818 ] +, "main_type": "NONE", "name": "Shape7502", "outputIndexes": +[ 1819 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1818 ] +, "main_type": "NONE", "name": "Rank7504", "outputIndexes": +[ 1820 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1820, 1820 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7506", "outputIndexes": +[ 1821 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1821 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7507", "outputIndexes": +[ 1822 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1821, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7509", "outputIndexes": +[ 1823 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1823 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7510", "outputIndexes": +[ 1824 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1819, 1822, 1824, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7513", "outputIndexes": +[ 1825 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1825 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7514", "outputIndexes": +[ 1826 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1826, 1826 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7516", "outputIndexes": +[ 1827 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1818, 1827, 1821 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_288_output_0", "outputIndexes": +[ 1828 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1828, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_608_output_0", "outputIndexes": +[ 1829 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1821 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7524", "outputIndexes": +[ 1830 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1823 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7527", "outputIndexes": +[ 1831 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1819, 1830, 1831, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7530", "outputIndexes": +[ 1832 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1832 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7531", "outputIndexes": +[ 1833 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1833 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7532", "outputIndexes": +[ 1834 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1834, 1833 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7533", "outputIndexes": +[ 1835 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1818, 1835, 1821 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_289_output_0", "outputIndexes": +[ 1836 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1836, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_609_output_0", "outputIndexes": +[ 1837 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1829, 1837, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_247_output_0", "outputIndexes": +[ 1838 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1817, 1838 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_172_output_0", "outputIndexes": +[ 1839 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1839 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2302944132, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_19/Mul_1_output_0", "outputIndexes": +[ 1841 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1841, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_152_output_0", "outputIndexes": +[ 1842 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1841 ] +, "main_type": "NONE", "name": "/Shape_268_output_0", "outputIndexes": +[ 1843 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1843 ] +, "main_type": "NONE", "name": "Shape7652", "outputIndexes": +[ 1844 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1843 ] +, "main_type": "NONE", "name": "Rank7654", "outputIndexes": +[ 1845 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1845, 1845 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7656", "outputIndexes": +[ 1846 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1846 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7657", "outputIndexes": +[ 1847 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1846, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7659", "outputIndexes": +[ 1848 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1848 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7660", "outputIndexes": +[ 1849 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1844, 1847, 1849, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7663", "outputIndexes": +[ 1850 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1850 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7664", "outputIndexes": +[ 1851 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1851 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7665", "outputIndexes": +[ 1852 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1852, 1851 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7666", "outputIndexes": +[ 1853 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1843, 1853, 1846 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_290_output_0", "outputIndexes": +[ 1854 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1854, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_57_output_0", "outputIndexes": +[ 1855 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1855, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_615_output_0", "outputIndexes": +[ 1856 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1841, 1856, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_77_output_0", "outputIndexes": +[ 1857 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1857 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_38_output_0", "outputIndexes": +[ 1858 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1855, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_614_output_0", "outputIndexes": +[ 1859 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1841, 22, 1859, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_76_output_0", "outputIndexes": +[ 1860 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1858, 1860 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_250_output_0", "outputIndexes": +[ 1861 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1861, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_153_output_0", "outputIndexes": +[ 1862 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1842, 1862 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_95_output_0", "outputIndexes": +[ 1863 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3932 ] +, "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": +[ 3937 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3937 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3938 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3938 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1864 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1828, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_610_output_0", "outputIndexes": +[ 1865 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1836, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_611_output_0", "outputIndexes": +[ 1866 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1865, 1866, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_248_output_0", "outputIndexes": +[ 1867 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1864, 1867 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_173_output_0", "outputIndexes": +[ 1868 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1868 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2305566618, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_19/Mul_1_output_0", "outputIndexes": +[ 1870 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1870, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_154_output_0", "outputIndexes": +[ 1871 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1870 ] +, "main_type": "NONE", "name": "/Shape_269_output_0", "outputIndexes": +[ 1872 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1872 ] +, "main_type": "NONE", "name": "Shape7548", "outputIndexes": +[ 1873 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1872 ] +, "main_type": "NONE", "name": "Rank7550", "outputIndexes": +[ 1874 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1874, 1874 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7552", "outputIndexes": +[ 1875 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1875 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7553", "outputIndexes": +[ 1876 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1875, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7555", "outputIndexes": +[ 1877 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1877 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7556", "outputIndexes": +[ 1878 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1873, 1876, 1878, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7559", "outputIndexes": +[ 1879 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1879 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7560", "outputIndexes": +[ 1880 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1880 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7561", "outputIndexes": +[ 1881 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1881, 1880 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7562", "outputIndexes": +[ 1882 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1872, 1882, 1875 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_291_output_0", "outputIndexes": +[ 1883 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1883, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_58_output_0", "outputIndexes": +[ 1884 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1884, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_617_output_0", "outputIndexes": +[ 1885 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1870, 1885, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_79_output_0", "outputIndexes": +[ 1886 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1886 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_39_output_0", "outputIndexes": +[ 1887 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1884, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_616_output_0", "outputIndexes": +[ 1888 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1870, 22, 1888, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_78_output_0", "outputIndexes": +[ 1889 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1887, 1889 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_251_output_0", "outputIndexes": +[ 1890 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1890, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_155_output_0", "outputIndexes": +[ 1891 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1871, 1891 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_96_output_0", "outputIndexes": +[ 1892 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3932 ] +, "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": +[ 3941 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3941 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3942 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3942 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1893 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1828, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_612_output_0", "outputIndexes": +[ 1894 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1836, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_613_output_0", "outputIndexes": +[ 1895 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1894, 1895, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_249_output_0", "outputIndexes": +[ 1896 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1893, 1896 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_174_output_0", "outputIndexes": +[ 1897 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1863, 1892, 1897, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_179_output_0", "outputIndexes": +[ 1898 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1898 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3943 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3943 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3944 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3944 ] +, "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": +[ 3945 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3945 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3946 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3946 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.19/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1899 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1815, 1899 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_98_output_0", "outputIndexes": +[ 1900 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1900 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2318674886, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_19/Mul_1_output_0", "outputIndexes": +[ 1901 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1901 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3947 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3947 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3948 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3948 ] +, "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": +[ 3949 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3949 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3950 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3950 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1902 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1902 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_19/Mul_output_0", "outputIndexes": +[ 1903 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3948 ] +, "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": +[ 3953 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3953 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3954 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3954 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1904 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1903, 1904 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_19/Mul_output_0", "outputIndexes": +[ 1905 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1905 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3955 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3955 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.19/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3956 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3956 ] +, "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": +[ 3957 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3957 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.19/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3958 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3958 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.19/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1906 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1900, 1906 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_99_output_0", "outputIndexes": +[ 1907 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1907, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_180_output_0", "outputIndexes": +[ 1908 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1908 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2413079560, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_20/Mul_1_output_0", "outputIndexes": +[ 1909 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1909 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3959 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3959 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3960 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3960 ] +, "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": +[ 3961 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3961 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3962 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3962 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 1910 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1909 ] +, "main_type": "NONE", "name": "/Shape_280_output_0", "outputIndexes": +[ 1911 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1911 ] +, "main_type": "NONE", "name": "Shape7891", "outputIndexes": +[ 1912 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1911 ] +, "main_type": "NONE", "name": "Rank7893", "outputIndexes": +[ 1913 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1913, 1913 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7895", "outputIndexes": +[ 1914 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1914 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7896", "outputIndexes": +[ 1915 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1914, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7898", "outputIndexes": +[ 1916 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1916 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7899", "outputIndexes": +[ 1917 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1912, 1915, 1917, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7902", "outputIndexes": +[ 1918 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1918 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7903", "outputIndexes": +[ 1919 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1919, 1919 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7905", "outputIndexes": +[ 1920 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1911, 1920, 1914 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_303_output_0", "outputIndexes": +[ 1921 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1921, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_640_output_0", "outputIndexes": +[ 1922 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1914 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7913", "outputIndexes": +[ 1923 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1916 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7916", "outputIndexes": +[ 1924 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1912, 1923, 1924, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7919", "outputIndexes": +[ 1925 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1925 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7920", "outputIndexes": +[ 1926 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 1926 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7921", "outputIndexes": +[ 1927 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1927, 1926 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7922", "outputIndexes": +[ 1928 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1911, 1928, 1914 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_304_output_0", "outputIndexes": +[ 1929 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1929, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_641_output_0", "outputIndexes": +[ 1930 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1922, 1930, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_260_output_0", "outputIndexes": +[ 1931 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1910, 1931 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_181_output_0", "outputIndexes": +[ 1932 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1932 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2423598110, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_20/Mul_1_output_0", "outputIndexes": +[ 1934 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1934, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_160_output_0", "outputIndexes": +[ 1935 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1934 ] +, "main_type": "NONE", "name": "/Shape_282_output_0", "outputIndexes": +[ 1936 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1936 ] +, "main_type": "NONE", "name": "Shape8041", "outputIndexes": +[ 1937 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1936 ] +, "main_type": "NONE", "name": "Rank8043", "outputIndexes": +[ 1938 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1938, 1938 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8045", "outputIndexes": +[ 1939 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1939 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8046", "outputIndexes": +[ 1940 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1939, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8048", "outputIndexes": +[ 1941 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1941 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8049", "outputIndexes": +[ 1942 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1937, 1940, 1942, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8052", "outputIndexes": +[ 1943 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1943 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8053", "outputIndexes": +[ 1944 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1944 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8054", "outputIndexes": +[ 1945 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1945, 1944 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8055", "outputIndexes": +[ 1946 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1936, 1946, 1939 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_305_output_0", "outputIndexes": +[ 1947 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1947, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_60_output_0", "outputIndexes": +[ 1948 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1948, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_647_output_0", "outputIndexes": +[ 1949 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1934, 1949, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_81_output_0", "outputIndexes": +[ 1950 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1950 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_40_output_0", "outputIndexes": +[ 1951 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1948, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_646_output_0", "outputIndexes": +[ 1952 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1934, 22, 1952, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_80_output_0", "outputIndexes": +[ 1953 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1951, 1953 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_263_output_0", "outputIndexes": +[ 1954 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1954, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_161_output_0", "outputIndexes": +[ 1955 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1935, 1955 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_100_output_0", "outputIndexes": +[ 1956 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3960 ] +, "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": +[ 3965 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3965 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3966 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3966 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 1957 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1921, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_642_output_0", "outputIndexes": +[ 1958 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1929, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_643_output_0", "outputIndexes": +[ 1959 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1958, 1959, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_261_output_0", "outputIndexes": +[ 1960 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1957, 1960 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_182_output_0", "outputIndexes": +[ 1961 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1961 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2426220596, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_20/Mul_1_output_0", "outputIndexes": +[ 1963 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1963, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_162_output_0", "outputIndexes": +[ 1964 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1963 ] +, "main_type": "NONE", "name": "/Shape_283_output_0", "outputIndexes": +[ 1965 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1965 ] +, "main_type": "NONE", "name": "Shape7937", "outputIndexes": +[ 1966 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 1965 ] +, "main_type": "NONE", "name": "Rank7939", "outputIndexes": +[ 1967 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1967, 1967 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7941", "outputIndexes": +[ 1968 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1968 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7942", "outputIndexes": +[ 1969 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1968, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7944", "outputIndexes": +[ 1970 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1970 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze7945", "outputIndexes": +[ 1971 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1966, 1969, 1971, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice7948", "outputIndexes": +[ 1972 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1972 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze7949", "outputIndexes": +[ 1973 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 1973 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7950", "outputIndexes": +[ 1974 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1974, 1973 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp7951", "outputIndexes": +[ 1975 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1965, 1975, 1968 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_306_output_0", "outputIndexes": +[ 1976 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1976, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_61_output_0", "outputIndexes": +[ 1977 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1977, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_649_output_0", "outputIndexes": +[ 1978 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1963, 1978, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_83_output_0", "outputIndexes": +[ 1979 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1979 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_41_output_0", "outputIndexes": +[ 1980 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1977, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_648_output_0", "outputIndexes": +[ 1981 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1963, 22, 1981, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_82_output_0", "outputIndexes": +[ 1982 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1980, 1982 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_264_output_0", "outputIndexes": +[ 1983 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1983, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_163_output_0", "outputIndexes": +[ 1984 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1964, 1984 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_101_output_0", "outputIndexes": +[ 1985 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3960 ] +, "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": +[ 3969 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3969 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3970 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3970 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 1986 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1921, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_644_output_0", "outputIndexes": +[ 1987 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1929, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_645_output_0", "outputIndexes": +[ 1988 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1987, 1988, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_262_output_0", "outputIndexes": +[ 1989 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1986, 1989 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_183_output_0", "outputIndexes": +[ 1990 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1956, 1985, 1990, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_188_output_0", "outputIndexes": +[ 1991 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1991 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3971 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3971 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 3972 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3972 ] +, "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": +[ 3973 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3973 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 3974 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3974 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.20/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 1992 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1908, 1992 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_103_output_0", "outputIndexes": +[ 1993 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1993 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2439328864, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_20/Mul_1_output_0", "outputIndexes": +[ 1994 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1994 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 3975 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3975 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 3976 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3976 ] +, "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": +[ 3977 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3977 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 3978 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3978 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 1995 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1995 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_20/Mul_output_0", "outputIndexes": +[ 1996 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3976 ] +, "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": +[ 3981 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3981 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 3982 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3982 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 1997 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1996, 1997 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_20/Mul_output_0", "outputIndexes": +[ 1998 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1998 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 3983 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3983 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.20/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 3984 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3984 ] +, "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": +[ 3985 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3985 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.20/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 3986 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3986 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.20/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 1999 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 1993, 1999 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_104_output_0", "outputIndexes": +[ 2000 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2000, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_189_output_0", "outputIndexes": +[ 2001 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2001 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2533733538, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_21/Mul_1_output_0", "outputIndexes": +[ 2002 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2002 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 3987 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3987 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 3988 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3988 ] +, "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": +[ 3989 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3989 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 3990 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3990 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2003 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2002 ] +, "main_type": "NONE", "name": "/Shape_294_output_0", "outputIndexes": +[ 2004 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2004 ] +, "main_type": "NONE", "name": "Shape8280", "outputIndexes": +[ 2005 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2004 ] +, "main_type": "NONE", "name": "Rank8282", "outputIndexes": +[ 2006 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2006, 2006 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8284", "outputIndexes": +[ 2007 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2007 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8285", "outputIndexes": +[ 2008 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2007, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8287", "outputIndexes": +[ 2009 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2009 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8288", "outputIndexes": +[ 2010 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2005, 2008, 2010, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8291", "outputIndexes": +[ 2011 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2011 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8292", "outputIndexes": +[ 2012 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2012, 2012 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8294", "outputIndexes": +[ 2013 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2004, 2013, 2007 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_318_output_0", "outputIndexes": +[ 2014 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2014, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_672_output_0", "outputIndexes": +[ 2015 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2007 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8302", "outputIndexes": +[ 2016 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2009 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8305", "outputIndexes": +[ 2017 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2005, 2016, 2017, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8308", "outputIndexes": +[ 2018 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2018 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8309", "outputIndexes": +[ 2019 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2019 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8310", "outputIndexes": +[ 2020 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2020, 2019 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8311", "outputIndexes": +[ 2021 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2004, 2021, 2007 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_319_output_0", "outputIndexes": +[ 2022 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2022, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_673_output_0", "outputIndexes": +[ 2023 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2015, 2023, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_273_output_0", "outputIndexes": +[ 2024 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2003, 2024 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_190_output_0", "outputIndexes": +[ 2025 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2025 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2544252088, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_21/Mul_1_output_0", "outputIndexes": +[ 2027 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2027, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_168_output_0", "outputIndexes": +[ 2028 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2027 ] +, "main_type": "NONE", "name": "/Shape_296_output_0", "outputIndexes": +[ 2029 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2029 ] +, "main_type": "NONE", "name": "Shape8430", "outputIndexes": +[ 2030 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2029 ] +, "main_type": "NONE", "name": "Rank8432", "outputIndexes": +[ 2031 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2031, 2031 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8434", "outputIndexes": +[ 2032 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2032 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8435", "outputIndexes": +[ 2033 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2032, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8437", "outputIndexes": +[ 2034 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2034 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8438", "outputIndexes": +[ 2035 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2030, 2033, 2035, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8441", "outputIndexes": +[ 2036 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2036 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8442", "outputIndexes": +[ 2037 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2037 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8443", "outputIndexes": +[ 2038 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2038, 2037 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8444", "outputIndexes": +[ 2039 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2029, 2039, 2032 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_320_output_0", "outputIndexes": +[ 2040 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2040, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_63_output_0", "outputIndexes": +[ 2041 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2041, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_679_output_0", "outputIndexes": +[ 2042 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2027, 2042, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_85_output_0", "outputIndexes": +[ 2043 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2043 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_42_output_0", "outputIndexes": +[ 2044 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2041, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_678_output_0", "outputIndexes": +[ 2045 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2027, 22, 2045, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_84_output_0", "outputIndexes": +[ 2046 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2044, 2046 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_276_output_0", "outputIndexes": +[ 2047 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2047, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_169_output_0", "outputIndexes": +[ 2048 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2028, 2048 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_105_output_0", "outputIndexes": +[ 2049 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3988 ] +, "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": +[ 3993 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3993 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 3994 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3994 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2050 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2014, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_674_output_0", "outputIndexes": +[ 2051 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2022, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_675_output_0", "outputIndexes": +[ 2052 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2051, 2052, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_274_output_0", "outputIndexes": +[ 2053 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2050, 2053 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_191_output_0", "outputIndexes": +[ 2054 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2054 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2546874574, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_21/Mul_1_output_0", "outputIndexes": +[ 2056 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2056, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_170_output_0", "outputIndexes": +[ 2057 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2056 ] +, "main_type": "NONE", "name": "/Shape_297_output_0", "outputIndexes": +[ 2058 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2058 ] +, "main_type": "NONE", "name": "Shape8326", "outputIndexes": +[ 2059 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2058 ] +, "main_type": "NONE", "name": "Rank8328", "outputIndexes": +[ 2060 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2060, 2060 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8330", "outputIndexes": +[ 2061 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2061 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8331", "outputIndexes": +[ 2062 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2061, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8333", "outputIndexes": +[ 2063 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2063 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8334", "outputIndexes": +[ 2064 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2059, 2062, 2064, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8337", "outputIndexes": +[ 2065 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2065 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8338", "outputIndexes": +[ 2066 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2066 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8339", "outputIndexes": +[ 2067 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2067, 2066 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8340", "outputIndexes": +[ 2068 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2058, 2068, 2061 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_321_output_0", "outputIndexes": +[ 2069 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2069, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_64_output_0", "outputIndexes": +[ 2070 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2070, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_681_output_0", "outputIndexes": +[ 2071 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2056, 2071, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_87_output_0", "outputIndexes": +[ 2072 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2072 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_43_output_0", "outputIndexes": +[ 2073 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2070, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_680_output_0", "outputIndexes": +[ 2074 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2056, 22, 2074, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_86_output_0", "outputIndexes": +[ 2075 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2073, 2075 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_277_output_0", "outputIndexes": +[ 2076 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2076, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_171_output_0", "outputIndexes": +[ 2077 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2057, 2077 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_106_output_0", "outputIndexes": +[ 2078 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3988 ] +, "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": +[ 3997 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3997 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 3998 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3998 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2079 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2014, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_676_output_0", "outputIndexes": +[ 2080 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2022, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_677_output_0", "outputIndexes": +[ 2081 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2080, 2081, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_275_output_0", "outputIndexes": +[ 2082 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2079, 2082 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_192_output_0", "outputIndexes": +[ 2083 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2049, 2078, 2083, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_197_output_0", "outputIndexes": +[ 2084 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2084 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 3999 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3999 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4000 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4000 ] +, "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": +[ 4001 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4001 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4002 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4002 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.21/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2085 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2001, 2085 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_108_output_0", "outputIndexes": +[ 2086 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2086 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2559982842, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_21/Mul_1_output_0", "outputIndexes": +[ 2087 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2087 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4003 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4003 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4004 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4004 ] +, "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": +[ 4005 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4005 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4006 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4006 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2088 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2088 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_21/Mul_output_0", "outputIndexes": +[ 2089 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4004 ] +, "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": +[ 4009 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4009 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4010 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4010 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2090 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2089, 2090 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_21/Mul_output_0", "outputIndexes": +[ 2091 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2091 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4011 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4011 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.21/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4012 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4012 ] +, "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": +[ 4013 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4013 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.21/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4014 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4014 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.21/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2092 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2086, 2092 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_109_output_0", "outputIndexes": +[ 2093 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2093, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_198_output_0", "outputIndexes": +[ 2094 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2094 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2654387516, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_22/Mul_1_output_0", "outputIndexes": +[ 2095 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2095 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4015 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4015 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4016 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4016 ] +, "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": +[ 4017 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4017 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4018 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4018 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2096 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2095 ] +, "main_type": "NONE", "name": "/Shape_308_output_0", "outputIndexes": +[ 2097 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2097 ] +, "main_type": "NONE", "name": "Shape8669", "outputIndexes": +[ 2098 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2097 ] +, "main_type": "NONE", "name": "Rank8671", "outputIndexes": +[ 2099 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2099, 2099 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8673", "outputIndexes": +[ 2100 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2100 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8674", "outputIndexes": +[ 2101 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2100, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8676", "outputIndexes": +[ 2102 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2102 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8677", "outputIndexes": +[ 2103 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2098, 2101, 2103, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8680", "outputIndexes": +[ 2104 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2104 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8681", "outputIndexes": +[ 2105 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2105, 2105 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8683", "outputIndexes": +[ 2106 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2097, 2106, 2100 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_333_output_0", "outputIndexes": +[ 2107 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2107, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_704_output_0", "outputIndexes": +[ 2108 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2100 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8691", "outputIndexes": +[ 2109 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2102 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8694", "outputIndexes": +[ 2110 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2098, 2109, 2110, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8697", "outputIndexes": +[ 2111 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2111 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8698", "outputIndexes": +[ 2112 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2112 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8699", "outputIndexes": +[ 2113 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2113, 2112 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8700", "outputIndexes": +[ 2114 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2097, 2114, 2100 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_334_output_0", "outputIndexes": +[ 2115 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2115, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_705_output_0", "outputIndexes": +[ 2116 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2108, 2116, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_286_output_0", "outputIndexes": +[ 2117 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2096, 2117 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_199_output_0", "outputIndexes": +[ 2118 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2118 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2664906066, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_22/Mul_1_output_0", "outputIndexes": +[ 2120 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2120, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_176_output_0", "outputIndexes": +[ 2121 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2120 ] +, "main_type": "NONE", "name": "/Shape_310_output_0", "outputIndexes": +[ 2122 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2122 ] +, "main_type": "NONE", "name": "Shape8819", "outputIndexes": +[ 2123 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2122 ] +, "main_type": "NONE", "name": "Rank8821", "outputIndexes": +[ 2124 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2124, 2124 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8823", "outputIndexes": +[ 2125 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2125 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8824", "outputIndexes": +[ 2126 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2125, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8826", "outputIndexes": +[ 2127 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2127 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8827", "outputIndexes": +[ 2128 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2123, 2126, 2128, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8830", "outputIndexes": +[ 2129 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2129 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8831", "outputIndexes": +[ 2130 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2130 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8832", "outputIndexes": +[ 2131 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2131, 2130 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8833", "outputIndexes": +[ 2132 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2122, 2132, 2125 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_335_output_0", "outputIndexes": +[ 2133 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2133, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_66_output_0", "outputIndexes": +[ 2134 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2134, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_711_output_0", "outputIndexes": +[ 2135 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2120, 2135, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_89_output_0", "outputIndexes": +[ 2136 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2136 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_44_output_0", "outputIndexes": +[ 2137 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2134, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_710_output_0", "outputIndexes": +[ 2138 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2120, 22, 2138, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_88_output_0", "outputIndexes": +[ 2139 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2137, 2139 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_289_output_0", "outputIndexes": +[ 2140 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2140, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_177_output_0", "outputIndexes": +[ 2141 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2121, 2141 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_110_output_0", "outputIndexes": +[ 2142 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4016 ] +, "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": +[ 4021 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4021 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4022 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4022 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2143 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2107, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_706_output_0", "outputIndexes": +[ 2144 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2115, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_707_output_0", "outputIndexes": +[ 2145 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2144, 2145, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_287_output_0", "outputIndexes": +[ 2146 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2143, 2146 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_200_output_0", "outputIndexes": +[ 2147 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2147 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2667528552, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_22/Mul_1_output_0", "outputIndexes": +[ 2149 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2149, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_178_output_0", "outputIndexes": +[ 2150 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2149 ] +, "main_type": "NONE", "name": "/Shape_311_output_0", "outputIndexes": +[ 2151 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2151 ] +, "main_type": "NONE", "name": "Shape8715", "outputIndexes": +[ 2152 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2151 ] +, "main_type": "NONE", "name": "Rank8717", "outputIndexes": +[ 2153 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2153, 2153 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8719", "outputIndexes": +[ 2154 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2154 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8720", "outputIndexes": +[ 2155 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2154, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8722", "outputIndexes": +[ 2156 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2156 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze8723", "outputIndexes": +[ 2157 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2152, 2155, 2157, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice8726", "outputIndexes": +[ 2158 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2158 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze8727", "outputIndexes": +[ 2159 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2159 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8728", "outputIndexes": +[ 2160 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2160, 2159 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp8729", "outputIndexes": +[ 2161 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2151, 2161, 2154 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_336_output_0", "outputIndexes": +[ 2162 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2162, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_67_output_0", "outputIndexes": +[ 2163 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2163, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_713_output_0", "outputIndexes": +[ 2164 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2149, 2164, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_91_output_0", "outputIndexes": +[ 2165 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2165 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_45_output_0", "outputIndexes": +[ 2166 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2163, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_712_output_0", "outputIndexes": +[ 2167 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2149, 22, 2167, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_90_output_0", "outputIndexes": +[ 2168 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2166, 2168 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_290_output_0", "outputIndexes": +[ 2169 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2169, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_179_output_0", "outputIndexes": +[ 2170 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2150, 2170 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_111_output_0", "outputIndexes": +[ 2171 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4016 ] +, "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": +[ 4025 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4025 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4026 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4026 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2172 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2107, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_708_output_0", "outputIndexes": +[ 2173 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2115, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_709_output_0", "outputIndexes": +[ 2174 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2173, 2174, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_288_output_0", "outputIndexes": +[ 2175 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2172, 2175 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_201_output_0", "outputIndexes": +[ 2176 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2142, 2171, 2176, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_206_output_0", "outputIndexes": +[ 2177 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2177 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4027 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4027 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4028 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4028 ] +, "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": +[ 4029 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4029 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4030 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4030 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.22/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2178 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2094, 2178 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_113_output_0", "outputIndexes": +[ 2179 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2179 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2680636820, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_22/Mul_1_output_0", "outputIndexes": +[ 2180 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2180 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4031 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4031 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4032 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4032 ] +, "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": +[ 4033 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4033 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4034 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4034 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2181 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2181 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_22/Mul_output_0", "outputIndexes": +[ 2182 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4032 ] +, "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": +[ 4037 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4037 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4038 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4038 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2183 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2182, 2183 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_22/Mul_output_0", "outputIndexes": +[ 2184 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2184 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4039 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4039 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.22/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4040 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4040 ] +, "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": +[ 4041 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4041 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.22/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4042 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4042 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.22/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2185 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2179, 2185 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_114_output_0", "outputIndexes": +[ 2186 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2186, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_207_output_0", "outputIndexes": +[ 2187 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2187 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2775041494, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_23/Mul_1_output_0", "outputIndexes": +[ 2188 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2188 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4043 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4043 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4044 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4044 ] +, "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": +[ 4045 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4045 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4046 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4046 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2189 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2188 ] +, "main_type": "NONE", "name": "/Shape_322_output_0", "outputIndexes": +[ 2190 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2190 ] +, "main_type": "NONE", "name": "Shape9058", "outputIndexes": +[ 2191 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2190 ] +, "main_type": "NONE", "name": "Rank9060", "outputIndexes": +[ 2192 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2192, 2192 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9062", "outputIndexes": +[ 2193 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2193 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9063", "outputIndexes": +[ 2194 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2193, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9065", "outputIndexes": +[ 2195 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2195 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9066", "outputIndexes": +[ 2196 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2191, 2194, 2196, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9069", "outputIndexes": +[ 2197 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2197 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9070", "outputIndexes": +[ 2198 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2198, 2198 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9072", "outputIndexes": +[ 2199 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2190, 2199, 2193 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_348_output_0", "outputIndexes": +[ 2200 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2200, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_736_output_0", "outputIndexes": +[ 2201 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2193 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9080", "outputIndexes": +[ 2202 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2195 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9083", "outputIndexes": +[ 2203 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2191, 2202, 2203, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9086", "outputIndexes": +[ 2204 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2204 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9087", "outputIndexes": +[ 2205 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2205 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9088", "outputIndexes": +[ 2206 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2206, 2205 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9089", "outputIndexes": +[ 2207 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2190, 2207, 2193 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_349_output_0", "outputIndexes": +[ 2208 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2208, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_737_output_0", "outputIndexes": +[ 2209 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2201, 2209, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_299_output_0", "outputIndexes": +[ 2210 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2189, 2210 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_208_output_0", "outputIndexes": +[ 2211 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2211 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2785560044, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_23/Mul_1_output_0", "outputIndexes": +[ 2213 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2213, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_184_output_0", "outputIndexes": +[ 2214 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2213 ] +, "main_type": "NONE", "name": "/Shape_324_output_0", "outputIndexes": +[ 2215 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2215 ] +, "main_type": "NONE", "name": "Shape9208", "outputIndexes": +[ 2216 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2215 ] +, "main_type": "NONE", "name": "Rank9210", "outputIndexes": +[ 2217 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2217, 2217 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9212", "outputIndexes": +[ 2218 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2218 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9213", "outputIndexes": +[ 2219 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2218, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9215", "outputIndexes": +[ 2220 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2220 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9216", "outputIndexes": +[ 2221 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2216, 2219, 2221, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9219", "outputIndexes": +[ 2222 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2222 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9220", "outputIndexes": +[ 2223 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2223 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9221", "outputIndexes": +[ 2224 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2224, 2223 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9222", "outputIndexes": +[ 2225 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2215, 2225, 2218 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_350_output_0", "outputIndexes": +[ 2226 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2226, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_69_output_0", "outputIndexes": +[ 2227 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2227, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_743_output_0", "outputIndexes": +[ 2228 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2213, 2228, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_93_output_0", "outputIndexes": +[ 2229 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2229 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_46_output_0", "outputIndexes": +[ 2230 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2227, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_742_output_0", "outputIndexes": +[ 2231 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2213, 22, 2231, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_92_output_0", "outputIndexes": +[ 2232 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2230, 2232 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_302_output_0", "outputIndexes": +[ 2233 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2233, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_185_output_0", "outputIndexes": +[ 2234 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2214, 2234 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_115_output_0", "outputIndexes": +[ 2235 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4044 ] +, "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": +[ 4049 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4049 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4050 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4050 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2236 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2200, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_738_output_0", "outputIndexes": +[ 2237 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2208, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_739_output_0", "outputIndexes": +[ 2238 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2237, 2238, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_300_output_0", "outputIndexes": +[ 2239 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2236, 2239 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_209_output_0", "outputIndexes": +[ 2240 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2240 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2788182530, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_23/Mul_1_output_0", "outputIndexes": +[ 2242 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2242, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_186_output_0", "outputIndexes": +[ 2243 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2242 ] +, "main_type": "NONE", "name": "/Shape_325_output_0", "outputIndexes": +[ 2244 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2244 ] +, "main_type": "NONE", "name": "Shape9104", "outputIndexes": +[ 2245 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2244 ] +, "main_type": "NONE", "name": "Rank9106", "outputIndexes": +[ 2246 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2246, 2246 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9108", "outputIndexes": +[ 2247 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2247 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9109", "outputIndexes": +[ 2248 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2247, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9111", "outputIndexes": +[ 2249 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2249 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9112", "outputIndexes": +[ 2250 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2245, 2248, 2250, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9115", "outputIndexes": +[ 2251 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2251 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9116", "outputIndexes": +[ 2252 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2252 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9117", "outputIndexes": +[ 2253 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2253, 2252 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9118", "outputIndexes": +[ 2254 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2244, 2254, 2247 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_351_output_0", "outputIndexes": +[ 2255 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2255, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_70_output_0", "outputIndexes": +[ 2256 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2256, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_745_output_0", "outputIndexes": +[ 2257 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2242, 2257, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_95_output_0", "outputIndexes": +[ 2258 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2258 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_47_output_0", "outputIndexes": +[ 2259 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2256, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_744_output_0", "outputIndexes": +[ 2260 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2242, 22, 2260, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_94_output_0", "outputIndexes": +[ 2261 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2259, 2261 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_303_output_0", "outputIndexes": +[ 2262 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2262, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_187_output_0", "outputIndexes": +[ 2263 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2243, 2263 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_116_output_0", "outputIndexes": +[ 2264 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4044 ] +, "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": +[ 4053 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4053 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4054 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4054 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2265 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2200, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_740_output_0", "outputIndexes": +[ 2266 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2208, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_741_output_0", "outputIndexes": +[ 2267 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2266, 2267, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_301_output_0", "outputIndexes": +[ 2268 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2265, 2268 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_210_output_0", "outputIndexes": +[ 2269 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2235, 2264, 2269, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_215_output_0", "outputIndexes": +[ 2270 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2270 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4055 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4055 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4056 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4056 ] +, "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": +[ 4057 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4057 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4058 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4058 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.23/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2271 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2187, 2271 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_118_output_0", "outputIndexes": +[ 2272 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2272 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2801290798, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_23/Mul_1_output_0", "outputIndexes": +[ 2273 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2273 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4059 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4059 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4060 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4060 ] +, "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": +[ 4061 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4061 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4062 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4062 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2274 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2274 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_23/Mul_output_0", "outputIndexes": +[ 2275 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4060 ] +, "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": +[ 4065 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4065 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4066 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4066 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2276 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2275, 2276 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_23/Mul_output_0", "outputIndexes": +[ 2277 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2277 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4067 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4067 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.23/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4068 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4068 ] +, "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": +[ 4069 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4069 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.23/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4070 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4070 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.23/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2278 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2272, 2278 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_119_output_0", "outputIndexes": +[ 2279 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2279, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_216_output_0", "outputIndexes": +[ 2280 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2280 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2895695472, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_24/Mul_1_output_0", "outputIndexes": +[ 2281 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2281 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4071 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4071 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4072 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4072 ] +, "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": +[ 4073 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4073 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4074 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4074 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2282 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2281 ] +, "main_type": "NONE", "name": "/Shape_336_output_0", "outputIndexes": +[ 2283 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2283 ] +, "main_type": "NONE", "name": "Shape9447", "outputIndexes": +[ 2284 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2283 ] +, "main_type": "NONE", "name": "Rank9449", "outputIndexes": +[ 2285 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2285, 2285 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9451", "outputIndexes": +[ 2286 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2286 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9452", "outputIndexes": +[ 2287 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2286, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9454", "outputIndexes": +[ 2288 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2288 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9455", "outputIndexes": +[ 2289 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2284, 2287, 2289, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9458", "outputIndexes": +[ 2290 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2290 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9459", "outputIndexes": +[ 2291 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2291, 2291 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9461", "outputIndexes": +[ 2292 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2283, 2292, 2286 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_363_output_0", "outputIndexes": +[ 2293 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2293, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_768_output_0", "outputIndexes": +[ 2294 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2286 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9469", "outputIndexes": +[ 2295 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2288 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9472", "outputIndexes": +[ 2296 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2284, 2295, 2296, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9475", "outputIndexes": +[ 2297 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2297 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9476", "outputIndexes": +[ 2298 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2298 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9477", "outputIndexes": +[ 2299 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2299, 2298 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9478", "outputIndexes": +[ 2300 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2283, 2300, 2286 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_364_output_0", "outputIndexes": +[ 2301 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2301, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_769_output_0", "outputIndexes": +[ 2302 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2294, 2302, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_312_output_0", "outputIndexes": +[ 2303 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2282, 2303 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_217_output_0", "outputIndexes": +[ 2304 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2304 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2906214022, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_24/Mul_1_output_0", "outputIndexes": +[ 2306 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2306, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_192_output_0", "outputIndexes": +[ 2307 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2306 ] +, "main_type": "NONE", "name": "/Shape_338_output_0", "outputIndexes": +[ 2308 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2308 ] +, "main_type": "NONE", "name": "Shape9597", "outputIndexes": +[ 2309 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2308 ] +, "main_type": "NONE", "name": "Rank9599", "outputIndexes": +[ 2310 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2310, 2310 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9601", "outputIndexes": +[ 2311 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2311 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9602", "outputIndexes": +[ 2312 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2311, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9604", "outputIndexes": +[ 2313 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2313 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9605", "outputIndexes": +[ 2314 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2309, 2312, 2314, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9608", "outputIndexes": +[ 2315 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2315 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9609", "outputIndexes": +[ 2316 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2316 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9610", "outputIndexes": +[ 2317 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2317, 2316 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9611", "outputIndexes": +[ 2318 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2308, 2318, 2311 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_365_output_0", "outputIndexes": +[ 2319 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2319, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_72_output_0", "outputIndexes": +[ 2320 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2320, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_775_output_0", "outputIndexes": +[ 2321 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2306, 2321, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_97_output_0", "outputIndexes": +[ 2322 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2322 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_48_output_0", "outputIndexes": +[ 2323 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2320, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_774_output_0", "outputIndexes": +[ 2324 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2306, 22, 2324, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_96_output_0", "outputIndexes": +[ 2325 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2323, 2325 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_315_output_0", "outputIndexes": +[ 2326 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2326, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_193_output_0", "outputIndexes": +[ 2327 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2307, 2327 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_120_output_0", "outputIndexes": +[ 2328 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4072 ] +, "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": +[ 4077 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4077 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4078 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4078 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2329 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2293, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_770_output_0", "outputIndexes": +[ 2330 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2301, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_771_output_0", "outputIndexes": +[ 2331 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2330, 2331, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_313_output_0", "outputIndexes": +[ 2332 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2329, 2332 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_218_output_0", "outputIndexes": +[ 2333 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2333 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2908836508, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_24/Mul_1_output_0", "outputIndexes": +[ 2335 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2335, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_194_output_0", "outputIndexes": +[ 2336 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2335 ] +, "main_type": "NONE", "name": "/Shape_339_output_0", "outputIndexes": +[ 2337 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2337 ] +, "main_type": "NONE", "name": "Shape9493", "outputIndexes": +[ 2338 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2337 ] +, "main_type": "NONE", "name": "Rank9495", "outputIndexes": +[ 2339 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2339, 2339 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9497", "outputIndexes": +[ 2340 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2340 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9498", "outputIndexes": +[ 2341 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2340, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9500", "outputIndexes": +[ 2342 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2342 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9501", "outputIndexes": +[ 2343 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2338, 2341, 2343, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9504", "outputIndexes": +[ 2344 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2344 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9505", "outputIndexes": +[ 2345 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2345 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9506", "outputIndexes": +[ 2346 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2346, 2345 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9507", "outputIndexes": +[ 2347 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2337, 2347, 2340 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_366_output_0", "outputIndexes": +[ 2348 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2348, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_73_output_0", "outputIndexes": +[ 2349 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2349, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_777_output_0", "outputIndexes": +[ 2350 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2335, 2350, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_99_output_0", "outputIndexes": +[ 2351 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2351 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_49_output_0", "outputIndexes": +[ 2352 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2349, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_776_output_0", "outputIndexes": +[ 2353 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2335, 22, 2353, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_98_output_0", "outputIndexes": +[ 2354 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2352, 2354 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_316_output_0", "outputIndexes": +[ 2355 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2355, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_195_output_0", "outputIndexes": +[ 2356 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2336, 2356 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_121_output_0", "outputIndexes": +[ 2357 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4072 ] +, "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": +[ 4081 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4081 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4082 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4082 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2358 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2293, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_772_output_0", "outputIndexes": +[ 2359 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2301, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_773_output_0", "outputIndexes": +[ 2360 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2359, 2360, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_314_output_0", "outputIndexes": +[ 2361 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2358, 2361 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_219_output_0", "outputIndexes": +[ 2362 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2328, 2357, 2362, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_224_output_0", "outputIndexes": +[ 2363 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2363 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4083 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4083 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4084 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4084 ] +, "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": +[ 4085 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4085 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4086 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4086 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.24/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2364 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2280, 2364 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_123_output_0", "outputIndexes": +[ 2365 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2365 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 2921944776, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_24/Mul_1_output_0", "outputIndexes": +[ 2366 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2366 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4087 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4087 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4088 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4088 ] +, "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": +[ 4089 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4089 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4090 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4090 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2367 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2367 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_24/Mul_output_0", "outputIndexes": +[ 2368 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4088 ] +, "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": +[ 4093 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4093 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4094 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4094 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2369 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2368, 2369 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_24/Mul_output_0", "outputIndexes": +[ 2370 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2370 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4095 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4095 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.24/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4096 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4096 ] +, "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": +[ 4097 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4097 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.24/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4098 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4098 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.24/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2371 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2365, 2371 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_124_output_0", "outputIndexes": +[ 2372 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2372, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_225_output_0", "outputIndexes": +[ 2373 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2373 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3016349450, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_25/Mul_1_output_0", "outputIndexes": +[ 2374 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2374 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4099 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4099 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4100 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4100 ] +, "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": +[ 4101 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4101 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4102 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4102 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2375 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2374 ] +, "main_type": "NONE", "name": "/Shape_350_output_0", "outputIndexes": +[ 2376 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2376 ] +, "main_type": "NONE", "name": "Shape9836", "outputIndexes": +[ 2377 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2376 ] +, "main_type": "NONE", "name": "Rank9838", "outputIndexes": +[ 2378 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2378, 2378 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9840", "outputIndexes": +[ 2379 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2379 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9841", "outputIndexes": +[ 2380 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2379, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9843", "outputIndexes": +[ 2381 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2381 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9844", "outputIndexes": +[ 2382 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2377, 2380, 2382, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9847", "outputIndexes": +[ 2383 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2383 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9848", "outputIndexes": +[ 2384 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2384, 2384 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9850", "outputIndexes": +[ 2385 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2376, 2385, 2379 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_378_output_0", "outputIndexes": +[ 2386 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2386, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_800_output_0", "outputIndexes": +[ 2387 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2379 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9858", "outputIndexes": +[ 2388 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2381 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9861", "outputIndexes": +[ 2389 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2377, 2388, 2389, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9864", "outputIndexes": +[ 2390 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2390 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9865", "outputIndexes": +[ 2391 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2391 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9866", "outputIndexes": +[ 2392 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2392, 2391 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9867", "outputIndexes": +[ 2393 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2376, 2393, 2379 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_379_output_0", "outputIndexes": +[ 2394 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2394, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_801_output_0", "outputIndexes": +[ 2395 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2387, 2395, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_325_output_0", "outputIndexes": +[ 2396 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2375, 2396 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_226_output_0", "outputIndexes": +[ 2397 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2397 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3026868000, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_25/Mul_1_output_0", "outputIndexes": +[ 2399 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2399, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_200_output_0", "outputIndexes": +[ 2400 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2399 ] +, "main_type": "NONE", "name": "/Shape_352_output_0", "outputIndexes": +[ 2401 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2401 ] +, "main_type": "NONE", "name": "Shape9986", "outputIndexes": +[ 2402 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2401 ] +, "main_type": "NONE", "name": "Rank9988", "outputIndexes": +[ 2403 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2403, 2403 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9990", "outputIndexes": +[ 2404 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2404 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9991", "outputIndexes": +[ 2405 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2404, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9993", "outputIndexes": +[ 2406 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2406 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9994", "outputIndexes": +[ 2407 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2402, 2405, 2407, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9997", "outputIndexes": +[ 2408 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2408 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9998", "outputIndexes": +[ 2409 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2409 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9999", "outputIndexes": +[ 2410 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2410, 2409 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10000", "outputIndexes": +[ 2411 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2401, 2411, 2404 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_380_output_0", "outputIndexes": +[ 2412 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2412, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_75_output_0", "outputIndexes": +[ 2413 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2413, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_807_output_0", "outputIndexes": +[ 2414 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2399, 2414, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_101_output_0", "outputIndexes": +[ 2415 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2415 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_50_output_0", "outputIndexes": +[ 2416 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2413, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_806_output_0", "outputIndexes": +[ 2417 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2399, 22, 2417, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_100_output_0", "outputIndexes": +[ 2418 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2416, 2418 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_328_output_0", "outputIndexes": +[ 2419 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2419, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_201_output_0", "outputIndexes": +[ 2420 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2400, 2420 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_125_output_0", "outputIndexes": +[ 2421 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4100 ] +, "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": +[ 4105 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4105 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4106 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4106 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2422 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2386, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_802_output_0", "outputIndexes": +[ 2423 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2394, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_803_output_0", "outputIndexes": +[ 2424 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2423, 2424, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_326_output_0", "outputIndexes": +[ 2425 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2422, 2425 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_227_output_0", "outputIndexes": +[ 2426 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2426 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3029490486, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_25/Mul_1_output_0", "outputIndexes": +[ 2428 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2428, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_202_output_0", "outputIndexes": +[ 2429 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2428 ] +, "main_type": "NONE", "name": "/Shape_353_output_0", "outputIndexes": +[ 2430 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2430 ] +, "main_type": "NONE", "name": "Shape9882", "outputIndexes": +[ 2431 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2430 ] +, "main_type": "NONE", "name": "Rank9884", "outputIndexes": +[ 2432 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2432, 2432 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9886", "outputIndexes": +[ 2433 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2433 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9887", "outputIndexes": +[ 2434 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2433, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9889", "outputIndexes": +[ 2435 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2435 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze9890", "outputIndexes": +[ 2436 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2431, 2434, 2436, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice9893", "outputIndexes": +[ 2437 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2437 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze9894", "outputIndexes": +[ 2438 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2438 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9895", "outputIndexes": +[ 2439 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2439, 2438 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp9896", "outputIndexes": +[ 2440 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2430, 2440, 2433 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_381_output_0", "outputIndexes": +[ 2441 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2441, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_76_output_0", "outputIndexes": +[ 2442 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2442, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_809_output_0", "outputIndexes": +[ 2443 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2428, 2443, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_103_output_0", "outputIndexes": +[ 2444 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2444 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_51_output_0", "outputIndexes": +[ 2445 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2442, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_808_output_0", "outputIndexes": +[ 2446 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2428, 22, 2446, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_102_output_0", "outputIndexes": +[ 2447 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2445, 2447 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_329_output_0", "outputIndexes": +[ 2448 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2448, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_203_output_0", "outputIndexes": +[ 2449 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2429, 2449 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_126_output_0", "outputIndexes": +[ 2450 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4100 ] +, "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": +[ 4109 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4109 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4110 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4110 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2451 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2386, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_804_output_0", "outputIndexes": +[ 2452 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2394, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_805_output_0", "outputIndexes": +[ 2453 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2452, 2453, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_327_output_0", "outputIndexes": +[ 2454 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2451, 2454 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_228_output_0", "outputIndexes": +[ 2455 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2421, 2450, 2455, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_233_output_0", "outputIndexes": +[ 2456 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2456 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4111 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4111 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4112 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4112 ] +, "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": +[ 4113 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4113 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4114 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4114 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.25/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2457 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2373, 2457 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_128_output_0", "outputIndexes": +[ 2458 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2458 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3042598754, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_25/Mul_1_output_0", "outputIndexes": +[ 2459 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2459 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4115 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4115 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4116 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4116 ] +, "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": +[ 4117 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4117 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4118 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4118 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2460 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2460 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_25/Mul_output_0", "outputIndexes": +[ 2461 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4116 ] +, "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": +[ 4121 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4121 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4122 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4122 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2462 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2461, 2462 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_25/Mul_output_0", "outputIndexes": +[ 2463 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2463 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4123 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4123 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.25/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4124 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4124 ] +, "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": +[ 4125 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4125 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.25/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4126 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4126 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.25/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2464 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2458, 2464 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_129_output_0", "outputIndexes": +[ 2465 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2465, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_234_output_0", "outputIndexes": +[ 2466 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2466 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3137003428, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_26/Mul_1_output_0", "outputIndexes": +[ 2467 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2467 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4127 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4127 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4128 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4128 ] +, "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": +[ 4129 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4129 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4130 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4130 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2468 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2467 ] +, "main_type": "NONE", "name": "/Shape_364_output_0", "outputIndexes": +[ 2469 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2469 ] +, "main_type": "NONE", "name": "Shape10225", "outputIndexes": +[ 2470 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2469 ] +, "main_type": "NONE", "name": "Rank10227", "outputIndexes": +[ 2471 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2471, 2471 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10229", "outputIndexes": +[ 2472 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2472 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10230", "outputIndexes": +[ 2473 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2472, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10232", "outputIndexes": +[ 2474 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2474 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10233", "outputIndexes": +[ 2475 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2470, 2473, 2475, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10236", "outputIndexes": +[ 2476 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2476 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10237", "outputIndexes": +[ 2477 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2477, 2477 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10239", "outputIndexes": +[ 2478 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2469, 2478, 2472 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_393_output_0", "outputIndexes": +[ 2479 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2479, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_832_output_0", "outputIndexes": +[ 2480 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2472 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10247", "outputIndexes": +[ 2481 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2474 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10250", "outputIndexes": +[ 2482 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2470, 2481, 2482, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10253", "outputIndexes": +[ 2483 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2483 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10254", "outputIndexes": +[ 2484 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2484 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10255", "outputIndexes": +[ 2485 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2485, 2484 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10256", "outputIndexes": +[ 2486 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2469, 2486, 2472 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_394_output_0", "outputIndexes": +[ 2487 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2487, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_833_output_0", "outputIndexes": +[ 2488 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2480, 2488, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_338_output_0", "outputIndexes": +[ 2489 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2468, 2489 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_235_output_0", "outputIndexes": +[ 2490 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2490 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3147521978, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_26/Mul_1_output_0", "outputIndexes": +[ 2492 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2492, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_208_output_0", "outputIndexes": +[ 2493 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2492 ] +, "main_type": "NONE", "name": "/Shape_366_output_0", "outputIndexes": +[ 2494 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2494 ] +, "main_type": "NONE", "name": "Shape10375", "outputIndexes": +[ 2495 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2494 ] +, "main_type": "NONE", "name": "Rank10377", "outputIndexes": +[ 2496 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2496, 2496 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10379", "outputIndexes": +[ 2497 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2497 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10380", "outputIndexes": +[ 2498 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2497, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10382", "outputIndexes": +[ 2499 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2499 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10383", "outputIndexes": +[ 2500 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2495, 2498, 2500, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10386", "outputIndexes": +[ 2501 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2501 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10387", "outputIndexes": +[ 2502 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2502 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10388", "outputIndexes": +[ 2503 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2503, 2502 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10389", "outputIndexes": +[ 2504 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2494, 2504, 2497 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_395_output_0", "outputIndexes": +[ 2505 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2505, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_78_output_0", "outputIndexes": +[ 2506 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2506, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_839_output_0", "outputIndexes": +[ 2507 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2492, 2507, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_105_output_0", "outputIndexes": +[ 2508 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2508 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_52_output_0", "outputIndexes": +[ 2509 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2506, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_838_output_0", "outputIndexes": +[ 2510 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2492, 22, 2510, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_104_output_0", "outputIndexes": +[ 2511 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2509, 2511 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_341_output_0", "outputIndexes": +[ 2512 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2512, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_209_output_0", "outputIndexes": +[ 2513 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2493, 2513 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_130_output_0", "outputIndexes": +[ 2514 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4128 ] +, "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": +[ 4133 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4133 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4134 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4134 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2515 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2479, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_834_output_0", "outputIndexes": +[ 2516 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2487, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_835_output_0", "outputIndexes": +[ 2517 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2516, 2517, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_339_output_0", "outputIndexes": +[ 2518 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2515, 2518 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_236_output_0", "outputIndexes": +[ 2519 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2519 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3150144464, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_26/Mul_1_output_0", "outputIndexes": +[ 2521 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2521, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_210_output_0", "outputIndexes": +[ 2522 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2521 ] +, "main_type": "NONE", "name": "/Shape_367_output_0", "outputIndexes": +[ 2523 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2523 ] +, "main_type": "NONE", "name": "Shape10271", "outputIndexes": +[ 2524 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2523 ] +, "main_type": "NONE", "name": "Rank10273", "outputIndexes": +[ 2525 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2525, 2525 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10275", "outputIndexes": +[ 2526 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2526 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10276", "outputIndexes": +[ 2527 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2526, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10278", "outputIndexes": +[ 2528 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2528 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10279", "outputIndexes": +[ 2529 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2524, 2527, 2529, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10282", "outputIndexes": +[ 2530 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2530 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10283", "outputIndexes": +[ 2531 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2531 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10284", "outputIndexes": +[ 2532 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2532, 2531 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10285", "outputIndexes": +[ 2533 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2523, 2533, 2526 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_396_output_0", "outputIndexes": +[ 2534 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2534, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_79_output_0", "outputIndexes": +[ 2535 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2535, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_841_output_0", "outputIndexes": +[ 2536 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2521, 2536, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_107_output_0", "outputIndexes": +[ 2537 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2537 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_53_output_0", "outputIndexes": +[ 2538 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2535, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_840_output_0", "outputIndexes": +[ 2539 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2521, 22, 2539, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_106_output_0", "outputIndexes": +[ 2540 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2538, 2540 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_342_output_0", "outputIndexes": +[ 2541 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2541, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_211_output_0", "outputIndexes": +[ 2542 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2522, 2542 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_131_output_0", "outputIndexes": +[ 2543 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4128 ] +, "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": +[ 4137 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4137 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4138 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4138 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2544 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2479, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_836_output_0", "outputIndexes": +[ 2545 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2487, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_837_output_0", "outputIndexes": +[ 2546 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2545, 2546, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_340_output_0", "outputIndexes": +[ 2547 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2544, 2547 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_237_output_0", "outputIndexes": +[ 2548 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2514, 2543, 2548, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_242_output_0", "outputIndexes": +[ 2549 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2549 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4139 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4139 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4140 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4140 ] +, "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": +[ 4141 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4141 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4142 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4142 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.26/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2550 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2466, 2550 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_133_output_0", "outputIndexes": +[ 2551 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2551 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3163252732, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_26/Mul_1_output_0", "outputIndexes": +[ 2552 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2552 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4143 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4143 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4144 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4144 ] +, "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": +[ 4145 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4145 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4146 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4146 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2553 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2553 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_26/Mul_output_0", "outputIndexes": +[ 2554 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4144 ] +, "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": +[ 4149 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4149 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4150 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4150 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2555 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2554, 2555 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_26/Mul_output_0", "outputIndexes": +[ 2556 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2556 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4151 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4151 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.26/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4152 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4152 ] +, "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": +[ 4153 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4153 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.26/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4154 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4154 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.26/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2557 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2551, 2557 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_134_output_0", "outputIndexes": +[ 2558 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2558, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_243_output_0", "outputIndexes": +[ 2559 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2559 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3257657406, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_27/Mul_1_output_0", "outputIndexes": +[ 2560 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2560 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4155 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4155 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4156 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4156 ] +, "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": +[ 4157 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4157 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4158 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4158 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2561 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2560 ] +, "main_type": "NONE", "name": "/Shape_378_output_0", "outputIndexes": +[ 2562 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2562 ] +, "main_type": "NONE", "name": "Shape10614", "outputIndexes": +[ 2563 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2562 ] +, "main_type": "NONE", "name": "Rank10616", "outputIndexes": +[ 2564 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2564, 2564 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10618", "outputIndexes": +[ 2565 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2565 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10619", "outputIndexes": +[ 2566 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2565, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10621", "outputIndexes": +[ 2567 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2567 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10622", "outputIndexes": +[ 2568 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2563, 2566, 2568, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10625", "outputIndexes": +[ 2569 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2569 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10626", "outputIndexes": +[ 2570 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2570, 2570 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10628", "outputIndexes": +[ 2571 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2562, 2571, 2565 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_408_output_0", "outputIndexes": +[ 2572 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2572, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_864_output_0", "outputIndexes": +[ 2573 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2565 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10636", "outputIndexes": +[ 2574 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2567 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10639", "outputIndexes": +[ 2575 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2563, 2574, 2575, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10642", "outputIndexes": +[ 2576 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2576 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10643", "outputIndexes": +[ 2577 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2577 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10644", "outputIndexes": +[ 2578 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2578, 2577 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10645", "outputIndexes": +[ 2579 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2562, 2579, 2565 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_409_output_0", "outputIndexes": +[ 2580 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2580, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_865_output_0", "outputIndexes": +[ 2581 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2573, 2581, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_351_output_0", "outputIndexes": +[ 2582 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2561, 2582 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_244_output_0", "outputIndexes": +[ 2583 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2583 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3268175956, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_27/Mul_1_output_0", "outputIndexes": +[ 2585 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2585, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_216_output_0", "outputIndexes": +[ 2586 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2585 ] +, "main_type": "NONE", "name": "/Shape_380_output_0", "outputIndexes": +[ 2587 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2587 ] +, "main_type": "NONE", "name": "Shape10764", "outputIndexes": +[ 2588 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2587 ] +, "main_type": "NONE", "name": "Rank10766", "outputIndexes": +[ 2589 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2589, 2589 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10768", "outputIndexes": +[ 2590 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2590 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10769", "outputIndexes": +[ 2591 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2590, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10771", "outputIndexes": +[ 2592 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2592 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10772", "outputIndexes": +[ 2593 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2588, 2591, 2593, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10775", "outputIndexes": +[ 2594 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2594 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10776", "outputIndexes": +[ 2595 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2595 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10777", "outputIndexes": +[ 2596 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2596, 2595 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10778", "outputIndexes": +[ 2597 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2587, 2597, 2590 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_410_output_0", "outputIndexes": +[ 2598 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2598, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_81_output_0", "outputIndexes": +[ 2599 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2599, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_871_output_0", "outputIndexes": +[ 2600 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2585, 2600, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_109_output_0", "outputIndexes": +[ 2601 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2601 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_54_output_0", "outputIndexes": +[ 2602 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2599, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_870_output_0", "outputIndexes": +[ 2603 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2585, 22, 2603, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_108_output_0", "outputIndexes": +[ 2604 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2602, 2604 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_354_output_0", "outputIndexes": +[ 2605 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2605, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_217_output_0", "outputIndexes": +[ 2606 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2586, 2606 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_135_output_0", "outputIndexes": +[ 2607 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4156 ] +, "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": +[ 4161 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4161 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4162 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4162 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2608 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2572, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_866_output_0", "outputIndexes": +[ 2609 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2580, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_867_output_0", "outputIndexes": +[ 2610 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2609, 2610, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_352_output_0", "outputIndexes": +[ 2611 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2608, 2611 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_245_output_0", "outputIndexes": +[ 2612 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2612 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3270798442, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_27/Mul_1_output_0", "outputIndexes": +[ 2614 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2614, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_218_output_0", "outputIndexes": +[ 2615 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2614 ] +, "main_type": "NONE", "name": "/Shape_381_output_0", "outputIndexes": +[ 2616 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2616 ] +, "main_type": "NONE", "name": "Shape10660", "outputIndexes": +[ 2617 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2616 ] +, "main_type": "NONE", "name": "Rank10662", "outputIndexes": +[ 2618 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2618, 2618 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10664", "outputIndexes": +[ 2619 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2619 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10665", "outputIndexes": +[ 2620 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2619, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10667", "outputIndexes": +[ 2621 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2621 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze10668", "outputIndexes": +[ 2622 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2617, 2620, 2622, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice10671", "outputIndexes": +[ 2623 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2623 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze10672", "outputIndexes": +[ 2624 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2624 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10673", "outputIndexes": +[ 2625 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2625, 2624 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp10674", "outputIndexes": +[ 2626 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2616, 2626, 2619 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_411_output_0", "outputIndexes": +[ 2627 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2627, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_82_output_0", "outputIndexes": +[ 2628 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2628, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_873_output_0", "outputIndexes": +[ 2629 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2614, 2629, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_111_output_0", "outputIndexes": +[ 2630 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2630 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_55_output_0", "outputIndexes": +[ 2631 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2628, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_872_output_0", "outputIndexes": +[ 2632 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2614, 22, 2632, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_110_output_0", "outputIndexes": +[ 2633 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2631, 2633 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_355_output_0", "outputIndexes": +[ 2634 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2634, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_219_output_0", "outputIndexes": +[ 2635 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2615, 2635 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_136_output_0", "outputIndexes": +[ 2636 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4156 ] +, "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": +[ 4165 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4165 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4166 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4166 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2637 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2572, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_868_output_0", "outputIndexes": +[ 2638 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2580, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_869_output_0", "outputIndexes": +[ 2639 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2638, 2639, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_353_output_0", "outputIndexes": +[ 2640 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2637, 2640 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_246_output_0", "outputIndexes": +[ 2641 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2607, 2636, 2641, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_251_output_0", "outputIndexes": +[ 2642 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2642 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4167 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4167 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4168 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4168 ] +, "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": +[ 4169 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4169 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4170 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4170 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.27/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2643 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2559, 2643 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_138_output_0", "outputIndexes": +[ 2644 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2644 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3283906710, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_27/Mul_1_output_0", "outputIndexes": +[ 2645 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2645 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4171 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4171 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4172 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4172 ] +, "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": +[ 4173 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4173 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4174 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4174 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2646 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2646 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_27/Mul_output_0", "outputIndexes": +[ 2647 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4172 ] +, "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": +[ 4177 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4177 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4178 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4178 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2648 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2647, 2648 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_27/Mul_output_0", "outputIndexes": +[ 2649 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2649 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4179 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4179 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.27/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4180 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4180 ] +, "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": +[ 4181 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4181 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.27/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4182 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4182 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.27/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2650 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2644, 2650 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_139_output_0", "outputIndexes": +[ 2651 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2651, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_252_output_0", "outputIndexes": +[ 2652 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2652 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3378311384, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_28/Mul_1_output_0", "outputIndexes": +[ 2653 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2653 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4183 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4183 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4184 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4184 ] +, "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": +[ 4185 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4185 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4186 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4186 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2654 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2653 ] +, "main_type": "NONE", "name": "/Shape_392_output_0", "outputIndexes": +[ 2655 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2655 ] +, "main_type": "NONE", "name": "Shape11003", "outputIndexes": +[ 2656 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2655 ] +, "main_type": "NONE", "name": "Rank11005", "outputIndexes": +[ 2657 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2657, 2657 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11007", "outputIndexes": +[ 2658 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2658 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11008", "outputIndexes": +[ 2659 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2658, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11010", "outputIndexes": +[ 2660 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2660 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11011", "outputIndexes": +[ 2661 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2656, 2659, 2661, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11014", "outputIndexes": +[ 2662 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2662 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11015", "outputIndexes": +[ 2663 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2663, 2663 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11017", "outputIndexes": +[ 2664 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2655, 2664, 2658 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_423_output_0", "outputIndexes": +[ 2665 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2665, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_896_output_0", "outputIndexes": +[ 2666 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2658 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11025", "outputIndexes": +[ 2667 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2660 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11028", "outputIndexes": +[ 2668 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2656, 2667, 2668, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11031", "outputIndexes": +[ 2669 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2669 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11032", "outputIndexes": +[ 2670 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2670 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11033", "outputIndexes": +[ 2671 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2671, 2670 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11034", "outputIndexes": +[ 2672 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2655, 2672, 2658 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_424_output_0", "outputIndexes": +[ 2673 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2673, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_897_output_0", "outputIndexes": +[ 2674 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2666, 2674, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_364_output_0", "outputIndexes": +[ 2675 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2654, 2675 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_253_output_0", "outputIndexes": +[ 2676 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2676 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3388829934, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_28/Mul_1_output_0", "outputIndexes": +[ 2678 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2678, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_224_output_0", "outputIndexes": +[ 2679 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2678 ] +, "main_type": "NONE", "name": "/Shape_394_output_0", "outputIndexes": +[ 2680 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2680 ] +, "main_type": "NONE", "name": "Shape11153", "outputIndexes": +[ 2681 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2680 ] +, "main_type": "NONE", "name": "Rank11155", "outputIndexes": +[ 2682 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2682, 2682 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11157", "outputIndexes": +[ 2683 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2683 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11158", "outputIndexes": +[ 2684 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2683, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11160", "outputIndexes": +[ 2685 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2685 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11161", "outputIndexes": +[ 2686 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2681, 2684, 2686, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11164", "outputIndexes": +[ 2687 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2687 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11165", "outputIndexes": +[ 2688 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2688 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11166", "outputIndexes": +[ 2689 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2689, 2688 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11167", "outputIndexes": +[ 2690 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2680, 2690, 2683 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_425_output_0", "outputIndexes": +[ 2691 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2691, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_84_output_0", "outputIndexes": +[ 2692 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2692, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_903_output_0", "outputIndexes": +[ 2693 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2678, 2693, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_113_output_0", "outputIndexes": +[ 2694 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2694 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_56_output_0", "outputIndexes": +[ 2695 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2692, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_902_output_0", "outputIndexes": +[ 2696 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2678, 22, 2696, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_112_output_0", "outputIndexes": +[ 2697 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2695, 2697 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_367_output_0", "outputIndexes": +[ 2698 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2698, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_225_output_0", "outputIndexes": +[ 2699 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2679, 2699 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_140_output_0", "outputIndexes": +[ 2700 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4184 ] +, "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": +[ 4189 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4189 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4190 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4190 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2701 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2665, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_898_output_0", "outputIndexes": +[ 2702 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2673, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_899_output_0", "outputIndexes": +[ 2703 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2702, 2703, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_365_output_0", "outputIndexes": +[ 2704 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2701, 2704 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_254_output_0", "outputIndexes": +[ 2705 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2705 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3391452420, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_28/Mul_1_output_0", "outputIndexes": +[ 2707 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2707, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_226_output_0", "outputIndexes": +[ 2708 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2707 ] +, "main_type": "NONE", "name": "/Shape_395_output_0", "outputIndexes": +[ 2709 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2709 ] +, "main_type": "NONE", "name": "Shape11049", "outputIndexes": +[ 2710 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2709 ] +, "main_type": "NONE", "name": "Rank11051", "outputIndexes": +[ 2711 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2711, 2711 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11053", "outputIndexes": +[ 2712 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2712 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11054", "outputIndexes": +[ 2713 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2712, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11056", "outputIndexes": +[ 2714 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2714 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11057", "outputIndexes": +[ 2715 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2710, 2713, 2715, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11060", "outputIndexes": +[ 2716 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2716 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11061", "outputIndexes": +[ 2717 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2717 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11062", "outputIndexes": +[ 2718 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2718, 2717 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11063", "outputIndexes": +[ 2719 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2709, 2719, 2712 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_426_output_0", "outputIndexes": +[ 2720 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2720, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_85_output_0", "outputIndexes": +[ 2721 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2721, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_905_output_0", "outputIndexes": +[ 2722 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2707, 2722, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_115_output_0", "outputIndexes": +[ 2723 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2723 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_57_output_0", "outputIndexes": +[ 2724 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2721, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_904_output_0", "outputIndexes": +[ 2725 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2707, 22, 2725, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_114_output_0", "outputIndexes": +[ 2726 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2724, 2726 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_368_output_0", "outputIndexes": +[ 2727 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2727, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_227_output_0", "outputIndexes": +[ 2728 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2708, 2728 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_141_output_0", "outputIndexes": +[ 2729 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4184 ] +, "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": +[ 4193 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4193 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4194 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4194 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2730 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2665, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_900_output_0", "outputIndexes": +[ 2731 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2673, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_901_output_0", "outputIndexes": +[ 2732 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2731, 2732, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_366_output_0", "outputIndexes": +[ 2733 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2730, 2733 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_255_output_0", "outputIndexes": +[ 2734 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2700, 2729, 2734, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_260_output_0", "outputIndexes": +[ 2735 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2735 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4195 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4195 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4196 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4196 ] +, "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": +[ 4197 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4197 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4198 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4198 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.28/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2736 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2652, 2736 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_143_output_0", "outputIndexes": +[ 2737 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2737 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3404560688, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_28/Mul_1_output_0", "outputIndexes": +[ 2738 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2738 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4199 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4199 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4200 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4200 ] +, "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": +[ 4201 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4201 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4202 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4202 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2739 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2739 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_28/Mul_output_0", "outputIndexes": +[ 2740 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4200 ] +, "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": +[ 4205 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4205 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4206 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4206 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2741 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2740, 2741 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_28/Mul_output_0", "outputIndexes": +[ 2742 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2742 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4207 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4207 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.28/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4208 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4208 ] +, "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": +[ 4209 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4209 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.28/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4210 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4210 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.28/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2743 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2737, 2743 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_144_output_0", "outputIndexes": +[ 2744 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2744, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_261_output_0", "outputIndexes": +[ 2745 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2745 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3498965362, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_29/Mul_1_output_0", "outputIndexes": +[ 2746 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2746 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4211 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4211 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4212 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4212 ] +, "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": +[ 4213 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4213 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4214 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4214 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2747 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2746 ] +, "main_type": "NONE", "name": "/Shape_406_output_0", "outputIndexes": +[ 2748 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2748 ] +, "main_type": "NONE", "name": "Shape11392", "outputIndexes": +[ 2749 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2748 ] +, "main_type": "NONE", "name": "Rank11394", "outputIndexes": +[ 2750 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2750, 2750 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11396", "outputIndexes": +[ 2751 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2751 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11397", "outputIndexes": +[ 2752 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2751, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11399", "outputIndexes": +[ 2753 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2753 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11400", "outputIndexes": +[ 2754 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2749, 2752, 2754, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11403", "outputIndexes": +[ 2755 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2755 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11404", "outputIndexes": +[ 2756 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2756, 2756 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11406", "outputIndexes": +[ 2757 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2748, 2757, 2751 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_438_output_0", "outputIndexes": +[ 2758 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2758, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_928_output_0", "outputIndexes": +[ 2759 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2751 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11414", "outputIndexes": +[ 2760 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2753 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11417", "outputIndexes": +[ 2761 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2749, 2760, 2761, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11420", "outputIndexes": +[ 2762 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2762 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11421", "outputIndexes": +[ 2763 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2763 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11422", "outputIndexes": +[ 2764 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2764, 2763 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11423", "outputIndexes": +[ 2765 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2748, 2765, 2751 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_439_output_0", "outputIndexes": +[ 2766 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2766, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_929_output_0", "outputIndexes": +[ 2767 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2759, 2767, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_377_output_0", "outputIndexes": +[ 2768 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2747, 2768 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_262_output_0", "outputIndexes": +[ 2769 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2769 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3509483912, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_29/Mul_1_output_0", "outputIndexes": +[ 2771 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2771, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_232_output_0", "outputIndexes": +[ 2772 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2771 ] +, "main_type": "NONE", "name": "/Shape_408_output_0", "outputIndexes": +[ 2773 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2773 ] +, "main_type": "NONE", "name": "Shape11542", "outputIndexes": +[ 2774 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2773 ] +, "main_type": "NONE", "name": "Rank11544", "outputIndexes": +[ 2775 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2775, 2775 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11546", "outputIndexes": +[ 2776 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2776 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11547", "outputIndexes": +[ 2777 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2776, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11549", "outputIndexes": +[ 2778 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2778 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11550", "outputIndexes": +[ 2779 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2774, 2777, 2779, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11553", "outputIndexes": +[ 2780 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2780 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11554", "outputIndexes": +[ 2781 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2781 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11555", "outputIndexes": +[ 2782 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2782, 2781 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11556", "outputIndexes": +[ 2783 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2773, 2783, 2776 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_440_output_0", "outputIndexes": +[ 2784 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2784, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_87_output_0", "outputIndexes": +[ 2785 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2785, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_935_output_0", "outputIndexes": +[ 2786 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2771, 2786, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_117_output_0", "outputIndexes": +[ 2787 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2787 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_58_output_0", "outputIndexes": +[ 2788 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2785, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_934_output_0", "outputIndexes": +[ 2789 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2771, 22, 2789, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_116_output_0", "outputIndexes": +[ 2790 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2788, 2790 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_380_output_0", "outputIndexes": +[ 2791 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2791, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_233_output_0", "outputIndexes": +[ 2792 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2772, 2792 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_145_output_0", "outputIndexes": +[ 2793 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4212 ] +, "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": +[ 4217 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4217 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4218 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4218 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2794 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2758, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_930_output_0", "outputIndexes": +[ 2795 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2766, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_931_output_0", "outputIndexes": +[ 2796 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2795, 2796, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_378_output_0", "outputIndexes": +[ 2797 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2794, 2797 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_263_output_0", "outputIndexes": +[ 2798 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2798 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3512106398, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_29/Mul_1_output_0", "outputIndexes": +[ 2800 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2800, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_234_output_0", "outputIndexes": +[ 2801 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2800 ] +, "main_type": "NONE", "name": "/Shape_409_output_0", "outputIndexes": +[ 2802 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2802 ] +, "main_type": "NONE", "name": "Shape11438", "outputIndexes": +[ 2803 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2802 ] +, "main_type": "NONE", "name": "Rank11440", "outputIndexes": +[ 2804 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2804, 2804 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11442", "outputIndexes": +[ 2805 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2805 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11443", "outputIndexes": +[ 2806 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2805, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11445", "outputIndexes": +[ 2807 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2807 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11446", "outputIndexes": +[ 2808 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2803, 2806, 2808, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11449", "outputIndexes": +[ 2809 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2809 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11450", "outputIndexes": +[ 2810 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2810 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11451", "outputIndexes": +[ 2811 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2811, 2810 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11452", "outputIndexes": +[ 2812 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2802, 2812, 2805 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_441_output_0", "outputIndexes": +[ 2813 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2813, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_88_output_0", "outputIndexes": +[ 2814 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2814, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_937_output_0", "outputIndexes": +[ 2815 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2800, 2815, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_119_output_0", "outputIndexes": +[ 2816 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2816 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_59_output_0", "outputIndexes": +[ 2817 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2814, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_936_output_0", "outputIndexes": +[ 2818 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2800, 22, 2818, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_118_output_0", "outputIndexes": +[ 2819 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2817, 2819 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_381_output_0", "outputIndexes": +[ 2820 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2820, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_235_output_0", "outputIndexes": +[ 2821 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2801, 2821 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_146_output_0", "outputIndexes": +[ 2822 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4212 ] +, "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": +[ 4221 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4221 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4222 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4222 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2823 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2758, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_932_output_0", "outputIndexes": +[ 2824 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2766, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_933_output_0", "outputIndexes": +[ 2825 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2824, 2825, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_379_output_0", "outputIndexes": +[ 2826 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2823, 2826 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_264_output_0", "outputIndexes": +[ 2827 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2793, 2822, 2827, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_269_output_0", "outputIndexes": +[ 2828 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2828 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4223 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4223 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4224 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4224 ] +, "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": +[ 4225 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4225 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4226 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4226 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.29/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2829 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2745, 2829 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_148_output_0", "outputIndexes": +[ 2830 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2830 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3525214666, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_29/Mul_1_output_0", "outputIndexes": +[ 2831 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2831 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4227 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4227 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4228 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4228 ] +, "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": +[ 4229 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4229 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4230 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4230 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2832 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2832 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_29/Mul_output_0", "outputIndexes": +[ 2833 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4228 ] +, "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": +[ 4233 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4233 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4234 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4234 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2834 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2833, 2834 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_29/Mul_output_0", "outputIndexes": +[ 2835 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2835 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4235 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4235 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.29/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4236 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4236 ] +, "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": +[ 4237 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4237 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.29/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4238 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4238 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.29/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2836 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2830, 2836 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_149_output_0", "outputIndexes": +[ 2837 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2837, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_270_output_0", "outputIndexes": +[ 2838 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2838 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3619619340, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_30/Mul_1_output_0", "outputIndexes": +[ 2839 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2839 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4239 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4239 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4240 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4240 ] +, "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": +[ 4241 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4241 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4242 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4242 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2840 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2839 ] +, "main_type": "NONE", "name": "/Shape_420_output_0", "outputIndexes": +[ 2841 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2841 ] +, "main_type": "NONE", "name": "Shape11781", "outputIndexes": +[ 2842 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2841 ] +, "main_type": "NONE", "name": "Rank11783", "outputIndexes": +[ 2843 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2843, 2843 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11785", "outputIndexes": +[ 2844 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2844 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11786", "outputIndexes": +[ 2845 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2844, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11788", "outputIndexes": +[ 2846 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2846 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11789", "outputIndexes": +[ 2847 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2842, 2845, 2847, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11792", "outputIndexes": +[ 2848 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2848 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11793", "outputIndexes": +[ 2849 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2849, 2849 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11795", "outputIndexes": +[ 2850 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2841, 2850, 2844 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_453_output_0", "outputIndexes": +[ 2851 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2851, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_960_output_0", "outputIndexes": +[ 2852 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2844 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11803", "outputIndexes": +[ 2853 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2846 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11806", "outputIndexes": +[ 2854 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2842, 2853, 2854, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11809", "outputIndexes": +[ 2855 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2855 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11810", "outputIndexes": +[ 2856 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2856 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11811", "outputIndexes": +[ 2857 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2857, 2856 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11812", "outputIndexes": +[ 2858 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2841, 2858, 2844 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_454_output_0", "outputIndexes": +[ 2859 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2859, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_961_output_0", "outputIndexes": +[ 2860 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2852, 2860, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_390_output_0", "outputIndexes": +[ 2861 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2840, 2861 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_271_output_0", "outputIndexes": +[ 2862 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2862 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3630137890, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_30/Mul_1_output_0", "outputIndexes": +[ 2864 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2864, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_240_output_0", "outputIndexes": +[ 2865 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2864 ] +, "main_type": "NONE", "name": "/Shape_422_output_0", "outputIndexes": +[ 2866 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2866 ] +, "main_type": "NONE", "name": "Shape11931", "outputIndexes": +[ 2867 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2866 ] +, "main_type": "NONE", "name": "Rank11933", "outputIndexes": +[ 2868 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2868, 2868 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11935", "outputIndexes": +[ 2869 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2869 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11936", "outputIndexes": +[ 2870 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2869, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11938", "outputIndexes": +[ 2871 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2871 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11939", "outputIndexes": +[ 2872 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2867, 2870, 2872, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11942", "outputIndexes": +[ 2873 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2873 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11943", "outputIndexes": +[ 2874 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2874 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11944", "outputIndexes": +[ 2875 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2875, 2874 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11945", "outputIndexes": +[ 2876 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2866, 2876, 2869 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_455_output_0", "outputIndexes": +[ 2877 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2877, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_90_output_0", "outputIndexes": +[ 2878 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2878, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_967_output_0", "outputIndexes": +[ 2879 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2864, 2879, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_121_output_0", "outputIndexes": +[ 2880 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2880 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_60_output_0", "outputIndexes": +[ 2881 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2878, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_966_output_0", "outputIndexes": +[ 2882 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2864, 22, 2882, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_120_output_0", "outputIndexes": +[ 2883 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2881, 2883 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_393_output_0", "outputIndexes": +[ 2884 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2884, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_241_output_0", "outputIndexes": +[ 2885 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2865, 2885 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_150_output_0", "outputIndexes": +[ 2886 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4240 ] +, "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": +[ 4245 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4245 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4246 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4246 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2887 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2851, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_962_output_0", "outputIndexes": +[ 2888 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2859, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_963_output_0", "outputIndexes": +[ 2889 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2888, 2889, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_391_output_0", "outputIndexes": +[ 2890 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2887, 2890 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_272_output_0", "outputIndexes": +[ 2891 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2891 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3632760376, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_30/Mul_1_output_0", "outputIndexes": +[ 2893 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2893, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_242_output_0", "outputIndexes": +[ 2894 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2893 ] +, "main_type": "NONE", "name": "/Shape_423_output_0", "outputIndexes": +[ 2895 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2895 ] +, "main_type": "NONE", "name": "Shape11827", "outputIndexes": +[ 2896 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2895 ] +, "main_type": "NONE", "name": "Rank11829", "outputIndexes": +[ 2897 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2897, 2897 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11831", "outputIndexes": +[ 2898 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2898 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11832", "outputIndexes": +[ 2899 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2898, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11834", "outputIndexes": +[ 2900 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2900 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze11835", "outputIndexes": +[ 2901 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2896, 2899, 2901, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice11838", "outputIndexes": +[ 2902 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2902 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze11839", "outputIndexes": +[ 2903 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2903 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11840", "outputIndexes": +[ 2904 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2904, 2903 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp11841", "outputIndexes": +[ 2905 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2895, 2905, 2898 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_456_output_0", "outputIndexes": +[ 2906 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2906, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_91_output_0", "outputIndexes": +[ 2907 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2907, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_969_output_0", "outputIndexes": +[ 2908 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2893, 2908, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_123_output_0", "outputIndexes": +[ 2909 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2909 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_61_output_0", "outputIndexes": +[ 2910 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2907, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_968_output_0", "outputIndexes": +[ 2911 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2893, 22, 2911, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_122_output_0", "outputIndexes": +[ 2912 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2910, 2912 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_394_output_0", "outputIndexes": +[ 2913 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2913, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_243_output_0", "outputIndexes": +[ 2914 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2894, 2914 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_151_output_0", "outputIndexes": +[ 2915 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4240 ] +, "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": +[ 4249 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4249 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4250 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4250 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 2916 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2851, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_964_output_0", "outputIndexes": +[ 2917 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2859, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_965_output_0", "outputIndexes": +[ 2918 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2917, 2918, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_392_output_0", "outputIndexes": +[ 2919 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2916, 2919 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_273_output_0", "outputIndexes": +[ 2920 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2886, 2915, 2920, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_278_output_0", "outputIndexes": +[ 2921 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2921 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4251 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4251 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4252 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4252 ] +, "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": +[ 4253 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4253 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4254 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4254 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.30/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 2922 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2838, 2922 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_153_output_0", "outputIndexes": +[ 2923 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2923 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3645868644, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_30/Mul_1_output_0", "outputIndexes": +[ 2924 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2924 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4255 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4255 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4256 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4256 ] +, "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": +[ 4257 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4257 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4258 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4258 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 2925 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2925 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_30/Mul_output_0", "outputIndexes": +[ 2926 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4256 ] +, "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": +[ 4261 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4261 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4262 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4262 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 2927 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2926, 2927 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_30/Mul_output_0", "outputIndexes": +[ 2928 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2928 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4263 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4263 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.30/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4264 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4264 ] +, "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": +[ 4265 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4265 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.30/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4266 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4266 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.30/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 2929 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2923, 2929 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_154_output_0", "outputIndexes": +[ 2930 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2930, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_279_output_0", "outputIndexes": +[ 2931 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2931 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3740273318, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_31/Mul_1_output_0", "outputIndexes": +[ 2932 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2932 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4267 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4267 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4268 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4268 ] +, "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": +[ 4269 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4269 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4270 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4270 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 2933 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2932 ] +, "main_type": "NONE", "name": "/Shape_434_output_0", "outputIndexes": +[ 2934 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2934 ] +, "main_type": "NONE", "name": "Shape12170", "outputIndexes": +[ 2935 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2934 ] +, "main_type": "NONE", "name": "Rank12172", "outputIndexes": +[ 2936 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2936, 2936 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12174", "outputIndexes": +[ 2937 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2937 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12175", "outputIndexes": +[ 2938 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2937, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12177", "outputIndexes": +[ 2939 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2939 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12178", "outputIndexes": +[ 2940 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2935, 2938, 2940, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12181", "outputIndexes": +[ 2941 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2941 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12182", "outputIndexes": +[ 2942 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2942, 2942 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12184", "outputIndexes": +[ 2943 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2934, 2943, 2937 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_468_output_0", "outputIndexes": +[ 2944 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2944, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_992_output_0", "outputIndexes": +[ 2945 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2937 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12192", "outputIndexes": +[ 2946 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2939 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12195", "outputIndexes": +[ 2947 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2935, 2946, 2947, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12198", "outputIndexes": +[ 2948 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2948 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12199", "outputIndexes": +[ 2949 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 2949 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12200", "outputIndexes": +[ 2950 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2950, 2949 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12201", "outputIndexes": +[ 2951 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2934, 2951, 2937 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_469_output_0", "outputIndexes": +[ 2952 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2952, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_993_output_0", "outputIndexes": +[ 2953 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2945, 2953, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_403_output_0", "outputIndexes": +[ 2954 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2933, 2954 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_280_output_0", "outputIndexes": +[ 2955 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2955 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3750791868, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_31/Mul_1_output_0", "outputIndexes": +[ 2957 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2957, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_248_output_0", "outputIndexes": +[ 2958 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2957 ] +, "main_type": "NONE", "name": "/Shape_436_output_0", "outputIndexes": +[ 2959 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2959 ] +, "main_type": "NONE", "name": "Shape12320", "outputIndexes": +[ 2960 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2959 ] +, "main_type": "NONE", "name": "Rank12322", "outputIndexes": +[ 2961 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2961, 2961 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12324", "outputIndexes": +[ 2962 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2962 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12325", "outputIndexes": +[ 2963 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2962, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12327", "outputIndexes": +[ 2964 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2964 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12328", "outputIndexes": +[ 2965 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2960, 2963, 2965, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12331", "outputIndexes": +[ 2966 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2966 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12332", "outputIndexes": +[ 2967 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2967 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12333", "outputIndexes": +[ 2968 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2968, 2967 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12334", "outputIndexes": +[ 2969 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2959, 2969, 2962 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_470_output_0", "outputIndexes": +[ 2970 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2970, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_93_output_0", "outputIndexes": +[ 2971 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2971, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_999_output_0", "outputIndexes": +[ 2972 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2957, 2972, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_125_output_0", "outputIndexes": +[ 2973 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2973 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_62_output_0", "outputIndexes": +[ 2974 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2971, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_998_output_0", "outputIndexes": +[ 2975 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2957, 22, 2975, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_124_output_0", "outputIndexes": +[ 2976 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2974, 2976 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_406_output_0", "outputIndexes": +[ 2977 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2977, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_249_output_0", "outputIndexes": +[ 2978 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2958, 2978 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_155_output_0", "outputIndexes": +[ 2979 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4268 ] +, "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": +[ 4273 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4273 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4274 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4274 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 2980 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2944, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_994_output_0", "outputIndexes": +[ 2981 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2952, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_995_output_0", "outputIndexes": +[ 2982 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2981, 2982, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_404_output_0", "outputIndexes": +[ 2983 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2980, 2983 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_281_output_0", "outputIndexes": +[ 2984 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2984 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3753414354, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_31/Mul_1_output_0", "outputIndexes": +[ 2986 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2986, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_250_output_0", "outputIndexes": +[ 2987 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2986 ] +, "main_type": "NONE", "name": "/Shape_437_output_0", "outputIndexes": +[ 2988 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2988 ] +, "main_type": "NONE", "name": "Shape12216", "outputIndexes": +[ 2989 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 2988 ] +, "main_type": "NONE", "name": "Rank12218", "outputIndexes": +[ 2990 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2990, 2990 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12220", "outputIndexes": +[ 2991 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2991 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12221", "outputIndexes": +[ 2992 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2991, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12223", "outputIndexes": +[ 2993 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2993 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12224", "outputIndexes": +[ 2994 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2989, 2992, 2994, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12227", "outputIndexes": +[ 2995 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2995 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12228", "outputIndexes": +[ 2996 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 2996 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12229", "outputIndexes": +[ 2997 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2997, 2996 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12230", "outputIndexes": +[ 2998 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2988, 2998, 2991 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_471_output_0", "outputIndexes": +[ 2999 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2999, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_94_output_0", "outputIndexes": +[ 3000 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3000, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1001_output_0", "outputIndexes": +[ 3001 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2986, 3001, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_127_output_0", "outputIndexes": +[ 3002 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3002 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_63_output_0", "outputIndexes": +[ 3003 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3000, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1000_output_0", "outputIndexes": +[ 3004 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2986, 22, 3004, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_126_output_0", "outputIndexes": +[ 3005 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3003, 3005 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_407_output_0", "outputIndexes": +[ 3006 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3006, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_251_output_0", "outputIndexes": +[ 3007 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2987, 3007 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_156_output_0", "outputIndexes": +[ 3008 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4268 ] +, "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": +[ 4277 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4277 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4278 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4278 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3009 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2944, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_996_output_0", "outputIndexes": +[ 3010 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2952, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_997_output_0", "outputIndexes": +[ 3011 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3010, 3011, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_405_output_0", "outputIndexes": +[ 3012 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3009, 3012 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_282_output_0", "outputIndexes": +[ 3013 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2979, 3008, 3013, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_287_output_0", "outputIndexes": +[ 3014 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3014 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4279 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4279 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4280 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4280 ] +, "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": +[ 4281 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4281 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4282 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4282 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.31/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3015 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 2931, 3015 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_158_output_0", "outputIndexes": +[ 3016 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3016 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3766522622, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_31/Mul_1_output_0", "outputIndexes": +[ 3017 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3017 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4283 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4283 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4284 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4284 ] +, "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": +[ 4285 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4285 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4286 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4286 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3018 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3018 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_31/Mul_output_0", "outputIndexes": +[ 3019 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4284 ] +, "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": +[ 4289 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4289 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4290 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4290 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3020 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3019, 3020 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_31/Mul_output_0", "outputIndexes": +[ 3021 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3021 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4291 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4291 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.31/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4292 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4292 ] +, "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": +[ 4293 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4293 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.31/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4294 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4294 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.31/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3022 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3016, 3022 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_159_output_0", "outputIndexes": +[ 3023 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3023, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_288_output_0", "outputIndexes": +[ 3024 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3024 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3860927296, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_32/Mul_1_output_0", "outputIndexes": +[ 3025 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3025 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4295 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4295 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4296 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4296 ] +, "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": +[ 4297 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4297 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4298 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4298 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3026 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3025 ] +, "main_type": "NONE", "name": "/Shape_448_output_0", "outputIndexes": +[ 3027 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3027 ] +, "main_type": "NONE", "name": "Shape12559", "outputIndexes": +[ 3028 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3027 ] +, "main_type": "NONE", "name": "Rank12561", "outputIndexes": +[ 3029 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3029, 3029 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12563", "outputIndexes": +[ 3030 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3030 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12564", "outputIndexes": +[ 3031 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3030, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12566", "outputIndexes": +[ 3032 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3032 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12567", "outputIndexes": +[ 3033 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3028, 3031, 3033, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12570", "outputIndexes": +[ 3034 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3034 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12571", "outputIndexes": +[ 3035 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3035, 3035 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12573", "outputIndexes": +[ 3036 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3027, 3036, 3030 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_483_output_0", "outputIndexes": +[ 3037 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3037, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1024_output_0", "outputIndexes": +[ 3038 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3030 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12581", "outputIndexes": +[ 3039 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3032 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12584", "outputIndexes": +[ 3040 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3028, 3039, 3040, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12587", "outputIndexes": +[ 3041 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3041 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12588", "outputIndexes": +[ 3042 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 3042 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12589", "outputIndexes": +[ 3043 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3043, 3042 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12590", "outputIndexes": +[ 3044 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3027, 3044, 3030 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_484_output_0", "outputIndexes": +[ 3045 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3045, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1025_output_0", "outputIndexes": +[ 3046 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3038, 3046, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_416_output_0", "outputIndexes": +[ 3047 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3026, 3047 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_289_output_0", "outputIndexes": +[ 3048 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3048 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3871445846, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_32/Mul_1_output_0", "outputIndexes": +[ 3050 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3050, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_256_output_0", "outputIndexes": +[ 3051 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3050 ] +, "main_type": "NONE", "name": "/Shape_450_output_0", "outputIndexes": +[ 3052 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3052 ] +, "main_type": "NONE", "name": "Shape12709", "outputIndexes": +[ 3053 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3052 ] +, "main_type": "NONE", "name": "Rank12711", "outputIndexes": +[ 3054 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3054, 3054 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12713", "outputIndexes": +[ 3055 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3055 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12714", "outputIndexes": +[ 3056 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3055, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12716", "outputIndexes": +[ 3057 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3057 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12717", "outputIndexes": +[ 3058 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3053, 3056, 3058, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12720", "outputIndexes": +[ 3059 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3059 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12721", "outputIndexes": +[ 3060 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3060 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12722", "outputIndexes": +[ 3061 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3061, 3060 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12723", "outputIndexes": +[ 3062 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3052, 3062, 3055 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_485_output_0", "outputIndexes": +[ 3063 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3063, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_96_output_0", "outputIndexes": +[ 3064 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3064, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1031_output_0", "outputIndexes": +[ 3065 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3050, 3065, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_129_output_0", "outputIndexes": +[ 3066 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3066 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_64_output_0", "outputIndexes": +[ 3067 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3064, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1030_output_0", "outputIndexes": +[ 3068 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3050, 22, 3068, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_128_output_0", "outputIndexes": +[ 3069 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3067, 3069 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_419_output_0", "outputIndexes": +[ 3070 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3070, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_257_output_0", "outputIndexes": +[ 3071 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3051, 3071 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_160_output_0", "outputIndexes": +[ 3072 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4296 ] +, "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": +[ 4301 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4301 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4302 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4302 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3073 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3037, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1026_output_0", "outputIndexes": +[ 3074 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3045, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1027_output_0", "outputIndexes": +[ 3075 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3074, 3075, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_417_output_0", "outputIndexes": +[ 3076 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3073, 3076 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_290_output_0", "outputIndexes": +[ 3077 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3077 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3874068332, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_32/Mul_1_output_0", "outputIndexes": +[ 3079 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3079, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_258_output_0", "outputIndexes": +[ 3080 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3079 ] +, "main_type": "NONE", "name": "/Shape_451_output_0", "outputIndexes": +[ 3081 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3081 ] +, "main_type": "NONE", "name": "Shape12605", "outputIndexes": +[ 3082 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3081 ] +, "main_type": "NONE", "name": "Rank12607", "outputIndexes": +[ 3083 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3083, 3083 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12609", "outputIndexes": +[ 3084 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3084 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12610", "outputIndexes": +[ 3085 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3084, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12612", "outputIndexes": +[ 3086 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3086 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12613", "outputIndexes": +[ 3087 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3082, 3085, 3087, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12616", "outputIndexes": +[ 3088 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3088 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12617", "outputIndexes": +[ 3089 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3089 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12618", "outputIndexes": +[ 3090 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3090, 3089 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12619", "outputIndexes": +[ 3091 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3081, 3091, 3084 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_486_output_0", "outputIndexes": +[ 3092 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3092, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_97_output_0", "outputIndexes": +[ 3093 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3093, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1033_output_0", "outputIndexes": +[ 3094 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3079, 3094, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_131_output_0", "outputIndexes": +[ 3095 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3095 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_65_output_0", "outputIndexes": +[ 3096 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3093, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1032_output_0", "outputIndexes": +[ 3097 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3079, 22, 3097, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_130_output_0", "outputIndexes": +[ 3098 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3096, 3098 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_420_output_0", "outputIndexes": +[ 3099 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3099, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_259_output_0", "outputIndexes": +[ 3100 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3080, 3100 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_161_output_0", "outputIndexes": +[ 3101 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4296 ] +, "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": +[ 4305 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4305 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4306 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4306 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3102 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3037, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1028_output_0", "outputIndexes": +[ 3103 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3045, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1029_output_0", "outputIndexes": +[ 3104 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3103, 3104, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_418_output_0", "outputIndexes": +[ 3105 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3102, 3105 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_291_output_0", "outputIndexes": +[ 3106 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3072, 3101, 3106, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_296_output_0", "outputIndexes": +[ 3107 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3107 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4307 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4307 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4308 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4308 ] +, "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": +[ 4309 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4309 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4310 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4310 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.32/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3108 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3024, 3108 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_163_output_0", "outputIndexes": +[ 3109 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3109 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3887176600, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_32/Mul_1_output_0", "outputIndexes": +[ 3110 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3110 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4311 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4311 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4312 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4312 ] +, "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": +[ 4313 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4313 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4314 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4314 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3111 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3111 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_32/Mul_output_0", "outputIndexes": +[ 3112 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4312 ] +, "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": +[ 4317 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4317 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4318 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4318 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3113 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3112, 3113 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_32/Mul_output_0", "outputIndexes": +[ 3114 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3114 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4319 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4319 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.32/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4320 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4320 ] +, "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": +[ 4321 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4321 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.32/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4322 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4322 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.32/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3115 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3109, 3115 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_164_output_0", "outputIndexes": +[ 3116 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3116, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_297_output_0", "outputIndexes": +[ 3117 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3117 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3981581274, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_33/Mul_1_output_0", "outputIndexes": +[ 3118 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3118 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4323 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4323 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4324 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4324 ] +, "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": +[ 4325 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4325 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4326 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4326 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3119 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3118 ] +, "main_type": "NONE", "name": "/Shape_462_output_0", "outputIndexes": +[ 3120 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3120 ] +, "main_type": "NONE", "name": "Shape12948", "outputIndexes": +[ 3121 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3120 ] +, "main_type": "NONE", "name": "Rank12950", "outputIndexes": +[ 3122 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3122, 3122 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12952", "outputIndexes": +[ 3123 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3123 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12953", "outputIndexes": +[ 3124 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3123, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12955", "outputIndexes": +[ 3125 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3125 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12956", "outputIndexes": +[ 3126 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3121, 3124, 3126, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12959", "outputIndexes": +[ 3127 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3127 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12960", "outputIndexes": +[ 3128 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3128, 3128 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12962", "outputIndexes": +[ 3129 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3120, 3129, 3123 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_498_output_0", "outputIndexes": +[ 3130 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3130, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1056_output_0", "outputIndexes": +[ 3131 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3123 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12970", "outputIndexes": +[ 3132 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3125 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12973", "outputIndexes": +[ 3133 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3121, 3132, 3133, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice12976", "outputIndexes": +[ 3134 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3134 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze12977", "outputIndexes": +[ 3135 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 3135 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12978", "outputIndexes": +[ 3136 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3136, 3135 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12979", "outputIndexes": +[ 3137 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3120, 3137, 3123 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_499_output_0", "outputIndexes": +[ 3138 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3138, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1057_output_0", "outputIndexes": +[ 3139 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3131, 3139, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_429_output_0", "outputIndexes": +[ 3140 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3119, 3140 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_298_output_0", "outputIndexes": +[ 3141 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3141 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3992099824, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_33/Mul_1_output_0", "outputIndexes": +[ 3143 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3143, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_264_output_0", "outputIndexes": +[ 3144 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3143 ] +, "main_type": "NONE", "name": "/Shape_464_output_0", "outputIndexes": +[ 3145 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3145 ] +, "main_type": "NONE", "name": "Shape13098", "outputIndexes": +[ 3146 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3145 ] +, "main_type": "NONE", "name": "Rank13100", "outputIndexes": +[ 3147 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3147, 3147 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13102", "outputIndexes": +[ 3148 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3148 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13103", "outputIndexes": +[ 3149 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3148, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13105", "outputIndexes": +[ 3150 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3150 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13106", "outputIndexes": +[ 3151 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3146, 3149, 3151, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13109", "outputIndexes": +[ 3152 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3152 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13110", "outputIndexes": +[ 3153 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3153 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13111", "outputIndexes": +[ 3154 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3154, 3153 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13112", "outputIndexes": +[ 3155 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3145, 3155, 3148 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_500_output_0", "outputIndexes": +[ 3156 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3156, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_99_output_0", "outputIndexes": +[ 3157 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3157, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1063_output_0", "outputIndexes": +[ 3158 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3143, 3158, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_133_output_0", "outputIndexes": +[ 3159 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3159 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_66_output_0", "outputIndexes": +[ 3160 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3157, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1062_output_0", "outputIndexes": +[ 3161 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3143, 22, 3161, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_132_output_0", "outputIndexes": +[ 3162 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3160, 3162 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_432_output_0", "outputIndexes": +[ 3163 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3163, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_265_output_0", "outputIndexes": +[ 3164 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3144, 3164 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_165_output_0", "outputIndexes": +[ 3165 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4324 ] +, "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": +[ 4329 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4329 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4330 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4330 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3166 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3130, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1058_output_0", "outputIndexes": +[ 3167 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3138, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1059_output_0", "outputIndexes": +[ 3168 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3167, 3168, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_430_output_0", "outputIndexes": +[ 3169 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3166, 3169 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_299_output_0", "outputIndexes": +[ 3170 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3170 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 3994722310, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_33/Mul_1_output_0", "outputIndexes": +[ 3172 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3172, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_266_output_0", "outputIndexes": +[ 3173 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3172 ] +, "main_type": "NONE", "name": "/Shape_465_output_0", "outputIndexes": +[ 3174 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3174 ] +, "main_type": "NONE", "name": "Shape12994", "outputIndexes": +[ 3175 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3174 ] +, "main_type": "NONE", "name": "Rank12996", "outputIndexes": +[ 3176 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3176, 3176 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp12998", "outputIndexes": +[ 3177 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3177 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze12999", "outputIndexes": +[ 3178 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3177, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13001", "outputIndexes": +[ 3179 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3179 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13002", "outputIndexes": +[ 3180 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3175, 3178, 3180, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13005", "outputIndexes": +[ 3181 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3181 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13006", "outputIndexes": +[ 3182 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3182 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13007", "outputIndexes": +[ 3183 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3183, 3182 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13008", "outputIndexes": +[ 3184 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3174, 3184, 3177 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_501_output_0", "outputIndexes": +[ 3185 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3185, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_100_output_0", "outputIndexes": +[ 3186 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3186, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1065_output_0", "outputIndexes": +[ 3187 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3172, 3187, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_135_output_0", "outputIndexes": +[ 3188 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3188 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_67_output_0", "outputIndexes": +[ 3189 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3186, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1064_output_0", "outputIndexes": +[ 3190 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3172, 22, 3190, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_134_output_0", "outputIndexes": +[ 3191 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3189, 3191 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_433_output_0", "outputIndexes": +[ 3192 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3192, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_267_output_0", "outputIndexes": +[ 3193 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3173, 3193 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_166_output_0", "outputIndexes": +[ 3194 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4324 ] +, "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": +[ 4333 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4333 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4334 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4334 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3195 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3130, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1060_output_0", "outputIndexes": +[ 3196 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3138, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1061_output_0", "outputIndexes": +[ 3197 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3196, 3197, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_431_output_0", "outputIndexes": +[ 3198 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3195, 3198 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_300_output_0", "outputIndexes": +[ 3199 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3165, 3194, 3199, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_305_output_0", "outputIndexes": +[ 3200 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3200 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4335 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4335 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4336 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4336 ] +, "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": +[ 4337 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4337 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4338 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4338 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.33/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3201 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3117, 3201 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_168_output_0", "outputIndexes": +[ 3202 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3202 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4007830578, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_33/Mul_1_output_0", "outputIndexes": +[ 3203 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3203 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4339 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4339 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4340 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4340 ] +, "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": +[ 4341 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4341 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4342 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4342 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3204 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3204 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_33/Mul_output_0", "outputIndexes": +[ 3205 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4340 ] +, "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": +[ 4345 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4345 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4346 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4346 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3206 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3205, 3206 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_33/Mul_output_0", "outputIndexes": +[ 3207 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3207 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4347 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4347 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.33/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4348 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4348 ] +, "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": +[ 4349 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4349 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.33/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4350 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4350 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.33/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3208 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3202, 3208 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_169_output_0", "outputIndexes": +[ 3209 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3209, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_306_output_0", "outputIndexes": +[ 3210 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3210 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4102235252, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_34/Mul_1_output_0", "outputIndexes": +[ 3211 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3211 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4351 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4351 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4352 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4352 ] +, "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": +[ 4353 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4353 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4354 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4354 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3212 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3211 ] +, "main_type": "NONE", "name": "/Shape_476_output_0", "outputIndexes": +[ 3213 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3213 ] +, "main_type": "NONE", "name": "Shape13337", "outputIndexes": +[ 3214 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3213 ] +, "main_type": "NONE", "name": "Rank13339", "outputIndexes": +[ 3215 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3215, 3215 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13341", "outputIndexes": +[ 3216 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3216 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13342", "outputIndexes": +[ 3217 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3216, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13344", "outputIndexes": +[ 3218 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3218 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13345", "outputIndexes": +[ 3219 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3214, 3217, 3219, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13348", "outputIndexes": +[ 3220 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3220 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13349", "outputIndexes": +[ 3221 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3221, 3221 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13351", "outputIndexes": +[ 3222 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3213, 3222, 3216 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_513_output_0", "outputIndexes": +[ 3223 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3223, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1088_output_0", "outputIndexes": +[ 3224 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3216 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13359", "outputIndexes": +[ 3225 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3218 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13362", "outputIndexes": +[ 3226 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3214, 3225, 3226, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13365", "outputIndexes": +[ 3227 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3227 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13366", "outputIndexes": +[ 3228 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 3228 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13367", "outputIndexes": +[ 3229 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3229, 3228 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13368", "outputIndexes": +[ 3230 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3213, 3230, 3216 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_514_output_0", "outputIndexes": +[ 3231 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3231, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1089_output_0", "outputIndexes": +[ 3232 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3224, 3232, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_442_output_0", "outputIndexes": +[ 3233 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3212, 3233 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_307_output_0", "outputIndexes": +[ 3234 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3234 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4112753802, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_34/Mul_1_output_0", "outputIndexes": +[ 3236 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3236, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_272_output_0", "outputIndexes": +[ 3237 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3236 ] +, "main_type": "NONE", "name": "/Shape_478_output_0", "outputIndexes": +[ 3238 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3238 ] +, "main_type": "NONE", "name": "Shape13487", "outputIndexes": +[ 3239 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3238 ] +, "main_type": "NONE", "name": "Rank13489", "outputIndexes": +[ 3240 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3240, 3240 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13491", "outputIndexes": +[ 3241 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3241 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13492", "outputIndexes": +[ 3242 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3241, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13494", "outputIndexes": +[ 3243 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3243 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13495", "outputIndexes": +[ 3244 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3239, 3242, 3244, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13498", "outputIndexes": +[ 3245 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3245 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13499", "outputIndexes": +[ 3246 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3246 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13500", "outputIndexes": +[ 3247 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3247, 3246 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13501", "outputIndexes": +[ 3248 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3238, 3248, 3241 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_515_output_0", "outputIndexes": +[ 3249 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3249, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_102_output_0", "outputIndexes": +[ 3250 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3250, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1095_output_0", "outputIndexes": +[ 3251 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3236, 3251, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_137_output_0", "outputIndexes": +[ 3252 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3252 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_68_output_0", "outputIndexes": +[ 3253 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3250, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1094_output_0", "outputIndexes": +[ 3254 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3236, 22, 3254, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_136_output_0", "outputIndexes": +[ 3255 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3253, 3255 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_445_output_0", "outputIndexes": +[ 3256 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3256, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_273_output_0", "outputIndexes": +[ 3257 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3237, 3257 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_170_output_0", "outputIndexes": +[ 3258 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4352 ] +, "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": +[ 4357 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4357 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4358 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4358 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3259 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3223, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1090_output_0", "outputIndexes": +[ 3260 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3231, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1091_output_0", "outputIndexes": +[ 3261 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3260, 3261, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_443_output_0", "outputIndexes": +[ 3262 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3259, 3262 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_308_output_0", "outputIndexes": +[ 3263 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3263 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4115376288, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_34/Mul_1_output_0", "outputIndexes": +[ 3265 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3265, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_274_output_0", "outputIndexes": +[ 3266 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3265 ] +, "main_type": "NONE", "name": "/Shape_479_output_0", "outputIndexes": +[ 3267 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3267 ] +, "main_type": "NONE", "name": "Shape13383", "outputIndexes": +[ 3268 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3267 ] +, "main_type": "NONE", "name": "Rank13385", "outputIndexes": +[ 3269 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3269, 3269 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13387", "outputIndexes": +[ 3270 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3270 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13388", "outputIndexes": +[ 3271 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3270, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13390", "outputIndexes": +[ 3272 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3272 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13391", "outputIndexes": +[ 3273 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3268, 3271, 3273, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13394", "outputIndexes": +[ 3274 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3274 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13395", "outputIndexes": +[ 3275 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3275 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13396", "outputIndexes": +[ 3276 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3276, 3275 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13397", "outputIndexes": +[ 3277 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3267, 3277, 3270 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_516_output_0", "outputIndexes": +[ 3278 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3278, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_103_output_0", "outputIndexes": +[ 3279 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3279, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1097_output_0", "outputIndexes": +[ 3280 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3265, 3280, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_139_output_0", "outputIndexes": +[ 3281 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3281 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_69_output_0", "outputIndexes": +[ 3282 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3279, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1096_output_0", "outputIndexes": +[ 3283 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3265, 22, 3283, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_138_output_0", "outputIndexes": +[ 3284 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3282, 3284 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_446_output_0", "outputIndexes": +[ 3285 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3285, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_275_output_0", "outputIndexes": +[ 3286 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3266, 3286 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_171_output_0", "outputIndexes": +[ 3287 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4352 ] +, "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": +[ 4361 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4361 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4362 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4362 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3288 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3223, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1092_output_0", "outputIndexes": +[ 3289 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3231, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1093_output_0", "outputIndexes": +[ 3290 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3289, 3290, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_444_output_0", "outputIndexes": +[ 3291 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3288, 3291 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_309_output_0", "outputIndexes": +[ 3292 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3258, 3287, 3292, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_314_output_0", "outputIndexes": +[ 3293 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3293 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4363 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4363 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4364 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4364 ] +, "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": +[ 4365 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4365 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4366 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4366 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.34/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3294 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3210, 3294 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_173_output_0", "outputIndexes": +[ 3295 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3295 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4128484556, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_34/Mul_1_output_0", "outputIndexes": +[ 3296 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3296 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4367 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4367 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4368 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4368 ] +, "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": +[ 4369 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4369 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4370 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4370 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3297 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3297 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_34/Mul_output_0", "outputIndexes": +[ 3298 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4368 ] +, "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": +[ 4373 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4373 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4374 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4374 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3299 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3298, 3299 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_34/Mul_output_0", "outputIndexes": +[ 3300 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3300 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4375 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4375 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.34/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4376 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4376 ] +, "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": +[ 4377 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4377 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.34/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4378 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4378 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.34/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3301 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3295, 3301 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_174_output_0", "outputIndexes": +[ 3302 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3302, 5 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_315_output_0", "outputIndexes": +[ 3303 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3303 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4222889230, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/input_layernorm_35/Mul_1_output_0", "outputIndexes": +[ 3304 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3304 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/q_proj/Linear/pre_reshape", "outputIndexes": +[ 4379 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4379 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/self_attn/q_proj/Linear/pre_convert", "outputIndexes": +[ 4380 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4380 ] +, "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": +[ 4381 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4381 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/q_proj/Linear/post_convert", "outputIndexes": +[ 4382 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4382 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/q_proj/Linear/post_reshape", "outputIndexes": +[ 3305 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3304 ] +, "main_type": "NONE", "name": "/Shape_490_output_0", "outputIndexes": +[ 3306 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3306 ] +, "main_type": "NONE", "name": "Shape13726", "outputIndexes": +[ 3307 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3306 ] +, "main_type": "NONE", "name": "Rank13728", "outputIndexes": +[ 3308 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3308, 3308 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13730", "outputIndexes": +[ 3309 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3309 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13731", "outputIndexes": +[ 3310 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3309, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13733", "outputIndexes": +[ 3311 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3311 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13734", "outputIndexes": +[ 3312 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3307, 3310, 3312, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13737", "outputIndexes": +[ 3313 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3313 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13738", "outputIndexes": +[ 3314 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3314, 3314 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13740", "outputIndexes": +[ 3315 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3306, 3315, 3309 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_528_output_0", "outputIndexes": +[ 3316 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3316, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1120_output_0", "outputIndexes": +[ 3317 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3309 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13748", "outputIndexes": +[ 3318 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3311 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13751", "outputIndexes": +[ 3319 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3307, 3318, 3319, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13754", "outputIndexes": +[ 3320 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3320 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13755", "outputIndexes": +[ 3321 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 14, 3321 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13756", "outputIndexes": +[ 3322 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3322, 3321 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13757", "outputIndexes": +[ 3323 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3306, 3323, 3309 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_529_output_0", "outputIndexes": +[ 3324 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3324, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1121_output_0", "outputIndexes": +[ 3325 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3317, 3325, 32, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_455_output_0", "outputIndexes": +[ 3326 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3305, 3326 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_316_output_0", "outputIndexes": +[ 3327 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3327 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4233407780, 512, 512 ] +, "useRMSNorm": true } +, "name": "/q_norm_35/Mul_1_output_0", "outputIndexes": +[ 3329 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3329, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_280_output_0", "outputIndexes": +[ 3330 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3329 ] +, "main_type": "NONE", "name": "/Shape_492_output_0", "outputIndexes": +[ 3331 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3331 ] +, "main_type": "NONE", "name": "Shape13844", "outputIndexes": +[ 3332 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3331 ] +, "main_type": "NONE", "name": "Rank13846", "outputIndexes": +[ 3333 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3333, 3333 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13848", "outputIndexes": +[ 3334 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3334 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13849", "outputIndexes": +[ 3335 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3334, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13851", "outputIndexes": +[ 3336 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3336 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13852", "outputIndexes": +[ 3337 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3332, 3335, 3337, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13855", "outputIndexes": +[ 3338 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3338 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13856", "outputIndexes": +[ 3339 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3339 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13857", "outputIndexes": +[ 3340 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3340, 3339 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13858", "outputIndexes": +[ 3341 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3331, 3341, 3334 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_530_output_0", "outputIndexes": +[ 3342 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3342, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_105_output_0", "outputIndexes": +[ 3343 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3343, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1127_output_0", "outputIndexes": +[ 3344 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3329, 3344, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_141_output_0", "outputIndexes": +[ 3345 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3345 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_70_output_0", "outputIndexes": +[ 3346 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3343, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1126_output_0", "outputIndexes": +[ 3347 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3329, 22, 3347, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_140_output_0", "outputIndexes": +[ 3348 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3346, 3348 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_458_output_0", "outputIndexes": +[ 3349 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3349, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_281_output_0", "outputIndexes": +[ 3350 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3330, 3350 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_175_output_0", "outputIndexes": +[ 3351 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4380 ] +, "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": +[ 4385 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4385 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/k_proj/Linear/post_convert", "outputIndexes": +[ 4386 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4386 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/k_proj/Linear/post_reshape", "outputIndexes": +[ 3352 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3316, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1122_output_0", "outputIndexes": +[ 3353 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3324, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1123_output_0", "outputIndexes": +[ 3354 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3353, 3354, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_456_output_0", "outputIndexes": +[ 3355 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3352, 3355 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_317_output_0", "outputIndexes": +[ 3356 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3356 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4236030266, 512, 512 ] +, "useRMSNorm": true } +, "name": "/k_norm_35/Mul_1_output_0", "outputIndexes": +[ 3358 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3358, 61 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_282_output_0", "outputIndexes": +[ 3359 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3358 ] +, "main_type": "NONE", "name": "/Shape_493_output_0", "outputIndexes": +[ 3360 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3360 ] +, "main_type": "NONE", "name": "Shape13772", "outputIndexes": +[ 3361 ] +, "type": "Shape", "defaultDimentionFormat": "NCHW" } +, +{ "inputIndexes": +[ 3360 ] +, "main_type": "NONE", "name": "Rank13774", "outputIndexes": +[ 3362 ] +, "type": "Rank", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3362, 3362 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13776", "outputIndexes": +[ 3363 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3363 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13777", "outputIndexes": +[ 3364 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3363, 14 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13779", "outputIndexes": +[ 3365 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3365 ] +, "main_type": "SqueezeParam", "main": +{ "squeezeDims": +[ 0 ] + } +, "name": "Unsqueeze13780", "outputIndexes": +[ 3366 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3361, 3364, 3366, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 } +, "name": "StridedSlice13783", "outputIndexes": +[ 3367 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3367 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "Squeeze13784", "outputIndexes": +[ 3368 ] +, "type": "Squeeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 64, 3368 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13785", "outputIndexes": +[ 3369 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3369, 3368 ] +, "main_type": "BinaryOp", "main": +{ "opType": 19, "T": "DT_FLOAT", "activationType": 0 } +, "name": "BinaryOp13786", "outputIndexes": +[ 3370 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3360, 3370, 3363 ] +, "main_type": "GatherV2", "main": +{ "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" } +, "name": "/Gather_531_output_0", "outputIndexes": +[ 3371 ] +, "type": "GatherV2", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3371, 76 ] +, "main_type": "BinaryOp", "main": +{ "opType": 7, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Div_106_output_0", "outputIndexes": +[ 3372 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3372, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1129_output_0", "outputIndexes": +[ 3373 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3358, 3373, 79, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_143_output_0", "outputIndexes": +[ 3374 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3374 ] +, "main_type": "UnaryOp", "main": +{ "opType": "NEG", "T": "DT_FLOAT" } +, "name": "/Neg_71_output_0", "outputIndexes": +[ 3375 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3372, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1128_output_0", "outputIndexes": +[ 3376 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3358, 22, 3376, 80, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_142_output_0", "outputIndexes": +[ 3377 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3375, 3377 ] +, "main_type": "Axis", "main": +{ "axis": -1 } +, "name": "/Concat_459_output_0", "outputIndexes": +[ 3378 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3378, 92 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Mul_283_output_0", "outputIndexes": +[ 3379 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3359, 3379 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_176_output_0", "outputIndexes": +[ 3380 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4380 ] +, "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": +[ 4389 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4389 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/v_proj/Linear/post_convert", "outputIndexes": +[ 4390 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4390 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 1024 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/v_proj/Linear/post_reshape", "outputIndexes": +[ 3381 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3316, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1124_output_0", "outputIndexes": +[ 3382 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3324, 22 ] +, "main_type": "SqueezeParam", "main": +{ } +, "name": "/Unsqueeze_1125_output_0", "outputIndexes": +[ 3383 ] +, "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3382, 3383, 98, 33 ] +, "main_type": "Axis", "main": +{ "axis": 0 } +, "name": "/Concat_457_output_0", "outputIndexes": +[ 3384 ] +, "type": "Concat", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3381, 3384 ] +, "main_type": "Reshape", "main": +{ "dimType": "NCHW" } +, "name": "/Reshape_318_output_0", "outputIndexes": +[ 3385 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3351, 3380, 3385, 130 ] +, "main_type": "AttentionParam", "main": +{ "kv_cache": true } +, "name": "Attention/Reshape_323_output_0", "outputIndexes": +[ 3386 ] +, "type": "Attention", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3386 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/o_proj/Linear/pre_reshape", "outputIndexes": +[ 4391 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4391 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/self_attn/o_proj/Linear/pre_convert", "outputIndexes": +[ 4392 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4392 ] +, "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": +[ 4393 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4393 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/self_attn/o_proj/Linear/post_convert", "outputIndexes": +[ 4394 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4394 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.35/self_attn/o_proj/Linear/post_reshape", "outputIndexes": +[ 3387 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3303, 3387 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_178_output_0", "outputIndexes": +[ 3388 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3388 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4249138534, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "/post_attention_layernorm_35/Mul_1_output_0", "outputIndexes": +[ 3389 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3389 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/gate_proj/Linear/pre_reshape", "outputIndexes": +[ 4395 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4395 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/mlp/gate_proj/Linear/pre_convert", "outputIndexes": +[ 4396 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4396 ] +, "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": +[ 4397 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4397 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/mlp/gate_proj/Linear/post_convert", "outputIndexes": +[ 4398 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4398 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/gate_proj/Linear/post_reshape", "outputIndexes": +[ 3390 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3390 ] +, "main_type": "UnaryOp", "main": +{ "opType": "SILU", "T": "DT_INVALID" } +, "name": "/mlp/act_fn_35/Mul_output_0", "outputIndexes": +[ 3391 ] +, "type": "UnaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4396 ] +, "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": +[ 4401 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4401 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/mlp/up_proj/Linear/post_convert", "outputIndexes": +[ 4402 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4402 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 12288 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/up_proj/Linear/post_reshape", "outputIndexes": +[ 3392 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3391, 3392 ] +, "main_type": "BinaryOp", "main": +{ "opType": 2, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/mlp_35/Mul_output_0", "outputIndexes": +[ 3393 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3393 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 12288, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/down_proj/Linear/pre_reshape", "outputIndexes": +[ 4403 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4403 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/layers.35/mlp/down_proj/Linear/pre_convert", "outputIndexes": +[ 4404 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4404 ] +, "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": +[ 4405 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4405 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/layers.35/mlp/down_proj/Linear/post_convert", "outputIndexes": +[ 4406 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4406 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 4096 ] +, "dimType": "NCHW" } +, "name": "/layers.35/mlp/down_proj/Linear/post_reshape", "outputIndexes": +[ 3394 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3388, 3394 ] +, "main_type": "BinaryOp", "main": +{ "opType": 0, "T": "DT_FLOAT", "activationType": 0 } +, "name": "/Add_179_output_0", "outputIndexes": +[ 3395 ] +, "type": "BinaryOp", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3395, 3, 79, 17, 17 ] +, "main_type": "StridedSliceParam", "main": +{ "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 } +, "name": "/Slice_144_output_0", "outputIndexes": +[ 3396 ] +, "type": "StridedSlice", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3396 ] +, "main_type": "LayerNorm", "main": +{ "axis": +[ -1 ] +, "epsilon": 0.000001, "group": 1, "external": +[ 4343543208, 16384, 16384 ] +, "useRMSNorm": true } +, "name": "hidden_states", "outputIndexes": +[ 3397 ] +, "type": "LayerNorm", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 3397 ] +, "main_type": "Reshape", "main": +{ "dims": +[ -1, 4096, 1, 1 ] +, "dimType": "NCHW" } +, "name": "/lm/lm_head/Linear/pre_reshape", "outputIndexes": +[ 4407 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4407 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NCHW", "dest": "NC4HW4" } +, "name": "/lm/lm_head/Linear/pre_convert", "outputIndexes": +[ 4408 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4408 ] +, "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": +[ 4409 ] +, "type": "Convolution", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4409 ] +, "main_type": "TensorConvertInfo", "main": +{ "source": "NC4HW4", "dest": "NCHW" } +, "name": "/lm/lm_head/Linear/post_convert", "outputIndexes": +[ 4410 ] +, "type": "ConvertTensor", "defaultDimentionFormat": "NHWC" } +, +{ "inputIndexes": +[ 4410 ] +, "main_type": "Reshape", "main": +{ "dims": +[ 1, -1, 151936 ] +, "dimType": "NCHW" } +, "name": "/lm/lm_head/Linear/post_reshape", "outputIndexes": +[ 3398 ] +, "type": "Reshape", "defaultDimentionFormat": "NHWC" } + ] +, "outputName": +[ "logits", "hidden_states", "presents" ] +, "preferForwardType": "CPU", "sourceType": "ONNX", "tensorName": +[ "past_key_values", "presents", "position_ids", "logits_index", "input_ids", "/Constant_output_0", "/Reshape_output_0", "/input_layernorm/Mul_1_output_0", "/FakeLinear_output_0", "/Shape_output_0", "Shape52", "Rank54", "BinaryOp56", "Unsqueeze57", "Const9", "BinaryOp59", "Unsqueeze60", "Unsqueeze62", "StridedSlice63", "Squeeze64", "BinaryOp66", "/Gather_1_output_0", "/rotary/Constant_4_output_0", "/Unsqueeze_output_0", "Unsqueeze76", "Unsqueeze79", "StridedSlice82", "Squeeze83", "BinaryOp84", "BinaryOp85", "/Gather_2_output_0", "/Unsqueeze_1_output_0", "/Constant_4_output_0", "/Constant_5_output_0", "/Concat_output_0", "/Reshape_1_output_0", "/q_norm/Cast_output_0", "/q_norm/Mul_1_output_0", "/rotary/Cast_output_0", "/rotary/Constant_1_output_0", "/rotary/Reshape_output_0", "/rotary/Constant_2_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_7_output_0", "/rotary/Unsqueeze_2_output_0", "/rotary/Unsqueeze_3_output_0", "Shape116", "Rank118", "BinaryOp120", "Unsqueeze121", "BinaryOp123", "Unsqueeze124", "StridedSlice127", "Squeeze128", "BinaryOp130", "/Gather_3_output_0", "/Mul_output_0", "/Shape_2_output_0", "/Constant_11_output_0", "Shape258", "Rank260", "BinaryOp262", "Unsqueeze263", "BinaryOp265", "Unsqueeze266", "StridedSlice269", "Squeeze270", "BinaryOp271", "BinaryOp272", "/Gather_5_output_0", "/rotary/Constant_6_output_0", "/Div_output_0", "/Unsqueeze_7_output_0", "/Constant_20_output_0", "/Constant_14_output_0", "/Slice_1_output_0", "/Neg_output_0", "/Unsqueeze_6_output_0", "/Slice_output_0", "/Concat_3_output_0", "Unsqueeze167", "Unsqueeze170", "StridedSlice173", "Squeeze174", "BinaryOp175", "BinaryOp176", "/Gather_4_output_0", "/Mul_1_output_0", "/Add_output_0", "/FakeLinear_1_output_0", "/Unsqueeze_2_output_0", "/Unsqueeze_3_output_0", "/Constant_7_output_0", "/Concat_1_output_0", "/Reshape_2_output_0", "/k_norm/Cast_output_0", "/k_norm/Mul_1_output_0", "/Mul_2_output_0", "/Shape_3_output_0", "Shape135", "Rank137", "BinaryOp139", "Unsqueeze140", "BinaryOp142", "Unsqueeze143", "StridedSlice146", "Squeeze147", "BinaryOp148", "BinaryOp149", "/Gather_6_output_0", "/Div_1_output_0", "/Unsqueeze_9_output_0", "/Slice_3_output_0", "/Neg_1_output_0", "/Unsqueeze_8_output_0", "/Slice_2_output_0", "/Concat_4_output_0", "/Mul_3_output_0", "/Add_1_output_0", "/FakeLinear_2_output_0", "/Unsqueeze_4_output_0", "/Unsqueeze_5_output_0", "/Concat_2_output_0", "/Reshape_3_output_0", "attention_mask", "/Reshape_8_output_0", "/FakeLinear_3_output_0", "/Add_3_output_0", "/post_attention_layernorm/Mul_1_output_0", "/mlp/gate_proj/FakeLinear_output_0", "/mlp/act_fn/Mul_output_0", "/mlp/up_proj/FakeLinear_output_0", "/mlp/Mul_output_0", "/mlp/down_proj/FakeLinear_output_0", "/Add_4_output_0", "/Reshape_9_output_0", "/input_layernorm_1/Mul_1_output_0", "/FakeLinear_4_output_0", "/Shape_14_output_0", "Shape503", "Rank505", "BinaryOp507", "Unsqueeze508", "BinaryOp510", "Unsqueeze511", "StridedSlice514", "Squeeze515", "BinaryOp517", "/Gather_18_output_0", "/Unsqueeze_32_output_0", "Unsqueeze525", "Unsqueeze528", "StridedSlice531", "Squeeze532", "BinaryOp533", "BinaryOp534", "/Gather_19_output_0", "/Unsqueeze_33_output_0", "/Concat_13_output_0", "/Reshape_10_output_0", "/q_norm_1/Cast_output_0", "/q_norm_1/Mul_1_output_0", "/Mul_8_output_0", "/Shape_16_output_0", "Shape652", "Rank654", "BinaryOp656", "Unsqueeze657", "BinaryOp659", "Unsqueeze660", "StridedSlice663", "Squeeze664", "BinaryOp665", "BinaryOp666", "/Gather_20_output_0", "/Div_3_output_0", "/Unsqueeze_39_output_0", "/Slice_5_output_0", "/Neg_2_output_0", "/Unsqueeze_38_output_0", "/Slice_4_output_0", "/Concat_16_output_0", "/Mul_9_output_0", "/Add_5_output_0", "/FakeLinear_5_output_0", "/Unsqueeze_34_output_0", "/Unsqueeze_35_output_0", "/Concat_14_output_0", "/Reshape_11_output_0", "/k_norm_1/Cast_output_0", "/k_norm_1/Mul_1_output_0", "/Mul_10_output_0", "/Shape_17_output_0", "Shape549", "Rank551", "BinaryOp553", "Unsqueeze554", "BinaryOp556", "Unsqueeze557", "StridedSlice560", "Squeeze561", "BinaryOp562", "BinaryOp563", "/Gather_21_output_0", "/Div_4_output_0", "/Unsqueeze_41_output_0", "/Slice_7_output_0", "/Neg_3_output_0", "/Unsqueeze_40_output_0", "/Slice_6_output_0", "/Concat_17_output_0", "/Mul_11_output_0", "/Add_6_output_0", "/FakeLinear_6_output_0", "/Unsqueeze_36_output_0", "/Unsqueeze_37_output_0", "/Concat_15_output_0", "/Reshape_12_output_0", "/Reshape_17_output_0", "/FakeLinear_7_output_0", "/Add_8_output_0", "/post_attention_layernorm_1/Mul_1_output_0", "/mlp/gate_proj_1/FakeLinear_output_0", "/mlp/act_fn_1/Mul_output_0", "/mlp/up_proj_1/FakeLinear_output_0", "/mlp_1/Mul_output_0", "/mlp/down_proj_1/FakeLinear_output_0", "/Add_9_output_0", "/Reshape_18_output_0", "/input_layernorm_2/Mul_1_output_0", "/FakeLinear_8_output_0", "/Shape_28_output_0", "Shape891", "Rank893", "BinaryOp895", "Unsqueeze896", "BinaryOp898", "Unsqueeze899", "StridedSlice902", "Squeeze903", "BinaryOp905", "/Gather_33_output_0", "/Unsqueeze_64_output_0", "Unsqueeze913", "Unsqueeze916", "StridedSlice919", "Squeeze920", "BinaryOp921", "BinaryOp922", "/Gather_34_output_0", "/Unsqueeze_65_output_0", "/Concat_26_output_0", "/Reshape_19_output_0", "/q_norm_2/Cast_output_0", "/q_norm_2/Mul_1_output_0", "/Mul_16_output_0", "/Shape_30_output_0", "Shape1040", "Rank1042", "BinaryOp1044", "Unsqueeze1045", "BinaryOp1047", "Unsqueeze1048", "StridedSlice1051", "Squeeze1052", "BinaryOp1053", "BinaryOp1054", "/Gather_35_output_0", "/Div_6_output_0", "/Unsqueeze_71_output_0", "/Slice_9_output_0", "/Neg_4_output_0", "/Unsqueeze_70_output_0", "/Slice_8_output_0", "/Concat_29_output_0", "/Mul_17_output_0", "/Add_10_output_0", "/FakeLinear_9_output_0", "/Unsqueeze_66_output_0", "/Unsqueeze_67_output_0", "/Concat_27_output_0", "/Reshape_20_output_0", "/k_norm_2/Cast_output_0", "/k_norm_2/Mul_1_output_0", "/Mul_18_output_0", "/Shape_31_output_0", "Shape937", "Rank939", "BinaryOp941", "Unsqueeze942", "BinaryOp944", "Unsqueeze945", "StridedSlice948", "Squeeze949", "BinaryOp950", "BinaryOp951", "/Gather_36_output_0", "/Div_7_output_0", "/Unsqueeze_73_output_0", "/Slice_11_output_0", "/Neg_5_output_0", "/Unsqueeze_72_output_0", "/Slice_10_output_0", "/Concat_30_output_0", "/Mul_19_output_0", "/Add_11_output_0", "/FakeLinear_10_output_0", "/Unsqueeze_68_output_0", "/Unsqueeze_69_output_0", "/Concat_28_output_0", "/Reshape_21_output_0", "/Reshape_26_output_0", "/FakeLinear_11_output_0", "/Add_13_output_0", "/post_attention_layernorm_2/Mul_1_output_0", "/mlp/gate_proj_2/FakeLinear_output_0", "/mlp/act_fn_2/Mul_output_0", "/mlp/up_proj_2/FakeLinear_output_0", "/mlp_2/Mul_output_0", "/mlp/down_proj_2/FakeLinear_output_0", "/Add_14_output_0", "/Reshape_27_output_0", "/input_layernorm_3/Mul_1_output_0", "/FakeLinear_12_output_0", "/Shape_42_output_0", "Shape1279", "Rank1281", "BinaryOp1283", "Unsqueeze1284", "BinaryOp1286", "Unsqueeze1287", "StridedSlice1290", "Squeeze1291", "BinaryOp1293", "/Gather_48_output_0", "/Unsqueeze_96_output_0", "Unsqueeze1301", "Unsqueeze1304", "StridedSlice1307", "Squeeze1308", "BinaryOp1309", "BinaryOp1310", "/Gather_49_output_0", "/Unsqueeze_97_output_0", "/Concat_39_output_0", "/Reshape_28_output_0", "/q_norm_3/Cast_output_0", "/q_norm_3/Mul_1_output_0", "/Mul_24_output_0", "/Shape_44_output_0", "Shape1428", "Rank1430", "BinaryOp1432", "Unsqueeze1433", "BinaryOp1435", "Unsqueeze1436", "StridedSlice1439", "Squeeze1440", "BinaryOp1441", "BinaryOp1442", "/Gather_50_output_0", "/Div_9_output_0", "/Unsqueeze_103_output_0", "/Slice_13_output_0", "/Neg_6_output_0", "/Unsqueeze_102_output_0", "/Slice_12_output_0", "/Concat_42_output_0", "/Mul_25_output_0", "/Add_15_output_0", "/FakeLinear_13_output_0", "/Unsqueeze_98_output_0", "/Unsqueeze_99_output_0", "/Concat_40_output_0", "/Reshape_29_output_0", "/k_norm_3/Cast_output_0", "/k_norm_3/Mul_1_output_0", "/Mul_26_output_0", "/Shape_45_output_0", "Shape1325", "Rank1327", "BinaryOp1329", "Unsqueeze1330", "BinaryOp1332", "Unsqueeze1333", "StridedSlice1336", "Squeeze1337", "BinaryOp1338", "BinaryOp1339", "/Gather_51_output_0", "/Div_10_output_0", "/Unsqueeze_105_output_0", "/Slice_15_output_0", "/Neg_7_output_0", "/Unsqueeze_104_output_0", "/Slice_14_output_0", "/Concat_43_output_0", "/Mul_27_output_0", "/Add_16_output_0", "/FakeLinear_14_output_0", "/Unsqueeze_100_output_0", "/Unsqueeze_101_output_0", "/Concat_41_output_0", "/Reshape_30_output_0", "/Reshape_35_output_0", "/FakeLinear_15_output_0", "/Add_18_output_0", "/post_attention_layernorm_3/Mul_1_output_0", "/mlp/gate_proj_3/FakeLinear_output_0", "/mlp/act_fn_3/Mul_output_0", "/mlp/up_proj_3/FakeLinear_output_0", "/mlp_3/Mul_output_0", "/mlp/down_proj_3/FakeLinear_output_0", "/Add_19_output_0", "/Reshape_36_output_0", "/input_layernorm_4/Mul_1_output_0", "/FakeLinear_16_output_0", "/Shape_56_output_0", "Shape1667", "Rank1669", "BinaryOp1671", "Unsqueeze1672", "BinaryOp1674", "Unsqueeze1675", "StridedSlice1678", "Squeeze1679", "BinaryOp1681", "/Gather_63_output_0", "/Unsqueeze_128_output_0", "Unsqueeze1689", "Unsqueeze1692", "StridedSlice1695", "Squeeze1696", "BinaryOp1697", "BinaryOp1698", "/Gather_64_output_0", "/Unsqueeze_129_output_0", "/Concat_52_output_0", "/Reshape_37_output_0", "/q_norm_4/Cast_output_0", "/q_norm_4/Mul_1_output_0", "/Mul_32_output_0", "/Shape_58_output_0", "Shape1817", "Rank1819", "BinaryOp1821", "Unsqueeze1822", "BinaryOp1824", "Unsqueeze1825", "StridedSlice1828", "Squeeze1829", "BinaryOp1830", "BinaryOp1831", "/Gather_65_output_0", "/Div_12_output_0", "/Unsqueeze_135_output_0", "/Slice_17_output_0", "/Neg_8_output_0", "/Unsqueeze_134_output_0", "/Slice_16_output_0", "/Concat_55_output_0", "/Mul_33_output_0", "/Add_20_output_0", "/FakeLinear_17_output_0", "/Unsqueeze_130_output_0", "/Unsqueeze_131_output_0", "/Concat_53_output_0", "/Reshape_38_output_0", "/k_norm_4/Cast_output_0", "/k_norm_4/Mul_1_output_0", "/Mul_34_output_0", "/Shape_59_output_0", "Shape1713", "Rank1715", "BinaryOp1717", "Unsqueeze1718", "BinaryOp1720", "Unsqueeze1721", "StridedSlice1724", "Squeeze1725", "BinaryOp1726", "BinaryOp1727", "/Gather_66_output_0", "/Div_13_output_0", "/Unsqueeze_137_output_0", "/Slice_19_output_0", "/Neg_9_output_0", "/Unsqueeze_136_output_0", "/Slice_18_output_0", "/Concat_56_output_0", "/Mul_35_output_0", "/Add_21_output_0", "/FakeLinear_18_output_0", "/Unsqueeze_132_output_0", "/Unsqueeze_133_output_0", "/Concat_54_output_0", "/Reshape_39_output_0", "/Reshape_44_output_0", "/FakeLinear_19_output_0", "/Add_23_output_0", "/post_attention_layernorm_4/Mul_1_output_0", "/mlp/gate_proj_4/FakeLinear_output_0", "/mlp/act_fn_4/Mul_output_0", "/mlp/up_proj_4/FakeLinear_output_0", "/mlp_4/Mul_output_0", "/mlp/down_proj_4/FakeLinear_output_0", "/Add_24_output_0", "/Reshape_45_output_0", "/input_layernorm_5/Mul_1_output_0", "/FakeLinear_20_output_0", "/Shape_70_output_0", "Shape2056", "Rank2058", "BinaryOp2060", "Unsqueeze2061", "BinaryOp2063", "Unsqueeze2064", "StridedSlice2067", "Squeeze2068", "BinaryOp2070", "/Gather_78_output_0", "/Unsqueeze_160_output_0", "Unsqueeze2078", "Unsqueeze2081", "StridedSlice2084", "Squeeze2085", "BinaryOp2086", "BinaryOp2087", "/Gather_79_output_0", "/Unsqueeze_161_output_0", "/Concat_65_output_0", "/Reshape_46_output_0", "/q_norm_5/Cast_output_0", "/q_norm_5/Mul_1_output_0", "/Mul_40_output_0", "/Shape_72_output_0", "Shape2206", "Rank2208", "BinaryOp2210", "Unsqueeze2211", "BinaryOp2213", "Unsqueeze2214", "StridedSlice2217", "Squeeze2218", "BinaryOp2219", "BinaryOp2220", "/Gather_80_output_0", "/Div_15_output_0", "/Unsqueeze_167_output_0", "/Slice_21_output_0", "/Neg_10_output_0", "/Unsqueeze_166_output_0", "/Slice_20_output_0", "/Concat_68_output_0", "/Mul_41_output_0", "/Add_25_output_0", "/FakeLinear_21_output_0", "/Unsqueeze_162_output_0", "/Unsqueeze_163_output_0", "/Concat_66_output_0", "/Reshape_47_output_0", "/k_norm_5/Cast_output_0", "/k_norm_5/Mul_1_output_0", "/Mul_42_output_0", "/Shape_73_output_0", "Shape2102", "Rank2104", "BinaryOp2106", "Unsqueeze2107", "BinaryOp2109", "Unsqueeze2110", "StridedSlice2113", "Squeeze2114", "BinaryOp2115", "BinaryOp2116", "/Gather_81_output_0", "/Div_16_output_0", "/Unsqueeze_169_output_0", "/Slice_23_output_0", "/Neg_11_output_0", "/Unsqueeze_168_output_0", "/Slice_22_output_0", "/Concat_69_output_0", "/Mul_43_output_0", "/Add_26_output_0", "/FakeLinear_22_output_0", "/Unsqueeze_164_output_0", "/Unsqueeze_165_output_0", "/Concat_67_output_0", "/Reshape_48_output_0", "/Reshape_53_output_0", "/FakeLinear_23_output_0", "/Add_28_output_0", "/post_attention_layernorm_5/Mul_1_output_0", "/mlp/gate_proj_5/FakeLinear_output_0", "/mlp/act_fn_5/Mul_output_0", "/mlp/up_proj_5/FakeLinear_output_0", "/mlp_5/Mul_output_0", "/mlp/down_proj_5/FakeLinear_output_0", "/Add_29_output_0", "/Reshape_54_output_0", "/input_layernorm_6/Mul_1_output_0", "/FakeLinear_24_output_0", "/Shape_84_output_0", "Shape2445", "Rank2447", "BinaryOp2449", "Unsqueeze2450", "BinaryOp2452", "Unsqueeze2453", "StridedSlice2456", "Squeeze2457", "BinaryOp2459", "/Gather_93_output_0", "/Unsqueeze_192_output_0", "Unsqueeze2467", "Unsqueeze2470", "StridedSlice2473", "Squeeze2474", "BinaryOp2475", "BinaryOp2476", "/Gather_94_output_0", "/Unsqueeze_193_output_0", "/Concat_78_output_0", "/Reshape_55_output_0", "/q_norm_6/Cast_output_0", "/q_norm_6/Mul_1_output_0", "/Mul_48_output_0", "/Shape_86_output_0", "Shape2595", "Rank2597", "BinaryOp2599", "Unsqueeze2600", "BinaryOp2602", "Unsqueeze2603", "StridedSlice2606", "Squeeze2607", "BinaryOp2608", "BinaryOp2609", "/Gather_95_output_0", "/Div_18_output_0", "/Unsqueeze_199_output_0", "/Slice_25_output_0", "/Neg_12_output_0", "/Unsqueeze_198_output_0", "/Slice_24_output_0", "/Concat_81_output_0", "/Mul_49_output_0", "/Add_30_output_0", "/FakeLinear_25_output_0", "/Unsqueeze_194_output_0", "/Unsqueeze_195_output_0", "/Concat_79_output_0", "/Reshape_56_output_0", "/k_norm_6/Cast_output_0", "/k_norm_6/Mul_1_output_0", "/Mul_50_output_0", "/Shape_87_output_0", "Shape2491", "Rank2493", "BinaryOp2495", "Unsqueeze2496", "BinaryOp2498", "Unsqueeze2499", "StridedSlice2502", "Squeeze2503", "BinaryOp2504", "BinaryOp2505", "/Gather_96_output_0", "/Div_19_output_0", "/Unsqueeze_201_output_0", "/Slice_27_output_0", "/Neg_13_output_0", "/Unsqueeze_200_output_0", "/Slice_26_output_0", "/Concat_82_output_0", "/Mul_51_output_0", "/Add_31_output_0", "/FakeLinear_26_output_0", "/Unsqueeze_196_output_0", "/Unsqueeze_197_output_0", "/Concat_80_output_0", "/Reshape_57_output_0", "/Reshape_62_output_0", "/FakeLinear_27_output_0", "/Add_33_output_0", "/post_attention_layernorm_6/Mul_1_output_0", "/mlp/gate_proj_6/FakeLinear_output_0", "/mlp/act_fn_6/Mul_output_0", "/mlp/up_proj_6/FakeLinear_output_0", "/mlp_6/Mul_output_0", "/mlp/down_proj_6/FakeLinear_output_0", "/Add_34_output_0", "/Reshape_63_output_0", "/input_layernorm_7/Mul_1_output_0", "/FakeLinear_28_output_0", "/Shape_98_output_0", "Shape2834", "Rank2836", "BinaryOp2838", "Unsqueeze2839", "BinaryOp2841", "Unsqueeze2842", "StridedSlice2845", "Squeeze2846", "BinaryOp2848", "/Gather_108_output_0", "/Unsqueeze_224_output_0", "Unsqueeze2856", "Unsqueeze2859", "StridedSlice2862", "Squeeze2863", "BinaryOp2864", "BinaryOp2865", "/Gather_109_output_0", "/Unsqueeze_225_output_0", "/Concat_91_output_0", "/Reshape_64_output_0", "/q_norm_7/Cast_output_0", "/q_norm_7/Mul_1_output_0", "/Mul_56_output_0", "/Shape_100_output_0", "Shape2984", "Rank2986", "BinaryOp2988", "Unsqueeze2989", "BinaryOp2991", "Unsqueeze2992", "StridedSlice2995", "Squeeze2996", "BinaryOp2997", "BinaryOp2998", "/Gather_110_output_0", "/Div_21_output_0", "/Unsqueeze_231_output_0", "/Slice_29_output_0", "/Neg_14_output_0", "/Unsqueeze_230_output_0", "/Slice_28_output_0", "/Concat_94_output_0", "/Mul_57_output_0", "/Add_35_output_0", "/FakeLinear_29_output_0", "/Unsqueeze_226_output_0", "/Unsqueeze_227_output_0", "/Concat_92_output_0", "/Reshape_65_output_0", "/k_norm_7/Cast_output_0", "/k_norm_7/Mul_1_output_0", "/Mul_58_output_0", "/Shape_101_output_0", "Shape2880", "Rank2882", "BinaryOp2884", "Unsqueeze2885", "BinaryOp2887", "Unsqueeze2888", "StridedSlice2891", "Squeeze2892", "BinaryOp2893", "BinaryOp2894", "/Gather_111_output_0", "/Div_22_output_0", "/Unsqueeze_233_output_0", "/Slice_31_output_0", "/Neg_15_output_0", "/Unsqueeze_232_output_0", "/Slice_30_output_0", "/Concat_95_output_0", "/Mul_59_output_0", "/Add_36_output_0", "/FakeLinear_30_output_0", "/Unsqueeze_228_output_0", "/Unsqueeze_229_output_0", "/Concat_93_output_0", "/Reshape_66_output_0", "/Reshape_71_output_0", "/FakeLinear_31_output_0", "/Add_38_output_0", "/post_attention_layernorm_7/Mul_1_output_0", "/mlp/gate_proj_7/FakeLinear_output_0", "/mlp/act_fn_7/Mul_output_0", "/mlp/up_proj_7/FakeLinear_output_0", "/mlp_7/Mul_output_0", "/mlp/down_proj_7/FakeLinear_output_0", "/Add_39_output_0", "/Reshape_72_output_0", "/input_layernorm_8/Mul_1_output_0", "/FakeLinear_32_output_0", "/Shape_112_output_0", "Shape3223", "Rank3225", "BinaryOp3227", "Unsqueeze3228", "BinaryOp3230", "Unsqueeze3231", "StridedSlice3234", "Squeeze3235", "BinaryOp3237", "/Gather_123_output_0", "/Unsqueeze_256_output_0", "Unsqueeze3245", "Unsqueeze3248", "StridedSlice3251", "Squeeze3252", "BinaryOp3253", "BinaryOp3254", "/Gather_124_output_0", "/Unsqueeze_257_output_0", "/Concat_104_output_0", "/Reshape_73_output_0", "/q_norm_8/Cast_output_0", "/q_norm_8/Mul_1_output_0", "/Mul_64_output_0", "/Shape_114_output_0", "Shape3373", "Rank3375", "BinaryOp3377", "Unsqueeze3378", "BinaryOp3380", "Unsqueeze3381", "StridedSlice3384", "Squeeze3385", "BinaryOp3386", "BinaryOp3387", "/Gather_125_output_0", "/Div_24_output_0", "/Unsqueeze_263_output_0", "/Slice_33_output_0", "/Neg_16_output_0", "/Unsqueeze_262_output_0", "/Slice_32_output_0", "/Concat_107_output_0", "/Mul_65_output_0", "/Add_40_output_0", "/FakeLinear_33_output_0", "/Unsqueeze_258_output_0", "/Unsqueeze_259_output_0", "/Concat_105_output_0", "/Reshape_74_output_0", "/k_norm_8/Cast_output_0", "/k_norm_8/Mul_1_output_0", "/Mul_66_output_0", "/Shape_115_output_0", "Shape3269", "Rank3271", "BinaryOp3273", "Unsqueeze3274", "BinaryOp3276", "Unsqueeze3277", "StridedSlice3280", "Squeeze3281", "BinaryOp3282", "BinaryOp3283", "/Gather_126_output_0", "/Div_25_output_0", "/Unsqueeze_265_output_0", "/Slice_35_output_0", "/Neg_17_output_0", "/Unsqueeze_264_output_0", "/Slice_34_output_0", "/Concat_108_output_0", "/Mul_67_output_0", "/Add_41_output_0", "/FakeLinear_34_output_0", "/Unsqueeze_260_output_0", "/Unsqueeze_261_output_0", "/Concat_106_output_0", "/Reshape_75_output_0", "/Reshape_80_output_0", "/FakeLinear_35_output_0", "/Add_43_output_0", "/post_attention_layernorm_8/Mul_1_output_0", "/mlp/gate_proj_8/FakeLinear_output_0", "/mlp/act_fn_8/Mul_output_0", "/mlp/up_proj_8/FakeLinear_output_0", "/mlp_8/Mul_output_0", "/mlp/down_proj_8/FakeLinear_output_0", "/Add_44_output_0", "/Reshape_81_output_0", "/input_layernorm_9/Mul_1_output_0", "/FakeLinear_36_output_0", "/Shape_126_output_0", "Shape3612", "Rank3614", "BinaryOp3616", "Unsqueeze3617", "BinaryOp3619", "Unsqueeze3620", "StridedSlice3623", "Squeeze3624", "BinaryOp3626", "/Gather_138_output_0", "/Unsqueeze_288_output_0", "Unsqueeze3634", "Unsqueeze3637", "StridedSlice3640", "Squeeze3641", "BinaryOp3642", "BinaryOp3643", "/Gather_139_output_0", "/Unsqueeze_289_output_0", "/Concat_117_output_0", "/Reshape_82_output_0", "/q_norm_9/Cast_output_0", "/q_norm_9/Mul_1_output_0", "/Mul_72_output_0", "/Shape_128_output_0", "Shape3762", "Rank3764", "BinaryOp3766", "Unsqueeze3767", "BinaryOp3769", "Unsqueeze3770", "StridedSlice3773", "Squeeze3774", "BinaryOp3775", "BinaryOp3776", "/Gather_140_output_0", "/Div_27_output_0", "/Unsqueeze_295_output_0", "/Slice_37_output_0", "/Neg_18_output_0", "/Unsqueeze_294_output_0", "/Slice_36_output_0", "/Concat_120_output_0", "/Mul_73_output_0", "/Add_45_output_0", "/FakeLinear_37_output_0", "/Unsqueeze_290_output_0", "/Unsqueeze_291_output_0", "/Concat_118_output_0", "/Reshape_83_output_0", "/k_norm_9/Cast_output_0", "/k_norm_9/Mul_1_output_0", "/Mul_74_output_0", "/Shape_129_output_0", "Shape3658", "Rank3660", "BinaryOp3662", "Unsqueeze3663", "BinaryOp3665", "Unsqueeze3666", "StridedSlice3669", "Squeeze3670", "BinaryOp3671", "BinaryOp3672", "/Gather_141_output_0", "/Div_28_output_0", "/Unsqueeze_297_output_0", "/Slice_39_output_0", "/Neg_19_output_0", "/Unsqueeze_296_output_0", "/Slice_38_output_0", "/Concat_121_output_0", "/Mul_75_output_0", "/Add_46_output_0", "/FakeLinear_38_output_0", "/Unsqueeze_292_output_0", "/Unsqueeze_293_output_0", "/Concat_119_output_0", "/Reshape_84_output_0", "/Reshape_89_output_0", "/FakeLinear_39_output_0", "/Add_48_output_0", "/post_attention_layernorm_9/Mul_1_output_0", "/mlp/gate_proj_9/FakeLinear_output_0", "/mlp/act_fn_9/Mul_output_0", "/mlp/up_proj_9/FakeLinear_output_0", "/mlp_9/Mul_output_0", "/mlp/down_proj_9/FakeLinear_output_0", "/Add_49_output_0", "/Reshape_90_output_0", "/input_layernorm_10/Mul_1_output_0", "/FakeLinear_40_output_0", "/Shape_140_output_0", "Shape4001", "Rank4003", "BinaryOp4005", "Unsqueeze4006", "BinaryOp4008", "Unsqueeze4009", "StridedSlice4012", "Squeeze4013", "BinaryOp4015", "/Gather_153_output_0", "/Unsqueeze_320_output_0", "Unsqueeze4023", "Unsqueeze4026", "StridedSlice4029", "Squeeze4030", "BinaryOp4031", "BinaryOp4032", "/Gather_154_output_0", "/Unsqueeze_321_output_0", "/Concat_130_output_0", "/Reshape_91_output_0", "/q_norm_10/Cast_output_0", "/q_norm_10/Mul_1_output_0", "/Mul_80_output_0", "/Shape_142_output_0", "Shape4151", "Rank4153", "BinaryOp4155", "Unsqueeze4156", "BinaryOp4158", "Unsqueeze4159", "StridedSlice4162", "Squeeze4163", "BinaryOp4164", "BinaryOp4165", "/Gather_155_output_0", "/Div_30_output_0", "/Unsqueeze_327_output_0", "/Slice_41_output_0", "/Neg_20_output_0", "/Unsqueeze_326_output_0", "/Slice_40_output_0", "/Concat_133_output_0", "/Mul_81_output_0", "/Add_50_output_0", "/FakeLinear_41_output_0", "/Unsqueeze_322_output_0", "/Unsqueeze_323_output_0", "/Concat_131_output_0", "/Reshape_92_output_0", "/k_norm_10/Cast_output_0", "/k_norm_10/Mul_1_output_0", "/Mul_82_output_0", "/Shape_143_output_0", "Shape4047", "Rank4049", "BinaryOp4051", "Unsqueeze4052", "BinaryOp4054", "Unsqueeze4055", "StridedSlice4058", "Squeeze4059", "BinaryOp4060", "BinaryOp4061", "/Gather_156_output_0", "/Div_31_output_0", "/Unsqueeze_329_output_0", "/Slice_43_output_0", "/Neg_21_output_0", "/Unsqueeze_328_output_0", "/Slice_42_output_0", "/Concat_134_output_0", "/Mul_83_output_0", "/Add_51_output_0", "/FakeLinear_42_output_0", "/Unsqueeze_324_output_0", "/Unsqueeze_325_output_0", "/Concat_132_output_0", "/Reshape_93_output_0", "/Reshape_98_output_0", "/FakeLinear_43_output_0", "/Add_53_output_0", "/post_attention_layernorm_10/Mul_1_output_0", "/mlp/gate_proj_10/FakeLinear_output_0", "/mlp/act_fn_10/Mul_output_0", "/mlp/up_proj_10/FakeLinear_output_0", "/mlp_10/Mul_output_0", "/mlp/down_proj_10/FakeLinear_output_0", "/Add_54_output_0", "/Reshape_99_output_0", "/input_layernorm_11/Mul_1_output_0", "/FakeLinear_44_output_0", "/Shape_154_output_0", "Shape4390", "Rank4392", "BinaryOp4394", "Unsqueeze4395", "BinaryOp4397", "Unsqueeze4398", "StridedSlice4401", "Squeeze4402", "BinaryOp4404", "/Gather_168_output_0", "/Unsqueeze_352_output_0", "Unsqueeze4412", "Unsqueeze4415", "StridedSlice4418", "Squeeze4419", "BinaryOp4420", "BinaryOp4421", "/Gather_169_output_0", "/Unsqueeze_353_output_0", "/Concat_143_output_0", "/Reshape_100_output_0", "/q_norm_11/Cast_output_0", "/q_norm_11/Mul_1_output_0", "/Mul_88_output_0", "/Shape_156_output_0", "Shape4540", "Rank4542", "BinaryOp4544", "Unsqueeze4545", "BinaryOp4547", "Unsqueeze4548", "StridedSlice4551", "Squeeze4552", "BinaryOp4553", "BinaryOp4554", "/Gather_170_output_0", "/Div_33_output_0", "/Unsqueeze_359_output_0", "/Slice_45_output_0", "/Neg_22_output_0", "/Unsqueeze_358_output_0", "/Slice_44_output_0", "/Concat_146_output_0", "/Mul_89_output_0", "/Add_55_output_0", "/FakeLinear_45_output_0", "/Unsqueeze_354_output_0", "/Unsqueeze_355_output_0", "/Concat_144_output_0", "/Reshape_101_output_0", "/k_norm_11/Cast_output_0", "/k_norm_11/Mul_1_output_0", "/Mul_90_output_0", "/Shape_157_output_0", "Shape4436", "Rank4438", "BinaryOp4440", "Unsqueeze4441", "BinaryOp4443", "Unsqueeze4444", "StridedSlice4447", "Squeeze4448", "BinaryOp4449", "BinaryOp4450", "/Gather_171_output_0", "/Div_34_output_0", "/Unsqueeze_361_output_0", "/Slice_47_output_0", "/Neg_23_output_0", "/Unsqueeze_360_output_0", "/Slice_46_output_0", "/Concat_147_output_0", "/Mul_91_output_0", "/Add_56_output_0", "/FakeLinear_46_output_0", "/Unsqueeze_356_output_0", "/Unsqueeze_357_output_0", "/Concat_145_output_0", "/Reshape_102_output_0", "/Reshape_107_output_0", "/FakeLinear_47_output_0", "/Add_58_output_0", "/post_attention_layernorm_11/Mul_1_output_0", "/mlp/gate_proj_11/FakeLinear_output_0", "/mlp/act_fn_11/Mul_output_0", "/mlp/up_proj_11/FakeLinear_output_0", "/mlp_11/Mul_output_0", "/mlp/down_proj_11/FakeLinear_output_0", "/Add_59_output_0", "/Reshape_108_output_0", "/input_layernorm_12/Mul_1_output_0", "/FakeLinear_48_output_0", "/Shape_168_output_0", "Shape4779", "Rank4781", "BinaryOp4783", "Unsqueeze4784", "BinaryOp4786", "Unsqueeze4787", "StridedSlice4790", "Squeeze4791", "BinaryOp4793", "/Gather_183_output_0", "/Unsqueeze_384_output_0", "Unsqueeze4801", "Unsqueeze4804", "StridedSlice4807", "Squeeze4808", "BinaryOp4809", "BinaryOp4810", "/Gather_184_output_0", "/Unsqueeze_385_output_0", "/Concat_156_output_0", "/Reshape_109_output_0", "/q_norm_12/Cast_output_0", "/q_norm_12/Mul_1_output_0", "/Mul_96_output_0", "/Shape_170_output_0", "Shape4929", "Rank4931", "BinaryOp4933", "Unsqueeze4934", "BinaryOp4936", "Unsqueeze4937", "StridedSlice4940", "Squeeze4941", "BinaryOp4942", "BinaryOp4943", "/Gather_185_output_0", "/Div_36_output_0", "/Unsqueeze_391_output_0", "/Slice_49_output_0", "/Neg_24_output_0", "/Unsqueeze_390_output_0", "/Slice_48_output_0", "/Concat_159_output_0", "/Mul_97_output_0", "/Add_60_output_0", "/FakeLinear_49_output_0", "/Unsqueeze_386_output_0", "/Unsqueeze_387_output_0", "/Concat_157_output_0", "/Reshape_110_output_0", "/k_norm_12/Cast_output_0", "/k_norm_12/Mul_1_output_0", "/Mul_98_output_0", "/Shape_171_output_0", "Shape4825", "Rank4827", "BinaryOp4829", "Unsqueeze4830", "BinaryOp4832", "Unsqueeze4833", "StridedSlice4836", "Squeeze4837", "BinaryOp4838", "BinaryOp4839", "/Gather_186_output_0", "/Div_37_output_0", "/Unsqueeze_393_output_0", "/Slice_51_output_0", "/Neg_25_output_0", "/Unsqueeze_392_output_0", "/Slice_50_output_0", "/Concat_160_output_0", "/Mul_99_output_0", "/Add_61_output_0", "/FakeLinear_50_output_0", "/Unsqueeze_388_output_0", "/Unsqueeze_389_output_0", "/Concat_158_output_0", "/Reshape_111_output_0", "/Reshape_116_output_0", "/FakeLinear_51_output_0", "/Add_63_output_0", "/post_attention_layernorm_12/Mul_1_output_0", "/mlp/gate_proj_12/FakeLinear_output_0", "/mlp/act_fn_12/Mul_output_0", "/mlp/up_proj_12/FakeLinear_output_0", "/mlp_12/Mul_output_0", "/mlp/down_proj_12/FakeLinear_output_0", "/Add_64_output_0", "/Reshape_117_output_0", "/input_layernorm_13/Mul_1_output_0", "/FakeLinear_52_output_0", "/Shape_182_output_0", "Shape5168", "Rank5170", "BinaryOp5172", "Unsqueeze5173", "BinaryOp5175", "Unsqueeze5176", "StridedSlice5179", "Squeeze5180", "BinaryOp5182", "/Gather_198_output_0", "/Unsqueeze_416_output_0", "Unsqueeze5190", "Unsqueeze5193", "StridedSlice5196", "Squeeze5197", "BinaryOp5198", "BinaryOp5199", "/Gather_199_output_0", "/Unsqueeze_417_output_0", "/Concat_169_output_0", "/Reshape_118_output_0", "/q_norm_13/Cast_output_0", "/q_norm_13/Mul_1_output_0", "/Mul_104_output_0", "/Shape_184_output_0", "Shape5318", "Rank5320", "BinaryOp5322", "Unsqueeze5323", "BinaryOp5325", "Unsqueeze5326", "StridedSlice5329", "Squeeze5330", "BinaryOp5331", "BinaryOp5332", "/Gather_200_output_0", "/Div_39_output_0", "/Unsqueeze_423_output_0", "/Slice_53_output_0", "/Neg_26_output_0", "/Unsqueeze_422_output_0", "/Slice_52_output_0", "/Concat_172_output_0", "/Mul_105_output_0", "/Add_65_output_0", "/FakeLinear_53_output_0", "/Unsqueeze_418_output_0", "/Unsqueeze_419_output_0", "/Concat_170_output_0", "/Reshape_119_output_0", "/k_norm_13/Cast_output_0", "/k_norm_13/Mul_1_output_0", "/Mul_106_output_0", "/Shape_185_output_0", "Shape5214", "Rank5216", "BinaryOp5218", "Unsqueeze5219", "BinaryOp5221", "Unsqueeze5222", "StridedSlice5225", "Squeeze5226", "BinaryOp5227", "BinaryOp5228", "/Gather_201_output_0", "/Div_40_output_0", "/Unsqueeze_425_output_0", "/Slice_55_output_0", "/Neg_27_output_0", "/Unsqueeze_424_output_0", "/Slice_54_output_0", "/Concat_173_output_0", "/Mul_107_output_0", "/Add_66_output_0", "/FakeLinear_54_output_0", "/Unsqueeze_420_output_0", "/Unsqueeze_421_output_0", "/Concat_171_output_0", "/Reshape_120_output_0", "/Reshape_125_output_0", "/FakeLinear_55_output_0", "/Add_68_output_0", "/post_attention_layernorm_13/Mul_1_output_0", "/mlp/gate_proj_13/FakeLinear_output_0", "/mlp/act_fn_13/Mul_output_0", "/mlp/up_proj_13/FakeLinear_output_0", "/mlp_13/Mul_output_0", "/mlp/down_proj_13/FakeLinear_output_0", "/Add_69_output_0", "/Reshape_126_output_0", "/input_layernorm_14/Mul_1_output_0", "/FakeLinear_56_output_0", "/Shape_196_output_0", "Shape5557", "Rank5559", "BinaryOp5561", "Unsqueeze5562", "BinaryOp5564", "Unsqueeze5565", "StridedSlice5568", "Squeeze5569", "BinaryOp5571", "/Gather_213_output_0", "/Unsqueeze_448_output_0", "Unsqueeze5579", "Unsqueeze5582", "StridedSlice5585", "Squeeze5586", "BinaryOp5587", "BinaryOp5588", "/Gather_214_output_0", "/Unsqueeze_449_output_0", "/Concat_182_output_0", "/Reshape_127_output_0", "/q_norm_14/Cast_output_0", "/q_norm_14/Mul_1_output_0", "/Mul_112_output_0", "/Shape_198_output_0", "Shape5707", "Rank5709", "BinaryOp5711", "Unsqueeze5712", "BinaryOp5714", "Unsqueeze5715", "StridedSlice5718", "Squeeze5719", "BinaryOp5720", "BinaryOp5721", "/Gather_215_output_0", "/Div_42_output_0", "/Unsqueeze_455_output_0", "/Slice_57_output_0", "/Neg_28_output_0", "/Unsqueeze_454_output_0", "/Slice_56_output_0", "/Concat_185_output_0", "/Mul_113_output_0", "/Add_70_output_0", "/FakeLinear_57_output_0", "/Unsqueeze_450_output_0", "/Unsqueeze_451_output_0", "/Concat_183_output_0", "/Reshape_128_output_0", "/k_norm_14/Cast_output_0", "/k_norm_14/Mul_1_output_0", "/Mul_114_output_0", "/Shape_199_output_0", "Shape5603", "Rank5605", "BinaryOp5607", "Unsqueeze5608", "BinaryOp5610", "Unsqueeze5611", "StridedSlice5614", "Squeeze5615", "BinaryOp5616", "BinaryOp5617", "/Gather_216_output_0", "/Div_43_output_0", "/Unsqueeze_457_output_0", "/Slice_59_output_0", "/Neg_29_output_0", "/Unsqueeze_456_output_0", "/Slice_58_output_0", "/Concat_186_output_0", "/Mul_115_output_0", "/Add_71_output_0", "/FakeLinear_58_output_0", "/Unsqueeze_452_output_0", "/Unsqueeze_453_output_0", "/Concat_184_output_0", "/Reshape_129_output_0", "/Reshape_134_output_0", "/FakeLinear_59_output_0", "/Add_73_output_0", "/post_attention_layernorm_14/Mul_1_output_0", "/mlp/gate_proj_14/FakeLinear_output_0", "/mlp/act_fn_14/Mul_output_0", "/mlp/up_proj_14/FakeLinear_output_0", "/mlp_14/Mul_output_0", "/mlp/down_proj_14/FakeLinear_output_0", "/Add_74_output_0", "/Reshape_135_output_0", "/input_layernorm_15/Mul_1_output_0", "/FakeLinear_60_output_0", "/Shape_210_output_0", "Shape5946", "Rank5948", "BinaryOp5950", "Unsqueeze5951", "BinaryOp5953", "Unsqueeze5954", "StridedSlice5957", "Squeeze5958", "BinaryOp5960", "/Gather_228_output_0", "/Unsqueeze_480_output_0", "Unsqueeze5968", "Unsqueeze5971", "StridedSlice5974", "Squeeze5975", "BinaryOp5976", "BinaryOp5977", "/Gather_229_output_0", "/Unsqueeze_481_output_0", "/Concat_195_output_0", "/Reshape_136_output_0", "/q_norm_15/Cast_output_0", "/q_norm_15/Mul_1_output_0", "/Mul_120_output_0", "/Shape_212_output_0", "Shape6096", "Rank6098", "BinaryOp6100", "Unsqueeze6101", "BinaryOp6103", "Unsqueeze6104", "StridedSlice6107", "Squeeze6108", "BinaryOp6109", "BinaryOp6110", "/Gather_230_output_0", "/Div_45_output_0", "/Unsqueeze_487_output_0", "/Slice_61_output_0", "/Neg_30_output_0", "/Unsqueeze_486_output_0", "/Slice_60_output_0", "/Concat_198_output_0", "/Mul_121_output_0", "/Add_75_output_0", "/FakeLinear_61_output_0", "/Unsqueeze_482_output_0", "/Unsqueeze_483_output_0", "/Concat_196_output_0", "/Reshape_137_output_0", "/k_norm_15/Cast_output_0", "/k_norm_15/Mul_1_output_0", "/Mul_122_output_0", "/Shape_213_output_0", "Shape5992", "Rank5994", "BinaryOp5996", "Unsqueeze5997", "BinaryOp5999", "Unsqueeze6000", "StridedSlice6003", "Squeeze6004", "BinaryOp6005", "BinaryOp6006", "/Gather_231_output_0", "/Div_46_output_0", "/Unsqueeze_489_output_0", "/Slice_63_output_0", "/Neg_31_output_0", "/Unsqueeze_488_output_0", "/Slice_62_output_0", "/Concat_199_output_0", "/Mul_123_output_0", "/Add_76_output_0", "/FakeLinear_62_output_0", "/Unsqueeze_484_output_0", "/Unsqueeze_485_output_0", "/Concat_197_output_0", "/Reshape_138_output_0", "/Reshape_143_output_0", "/FakeLinear_63_output_0", "/Add_78_output_0", "/post_attention_layernorm_15/Mul_1_output_0", "/mlp/gate_proj_15/FakeLinear_output_0", "/mlp/act_fn_15/Mul_output_0", "/mlp/up_proj_15/FakeLinear_output_0", "/mlp_15/Mul_output_0", "/mlp/down_proj_15/FakeLinear_output_0", "/Add_79_output_0", "/Reshape_144_output_0", "/input_layernorm_16/Mul_1_output_0", "/FakeLinear_64_output_0", "/Shape_224_output_0", "Shape6335", "Rank6337", "BinaryOp6339", "Unsqueeze6340", "BinaryOp6342", "Unsqueeze6343", "StridedSlice6346", "Squeeze6347", "BinaryOp6349", "/Gather_243_output_0", "/Unsqueeze_512_output_0", "Unsqueeze6357", "Unsqueeze6360", "StridedSlice6363", "Squeeze6364", "BinaryOp6365", "BinaryOp6366", "/Gather_244_output_0", "/Unsqueeze_513_output_0", "/Concat_208_output_0", "/Reshape_145_output_0", "/q_norm_16/Cast_output_0", "/q_norm_16/Mul_1_output_0", "/Mul_128_output_0", "/Shape_226_output_0", "Shape6485", "Rank6487", "BinaryOp6489", "Unsqueeze6490", "BinaryOp6492", "Unsqueeze6493", "StridedSlice6496", "Squeeze6497", "BinaryOp6498", "BinaryOp6499", "/Gather_245_output_0", "/Div_48_output_0", "/Unsqueeze_519_output_0", "/Slice_65_output_0", "/Neg_32_output_0", "/Unsqueeze_518_output_0", "/Slice_64_output_0", "/Concat_211_output_0", "/Mul_129_output_0", "/Add_80_output_0", "/FakeLinear_65_output_0", "/Unsqueeze_514_output_0", "/Unsqueeze_515_output_0", "/Concat_209_output_0", "/Reshape_146_output_0", "/k_norm_16/Cast_output_0", "/k_norm_16/Mul_1_output_0", "/Mul_130_output_0", "/Shape_227_output_0", "Shape6381", "Rank6383", "BinaryOp6385", "Unsqueeze6386", "BinaryOp6388", "Unsqueeze6389", "StridedSlice6392", "Squeeze6393", "BinaryOp6394", "BinaryOp6395", "/Gather_246_output_0", "/Div_49_output_0", "/Unsqueeze_521_output_0", "/Slice_67_output_0", "/Neg_33_output_0", "/Unsqueeze_520_output_0", "/Slice_66_output_0", "/Concat_212_output_0", "/Mul_131_output_0", "/Add_81_output_0", "/FakeLinear_66_output_0", "/Unsqueeze_516_output_0", "/Unsqueeze_517_output_0", "/Concat_210_output_0", "/Reshape_147_output_0", "/Reshape_152_output_0", "/FakeLinear_67_output_0", "/Add_83_output_0", "/post_attention_layernorm_16/Mul_1_output_0", "/mlp/gate_proj_16/FakeLinear_output_0", "/mlp/act_fn_16/Mul_output_0", "/mlp/up_proj_16/FakeLinear_output_0", "/mlp_16/Mul_output_0", "/mlp/down_proj_16/FakeLinear_output_0", "/Add_84_output_0", "/Reshape_153_output_0", "/input_layernorm_17/Mul_1_output_0", "/FakeLinear_68_output_0", "/Shape_238_output_0", "Shape6724", "Rank6726", "BinaryOp6728", "Unsqueeze6729", "BinaryOp6731", "Unsqueeze6732", "StridedSlice6735", "Squeeze6736", "BinaryOp6738", "/Gather_258_output_0", "/Unsqueeze_544_output_0", "Unsqueeze6746", "Unsqueeze6749", "StridedSlice6752", "Squeeze6753", "BinaryOp6754", "BinaryOp6755", "/Gather_259_output_0", "/Unsqueeze_545_output_0", "/Concat_221_output_0", "/Reshape_154_output_0", "/q_norm_17/Cast_output_0", "/q_norm_17/Mul_1_output_0", "/Mul_136_output_0", "/Shape_240_output_0", "Shape6874", "Rank6876", "BinaryOp6878", "Unsqueeze6879", "BinaryOp6881", "Unsqueeze6882", "StridedSlice6885", "Squeeze6886", "BinaryOp6887", "BinaryOp6888", "/Gather_260_output_0", "/Div_51_output_0", "/Unsqueeze_551_output_0", "/Slice_69_output_0", "/Neg_34_output_0", "/Unsqueeze_550_output_0", "/Slice_68_output_0", "/Concat_224_output_0", "/Mul_137_output_0", "/Add_85_output_0", "/FakeLinear_69_output_0", "/Unsqueeze_546_output_0", "/Unsqueeze_547_output_0", "/Concat_222_output_0", "/Reshape_155_output_0", "/k_norm_17/Cast_output_0", "/k_norm_17/Mul_1_output_0", "/Mul_138_output_0", "/Shape_241_output_0", "Shape6770", "Rank6772", "BinaryOp6774", "Unsqueeze6775", "BinaryOp6777", "Unsqueeze6778", "StridedSlice6781", "Squeeze6782", "BinaryOp6783", "BinaryOp6784", "/Gather_261_output_0", "/Div_52_output_0", "/Unsqueeze_553_output_0", "/Slice_71_output_0", "/Neg_35_output_0", "/Unsqueeze_552_output_0", "/Slice_70_output_0", "/Concat_225_output_0", "/Mul_139_output_0", "/Add_86_output_0", "/FakeLinear_70_output_0", "/Unsqueeze_548_output_0", "/Unsqueeze_549_output_0", "/Concat_223_output_0", "/Reshape_156_output_0", "/Reshape_161_output_0", "/FakeLinear_71_output_0", "/Add_88_output_0", "/post_attention_layernorm_17/Mul_1_output_0", "/mlp/gate_proj_17/FakeLinear_output_0", "/mlp/act_fn_17/Mul_output_0", "/mlp/up_proj_17/FakeLinear_output_0", "/mlp_17/Mul_output_0", "/mlp/down_proj_17/FakeLinear_output_0", "/Add_89_output_0", "/Reshape_162_output_0", "/input_layernorm_18/Mul_1_output_0", "/FakeLinear_72_output_0", "/Shape_252_output_0", "Shape7113", "Rank7115", "BinaryOp7117", "Unsqueeze7118", "BinaryOp7120", "Unsqueeze7121", "StridedSlice7124", "Squeeze7125", "BinaryOp7127", "/Gather_273_output_0", "/Unsqueeze_576_output_0", "Unsqueeze7135", "Unsqueeze7138", "StridedSlice7141", "Squeeze7142", "BinaryOp7143", "BinaryOp7144", "/Gather_274_output_0", "/Unsqueeze_577_output_0", "/Concat_234_output_0", "/Reshape_163_output_0", "/q_norm_18/Cast_output_0", "/q_norm_18/Mul_1_output_0", "/Mul_144_output_0", "/Shape_254_output_0", "Shape7263", "Rank7265", "BinaryOp7267", "Unsqueeze7268", "BinaryOp7270", "Unsqueeze7271", "StridedSlice7274", "Squeeze7275", "BinaryOp7276", "BinaryOp7277", "/Gather_275_output_0", "/Div_54_output_0", "/Unsqueeze_583_output_0", "/Slice_73_output_0", "/Neg_36_output_0", "/Unsqueeze_582_output_0", "/Slice_72_output_0", "/Concat_237_output_0", "/Mul_145_output_0", "/Add_90_output_0", "/FakeLinear_73_output_0", "/Unsqueeze_578_output_0", "/Unsqueeze_579_output_0", "/Concat_235_output_0", "/Reshape_164_output_0", "/k_norm_18/Cast_output_0", "/k_norm_18/Mul_1_output_0", "/Mul_146_output_0", "/Shape_255_output_0", "Shape7159", "Rank7161", "BinaryOp7163", "Unsqueeze7164", "BinaryOp7166", "Unsqueeze7167", "StridedSlice7170", "Squeeze7171", "BinaryOp7172", "BinaryOp7173", "/Gather_276_output_0", "/Div_55_output_0", "/Unsqueeze_585_output_0", "/Slice_75_output_0", "/Neg_37_output_0", "/Unsqueeze_584_output_0", "/Slice_74_output_0", "/Concat_238_output_0", "/Mul_147_output_0", "/Add_91_output_0", "/FakeLinear_74_output_0", "/Unsqueeze_580_output_0", "/Unsqueeze_581_output_0", "/Concat_236_output_0", "/Reshape_165_output_0", "/Reshape_170_output_0", "/FakeLinear_75_output_0", "/Add_93_output_0", "/post_attention_layernorm_18/Mul_1_output_0", "/mlp/gate_proj_18/FakeLinear_output_0", "/mlp/act_fn_18/Mul_output_0", "/mlp/up_proj_18/FakeLinear_output_0", "/mlp_18/Mul_output_0", "/mlp/down_proj_18/FakeLinear_output_0", "/Add_94_output_0", "/Reshape_171_output_0", "/input_layernorm_19/Mul_1_output_0", "/FakeLinear_76_output_0", "/Shape_266_output_0", "Shape7502", "Rank7504", "BinaryOp7506", "Unsqueeze7507", "BinaryOp7509", "Unsqueeze7510", "StridedSlice7513", "Squeeze7514", "BinaryOp7516", "/Gather_288_output_0", "/Unsqueeze_608_output_0", "Unsqueeze7524", "Unsqueeze7527", "StridedSlice7530", "Squeeze7531", "BinaryOp7532", "BinaryOp7533", "/Gather_289_output_0", "/Unsqueeze_609_output_0", "/Concat_247_output_0", "/Reshape_172_output_0", "/q_norm_19/Cast_output_0", "/q_norm_19/Mul_1_output_0", "/Mul_152_output_0", "/Shape_268_output_0", "Shape7652", "Rank7654", "BinaryOp7656", "Unsqueeze7657", "BinaryOp7659", "Unsqueeze7660", "StridedSlice7663", "Squeeze7664", "BinaryOp7665", "BinaryOp7666", "/Gather_290_output_0", "/Div_57_output_0", "/Unsqueeze_615_output_0", "/Slice_77_output_0", "/Neg_38_output_0", "/Unsqueeze_614_output_0", "/Slice_76_output_0", "/Concat_250_output_0", "/Mul_153_output_0", "/Add_95_output_0", "/FakeLinear_77_output_0", "/Unsqueeze_610_output_0", "/Unsqueeze_611_output_0", "/Concat_248_output_0", "/Reshape_173_output_0", "/k_norm_19/Cast_output_0", "/k_norm_19/Mul_1_output_0", "/Mul_154_output_0", "/Shape_269_output_0", "Shape7548", "Rank7550", "BinaryOp7552", "Unsqueeze7553", "BinaryOp7555", "Unsqueeze7556", "StridedSlice7559", "Squeeze7560", "BinaryOp7561", "BinaryOp7562", "/Gather_291_output_0", "/Div_58_output_0", "/Unsqueeze_617_output_0", "/Slice_79_output_0", "/Neg_39_output_0", "/Unsqueeze_616_output_0", "/Slice_78_output_0", "/Concat_251_output_0", "/Mul_155_output_0", "/Add_96_output_0", "/FakeLinear_78_output_0", "/Unsqueeze_612_output_0", "/Unsqueeze_613_output_0", "/Concat_249_output_0", "/Reshape_174_output_0", "/Reshape_179_output_0", "/FakeLinear_79_output_0", "/Add_98_output_0", "/post_attention_layernorm_19/Mul_1_output_0", "/mlp/gate_proj_19/FakeLinear_output_0", "/mlp/act_fn_19/Mul_output_0", "/mlp/up_proj_19/FakeLinear_output_0", "/mlp_19/Mul_output_0", "/mlp/down_proj_19/FakeLinear_output_0", "/Add_99_output_0", "/Reshape_180_output_0", "/input_layernorm_20/Mul_1_output_0", "/FakeLinear_80_output_0", "/Shape_280_output_0", "Shape7891", "Rank7893", "BinaryOp7895", "Unsqueeze7896", "BinaryOp7898", "Unsqueeze7899", "StridedSlice7902", "Squeeze7903", "BinaryOp7905", "/Gather_303_output_0", "/Unsqueeze_640_output_0", "Unsqueeze7913", "Unsqueeze7916", "StridedSlice7919", "Squeeze7920", "BinaryOp7921", "BinaryOp7922", "/Gather_304_output_0", "/Unsqueeze_641_output_0", "/Concat_260_output_0", "/Reshape_181_output_0", "/q_norm_20/Cast_output_0", "/q_norm_20/Mul_1_output_0", "/Mul_160_output_0", "/Shape_282_output_0", "Shape8041", "Rank8043", "BinaryOp8045", "Unsqueeze8046", "BinaryOp8048", "Unsqueeze8049", "StridedSlice8052", "Squeeze8053", "BinaryOp8054", "BinaryOp8055", "/Gather_305_output_0", "/Div_60_output_0", "/Unsqueeze_647_output_0", "/Slice_81_output_0", "/Neg_40_output_0", "/Unsqueeze_646_output_0", "/Slice_80_output_0", "/Concat_263_output_0", "/Mul_161_output_0", "/Add_100_output_0", "/FakeLinear_81_output_0", "/Unsqueeze_642_output_0", "/Unsqueeze_643_output_0", "/Concat_261_output_0", "/Reshape_182_output_0", "/k_norm_20/Cast_output_0", "/k_norm_20/Mul_1_output_0", "/Mul_162_output_0", "/Shape_283_output_0", "Shape7937", "Rank7939", "BinaryOp7941", "Unsqueeze7942", "BinaryOp7944", "Unsqueeze7945", "StridedSlice7948", "Squeeze7949", "BinaryOp7950", "BinaryOp7951", "/Gather_306_output_0", "/Div_61_output_0", "/Unsqueeze_649_output_0", "/Slice_83_output_0", "/Neg_41_output_0", "/Unsqueeze_648_output_0", "/Slice_82_output_0", "/Concat_264_output_0", "/Mul_163_output_0", "/Add_101_output_0", "/FakeLinear_82_output_0", "/Unsqueeze_644_output_0", "/Unsqueeze_645_output_0", "/Concat_262_output_0", "/Reshape_183_output_0", "/Reshape_188_output_0", "/FakeLinear_83_output_0", "/Add_103_output_0", "/post_attention_layernorm_20/Mul_1_output_0", "/mlp/gate_proj_20/FakeLinear_output_0", "/mlp/act_fn_20/Mul_output_0", "/mlp/up_proj_20/FakeLinear_output_0", "/mlp_20/Mul_output_0", "/mlp/down_proj_20/FakeLinear_output_0", "/Add_104_output_0", "/Reshape_189_output_0", "/input_layernorm_21/Mul_1_output_0", "/FakeLinear_84_output_0", "/Shape_294_output_0", "Shape8280", "Rank8282", "BinaryOp8284", "Unsqueeze8285", "BinaryOp8287", "Unsqueeze8288", "StridedSlice8291", "Squeeze8292", "BinaryOp8294", "/Gather_318_output_0", "/Unsqueeze_672_output_0", "Unsqueeze8302", "Unsqueeze8305", "StridedSlice8308", "Squeeze8309", "BinaryOp8310", "BinaryOp8311", "/Gather_319_output_0", "/Unsqueeze_673_output_0", "/Concat_273_output_0", "/Reshape_190_output_0", "/q_norm_21/Cast_output_0", "/q_norm_21/Mul_1_output_0", "/Mul_168_output_0", "/Shape_296_output_0", "Shape8430", "Rank8432", "BinaryOp8434", "Unsqueeze8435", "BinaryOp8437", "Unsqueeze8438", "StridedSlice8441", "Squeeze8442", "BinaryOp8443", "BinaryOp8444", "/Gather_320_output_0", "/Div_63_output_0", "/Unsqueeze_679_output_0", "/Slice_85_output_0", "/Neg_42_output_0", "/Unsqueeze_678_output_0", "/Slice_84_output_0", "/Concat_276_output_0", "/Mul_169_output_0", "/Add_105_output_0", "/FakeLinear_85_output_0", "/Unsqueeze_674_output_0", "/Unsqueeze_675_output_0", "/Concat_274_output_0", "/Reshape_191_output_0", "/k_norm_21/Cast_output_0", "/k_norm_21/Mul_1_output_0", "/Mul_170_output_0", "/Shape_297_output_0", "Shape8326", "Rank8328", "BinaryOp8330", "Unsqueeze8331", "BinaryOp8333", "Unsqueeze8334", "StridedSlice8337", "Squeeze8338", "BinaryOp8339", "BinaryOp8340", "/Gather_321_output_0", "/Div_64_output_0", "/Unsqueeze_681_output_0", "/Slice_87_output_0", "/Neg_43_output_0", "/Unsqueeze_680_output_0", "/Slice_86_output_0", "/Concat_277_output_0", "/Mul_171_output_0", "/Add_106_output_0", "/FakeLinear_86_output_0", "/Unsqueeze_676_output_0", "/Unsqueeze_677_output_0", "/Concat_275_output_0", "/Reshape_192_output_0", "/Reshape_197_output_0", "/FakeLinear_87_output_0", "/Add_108_output_0", "/post_attention_layernorm_21/Mul_1_output_0", "/mlp/gate_proj_21/FakeLinear_output_0", "/mlp/act_fn_21/Mul_output_0", "/mlp/up_proj_21/FakeLinear_output_0", "/mlp_21/Mul_output_0", "/mlp/down_proj_21/FakeLinear_output_0", "/Add_109_output_0", "/Reshape_198_output_0", "/input_layernorm_22/Mul_1_output_0", "/FakeLinear_88_output_0", "/Shape_308_output_0", "Shape8669", "Rank8671", "BinaryOp8673", "Unsqueeze8674", "BinaryOp8676", "Unsqueeze8677", "StridedSlice8680", "Squeeze8681", "BinaryOp8683", "/Gather_333_output_0", "/Unsqueeze_704_output_0", "Unsqueeze8691", "Unsqueeze8694", "StridedSlice8697", "Squeeze8698", "BinaryOp8699", "BinaryOp8700", "/Gather_334_output_0", "/Unsqueeze_705_output_0", "/Concat_286_output_0", "/Reshape_199_output_0", "/q_norm_22/Cast_output_0", "/q_norm_22/Mul_1_output_0", "/Mul_176_output_0", "/Shape_310_output_0", "Shape8819", "Rank8821", "BinaryOp8823", "Unsqueeze8824", "BinaryOp8826", "Unsqueeze8827", "StridedSlice8830", "Squeeze8831", "BinaryOp8832", "BinaryOp8833", "/Gather_335_output_0", "/Div_66_output_0", "/Unsqueeze_711_output_0", "/Slice_89_output_0", "/Neg_44_output_0", "/Unsqueeze_710_output_0", "/Slice_88_output_0", "/Concat_289_output_0", "/Mul_177_output_0", "/Add_110_output_0", "/FakeLinear_89_output_0", "/Unsqueeze_706_output_0", "/Unsqueeze_707_output_0", "/Concat_287_output_0", "/Reshape_200_output_0", "/k_norm_22/Cast_output_0", "/k_norm_22/Mul_1_output_0", "/Mul_178_output_0", "/Shape_311_output_0", "Shape8715", "Rank8717", "BinaryOp8719", "Unsqueeze8720", "BinaryOp8722", "Unsqueeze8723", "StridedSlice8726", "Squeeze8727", "BinaryOp8728", "BinaryOp8729", "/Gather_336_output_0", "/Div_67_output_0", "/Unsqueeze_713_output_0", "/Slice_91_output_0", "/Neg_45_output_0", "/Unsqueeze_712_output_0", "/Slice_90_output_0", "/Concat_290_output_0", "/Mul_179_output_0", "/Add_111_output_0", "/FakeLinear_90_output_0", "/Unsqueeze_708_output_0", "/Unsqueeze_709_output_0", "/Concat_288_output_0", "/Reshape_201_output_0", "/Reshape_206_output_0", "/FakeLinear_91_output_0", "/Add_113_output_0", "/post_attention_layernorm_22/Mul_1_output_0", "/mlp/gate_proj_22/FakeLinear_output_0", "/mlp/act_fn_22/Mul_output_0", "/mlp/up_proj_22/FakeLinear_output_0", "/mlp_22/Mul_output_0", "/mlp/down_proj_22/FakeLinear_output_0", "/Add_114_output_0", "/Reshape_207_output_0", "/input_layernorm_23/Mul_1_output_0", "/FakeLinear_92_output_0", "/Shape_322_output_0", "Shape9058", "Rank9060", "BinaryOp9062", "Unsqueeze9063", "BinaryOp9065", "Unsqueeze9066", "StridedSlice9069", "Squeeze9070", "BinaryOp9072", "/Gather_348_output_0", "/Unsqueeze_736_output_0", "Unsqueeze9080", "Unsqueeze9083", "StridedSlice9086", "Squeeze9087", "BinaryOp9088", "BinaryOp9089", "/Gather_349_output_0", "/Unsqueeze_737_output_0", "/Concat_299_output_0", "/Reshape_208_output_0", "/q_norm_23/Cast_output_0", "/q_norm_23/Mul_1_output_0", "/Mul_184_output_0", "/Shape_324_output_0", "Shape9208", "Rank9210", "BinaryOp9212", "Unsqueeze9213", "BinaryOp9215", "Unsqueeze9216", "StridedSlice9219", "Squeeze9220", "BinaryOp9221", "BinaryOp9222", "/Gather_350_output_0", "/Div_69_output_0", "/Unsqueeze_743_output_0", "/Slice_93_output_0", "/Neg_46_output_0", "/Unsqueeze_742_output_0", "/Slice_92_output_0", "/Concat_302_output_0", "/Mul_185_output_0", "/Add_115_output_0", "/FakeLinear_93_output_0", "/Unsqueeze_738_output_0", "/Unsqueeze_739_output_0", "/Concat_300_output_0", "/Reshape_209_output_0", "/k_norm_23/Cast_output_0", "/k_norm_23/Mul_1_output_0", "/Mul_186_output_0", "/Shape_325_output_0", "Shape9104", "Rank9106", "BinaryOp9108", "Unsqueeze9109", "BinaryOp9111", "Unsqueeze9112", "StridedSlice9115", "Squeeze9116", "BinaryOp9117", "BinaryOp9118", "/Gather_351_output_0", "/Div_70_output_0", "/Unsqueeze_745_output_0", "/Slice_95_output_0", "/Neg_47_output_0", "/Unsqueeze_744_output_0", "/Slice_94_output_0", "/Concat_303_output_0", "/Mul_187_output_0", "/Add_116_output_0", "/FakeLinear_94_output_0", "/Unsqueeze_740_output_0", "/Unsqueeze_741_output_0", "/Concat_301_output_0", "/Reshape_210_output_0", "/Reshape_215_output_0", "/FakeLinear_95_output_0", "/Add_118_output_0", "/post_attention_layernorm_23/Mul_1_output_0", "/mlp/gate_proj_23/FakeLinear_output_0", "/mlp/act_fn_23/Mul_output_0", "/mlp/up_proj_23/FakeLinear_output_0", "/mlp_23/Mul_output_0", "/mlp/down_proj_23/FakeLinear_output_0", "/Add_119_output_0", "/Reshape_216_output_0", "/input_layernorm_24/Mul_1_output_0", "/FakeLinear_96_output_0", "/Shape_336_output_0", "Shape9447", "Rank9449", "BinaryOp9451", "Unsqueeze9452", "BinaryOp9454", "Unsqueeze9455", "StridedSlice9458", "Squeeze9459", "BinaryOp9461", "/Gather_363_output_0", "/Unsqueeze_768_output_0", "Unsqueeze9469", "Unsqueeze9472", "StridedSlice9475", "Squeeze9476", "BinaryOp9477", "BinaryOp9478", "/Gather_364_output_0", "/Unsqueeze_769_output_0", "/Concat_312_output_0", "/Reshape_217_output_0", "/q_norm_24/Cast_output_0", "/q_norm_24/Mul_1_output_0", "/Mul_192_output_0", "/Shape_338_output_0", "Shape9597", "Rank9599", "BinaryOp9601", "Unsqueeze9602", "BinaryOp9604", "Unsqueeze9605", "StridedSlice9608", "Squeeze9609", "BinaryOp9610", "BinaryOp9611", "/Gather_365_output_0", "/Div_72_output_0", "/Unsqueeze_775_output_0", "/Slice_97_output_0", "/Neg_48_output_0", "/Unsqueeze_774_output_0", "/Slice_96_output_0", "/Concat_315_output_0", "/Mul_193_output_0", "/Add_120_output_0", "/FakeLinear_97_output_0", "/Unsqueeze_770_output_0", "/Unsqueeze_771_output_0", "/Concat_313_output_0", "/Reshape_218_output_0", "/k_norm_24/Cast_output_0", "/k_norm_24/Mul_1_output_0", "/Mul_194_output_0", "/Shape_339_output_0", "Shape9493", "Rank9495", "BinaryOp9497", "Unsqueeze9498", "BinaryOp9500", "Unsqueeze9501", "StridedSlice9504", "Squeeze9505", "BinaryOp9506", "BinaryOp9507", "/Gather_366_output_0", "/Div_73_output_0", "/Unsqueeze_777_output_0", "/Slice_99_output_0", "/Neg_49_output_0", "/Unsqueeze_776_output_0", "/Slice_98_output_0", "/Concat_316_output_0", "/Mul_195_output_0", "/Add_121_output_0", "/FakeLinear_98_output_0", "/Unsqueeze_772_output_0", "/Unsqueeze_773_output_0", "/Concat_314_output_0", "/Reshape_219_output_0", "/Reshape_224_output_0", "/FakeLinear_99_output_0", "/Add_123_output_0", "/post_attention_layernorm_24/Mul_1_output_0", "/mlp/gate_proj_24/FakeLinear_output_0", "/mlp/act_fn_24/Mul_output_0", "/mlp/up_proj_24/FakeLinear_output_0", "/mlp_24/Mul_output_0", "/mlp/down_proj_24/FakeLinear_output_0", "/Add_124_output_0", "/Reshape_225_output_0", "/input_layernorm_25/Mul_1_output_0", "/FakeLinear_100_output_0", "/Shape_350_output_0", "Shape9836", "Rank9838", "BinaryOp9840", "Unsqueeze9841", "BinaryOp9843", "Unsqueeze9844", "StridedSlice9847", "Squeeze9848", "BinaryOp9850", "/Gather_378_output_0", "/Unsqueeze_800_output_0", "Unsqueeze9858", "Unsqueeze9861", "StridedSlice9864", "Squeeze9865", "BinaryOp9866", "BinaryOp9867", "/Gather_379_output_0", "/Unsqueeze_801_output_0", "/Concat_325_output_0", "/Reshape_226_output_0", "/q_norm_25/Cast_output_0", "/q_norm_25/Mul_1_output_0", "/Mul_200_output_0", "/Shape_352_output_0", "Shape9986", "Rank9988", "BinaryOp9990", "Unsqueeze9991", "BinaryOp9993", "Unsqueeze9994", "StridedSlice9997", "Squeeze9998", "BinaryOp9999", "BinaryOp10000", "/Gather_380_output_0", "/Div_75_output_0", "/Unsqueeze_807_output_0", "/Slice_101_output_0", "/Neg_50_output_0", "/Unsqueeze_806_output_0", "/Slice_100_output_0", "/Concat_328_output_0", "/Mul_201_output_0", "/Add_125_output_0", "/FakeLinear_101_output_0", "/Unsqueeze_802_output_0", "/Unsqueeze_803_output_0", "/Concat_326_output_0", "/Reshape_227_output_0", "/k_norm_25/Cast_output_0", "/k_norm_25/Mul_1_output_0", "/Mul_202_output_0", "/Shape_353_output_0", "Shape9882", "Rank9884", "BinaryOp9886", "Unsqueeze9887", "BinaryOp9889", "Unsqueeze9890", "StridedSlice9893", "Squeeze9894", "BinaryOp9895", "BinaryOp9896", "/Gather_381_output_0", "/Div_76_output_0", "/Unsqueeze_809_output_0", "/Slice_103_output_0", "/Neg_51_output_0", "/Unsqueeze_808_output_0", "/Slice_102_output_0", "/Concat_329_output_0", "/Mul_203_output_0", "/Add_126_output_0", "/FakeLinear_102_output_0", "/Unsqueeze_804_output_0", "/Unsqueeze_805_output_0", "/Concat_327_output_0", "/Reshape_228_output_0", "/Reshape_233_output_0", "/FakeLinear_103_output_0", "/Add_128_output_0", "/post_attention_layernorm_25/Mul_1_output_0", "/mlp/gate_proj_25/FakeLinear_output_0", "/mlp/act_fn_25/Mul_output_0", "/mlp/up_proj_25/FakeLinear_output_0", "/mlp_25/Mul_output_0", "/mlp/down_proj_25/FakeLinear_output_0", "/Add_129_output_0", "/Reshape_234_output_0", "/input_layernorm_26/Mul_1_output_0", "/FakeLinear_104_output_0", "/Shape_364_output_0", "Shape10225", "Rank10227", "BinaryOp10229", "Unsqueeze10230", "BinaryOp10232", "Unsqueeze10233", "StridedSlice10236", "Squeeze10237", "BinaryOp10239", "/Gather_393_output_0", "/Unsqueeze_832_output_0", "Unsqueeze10247", "Unsqueeze10250", "StridedSlice10253", "Squeeze10254", "BinaryOp10255", "BinaryOp10256", "/Gather_394_output_0", "/Unsqueeze_833_output_0", "/Concat_338_output_0", "/Reshape_235_output_0", "/q_norm_26/Cast_output_0", "/q_norm_26/Mul_1_output_0", "/Mul_208_output_0", "/Shape_366_output_0", "Shape10375", "Rank10377", "BinaryOp10379", "Unsqueeze10380", "BinaryOp10382", "Unsqueeze10383", "StridedSlice10386", "Squeeze10387", "BinaryOp10388", "BinaryOp10389", "/Gather_395_output_0", "/Div_78_output_0", "/Unsqueeze_839_output_0", "/Slice_105_output_0", "/Neg_52_output_0", "/Unsqueeze_838_output_0", "/Slice_104_output_0", "/Concat_341_output_0", "/Mul_209_output_0", "/Add_130_output_0", "/FakeLinear_105_output_0", "/Unsqueeze_834_output_0", "/Unsqueeze_835_output_0", "/Concat_339_output_0", "/Reshape_236_output_0", "/k_norm_26/Cast_output_0", "/k_norm_26/Mul_1_output_0", "/Mul_210_output_0", "/Shape_367_output_0", "Shape10271", "Rank10273", "BinaryOp10275", "Unsqueeze10276", "BinaryOp10278", "Unsqueeze10279", "StridedSlice10282", "Squeeze10283", "BinaryOp10284", "BinaryOp10285", "/Gather_396_output_0", "/Div_79_output_0", "/Unsqueeze_841_output_0", "/Slice_107_output_0", "/Neg_53_output_0", "/Unsqueeze_840_output_0", "/Slice_106_output_0", "/Concat_342_output_0", "/Mul_211_output_0", "/Add_131_output_0", "/FakeLinear_106_output_0", "/Unsqueeze_836_output_0", "/Unsqueeze_837_output_0", "/Concat_340_output_0", "/Reshape_237_output_0", "/Reshape_242_output_0", "/FakeLinear_107_output_0", "/Add_133_output_0", "/post_attention_layernorm_26/Mul_1_output_0", "/mlp/gate_proj_26/FakeLinear_output_0", "/mlp/act_fn_26/Mul_output_0", "/mlp/up_proj_26/FakeLinear_output_0", "/mlp_26/Mul_output_0", "/mlp/down_proj_26/FakeLinear_output_0", "/Add_134_output_0", "/Reshape_243_output_0", "/input_layernorm_27/Mul_1_output_0", "/FakeLinear_108_output_0", "/Shape_378_output_0", "Shape10614", "Rank10616", "BinaryOp10618", "Unsqueeze10619", "BinaryOp10621", "Unsqueeze10622", "StridedSlice10625", "Squeeze10626", "BinaryOp10628", "/Gather_408_output_0", "/Unsqueeze_864_output_0", "Unsqueeze10636", "Unsqueeze10639", "StridedSlice10642", "Squeeze10643", "BinaryOp10644", "BinaryOp10645", "/Gather_409_output_0", "/Unsqueeze_865_output_0", "/Concat_351_output_0", "/Reshape_244_output_0", "/q_norm_27/Cast_output_0", "/q_norm_27/Mul_1_output_0", "/Mul_216_output_0", "/Shape_380_output_0", "Shape10764", "Rank10766", "BinaryOp10768", "Unsqueeze10769", "BinaryOp10771", "Unsqueeze10772", "StridedSlice10775", "Squeeze10776", "BinaryOp10777", "BinaryOp10778", "/Gather_410_output_0", "/Div_81_output_0", "/Unsqueeze_871_output_0", "/Slice_109_output_0", "/Neg_54_output_0", "/Unsqueeze_870_output_0", "/Slice_108_output_0", "/Concat_354_output_0", "/Mul_217_output_0", "/Add_135_output_0", "/FakeLinear_109_output_0", "/Unsqueeze_866_output_0", "/Unsqueeze_867_output_0", "/Concat_352_output_0", "/Reshape_245_output_0", "/k_norm_27/Cast_output_0", "/k_norm_27/Mul_1_output_0", "/Mul_218_output_0", "/Shape_381_output_0", "Shape10660", "Rank10662", "BinaryOp10664", "Unsqueeze10665", "BinaryOp10667", "Unsqueeze10668", "StridedSlice10671", "Squeeze10672", "BinaryOp10673", "BinaryOp10674", "/Gather_411_output_0", "/Div_82_output_0", "/Unsqueeze_873_output_0", "/Slice_111_output_0", "/Neg_55_output_0", "/Unsqueeze_872_output_0", "/Slice_110_output_0", "/Concat_355_output_0", "/Mul_219_output_0", "/Add_136_output_0", "/FakeLinear_110_output_0", "/Unsqueeze_868_output_0", "/Unsqueeze_869_output_0", "/Concat_353_output_0", "/Reshape_246_output_0", "/Reshape_251_output_0", "/FakeLinear_111_output_0", "/Add_138_output_0", "/post_attention_layernorm_27/Mul_1_output_0", "/mlp/gate_proj_27/FakeLinear_output_0", "/mlp/act_fn_27/Mul_output_0", "/mlp/up_proj_27/FakeLinear_output_0", "/mlp_27/Mul_output_0", "/mlp/down_proj_27/FakeLinear_output_0", "/Add_139_output_0", "/Reshape_252_output_0", "/input_layernorm_28/Mul_1_output_0", "/FakeLinear_112_output_0", "/Shape_392_output_0", "Shape11003", "Rank11005", "BinaryOp11007", "Unsqueeze11008", "BinaryOp11010", "Unsqueeze11011", "StridedSlice11014", "Squeeze11015", "BinaryOp11017", "/Gather_423_output_0", "/Unsqueeze_896_output_0", "Unsqueeze11025", "Unsqueeze11028", "StridedSlice11031", "Squeeze11032", "BinaryOp11033", "BinaryOp11034", "/Gather_424_output_0", "/Unsqueeze_897_output_0", "/Concat_364_output_0", "/Reshape_253_output_0", "/q_norm_28/Cast_output_0", "/q_norm_28/Mul_1_output_0", "/Mul_224_output_0", "/Shape_394_output_0", "Shape11153", "Rank11155", "BinaryOp11157", "Unsqueeze11158", "BinaryOp11160", "Unsqueeze11161", "StridedSlice11164", "Squeeze11165", "BinaryOp11166", "BinaryOp11167", "/Gather_425_output_0", "/Div_84_output_0", "/Unsqueeze_903_output_0", "/Slice_113_output_0", "/Neg_56_output_0", "/Unsqueeze_902_output_0", "/Slice_112_output_0", "/Concat_367_output_0", "/Mul_225_output_0", "/Add_140_output_0", "/FakeLinear_113_output_0", "/Unsqueeze_898_output_0", "/Unsqueeze_899_output_0", "/Concat_365_output_0", "/Reshape_254_output_0", "/k_norm_28/Cast_output_0", "/k_norm_28/Mul_1_output_0", "/Mul_226_output_0", "/Shape_395_output_0", "Shape11049", "Rank11051", "BinaryOp11053", "Unsqueeze11054", "BinaryOp11056", "Unsqueeze11057", "StridedSlice11060", "Squeeze11061", "BinaryOp11062", "BinaryOp11063", "/Gather_426_output_0", "/Div_85_output_0", "/Unsqueeze_905_output_0", "/Slice_115_output_0", "/Neg_57_output_0", "/Unsqueeze_904_output_0", "/Slice_114_output_0", "/Concat_368_output_0", "/Mul_227_output_0", "/Add_141_output_0", "/FakeLinear_114_output_0", "/Unsqueeze_900_output_0", "/Unsqueeze_901_output_0", "/Concat_366_output_0", "/Reshape_255_output_0", "/Reshape_260_output_0", "/FakeLinear_115_output_0", "/Add_143_output_0", "/post_attention_layernorm_28/Mul_1_output_0", "/mlp/gate_proj_28/FakeLinear_output_0", "/mlp/act_fn_28/Mul_output_0", "/mlp/up_proj_28/FakeLinear_output_0", "/mlp_28/Mul_output_0", "/mlp/down_proj_28/FakeLinear_output_0", "/Add_144_output_0", "/Reshape_261_output_0", "/input_layernorm_29/Mul_1_output_0", "/FakeLinear_116_output_0", "/Shape_406_output_0", "Shape11392", "Rank11394", "BinaryOp11396", "Unsqueeze11397", "BinaryOp11399", "Unsqueeze11400", "StridedSlice11403", "Squeeze11404", "BinaryOp11406", "/Gather_438_output_0", "/Unsqueeze_928_output_0", "Unsqueeze11414", "Unsqueeze11417", "StridedSlice11420", "Squeeze11421", "BinaryOp11422", "BinaryOp11423", "/Gather_439_output_0", "/Unsqueeze_929_output_0", "/Concat_377_output_0", "/Reshape_262_output_0", "/q_norm_29/Cast_output_0", "/q_norm_29/Mul_1_output_0", "/Mul_232_output_0", "/Shape_408_output_0", "Shape11542", "Rank11544", "BinaryOp11546", "Unsqueeze11547", "BinaryOp11549", "Unsqueeze11550", "StridedSlice11553", "Squeeze11554", "BinaryOp11555", "BinaryOp11556", "/Gather_440_output_0", "/Div_87_output_0", "/Unsqueeze_935_output_0", "/Slice_117_output_0", "/Neg_58_output_0", "/Unsqueeze_934_output_0", "/Slice_116_output_0", "/Concat_380_output_0", "/Mul_233_output_0", "/Add_145_output_0", "/FakeLinear_117_output_0", "/Unsqueeze_930_output_0", "/Unsqueeze_931_output_0", "/Concat_378_output_0", "/Reshape_263_output_0", "/k_norm_29/Cast_output_0", "/k_norm_29/Mul_1_output_0", "/Mul_234_output_0", "/Shape_409_output_0", "Shape11438", "Rank11440", "BinaryOp11442", "Unsqueeze11443", "BinaryOp11445", "Unsqueeze11446", "StridedSlice11449", "Squeeze11450", "BinaryOp11451", "BinaryOp11452", "/Gather_441_output_0", "/Div_88_output_0", "/Unsqueeze_937_output_0", "/Slice_119_output_0", "/Neg_59_output_0", "/Unsqueeze_936_output_0", "/Slice_118_output_0", "/Concat_381_output_0", "/Mul_235_output_0", "/Add_146_output_0", "/FakeLinear_118_output_0", "/Unsqueeze_932_output_0", "/Unsqueeze_933_output_0", "/Concat_379_output_0", "/Reshape_264_output_0", "/Reshape_269_output_0", "/FakeLinear_119_output_0", "/Add_148_output_0", "/post_attention_layernorm_29/Mul_1_output_0", "/mlp/gate_proj_29/FakeLinear_output_0", "/mlp/act_fn_29/Mul_output_0", "/mlp/up_proj_29/FakeLinear_output_0", "/mlp_29/Mul_output_0", "/mlp/down_proj_29/FakeLinear_output_0", "/Add_149_output_0", "/Reshape_270_output_0", "/input_layernorm_30/Mul_1_output_0", "/FakeLinear_120_output_0", "/Shape_420_output_0", "Shape11781", "Rank11783", "BinaryOp11785", "Unsqueeze11786", "BinaryOp11788", "Unsqueeze11789", "StridedSlice11792", "Squeeze11793", "BinaryOp11795", "/Gather_453_output_0", "/Unsqueeze_960_output_0", "Unsqueeze11803", "Unsqueeze11806", "StridedSlice11809", "Squeeze11810", "BinaryOp11811", "BinaryOp11812", "/Gather_454_output_0", "/Unsqueeze_961_output_0", "/Concat_390_output_0", "/Reshape_271_output_0", "/q_norm_30/Cast_output_0", "/q_norm_30/Mul_1_output_0", "/Mul_240_output_0", "/Shape_422_output_0", "Shape11931", "Rank11933", "BinaryOp11935", "Unsqueeze11936", "BinaryOp11938", "Unsqueeze11939", "StridedSlice11942", "Squeeze11943", "BinaryOp11944", "BinaryOp11945", "/Gather_455_output_0", "/Div_90_output_0", "/Unsqueeze_967_output_0", "/Slice_121_output_0", "/Neg_60_output_0", "/Unsqueeze_966_output_0", "/Slice_120_output_0", "/Concat_393_output_0", "/Mul_241_output_0", "/Add_150_output_0", "/FakeLinear_121_output_0", "/Unsqueeze_962_output_0", "/Unsqueeze_963_output_0", "/Concat_391_output_0", "/Reshape_272_output_0", "/k_norm_30/Cast_output_0", "/k_norm_30/Mul_1_output_0", "/Mul_242_output_0", "/Shape_423_output_0", "Shape11827", "Rank11829", "BinaryOp11831", "Unsqueeze11832", "BinaryOp11834", "Unsqueeze11835", "StridedSlice11838", "Squeeze11839", "BinaryOp11840", "BinaryOp11841", "/Gather_456_output_0", "/Div_91_output_0", "/Unsqueeze_969_output_0", "/Slice_123_output_0", "/Neg_61_output_0", "/Unsqueeze_968_output_0", "/Slice_122_output_0", "/Concat_394_output_0", "/Mul_243_output_0", "/Add_151_output_0", "/FakeLinear_122_output_0", "/Unsqueeze_964_output_0", "/Unsqueeze_965_output_0", "/Concat_392_output_0", "/Reshape_273_output_0", "/Reshape_278_output_0", "/FakeLinear_123_output_0", "/Add_153_output_0", "/post_attention_layernorm_30/Mul_1_output_0", "/mlp/gate_proj_30/FakeLinear_output_0", "/mlp/act_fn_30/Mul_output_0", "/mlp/up_proj_30/FakeLinear_output_0", "/mlp_30/Mul_output_0", "/mlp/down_proj_30/FakeLinear_output_0", "/Add_154_output_0", "/Reshape_279_output_0", "/input_layernorm_31/Mul_1_output_0", "/FakeLinear_124_output_0", "/Shape_434_output_0", "Shape12170", "Rank12172", "BinaryOp12174", "Unsqueeze12175", "BinaryOp12177", "Unsqueeze12178", "StridedSlice12181", "Squeeze12182", "BinaryOp12184", "/Gather_468_output_0", "/Unsqueeze_992_output_0", "Unsqueeze12192", "Unsqueeze12195", "StridedSlice12198", "Squeeze12199", "BinaryOp12200", "BinaryOp12201", "/Gather_469_output_0", "/Unsqueeze_993_output_0", "/Concat_403_output_0", "/Reshape_280_output_0", "/q_norm_31/Cast_output_0", "/q_norm_31/Mul_1_output_0", "/Mul_248_output_0", "/Shape_436_output_0", "Shape12320", "Rank12322", "BinaryOp12324", "Unsqueeze12325", "BinaryOp12327", "Unsqueeze12328", "StridedSlice12331", "Squeeze12332", "BinaryOp12333", "BinaryOp12334", "/Gather_470_output_0", "/Div_93_output_0", "/Unsqueeze_999_output_0", "/Slice_125_output_0", "/Neg_62_output_0", "/Unsqueeze_998_output_0", "/Slice_124_output_0", "/Concat_406_output_0", "/Mul_249_output_0", "/Add_155_output_0", "/FakeLinear_125_output_0", "/Unsqueeze_994_output_0", "/Unsqueeze_995_output_0", "/Concat_404_output_0", "/Reshape_281_output_0", "/k_norm_31/Cast_output_0", "/k_norm_31/Mul_1_output_0", "/Mul_250_output_0", "/Shape_437_output_0", "Shape12216", "Rank12218", "BinaryOp12220", "Unsqueeze12221", "BinaryOp12223", "Unsqueeze12224", "StridedSlice12227", "Squeeze12228", "BinaryOp12229", "BinaryOp12230", "/Gather_471_output_0", "/Div_94_output_0", "/Unsqueeze_1001_output_0", "/Slice_127_output_0", "/Neg_63_output_0", "/Unsqueeze_1000_output_0", "/Slice_126_output_0", "/Concat_407_output_0", "/Mul_251_output_0", "/Add_156_output_0", "/FakeLinear_126_output_0", "/Unsqueeze_996_output_0", "/Unsqueeze_997_output_0", "/Concat_405_output_0", "/Reshape_282_output_0", "/Reshape_287_output_0", "/FakeLinear_127_output_0", "/Add_158_output_0", "/post_attention_layernorm_31/Mul_1_output_0", "/mlp/gate_proj_31/FakeLinear_output_0", "/mlp/act_fn_31/Mul_output_0", "/mlp/up_proj_31/FakeLinear_output_0", "/mlp_31/Mul_output_0", "/mlp/down_proj_31/FakeLinear_output_0", "/Add_159_output_0", "/Reshape_288_output_0", "/input_layernorm_32/Mul_1_output_0", "/FakeLinear_128_output_0", "/Shape_448_output_0", "Shape12559", "Rank12561", "BinaryOp12563", "Unsqueeze12564", "BinaryOp12566", "Unsqueeze12567", "StridedSlice12570", "Squeeze12571", "BinaryOp12573", "/Gather_483_output_0", "/Unsqueeze_1024_output_0", "Unsqueeze12581", "Unsqueeze12584", "StridedSlice12587", "Squeeze12588", "BinaryOp12589", "BinaryOp12590", "/Gather_484_output_0", "/Unsqueeze_1025_output_0", "/Concat_416_output_0", "/Reshape_289_output_0", "/q_norm_32/Cast_output_0", "/q_norm_32/Mul_1_output_0", "/Mul_256_output_0", "/Shape_450_output_0", "Shape12709", "Rank12711", "BinaryOp12713", "Unsqueeze12714", "BinaryOp12716", "Unsqueeze12717", "StridedSlice12720", "Squeeze12721", "BinaryOp12722", "BinaryOp12723", "/Gather_485_output_0", "/Div_96_output_0", "/Unsqueeze_1031_output_0", "/Slice_129_output_0", "/Neg_64_output_0", "/Unsqueeze_1030_output_0", "/Slice_128_output_0", "/Concat_419_output_0", "/Mul_257_output_0", "/Add_160_output_0", "/FakeLinear_129_output_0", "/Unsqueeze_1026_output_0", "/Unsqueeze_1027_output_0", "/Concat_417_output_0", "/Reshape_290_output_0", "/k_norm_32/Cast_output_0", "/k_norm_32/Mul_1_output_0", "/Mul_258_output_0", "/Shape_451_output_0", "Shape12605", "Rank12607", "BinaryOp12609", "Unsqueeze12610", "BinaryOp12612", "Unsqueeze12613", "StridedSlice12616", "Squeeze12617", "BinaryOp12618", "BinaryOp12619", "/Gather_486_output_0", "/Div_97_output_0", "/Unsqueeze_1033_output_0", "/Slice_131_output_0", "/Neg_65_output_0", "/Unsqueeze_1032_output_0", "/Slice_130_output_0", "/Concat_420_output_0", "/Mul_259_output_0", "/Add_161_output_0", "/FakeLinear_130_output_0", "/Unsqueeze_1028_output_0", "/Unsqueeze_1029_output_0", "/Concat_418_output_0", "/Reshape_291_output_0", "/Reshape_296_output_0", "/FakeLinear_131_output_0", "/Add_163_output_0", "/post_attention_layernorm_32/Mul_1_output_0", "/mlp/gate_proj_32/FakeLinear_output_0", "/mlp/act_fn_32/Mul_output_0", "/mlp/up_proj_32/FakeLinear_output_0", "/mlp_32/Mul_output_0", "/mlp/down_proj_32/FakeLinear_output_0", "/Add_164_output_0", "/Reshape_297_output_0", "/input_layernorm_33/Mul_1_output_0", "/FakeLinear_132_output_0", "/Shape_462_output_0", "Shape12948", "Rank12950", "BinaryOp12952", "Unsqueeze12953", "BinaryOp12955", "Unsqueeze12956", "StridedSlice12959", "Squeeze12960", "BinaryOp12962", "/Gather_498_output_0", "/Unsqueeze_1056_output_0", "Unsqueeze12970", "Unsqueeze12973", "StridedSlice12976", "Squeeze12977", "BinaryOp12978", "BinaryOp12979", "/Gather_499_output_0", "/Unsqueeze_1057_output_0", "/Concat_429_output_0", "/Reshape_298_output_0", "/q_norm_33/Cast_output_0", "/q_norm_33/Mul_1_output_0", "/Mul_264_output_0", "/Shape_464_output_0", "Shape13098", "Rank13100", "BinaryOp13102", "Unsqueeze13103", "BinaryOp13105", "Unsqueeze13106", "StridedSlice13109", "Squeeze13110", "BinaryOp13111", "BinaryOp13112", "/Gather_500_output_0", "/Div_99_output_0", "/Unsqueeze_1063_output_0", "/Slice_133_output_0", "/Neg_66_output_0", "/Unsqueeze_1062_output_0", "/Slice_132_output_0", "/Concat_432_output_0", "/Mul_265_output_0", "/Add_165_output_0", "/FakeLinear_133_output_0", "/Unsqueeze_1058_output_0", "/Unsqueeze_1059_output_0", "/Concat_430_output_0", "/Reshape_299_output_0", "/k_norm_33/Cast_output_0", "/k_norm_33/Mul_1_output_0", "/Mul_266_output_0", "/Shape_465_output_0", "Shape12994", "Rank12996", "BinaryOp12998", "Unsqueeze12999", "BinaryOp13001", "Unsqueeze13002", "StridedSlice13005", "Squeeze13006", "BinaryOp13007", "BinaryOp13008", "/Gather_501_output_0", "/Div_100_output_0", "/Unsqueeze_1065_output_0", "/Slice_135_output_0", "/Neg_67_output_0", "/Unsqueeze_1064_output_0", "/Slice_134_output_0", "/Concat_433_output_0", "/Mul_267_output_0", "/Add_166_output_0", "/FakeLinear_134_output_0", "/Unsqueeze_1060_output_0", "/Unsqueeze_1061_output_0", "/Concat_431_output_0", "/Reshape_300_output_0", "/Reshape_305_output_0", "/FakeLinear_135_output_0", "/Add_168_output_0", "/post_attention_layernorm_33/Mul_1_output_0", "/mlp/gate_proj_33/FakeLinear_output_0", "/mlp/act_fn_33/Mul_output_0", "/mlp/up_proj_33/FakeLinear_output_0", "/mlp_33/Mul_output_0", "/mlp/down_proj_33/FakeLinear_output_0", "/Add_169_output_0", "/Reshape_306_output_0", "/input_layernorm_34/Mul_1_output_0", "/FakeLinear_136_output_0", "/Shape_476_output_0", "Shape13337", "Rank13339", "BinaryOp13341", "Unsqueeze13342", "BinaryOp13344", "Unsqueeze13345", "StridedSlice13348", "Squeeze13349", "BinaryOp13351", "/Gather_513_output_0", "/Unsqueeze_1088_output_0", "Unsqueeze13359", "Unsqueeze13362", "StridedSlice13365", "Squeeze13366", "BinaryOp13367", "BinaryOp13368", "/Gather_514_output_0", "/Unsqueeze_1089_output_0", "/Concat_442_output_0", "/Reshape_307_output_0", "/q_norm_34/Cast_output_0", "/q_norm_34/Mul_1_output_0", "/Mul_272_output_0", "/Shape_478_output_0", "Shape13487", "Rank13489", "BinaryOp13491", "Unsqueeze13492", "BinaryOp13494", "Unsqueeze13495", "StridedSlice13498", "Squeeze13499", "BinaryOp13500", "BinaryOp13501", "/Gather_515_output_0", "/Div_102_output_0", "/Unsqueeze_1095_output_0", "/Slice_137_output_0", "/Neg_68_output_0", "/Unsqueeze_1094_output_0", "/Slice_136_output_0", "/Concat_445_output_0", "/Mul_273_output_0", "/Add_170_output_0", "/FakeLinear_137_output_0", "/Unsqueeze_1090_output_0", "/Unsqueeze_1091_output_0", "/Concat_443_output_0", "/Reshape_308_output_0", "/k_norm_34/Cast_output_0", "/k_norm_34/Mul_1_output_0", "/Mul_274_output_0", "/Shape_479_output_0", "Shape13383", "Rank13385", "BinaryOp13387", "Unsqueeze13388", "BinaryOp13390", "Unsqueeze13391", "StridedSlice13394", "Squeeze13395", "BinaryOp13396", "BinaryOp13397", "/Gather_516_output_0", "/Div_103_output_0", "/Unsqueeze_1097_output_0", "/Slice_139_output_0", "/Neg_69_output_0", "/Unsqueeze_1096_output_0", "/Slice_138_output_0", "/Concat_446_output_0", "/Mul_275_output_0", "/Add_171_output_0", "/FakeLinear_138_output_0", "/Unsqueeze_1092_output_0", "/Unsqueeze_1093_output_0", "/Concat_444_output_0", "/Reshape_309_output_0", "/Reshape_314_output_0", "/FakeLinear_139_output_0", "/Add_173_output_0", "/post_attention_layernorm_34/Mul_1_output_0", "/mlp/gate_proj_34/FakeLinear_output_0", "/mlp/act_fn_34/Mul_output_0", "/mlp/up_proj_34/FakeLinear_output_0", "/mlp_34/Mul_output_0", "/mlp/down_proj_34/FakeLinear_output_0", "/Add_174_output_0", "/Reshape_315_output_0", "/input_layernorm_35/Mul_1_output_0", "/FakeLinear_140_output_0", "/Shape_490_output_0", "Shape13726", "Rank13728", "BinaryOp13730", "Unsqueeze13731", "BinaryOp13733", "Unsqueeze13734", "StridedSlice13737", "Squeeze13738", "BinaryOp13740", "/Gather_528_output_0", "/Unsqueeze_1120_output_0", "Unsqueeze13748", "Unsqueeze13751", "StridedSlice13754", "Squeeze13755", "BinaryOp13756", "BinaryOp13757", "/Gather_529_output_0", "/Unsqueeze_1121_output_0", "/Concat_455_output_0", "/Reshape_316_output_0", "/q_norm_35/Cast_output_0", "/q_norm_35/Mul_1_output_0", "/Mul_280_output_0", "/Shape_492_output_0", "Shape13844", "Rank13846", "BinaryOp13848", "Unsqueeze13849", "BinaryOp13851", "Unsqueeze13852", "StridedSlice13855", "Squeeze13856", "BinaryOp13857", "BinaryOp13858", "/Gather_530_output_0", "/Div_105_output_0", "/Unsqueeze_1127_output_0", "/Slice_141_output_0", "/Neg_70_output_0", "/Unsqueeze_1126_output_0", "/Slice_140_output_0", "/Concat_458_output_0", "/Mul_281_output_0", "/Add_175_output_0", "/FakeLinear_141_output_0", "/Unsqueeze_1122_output_0", "/Unsqueeze_1123_output_0", "/Concat_456_output_0", "/Reshape_317_output_0", "/k_norm_35/Cast_output_0", "/k_norm_35/Mul_1_output_0", "/Mul_282_output_0", "/Shape_493_output_0", "Shape13772", "Rank13774", "BinaryOp13776", "Unsqueeze13777", "BinaryOp13779", "Unsqueeze13780", "StridedSlice13783", "Squeeze13784", "BinaryOp13785", "BinaryOp13786", "/Gather_531_output_0", "/Div_106_output_0", "/Unsqueeze_1129_output_0", "/Slice_143_output_0", "/Neg_71_output_0", "/Unsqueeze_1128_output_0", "/Slice_142_output_0", "/Concat_459_output_0", "/Mul_283_output_0", "/Add_176_output_0", "/FakeLinear_142_output_0", "/Unsqueeze_1124_output_0", "/Unsqueeze_1125_output_0", "/Concat_457_output_0", "/Reshape_318_output_0", "/Reshape_323_output_0", "/FakeLinear_143_output_0", "/Add_178_output_0", "/post_attention_layernorm_35/Mul_1_output_0", "/mlp/gate_proj_35/FakeLinear_output_0", "/mlp/act_fn_35/Mul_output_0", "/mlp/up_proj_35/FakeLinear_output_0", "/mlp_35/Mul_output_0", "/mlp/down_proj_35/FakeLinear_output_0", "/Add_179_output_0", "/Slice_144_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": "948922b1-ed8a-4762-bb0f-cba84da5e041" }