{ "bizCode": "MNNTest", "extraInfo": { "version": "3.1.3" }, "oplists": [ { "main_type": "Input", "main": { "dims": [ 1, 1, 4096 ], "dtype": "DT_FLOAT", "dformat": "NCHW" }, "name": "input_ids", "outputIndexes": [ 2 ], "type": "Input", "defaultDimentionFormat": "NHWC" }, { "main_type": "Input", "main": { "dims": [ 1, 1, -1, -1 ], "dtype": "DT_FLOAT", "dformat": "NCHW" }, "name": "attention_mask", "outputIndexes": [ 125 ], "type": "Input", "defaultDimentionFormat": "NHWC" }, { "main_type": "Input", "main": { "dims": [ 1, -1 ], "dtype": "DT_INT32", "dformat": "NCHW" }, "name": "position_ids", "outputIndexes": [ 34 ], "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": [ 3251 ], "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": [ 3 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_output_0", "outputIndexes": [ 4 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm/Mul_1_output_0", "inputIndexes": [ 4 ], "outputIndexes": [ 5 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 0, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5 ], "outputIndexes": [ 3255 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3255 ], "outputIndexes": [ 3256 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear", "inputIndexes": [ 3256 ], "outputIndexes": [ 3257 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 32768, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3257 ], "outputIndexes": [ 3258 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3258 ], "outputIndexes": [ 6 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5 ], "main_type": "NONE", "name": "/Shape_output_0", "outputIndexes": [ 7 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 7 ], "main_type": "NONE", "name": "Shape51", "outputIndexes": [ 8 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 7 ], "main_type": "NONE", "name": "Rank53", "outputIndexes": [ 9 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 9, 9 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp55", "outputIndexes": [ 10 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 10 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze56", "outputIndexes": [ 11 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 1 ] }, "name": "Const9", "outputIndexes": [ 12 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 10, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp58", "outputIndexes": [ 13 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 13 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze59", "outputIndexes": [ 14 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 1 ] }, "name": "Unsqueeze61", "outputIndexes": [ 15 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 8, 11, 14, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice62", "outputIndexes": [ 16 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 16 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze63", "outputIndexes": [ 17 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 17, 17 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp65", "outputIndexes": [ 18 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 7, 18, 10 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_1_output_0", "outputIndexes": [ 19 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 0 ] }, "name": "/rotary/Constant_4_output_0", "outputIndexes": [ 20 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 19, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_output_0", "outputIndexes": [ 21 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 10 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze75", "outputIndexes": [ 22 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 13 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze78", "outputIndexes": [ 23 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 8, 22, 23, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice81", "outputIndexes": [ 24 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 24 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze82", "outputIndexes": [ 25 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 25 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp83", "outputIndexes": [ 26 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 26, 25 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp84", "outputIndexes": [ 27 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 7, 27, 10 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_2_output_0", "outputIndexes": [ 28 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 28, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1_output_0", "outputIndexes": [ 29 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 32 ] }, "name": "/Constant_4_output_0", "outputIndexes": [ 30 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 128 ] }, "name": "/Constant_5_output_0", "outputIndexes": [ 31 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 21, 29, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_output_0", "outputIndexes": [ 32 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 6, 32 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_1_output_0", "outputIndexes": [ 33 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 34 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/rotary/Cast_output_0", "outputIndexes": [ 35 ], "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": [ 36 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 35, 36 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/rotary/Reshape_output_0", "outputIndexes": [ 37 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 64 ], "dataFormat": "NCHW", "dataType": "DT_FLOAT", "float32s": [ 1.0, 0.811481, 0.658502, 0.534362, 0.433624, 0.351878, 0.285542, 0.231712, 0.18803, 0.152583, 0.123818, 0.100476, 0.081534, 0.066164, 0.053691, 0.043569, 0.035355, 0.02869, 0.023282, 0.018893, 0.015331, 0.012441, 0.010095, 0.008192, 0.006648, 0.005395, 0.004378, 0.003552, 0.002883, 0.002339, 0.001898, 0.00154, 0.00125, 0.001014, 0.000823, 0.000668, 0.000542, 0.00044, 0.000357, 0.00029, 0.000235, 0.000191, 0.000155, 0.000126, 0.000102, 8.3e-05, 6.7e-05, 5.4e-05, 4.4e-05, 3.6e-05, 2.9e-05, 2.4e-05, 1.9e-05, 1.6e-05, 1.3e-05, 1e-05, 8e-06, 7e-06, 5e-06, 4e-06, 4e-06, 3e-06, 2e-06, 2e-06 ] }, "name": "/rotary/Constant_2_output_0", "outputIndexes": [ 38 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 37, 38 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/rotary/Mul_output_0", "outputIndexes": [ 39 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 39 ], "main_type": "UnaryOp", "main": { "opType": "COS", "T": "DT_FLOAT" }, "name": "/rotary/Cos_output_0", "outputIndexes": [ 40 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 40, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_output_0", "outputIndexes": [ 41 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 39 ], "main_type": "UnaryOp", "main": { "opType": "SIN", "T": "DT_FLOAT" }, "name": "/rotary/Sin_output_0", "outputIndexes": [ 42 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 42, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_1_output_0", "outputIndexes": [ 43 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 41, 43 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/rotary/Concat_output_0", "outputIndexes": [ 44 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 44, 44 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/rotary/Concat_1_output_0", "outputIndexes": [ 45 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 2 ] }, "name": "/rotary/Constant_7_output_0", "outputIndexes": [ 46 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 45, 46 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_2_output_0", "outputIndexes": [ 47 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 47, 15 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_3_output_0", "outputIndexes": [ 48 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 48 ], "main_type": "NONE", "name": "Shape107", "outputIndexes": [ 49 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 48 ], "main_type": "NONE", "name": "Rank109", "outputIndexes": [ 50 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 50, 50 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp111", "outputIndexes": [ 51 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 51 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze112", "outputIndexes": [ 52 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 51, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp114", "outputIndexes": [ 53 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 53 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze115", "outputIndexes": [ 54 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 49, 52, 54, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice118", "outputIndexes": [ 55 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 55 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze119", "outputIndexes": [ 56 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 56, 56 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp121", "outputIndexes": [ 57 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 48, 57, 51 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_3_output_0", "outputIndexes": [ 58 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 33, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_output_0", "outputIndexes": [ 59 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 33 ], "main_type": "NONE", "name": "/Shape_2_output_0", "outputIndexes": [ 60 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "main_type": "Blob", "main": { "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 3 ] }, "name": "/Constant_11_output_0", "outputIndexes": [ 61 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 60 ], "main_type": "NONE", "name": "Shape239", "outputIndexes": [ 62 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 60 ], "main_type": "NONE", "name": "Rank241", "outputIndexes": [ 63 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 63 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp243", "outputIndexes": [ 64 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 64 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze244", "outputIndexes": [ 65 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 64, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp246", "outputIndexes": [ 66 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 66 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze247", "outputIndexes": [ 67 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 62, 65, 67, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice250", "outputIndexes": [ 68 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 68 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze251", "outputIndexes": [ 69 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 69 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp252", "outputIndexes": [ 70 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 70, 69 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp253", "outputIndexes": [ 71 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 60, 71, 64 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_5_output_0", "outputIndexes": [ 72 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 2 ] }, "name": "/rotary/Constant_6_output_0", "outputIndexes": [ 73 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 72, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_output_0", "outputIndexes": [ 74 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 74, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_7_output_0", "outputIndexes": [ 75 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 1073741824 ] }, "name": "/Constant_20_output_0", "outputIndexes": [ 76 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 3 ] }, "name": "/Constant_14_output_0", "outputIndexes": [ 77 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 33, 75, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_1_output_0", "outputIndexes": [ 78 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 78 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_output_0", "outputIndexes": [ 79 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 74, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_6_output_0", "outputIndexes": [ 80 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 33, 20, 80, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_output_0", "outputIndexes": [ 81 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 79, 81 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_3_output_0", "outputIndexes": [ 82 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 51 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze158", "outputIndexes": [ 83 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 53 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze161", "outputIndexes": [ 84 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 49, 83, 84, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice164", "outputIndexes": [ 85 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 85 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze165", "outputIndexes": [ 86 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 86 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp166", "outputIndexes": [ 87 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 87, 86 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp167", "outputIndexes": [ 88 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 48, 88, 51 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_4_output_0", "outputIndexes": [ 89 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 82, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_1_output_0", "outputIndexes": [ 90 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 59, 90 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_output_0", "outputIndexes": [ 91 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5 ], "outputIndexes": [ 3259 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3259 ], "outputIndexes": [ 3260 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear", "inputIndexes": [ 3260 ], "outputIndexes": [ 3261 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 9486358, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3261 ], "outputIndexes": [ 3262 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3262 ], "outputIndexes": [ 92 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 19, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2_output_0", "outputIndexes": [ 93 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 28, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_3_output_0", "outputIndexes": [ 94 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 8 ] }, "name": "/Constant_7_output_0", "outputIndexes": [ 95 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 93, 94, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_1_output_0", "outputIndexes": [ 96 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 92, 96 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_2_output_0", "outputIndexes": [ 97 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 97, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_2_output_0", "outputIndexes": [ 98 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 97 ], "main_type": "NONE", "name": "/Shape_3_output_0", "outputIndexes": [ 99 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 99 ], "main_type": "NONE", "name": "Shape126", "outputIndexes": [ 100 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 99 ], "main_type": "NONE", "name": "Rank128", "outputIndexes": [ 101 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 101, 101 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp130", "outputIndexes": [ 102 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 102 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze131", "outputIndexes": [ 103 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 102, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp133", "outputIndexes": [ 104 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 104 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze134", "outputIndexes": [ 105 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 100, 103, 105, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice137", "outputIndexes": [ 106 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 106 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze138", "outputIndexes": [ 107 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 107 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp139", "outputIndexes": [ 108 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 108, 107 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp140", "outputIndexes": [ 109 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 99, 109, 102 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_6_output_0", "outputIndexes": [ 110 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 110, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_1_output_0", "outputIndexes": [ 111 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 111, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_9_output_0", "outputIndexes": [ 112 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 97, 112, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_3_output_0", "outputIndexes": [ 113 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 113 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_1_output_0", "outputIndexes": [ 114 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 111, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_8_output_0", "outputIndexes": [ 115 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 97, 20, 115, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_2_output_0", "outputIndexes": [ 116 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 114, 116 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_4_output_0", "outputIndexes": [ 117 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 117, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_3_output_0", "outputIndexes": [ 118 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 98, 118 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_1_output_0", "outputIndexes": [ 119 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5 ], "outputIndexes": [ 3263 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3263 ], "outputIndexes": [ 3264 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear", "inputIndexes": [ 3264 ], "outputIndexes": [ 3265 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 11849772, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3265 ], "outputIndexes": [ 3266 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3266 ], "outputIndexes": [ 120 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 19, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_4_output_0", "outputIndexes": [ 121 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 28, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_5_output_0", "outputIndexes": [ 122 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 121, 122, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_2_output_0", "outputIndexes": [ 123 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 120, 123 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_3_output_0", "outputIndexes": [ 124 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 91, 119, 124, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_8_output_0", "outputIndexes": [ 126 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 126 ], "outputIndexes": [ 3267 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3267 ], "outputIndexes": [ 3268 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear", "inputIndexes": [ 3268 ], "outputIndexes": [ 3269 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 14213186, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3269 ], "outputIndexes": [ 3270 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3270 ], "outputIndexes": [ 127 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4, 127 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_3_output_0", "outputIndexes": [ 128 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm/Mul_1_output_0", "inputIndexes": [ 128 ], "outputIndexes": [ 129 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 23650392, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 129 ], "outputIndexes": [ 3271 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3271 ], "outputIndexes": [ 3272 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear", "inputIndexes": [ 3272 ], "outputIndexes": [ 3273 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 23683160, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3273 ], "outputIndexes": [ 3274 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3274 ], "outputIndexes": [ 130 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 130 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn/Mul_output_0", "outputIndexes": [ 131 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 129 ], "outputIndexes": [ 3275 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3275 ], "outputIndexes": [ 3276 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear", "inputIndexes": [ 3276 ], "outputIndexes": [ 3277 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 49045614, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3277 ], "outputIndexes": [ 3278 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3278 ], "outputIndexes": [ 132 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 131, 132 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp/Mul_output_0", "outputIndexes": [ 133 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 133 ], "outputIndexes": [ 3279 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3279 ], "outputIndexes": [ 3280 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear", "inputIndexes": [ 3280 ], "outputIndexes": [ 3281 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 74408068, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3281 ], "outputIndexes": [ 3282 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3282 ], "outputIndexes": [ 134 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 128, 134 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_4_output_0", "outputIndexes": [ 135 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 135, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_9_output_0", "outputIndexes": [ 136 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_1/Mul_1_output_0", "inputIndexes": [ 136 ], "outputIndexes": [ 137 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 99770522, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 137 ], "outputIndexes": [ 3283 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3283 ], "outputIndexes": [ 3284 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear", "inputIndexes": [ 3284 ], "outputIndexes": [ 3285 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 99803290, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3285 ], "outputIndexes": [ 3286 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3286 ], "outputIndexes": [ 138 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 137 ], "main_type": "NONE", "name": "/Shape_14_output_0", "outputIndexes": [ 139 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 139 ], "main_type": "NONE", "name": "Shape484", "outputIndexes": [ 140 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 139 ], "main_type": "NONE", "name": "Rank486", "outputIndexes": [ 141 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 141, 141 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp488", "outputIndexes": [ 142 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 142 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze489", "outputIndexes": [ 143 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 142, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp491", "outputIndexes": [ 144 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 144 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze492", "outputIndexes": [ 145 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 140, 143, 145, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice495", "outputIndexes": [ 146 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 146 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze496", "outputIndexes": [ 147 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 147, 147 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp498", "outputIndexes": [ 148 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 139, 148, 142 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_18_output_0", "outputIndexes": [ 149 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 149, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_32_output_0", "outputIndexes": [ 150 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 142 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze506", "outputIndexes": [ 151 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 144 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze509", "outputIndexes": [ 152 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 140, 151, 152, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice512", "outputIndexes": [ 153 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 153 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze513", "outputIndexes": [ 154 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 154 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp514", "outputIndexes": [ 155 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 155, 154 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp515", "outputIndexes": [ 156 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 139, 156, 142 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_19_output_0", "outputIndexes": [ 157 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 157, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_33_output_0", "outputIndexes": [ 158 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 150, 158, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_13_output_0", "outputIndexes": [ 159 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 138, 159 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_10_output_0", "outputIndexes": [ 160 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 160, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_8_output_0", "outputIndexes": [ 161 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 160 ], "main_type": "NONE", "name": "/Shape_16_output_0", "outputIndexes": [ 162 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 162 ], "main_type": "NONE", "name": "Shape615", "outputIndexes": [ 163 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 162 ], "main_type": "NONE", "name": "Rank617", "outputIndexes": [ 164 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 164, 164 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp619", "outputIndexes": [ 165 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 165 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze620", "outputIndexes": [ 166 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 165, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp622", "outputIndexes": [ 167 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 167 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze623", "outputIndexes": [ 168 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 163, 166, 168, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice626", "outputIndexes": [ 169 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 169 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze627", "outputIndexes": [ 170 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 170 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp628", "outputIndexes": [ 171 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 171, 170 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp629", "outputIndexes": [ 172 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 162, 172, 165 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_20_output_0", "outputIndexes": [ 173 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 173, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_3_output_0", "outputIndexes": [ 174 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 174, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_39_output_0", "outputIndexes": [ 175 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 160, 175, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_5_output_0", "outputIndexes": [ 176 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 176 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_2_output_0", "outputIndexes": [ 177 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 174, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_38_output_0", "outputIndexes": [ 178 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 160, 20, 178, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_4_output_0", "outputIndexes": [ 179 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 177, 179 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_16_output_0", "outputIndexes": [ 180 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 180, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_9_output_0", "outputIndexes": [ 181 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 161, 181 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_5_output_0", "outputIndexes": [ 182 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 137 ], "outputIndexes": [ 3287 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3287 ], "outputIndexes": [ 3288 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear", "inputIndexes": [ 3288 ], "outputIndexes": [ 3289 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 109256880, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3289 ], "outputIndexes": [ 3290 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3290 ], "outputIndexes": [ 183 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 149, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_34_output_0", "outputIndexes": [ 184 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 157, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_35_output_0", "outputIndexes": [ 185 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 184, 185, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_14_output_0", "outputIndexes": [ 186 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 183, 186 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_11_output_0", "outputIndexes": [ 187 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 187, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_10_output_0", "outputIndexes": [ 188 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 187 ], "main_type": "NONE", "name": "/Shape_17_output_0", "outputIndexes": [ 189 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 189 ], "main_type": "NONE", "name": "Shape522", "outputIndexes": [ 190 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 189 ], "main_type": "NONE", "name": "Rank524", "outputIndexes": [ 191 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 191, 191 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp526", "outputIndexes": [ 192 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 192 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze527", "outputIndexes": [ 193 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 192, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp529", "outputIndexes": [ 194 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 194 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze530", "outputIndexes": [ 195 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 190, 193, 195, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice533", "outputIndexes": [ 196 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 196 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze534", "outputIndexes": [ 197 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 197 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp535", "outputIndexes": [ 198 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 198, 197 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp536", "outputIndexes": [ 199 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 189, 199, 192 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_21_output_0", "outputIndexes": [ 200 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 200, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_4_output_0", "outputIndexes": [ 201 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 201, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_41_output_0", "outputIndexes": [ 202 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 187, 202, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_7_output_0", "outputIndexes": [ 203 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 203 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_3_output_0", "outputIndexes": [ 204 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 201, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_40_output_0", "outputIndexes": [ 205 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 187, 20, 205, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_6_output_0", "outputIndexes": [ 206 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 204, 206 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_17_output_0", "outputIndexes": [ 207 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 207, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_11_output_0", "outputIndexes": [ 208 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 188, 208 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_6_output_0", "outputIndexes": [ 209 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 137 ], "outputIndexes": [ 3291 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3291 ], "outputIndexes": [ 3292 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear", "inputIndexes": [ 3292 ], "outputIndexes": [ 3293 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 111620294, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3293 ], "outputIndexes": [ 3294 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3294 ], "outputIndexes": [ 210 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 149, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_36_output_0", "outputIndexes": [ 211 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 157, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_37_output_0", "outputIndexes": [ 212 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 211, 212, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_15_output_0", "outputIndexes": [ 213 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 210, 213 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_12_output_0", "outputIndexes": [ 214 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 182, 209, 214, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_17_output_0", "outputIndexes": [ 215 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 215 ], "outputIndexes": [ 3295 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3295 ], "outputIndexes": [ 3296 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear", "inputIndexes": [ 3296 ], "outputIndexes": [ 3297 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 113983708, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3297 ], "outputIndexes": [ 3298 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3298 ], "outputIndexes": [ 216 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 136, 216 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_8_output_0", "outputIndexes": [ 217 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_1/Mul_1_output_0", "inputIndexes": [ 217 ], "outputIndexes": [ 218 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 123420914, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 218 ], "outputIndexes": [ 3299 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3299 ], "outputIndexes": [ 3300 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear", "inputIndexes": [ 3300 ], "outputIndexes": [ 3301 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 123453682, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3301 ], "outputIndexes": [ 3302 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3302 ], "outputIndexes": [ 219 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 219 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_1/Mul_output_0", "outputIndexes": [ 220 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 218 ], "outputIndexes": [ 3303 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3303 ], "outputIndexes": [ 3304 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear", "inputIndexes": [ 3304 ], "outputIndexes": [ 3305 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 148816136, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3305 ], "outputIndexes": [ 3306 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3306 ], "outputIndexes": [ 221 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 220, 221 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_1/Mul_output_0", "outputIndexes": [ 222 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 222 ], "outputIndexes": [ 3307 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3307 ], "outputIndexes": [ 3308 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear", "inputIndexes": [ 3308 ], "outputIndexes": [ 3309 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 174178590, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3309 ], "outputIndexes": [ 3310 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3310 ], "outputIndexes": [ 223 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 217, 223 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_9_output_0", "outputIndexes": [ 224 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 224, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_18_output_0", "outputIndexes": [ 225 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_2/Mul_1_output_0", "inputIndexes": [ 225 ], "outputIndexes": [ 226 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 199541044, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 226 ], "outputIndexes": [ 3311 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3311 ], "outputIndexes": [ 3312 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear", "inputIndexes": [ 3312 ], "outputIndexes": [ 3313 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 199573812, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3313 ], "outputIndexes": [ 3314 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3314 ], "outputIndexes": [ 227 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 226 ], "main_type": "NONE", "name": "/Shape_28_output_0", "outputIndexes": [ 228 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 228 ], "main_type": "NONE", "name": "Shape854", "outputIndexes": [ 229 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 228 ], "main_type": "NONE", "name": "Rank856", "outputIndexes": [ 230 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 230, 230 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp858", "outputIndexes": [ 231 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 231 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze859", "outputIndexes": [ 232 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 231, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp861", "outputIndexes": [ 233 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 233 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze862", "outputIndexes": [ 234 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 229, 232, 234, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice865", "outputIndexes": [ 235 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 235 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze866", "outputIndexes": [ 236 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 236, 236 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp868", "outputIndexes": [ 237 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 228, 237, 231 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_33_output_0", "outputIndexes": [ 238 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 238, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_64_output_0", "outputIndexes": [ 239 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 231 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze876", "outputIndexes": [ 240 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 233 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze879", "outputIndexes": [ 241 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 229, 240, 241, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice882", "outputIndexes": [ 242 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 242 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze883", "outputIndexes": [ 243 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 243 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp884", "outputIndexes": [ 244 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 244, 243 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp885", "outputIndexes": [ 245 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 228, 245, 231 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_34_output_0", "outputIndexes": [ 246 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 246, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_65_output_0", "outputIndexes": [ 247 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 239, 247, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_26_output_0", "outputIndexes": [ 248 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 227, 248 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_19_output_0", "outputIndexes": [ 249 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 249, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_16_output_0", "outputIndexes": [ 250 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 249 ], "main_type": "NONE", "name": "/Shape_30_output_0", "outputIndexes": [ 251 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 251 ], "main_type": "NONE", "name": "Shape985", "outputIndexes": [ 252 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 251 ], "main_type": "NONE", "name": "Rank987", "outputIndexes": [ 253 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 253, 253 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp989", "outputIndexes": [ 254 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 254 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze990", "outputIndexes": [ 255 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 254, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp992", "outputIndexes": [ 256 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 256 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze993", "outputIndexes": [ 257 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 252, 255, 257, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice996", "outputIndexes": [ 258 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 258 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze997", "outputIndexes": [ 259 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 259 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp998", "outputIndexes": [ 260 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 260, 259 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp999", "outputIndexes": [ 261 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 251, 261, 254 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_35_output_0", "outputIndexes": [ 262 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 262, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_6_output_0", "outputIndexes": [ 263 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 263, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_71_output_0", "outputIndexes": [ 264 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 249, 264, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_9_output_0", "outputIndexes": [ 265 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 265 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_4_output_0", "outputIndexes": [ 266 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 263, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_70_output_0", "outputIndexes": [ 267 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 249, 20, 267, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_8_output_0", "outputIndexes": [ 268 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 266, 268 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_29_output_0", "outputIndexes": [ 269 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 269, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_17_output_0", "outputIndexes": [ 270 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 250, 270 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_10_output_0", "outputIndexes": [ 271 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 226 ], "outputIndexes": [ 3315 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3315 ], "outputIndexes": [ 3316 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear", "inputIndexes": [ 3316 ], "outputIndexes": [ 3317 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 209027402, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3317 ], "outputIndexes": [ 3318 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3318 ], "outputIndexes": [ 272 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 238, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_66_output_0", "outputIndexes": [ 273 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 246, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_67_output_0", "outputIndexes": [ 274 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 273, 274, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_27_output_0", "outputIndexes": [ 275 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 272, 275 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_20_output_0", "outputIndexes": [ 276 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 276, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_18_output_0", "outputIndexes": [ 277 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 276 ], "main_type": "NONE", "name": "/Shape_31_output_0", "outputIndexes": [ 278 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 278 ], "main_type": "NONE", "name": "Shape892", "outputIndexes": [ 279 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 278 ], "main_type": "NONE", "name": "Rank894", "outputIndexes": [ 280 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 280, 280 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp896", "outputIndexes": [ 281 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 281 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze897", "outputIndexes": [ 282 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 281, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp899", "outputIndexes": [ 283 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 283 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze900", "outputIndexes": [ 284 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 279, 282, 284, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice903", "outputIndexes": [ 285 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 285 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze904", "outputIndexes": [ 286 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 286 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp905", "outputIndexes": [ 287 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 287, 286 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp906", "outputIndexes": [ 288 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 278, 288, 281 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_36_output_0", "outputIndexes": [ 289 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 289, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_7_output_0", "outputIndexes": [ 290 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 290, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_73_output_0", "outputIndexes": [ 291 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 276, 291, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_11_output_0", "outputIndexes": [ 292 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 292 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_5_output_0", "outputIndexes": [ 293 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 290, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_72_output_0", "outputIndexes": [ 294 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 276, 20, 294, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_10_output_0", "outputIndexes": [ 295 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 293, 295 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_30_output_0", "outputIndexes": [ 296 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 296, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_19_output_0", "outputIndexes": [ 297 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 277, 297 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_11_output_0", "outputIndexes": [ 298 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 226 ], "outputIndexes": [ 3319 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3319 ], "outputIndexes": [ 3320 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear", "inputIndexes": [ 3320 ], "outputIndexes": [ 3321 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 211390816, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3321 ], "outputIndexes": [ 3322 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3322 ], "outputIndexes": [ 299 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 238, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_68_output_0", "outputIndexes": [ 300 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 246, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_69_output_0", "outputIndexes": [ 301 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 300, 301, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_28_output_0", "outputIndexes": [ 302 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 299, 302 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_21_output_0", "outputIndexes": [ 303 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 271, 298, 303, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_26_output_0", "outputIndexes": [ 304 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 304 ], "outputIndexes": [ 3323 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3323 ], "outputIndexes": [ 3324 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear", "inputIndexes": [ 3324 ], "outputIndexes": [ 3325 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 213754230, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3325 ], "outputIndexes": [ 3326 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3326 ], "outputIndexes": [ 305 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 225, 305 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_13_output_0", "outputIndexes": [ 306 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_2/Mul_1_output_0", "inputIndexes": [ 306 ], "outputIndexes": [ 307 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 223191436, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 307 ], "outputIndexes": [ 3327 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3327 ], "outputIndexes": [ 3328 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear", "inputIndexes": [ 3328 ], "outputIndexes": [ 3329 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 223224204, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3329 ], "outputIndexes": [ 3330 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3330 ], "outputIndexes": [ 308 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 308 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_2/Mul_output_0", "outputIndexes": [ 309 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 307 ], "outputIndexes": [ 3331 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3331 ], "outputIndexes": [ 3332 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear", "inputIndexes": [ 3332 ], "outputIndexes": [ 3333 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 248586658, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3333 ], "outputIndexes": [ 3334 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3334 ], "outputIndexes": [ 310 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 309, 310 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_2/Mul_output_0", "outputIndexes": [ 311 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 311 ], "outputIndexes": [ 3335 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3335 ], "outputIndexes": [ 3336 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear", "inputIndexes": [ 3336 ], "outputIndexes": [ 3337 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 273949112, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3337 ], "outputIndexes": [ 3338 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3338 ], "outputIndexes": [ 312 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 306, 312 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_14_output_0", "outputIndexes": [ 313 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 313, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_27_output_0", "outputIndexes": [ 314 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_3/Mul_1_output_0", "inputIndexes": [ 314 ], "outputIndexes": [ 315 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 299311566, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 315 ], "outputIndexes": [ 3339 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3339 ], "outputIndexes": [ 3340 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear", "inputIndexes": [ 3340 ], "outputIndexes": [ 3341 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 299344334, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3341 ], "outputIndexes": [ 3342 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3342 ], "outputIndexes": [ 316 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 315 ], "main_type": "NONE", "name": "/Shape_42_output_0", "outputIndexes": [ 317 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 317 ], "main_type": "NONE", "name": "Shape1224", "outputIndexes": [ 318 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 317 ], "main_type": "NONE", "name": "Rank1226", "outputIndexes": [ 319 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 319, 319 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1228", "outputIndexes": [ 320 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 320 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1229", "outputIndexes": [ 321 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 320, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1231", "outputIndexes": [ 322 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 322 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1232", "outputIndexes": [ 323 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 318, 321, 323, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1235", "outputIndexes": [ 324 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 324 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1236", "outputIndexes": [ 325 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 325, 325 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1238", "outputIndexes": [ 326 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 317, 326, 320 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_48_output_0", "outputIndexes": [ 327 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 327, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_96_output_0", "outputIndexes": [ 328 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 320 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1246", "outputIndexes": [ 329 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 322 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1249", "outputIndexes": [ 330 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 318, 329, 330, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1252", "outputIndexes": [ 331 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 331 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1253", "outputIndexes": [ 332 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 332 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1254", "outputIndexes": [ 333 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 333, 332 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1255", "outputIndexes": [ 334 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 317, 334, 320 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_49_output_0", "outputIndexes": [ 335 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 335, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_97_output_0", "outputIndexes": [ 336 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 328, 336, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_39_output_0", "outputIndexes": [ 337 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 316, 337 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_28_output_0", "outputIndexes": [ 338 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 338, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_24_output_0", "outputIndexes": [ 339 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 338 ], "main_type": "NONE", "name": "/Shape_44_output_0", "outputIndexes": [ 340 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 340 ], "main_type": "NONE", "name": "Shape1355", "outputIndexes": [ 341 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 340 ], "main_type": "NONE", "name": "Rank1357", "outputIndexes": [ 342 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 342, 342 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1359", "outputIndexes": [ 343 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 343 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1360", "outputIndexes": [ 344 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 343, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1362", "outputIndexes": [ 345 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 345 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1363", "outputIndexes": [ 346 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 341, 344, 346, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1366", "outputIndexes": [ 347 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 347 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1367", "outputIndexes": [ 348 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 348 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1368", "outputIndexes": [ 349 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 349, 348 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1369", "outputIndexes": [ 350 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 340, 350, 343 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_50_output_0", "outputIndexes": [ 351 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 351, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_9_output_0", "outputIndexes": [ 352 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 352, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_103_output_0", "outputIndexes": [ 353 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 338, 353, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_13_output_0", "outputIndexes": [ 354 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 354 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_6_output_0", "outputIndexes": [ 355 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 352, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_102_output_0", "outputIndexes": [ 356 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 338, 20, 356, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_12_output_0", "outputIndexes": [ 357 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 355, 357 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_42_output_0", "outputIndexes": [ 358 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 358, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_25_output_0", "outputIndexes": [ 359 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 339, 359 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_15_output_0", "outputIndexes": [ 360 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 315 ], "outputIndexes": [ 3343 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3343 ], "outputIndexes": [ 3344 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear", "inputIndexes": [ 3344 ], "outputIndexes": [ 3345 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 308797924, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3345 ], "outputIndexes": [ 3346 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3346 ], "outputIndexes": [ 361 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 327, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_98_output_0", "outputIndexes": [ 362 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 335, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_99_output_0", "outputIndexes": [ 363 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 362, 363, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_40_output_0", "outputIndexes": [ 364 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 361, 364 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_29_output_0", "outputIndexes": [ 365 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 365, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_26_output_0", "outputIndexes": [ 366 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 365 ], "main_type": "NONE", "name": "/Shape_45_output_0", "outputIndexes": [ 367 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 367 ], "main_type": "NONE", "name": "Shape1262", "outputIndexes": [ 368 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 367 ], "main_type": "NONE", "name": "Rank1264", "outputIndexes": [ 369 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 369, 369 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1266", "outputIndexes": [ 370 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 370 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1267", "outputIndexes": [ 371 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 370, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1269", "outputIndexes": [ 372 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 372 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1270", "outputIndexes": [ 373 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 368, 371, 373, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1273", "outputIndexes": [ 374 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 374 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1274", "outputIndexes": [ 375 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 375 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1275", "outputIndexes": [ 376 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 376, 375 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1276", "outputIndexes": [ 377 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 367, 377, 370 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_51_output_0", "outputIndexes": [ 378 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 378, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_10_output_0", "outputIndexes": [ 379 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 379, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_105_output_0", "outputIndexes": [ 380 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 365, 380, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_15_output_0", "outputIndexes": [ 381 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 381 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_7_output_0", "outputIndexes": [ 382 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 379, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_104_output_0", "outputIndexes": [ 383 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 365, 20, 383, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_14_output_0", "outputIndexes": [ 384 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 382, 384 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_43_output_0", "outputIndexes": [ 385 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 385, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_27_output_0", "outputIndexes": [ 386 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 366, 386 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_16_output_0", "outputIndexes": [ 387 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 315 ], "outputIndexes": [ 3347 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3347 ], "outputIndexes": [ 3348 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear", "inputIndexes": [ 3348 ], "outputIndexes": [ 3349 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 311161338, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3349 ], "outputIndexes": [ 3350 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3350 ], "outputIndexes": [ 388 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 327, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_100_output_0", "outputIndexes": [ 389 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 335, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_101_output_0", "outputIndexes": [ 390 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 389, 390, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_41_output_0", "outputIndexes": [ 391 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 388, 391 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_30_output_0", "outputIndexes": [ 392 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 360, 387, 392, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_35_output_0", "outputIndexes": [ 393 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 393 ], "outputIndexes": [ 3351 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3351 ], "outputIndexes": [ 3352 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear", "inputIndexes": [ 3352 ], "outputIndexes": [ 3353 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 313524752, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3353 ], "outputIndexes": [ 3354 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3354 ], "outputIndexes": [ 394 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 314, 394 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_18_output_0", "outputIndexes": [ 395 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_3/Mul_1_output_0", "inputIndexes": [ 395 ], "outputIndexes": [ 396 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 322961958, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 396 ], "outputIndexes": [ 3355 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3355 ], "outputIndexes": [ 3356 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear", "inputIndexes": [ 3356 ], "outputIndexes": [ 3357 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 322994726, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3357 ], "outputIndexes": [ 3358 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3358 ], "outputIndexes": [ 397 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 397 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_3/Mul_output_0", "outputIndexes": [ 398 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 396 ], "outputIndexes": [ 3359 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3359 ], "outputIndexes": [ 3360 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear", "inputIndexes": [ 3360 ], "outputIndexes": [ 3361 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 348357180, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3361 ], "outputIndexes": [ 3362 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3362 ], "outputIndexes": [ 399 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 398, 399 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_3/Mul_output_0", "outputIndexes": [ 400 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 400 ], "outputIndexes": [ 3363 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3363 ], "outputIndexes": [ 3364 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear", "inputIndexes": [ 3364 ], "outputIndexes": [ 3365 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 373719634, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3365 ], "outputIndexes": [ 3366 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3366 ], "outputIndexes": [ 401 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 395, 401 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_19_output_0", "outputIndexes": [ 402 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 402, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_36_output_0", "outputIndexes": [ 403 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_4/Mul_1_output_0", "inputIndexes": [ 403 ], "outputIndexes": [ 404 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 399082088, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 404 ], "outputIndexes": [ 3367 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3367 ], "outputIndexes": [ 3368 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear", "inputIndexes": [ 3368 ], "outputIndexes": [ 3369 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 399114856, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3369 ], "outputIndexes": [ 3370 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3370 ], "outputIndexes": [ 405 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 404 ], "main_type": "NONE", "name": "/Shape_56_output_0", "outputIndexes": [ 406 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 406 ], "main_type": "NONE", "name": "Shape1594", "outputIndexes": [ 407 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 406 ], "main_type": "NONE", "name": "Rank1596", "outputIndexes": [ 408 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 408, 408 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1598", "outputIndexes": [ 409 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 409 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1599", "outputIndexes": [ 410 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 409, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1601", "outputIndexes": [ 411 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 411 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1602", "outputIndexes": [ 412 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 407, 410, 412, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1605", "outputIndexes": [ 413 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 413 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1606", "outputIndexes": [ 414 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 414, 414 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1608", "outputIndexes": [ 415 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 406, 415, 409 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_63_output_0", "outputIndexes": [ 416 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 416, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_128_output_0", "outputIndexes": [ 417 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 409 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1616", "outputIndexes": [ 418 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 411 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1619", "outputIndexes": [ 419 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 407, 418, 419, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1622", "outputIndexes": [ 420 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 420 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1623", "outputIndexes": [ 421 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 421 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1624", "outputIndexes": [ 422 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 422, 421 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1625", "outputIndexes": [ 423 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 406, 423, 409 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_64_output_0", "outputIndexes": [ 424 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 424, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_129_output_0", "outputIndexes": [ 425 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 417, 425, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_52_output_0", "outputIndexes": [ 426 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 405, 426 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_37_output_0", "outputIndexes": [ 427 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 427, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_32_output_0", "outputIndexes": [ 428 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 427 ], "main_type": "NONE", "name": "/Shape_58_output_0", "outputIndexes": [ 429 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 429 ], "main_type": "NONE", "name": "Shape1726", "outputIndexes": [ 430 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 429 ], "main_type": "NONE", "name": "Rank1728", "outputIndexes": [ 431 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 431, 431 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1730", "outputIndexes": [ 432 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 432 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1731", "outputIndexes": [ 433 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 432, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1733", "outputIndexes": [ 434 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 434 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1734", "outputIndexes": [ 435 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 430, 433, 435, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1737", "outputIndexes": [ 436 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 436 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1738", "outputIndexes": [ 437 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 437 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1739", "outputIndexes": [ 438 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 438, 437 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1740", "outputIndexes": [ 439 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 429, 439, 432 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_65_output_0", "outputIndexes": [ 440 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 440, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_12_output_0", "outputIndexes": [ 441 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 441, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_135_output_0", "outputIndexes": [ 442 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 427, 442, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_17_output_0", "outputIndexes": [ 443 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 443 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_8_output_0", "outputIndexes": [ 444 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 441, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_134_output_0", "outputIndexes": [ 445 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 427, 20, 445, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_16_output_0", "outputIndexes": [ 446 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 444, 446 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_55_output_0", "outputIndexes": [ 447 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 447, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_33_output_0", "outputIndexes": [ 448 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 428, 448 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_20_output_0", "outputIndexes": [ 449 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 404 ], "outputIndexes": [ 3371 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3371 ], "outputIndexes": [ 3372 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear", "inputIndexes": [ 3372 ], "outputIndexes": [ 3373 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 408568446, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3373 ], "outputIndexes": [ 3374 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3374 ], "outputIndexes": [ 450 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 416, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_130_output_0", "outputIndexes": [ 451 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 424, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_131_output_0", "outputIndexes": [ 452 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 451, 452, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_53_output_0", "outputIndexes": [ 453 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 450, 453 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_38_output_0", "outputIndexes": [ 454 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 454, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_34_output_0", "outputIndexes": [ 455 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 454 ], "main_type": "NONE", "name": "/Shape_59_output_0", "outputIndexes": [ 456 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 456 ], "main_type": "NONE", "name": "Shape1632", "outputIndexes": [ 457 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 456 ], "main_type": "NONE", "name": "Rank1634", "outputIndexes": [ 458 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 458, 458 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1636", "outputIndexes": [ 459 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 459 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1637", "outputIndexes": [ 460 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 459, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1639", "outputIndexes": [ 461 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 461 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1640", "outputIndexes": [ 462 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 457, 460, 462, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1643", "outputIndexes": [ 463 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 463 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1644", "outputIndexes": [ 464 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 464 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1645", "outputIndexes": [ 465 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 465, 464 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1646", "outputIndexes": [ 466 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 456, 466, 459 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_66_output_0", "outputIndexes": [ 467 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 467, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_13_output_0", "outputIndexes": [ 468 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 468, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_137_output_0", "outputIndexes": [ 469 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 454, 469, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_19_output_0", "outputIndexes": [ 470 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 470 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_9_output_0", "outputIndexes": [ 471 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 468, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_136_output_0", "outputIndexes": [ 472 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 454, 20, 472, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_18_output_0", "outputIndexes": [ 473 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 471, 473 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_56_output_0", "outputIndexes": [ 474 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 474, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_35_output_0", "outputIndexes": [ 475 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 455, 475 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_21_output_0", "outputIndexes": [ 476 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 404 ], "outputIndexes": [ 3375 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3375 ], "outputIndexes": [ 3376 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear", "inputIndexes": [ 3376 ], "outputIndexes": [ 3377 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 410931860, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3377 ], "outputIndexes": [ 3378 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3378 ], "outputIndexes": [ 477 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 416, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_132_output_0", "outputIndexes": [ 478 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 424, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_133_output_0", "outputIndexes": [ 479 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 478, 479, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_54_output_0", "outputIndexes": [ 480 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 477, 480 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_39_output_0", "outputIndexes": [ 481 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 449, 476, 481, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_44_output_0", "outputIndexes": [ 482 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 482 ], "outputIndexes": [ 3379 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3379 ], "outputIndexes": [ 3380 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear", "inputIndexes": [ 3380 ], "outputIndexes": [ 3381 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 413295274, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3381 ], "outputIndexes": [ 3382 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3382 ], "outputIndexes": [ 483 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 403, 483 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_23_output_0", "outputIndexes": [ 484 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_4/Mul_1_output_0", "inputIndexes": [ 484 ], "outputIndexes": [ 485 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 422732480, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 485 ], "outputIndexes": [ 3383 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3383 ], "outputIndexes": [ 3384 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear", "inputIndexes": [ 3384 ], "outputIndexes": [ 3385 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 422765248, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3385 ], "outputIndexes": [ 3386 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3386 ], "outputIndexes": [ 486 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 486 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_4/Mul_output_0", "outputIndexes": [ 487 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 485 ], "outputIndexes": [ 3387 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3387 ], "outputIndexes": [ 3388 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear", "inputIndexes": [ 3388 ], "outputIndexes": [ 3389 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 448127702, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3389 ], "outputIndexes": [ 3390 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3390 ], "outputIndexes": [ 488 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 487, 488 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_4/Mul_output_0", "outputIndexes": [ 489 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 489 ], "outputIndexes": [ 3391 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3391 ], "outputIndexes": [ 3392 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear", "inputIndexes": [ 3392 ], "outputIndexes": [ 3393 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 473490156, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3393 ], "outputIndexes": [ 3394 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3394 ], "outputIndexes": [ 490 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 484, 490 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_24_output_0", "outputIndexes": [ 491 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 491, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_45_output_0", "outputIndexes": [ 492 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_5/Mul_1_output_0", "inputIndexes": [ 492 ], "outputIndexes": [ 493 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 498852610, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 493 ], "outputIndexes": [ 3395 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3395 ], "outputIndexes": [ 3396 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear", "inputIndexes": [ 3396 ], "outputIndexes": [ 3397 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 498885378, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3397 ], "outputIndexes": [ 3398 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3398 ], "outputIndexes": [ 494 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 493 ], "main_type": "NONE", "name": "/Shape_70_output_0", "outputIndexes": [ 495 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 495 ], "main_type": "NONE", "name": "Shape1965", "outputIndexes": [ 496 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 495 ], "main_type": "NONE", "name": "Rank1967", "outputIndexes": [ 497 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 497, 497 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1969", "outputIndexes": [ 498 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 498 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1970", "outputIndexes": [ 499 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 498, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1972", "outputIndexes": [ 500 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 500 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1973", "outputIndexes": [ 501 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 496, 499, 501, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1976", "outputIndexes": [ 502 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 502 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1977", "outputIndexes": [ 503 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 503, 503 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1979", "outputIndexes": [ 504 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 495, 504, 498 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_78_output_0", "outputIndexes": [ 505 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 505, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_160_output_0", "outputIndexes": [ 506 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 498 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1987", "outputIndexes": [ 507 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 500 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1990", "outputIndexes": [ 508 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 496, 507, 508, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1993", "outputIndexes": [ 509 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 509 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1994", "outputIndexes": [ 510 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 510 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1995", "outputIndexes": [ 511 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 511, 510 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1996", "outputIndexes": [ 512 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 495, 512, 498 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_79_output_0", "outputIndexes": [ 513 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 513, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_161_output_0", "outputIndexes": [ 514 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 506, 514, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_65_output_0", "outputIndexes": [ 515 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 494, 515 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_46_output_0", "outputIndexes": [ 516 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 516, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_40_output_0", "outputIndexes": [ 517 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 516 ], "main_type": "NONE", "name": "/Shape_72_output_0", "outputIndexes": [ 518 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 518 ], "main_type": "NONE", "name": "Shape2097", "outputIndexes": [ 519 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 518 ], "main_type": "NONE", "name": "Rank2099", "outputIndexes": [ 520 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 520, 520 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2101", "outputIndexes": [ 521 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 521 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2102", "outputIndexes": [ 522 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 521, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2104", "outputIndexes": [ 523 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 523 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2105", "outputIndexes": [ 524 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 519, 522, 524, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2108", "outputIndexes": [ 525 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 525 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2109", "outputIndexes": [ 526 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 526 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2110", "outputIndexes": [ 527 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 527, 526 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2111", "outputIndexes": [ 528 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 518, 528, 521 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_80_output_0", "outputIndexes": [ 529 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 529, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_15_output_0", "outputIndexes": [ 530 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 530, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_167_output_0", "outputIndexes": [ 531 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 516, 531, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_21_output_0", "outputIndexes": [ 532 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 532 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_10_output_0", "outputIndexes": [ 533 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 530, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_166_output_0", "outputIndexes": [ 534 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 516, 20, 534, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_20_output_0", "outputIndexes": [ 535 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 533, 535 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_68_output_0", "outputIndexes": [ 536 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 536, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_41_output_0", "outputIndexes": [ 537 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 517, 537 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_25_output_0", "outputIndexes": [ 538 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 493 ], "outputIndexes": [ 3399 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3399 ], "outputIndexes": [ 3400 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear", "inputIndexes": [ 3400 ], "outputIndexes": [ 3401 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 508338968, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3401 ], "outputIndexes": [ 3402 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3402 ], "outputIndexes": [ 539 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 505, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_162_output_0", "outputIndexes": [ 540 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 513, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_163_output_0", "outputIndexes": [ 541 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 540, 541, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_66_output_0", "outputIndexes": [ 542 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 539, 542 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_47_output_0", "outputIndexes": [ 543 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 543, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_42_output_0", "outputIndexes": [ 544 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 543 ], "main_type": "NONE", "name": "/Shape_73_output_0", "outputIndexes": [ 545 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 545 ], "main_type": "NONE", "name": "Shape2003", "outputIndexes": [ 546 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 545 ], "main_type": "NONE", "name": "Rank2005", "outputIndexes": [ 547 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 547, 547 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2007", "outputIndexes": [ 548 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 548 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2008", "outputIndexes": [ 549 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 548, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2010", "outputIndexes": [ 550 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 550 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2011", "outputIndexes": [ 551 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 546, 549, 551, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2014", "outputIndexes": [ 552 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 552 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2015", "outputIndexes": [ 553 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 553 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2016", "outputIndexes": [ 554 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 554, 553 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2017", "outputIndexes": [ 555 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 545, 555, 548 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_81_output_0", "outputIndexes": [ 556 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 556, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_16_output_0", "outputIndexes": [ 557 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 557, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_169_output_0", "outputIndexes": [ 558 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 543, 558, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_23_output_0", "outputIndexes": [ 559 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 559 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_11_output_0", "outputIndexes": [ 560 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 557, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_168_output_0", "outputIndexes": [ 561 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 543, 20, 561, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_22_output_0", "outputIndexes": [ 562 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 560, 562 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_69_output_0", "outputIndexes": [ 563 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 563, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_43_output_0", "outputIndexes": [ 564 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 544, 564 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_26_output_0", "outputIndexes": [ 565 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 493 ], "outputIndexes": [ 3403 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3403 ], "outputIndexes": [ 3404 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear", "inputIndexes": [ 3404 ], "outputIndexes": [ 3405 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 510702382, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3405 ], "outputIndexes": [ 3406 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3406 ], "outputIndexes": [ 566 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 505, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_164_output_0", "outputIndexes": [ 567 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 513, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_165_output_0", "outputIndexes": [ 568 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 567, 568, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_67_output_0", "outputIndexes": [ 569 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 566, 569 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_48_output_0", "outputIndexes": [ 570 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 538, 565, 570, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_53_output_0", "outputIndexes": [ 571 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 571 ], "outputIndexes": [ 3407 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3407 ], "outputIndexes": [ 3408 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear", "inputIndexes": [ 3408 ], "outputIndexes": [ 3409 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 513065796, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3409 ], "outputIndexes": [ 3410 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3410 ], "outputIndexes": [ 572 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 492, 572 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_28_output_0", "outputIndexes": [ 573 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_5/Mul_1_output_0", "inputIndexes": [ 573 ], "outputIndexes": [ 574 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 522503002, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 574 ], "outputIndexes": [ 3411 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3411 ], "outputIndexes": [ 3412 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear", "inputIndexes": [ 3412 ], "outputIndexes": [ 3413 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 522535770, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3413 ], "outputIndexes": [ 3414 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3414 ], "outputIndexes": [ 575 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 575 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_5/Mul_output_0", "outputIndexes": [ 576 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 574 ], "outputIndexes": [ 3415 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3415 ], "outputIndexes": [ 3416 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear", "inputIndexes": [ 3416 ], "outputIndexes": [ 3417 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 547898224, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3417 ], "outputIndexes": [ 3418 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3418 ], "outputIndexes": [ 577 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 576, 577 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_5/Mul_output_0", "outputIndexes": [ 578 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 578 ], "outputIndexes": [ 3419 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3419 ], "outputIndexes": [ 3420 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear", "inputIndexes": [ 3420 ], "outputIndexes": [ 3421 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 573260678, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3421 ], "outputIndexes": [ 3422 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3422 ], "outputIndexes": [ 579 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 573, 579 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_29_output_0", "outputIndexes": [ 580 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 580, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_54_output_0", "outputIndexes": [ 581 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_6/Mul_1_output_0", "inputIndexes": [ 581 ], "outputIndexes": [ 582 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 598623132, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 582 ], "outputIndexes": [ 3423 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3423 ], "outputIndexes": [ 3424 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear", "inputIndexes": [ 3424 ], "outputIndexes": [ 3425 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 598655900, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3425 ], "outputIndexes": [ 3426 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3426 ], "outputIndexes": [ 583 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 582 ], "main_type": "NONE", "name": "/Shape_84_output_0", "outputIndexes": [ 584 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 584 ], "main_type": "NONE", "name": "Shape2336", "outputIndexes": [ 585 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 584 ], "main_type": "NONE", "name": "Rank2338", "outputIndexes": [ 586 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 586, 586 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2340", "outputIndexes": [ 587 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 587 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2341", "outputIndexes": [ 588 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 587, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2343", "outputIndexes": [ 589 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 589 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2344", "outputIndexes": [ 590 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 585, 588, 590, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2347", "outputIndexes": [ 591 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 591 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2348", "outputIndexes": [ 592 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 592, 592 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2350", "outputIndexes": [ 593 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 584, 593, 587 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_93_output_0", "outputIndexes": [ 594 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 594, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_192_output_0", "outputIndexes": [ 595 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 587 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2358", "outputIndexes": [ 596 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 589 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2361", "outputIndexes": [ 597 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 585, 596, 597, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2364", "outputIndexes": [ 598 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 598 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2365", "outputIndexes": [ 599 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 599 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2366", "outputIndexes": [ 600 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 600, 599 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2367", "outputIndexes": [ 601 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 584, 601, 587 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_94_output_0", "outputIndexes": [ 602 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 602, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_193_output_0", "outputIndexes": [ 603 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 595, 603, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_78_output_0", "outputIndexes": [ 604 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 583, 604 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_55_output_0", "outputIndexes": [ 605 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 605, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_48_output_0", "outputIndexes": [ 606 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 605 ], "main_type": "NONE", "name": "/Shape_86_output_0", "outputIndexes": [ 607 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 607 ], "main_type": "NONE", "name": "Shape2468", "outputIndexes": [ 608 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 607 ], "main_type": "NONE", "name": "Rank2470", "outputIndexes": [ 609 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 609, 609 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2472", "outputIndexes": [ 610 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 610 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2473", "outputIndexes": [ 611 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 610, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2475", "outputIndexes": [ 612 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 612 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2476", "outputIndexes": [ 613 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 608, 611, 613, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2479", "outputIndexes": [ 614 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 614 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2480", "outputIndexes": [ 615 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 615 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2481", "outputIndexes": [ 616 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 616, 615 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2482", "outputIndexes": [ 617 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 607, 617, 610 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_95_output_0", "outputIndexes": [ 618 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 618, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_18_output_0", "outputIndexes": [ 619 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 619, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_199_output_0", "outputIndexes": [ 620 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 605, 620, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_25_output_0", "outputIndexes": [ 621 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 621 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_12_output_0", "outputIndexes": [ 622 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 619, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_198_output_0", "outputIndexes": [ 623 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 605, 20, 623, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_24_output_0", "outputIndexes": [ 624 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 622, 624 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_81_output_0", "outputIndexes": [ 625 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 625, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_49_output_0", "outputIndexes": [ 626 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 606, 626 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_30_output_0", "outputIndexes": [ 627 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 582 ], "outputIndexes": [ 3427 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3427 ], "outputIndexes": [ 3428 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear", "inputIndexes": [ 3428 ], "outputIndexes": [ 3429 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 608109490, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3429 ], "outputIndexes": [ 3430 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3430 ], "outputIndexes": [ 628 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 594, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_194_output_0", "outputIndexes": [ 629 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 602, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_195_output_0", "outputIndexes": [ 630 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 629, 630, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_79_output_0", "outputIndexes": [ 631 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 628, 631 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_56_output_0", "outputIndexes": [ 632 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 632, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_50_output_0", "outputIndexes": [ 633 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 632 ], "main_type": "NONE", "name": "/Shape_87_output_0", "outputIndexes": [ 634 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 634 ], "main_type": "NONE", "name": "Shape2374", "outputIndexes": [ 635 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 634 ], "main_type": "NONE", "name": "Rank2376", "outputIndexes": [ 636 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 636, 636 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2378", "outputIndexes": [ 637 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 637 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2379", "outputIndexes": [ 638 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 637, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2381", "outputIndexes": [ 639 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 639 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2382", "outputIndexes": [ 640 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 635, 638, 640, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2385", "outputIndexes": [ 641 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 641 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2386", "outputIndexes": [ 642 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 642 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2387", "outputIndexes": [ 643 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 643, 642 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2388", "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_96_output_0", "outputIndexes": [ 645 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 645, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_19_output_0", "outputIndexes": [ 646 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 646, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_201_output_0", "outputIndexes": [ 647 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 632, 647, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_27_output_0", "outputIndexes": [ 648 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 648 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_13_output_0", "outputIndexes": [ 649 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 646, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_200_output_0", "outputIndexes": [ 650 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 632, 20, 650, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_26_output_0", "outputIndexes": [ 651 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 649, 651 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_82_output_0", "outputIndexes": [ 652 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 652, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_51_output_0", "outputIndexes": [ 653 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 633, 653 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_31_output_0", "outputIndexes": [ 654 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 582 ], "outputIndexes": [ 3431 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3431 ], "outputIndexes": [ 3432 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear", "inputIndexes": [ 3432 ], "outputIndexes": [ 3433 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 610472904, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3433 ], "outputIndexes": [ 3434 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3434 ], "outputIndexes": [ 655 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 594, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_196_output_0", "outputIndexes": [ 656 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 602, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_197_output_0", "outputIndexes": [ 657 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 656, 657, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_80_output_0", "outputIndexes": [ 658 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 655, 658 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_57_output_0", "outputIndexes": [ 659 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 627, 654, 659, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_62_output_0", "outputIndexes": [ 660 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 660 ], "outputIndexes": [ 3435 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3435 ], "outputIndexes": [ 3436 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear", "inputIndexes": [ 3436 ], "outputIndexes": [ 3437 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 612836318, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3437 ], "outputIndexes": [ 3438 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3438 ], "outputIndexes": [ 661 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 581, 661 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_33_output_0", "outputIndexes": [ 662 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_6/Mul_1_output_0", "inputIndexes": [ 662 ], "outputIndexes": [ 663 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 622273524, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 663 ], "outputIndexes": [ 3439 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3439 ], "outputIndexes": [ 3440 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear", "inputIndexes": [ 3440 ], "outputIndexes": [ 3441 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 622306292, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3441 ], "outputIndexes": [ 3442 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3442 ], "outputIndexes": [ 664 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 664 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_6/Mul_output_0", "outputIndexes": [ 665 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 663 ], "outputIndexes": [ 3443 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3443 ], "outputIndexes": [ 3444 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear", "inputIndexes": [ 3444 ], "outputIndexes": [ 3445 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 647668746, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3445 ], "outputIndexes": [ 3446 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3446 ], "outputIndexes": [ 666 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 665, 666 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_6/Mul_output_0", "outputIndexes": [ 667 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 667 ], "outputIndexes": [ 3447 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3447 ], "outputIndexes": [ 3448 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear", "inputIndexes": [ 3448 ], "outputIndexes": [ 3449 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 673031200, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3449 ], "outputIndexes": [ 3450 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3450 ], "outputIndexes": [ 668 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 662, 668 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_34_output_0", "outputIndexes": [ 669 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 669, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_63_output_0", "outputIndexes": [ 670 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_7/Mul_1_output_0", "inputIndexes": [ 670 ], "outputIndexes": [ 671 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 698393654, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 671 ], "outputIndexes": [ 3451 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3451 ], "outputIndexes": [ 3452 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear", "inputIndexes": [ 3452 ], "outputIndexes": [ 3453 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 698426422, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3453 ], "outputIndexes": [ 3454 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3454 ], "outputIndexes": [ 672 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 671 ], "main_type": "NONE", "name": "/Shape_98_output_0", "outputIndexes": [ 673 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 673 ], "main_type": "NONE", "name": "Shape2707", "outputIndexes": [ 674 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 673 ], "main_type": "NONE", "name": "Rank2709", "outputIndexes": [ 675 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 675, 675 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2711", "outputIndexes": [ 676 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 676 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2712", "outputIndexes": [ 677 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 676, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2714", "outputIndexes": [ 678 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 678 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2715", "outputIndexes": [ 679 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 674, 677, 679, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2718", "outputIndexes": [ 680 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 680 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2719", "outputIndexes": [ 681 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 681, 681 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2721", "outputIndexes": [ 682 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 673, 682, 676 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_108_output_0", "outputIndexes": [ 683 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 683, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_224_output_0", "outputIndexes": [ 684 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 676 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2729", "outputIndexes": [ 685 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 678 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2732", "outputIndexes": [ 686 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 674, 685, 686, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2735", "outputIndexes": [ 687 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 687 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2736", "outputIndexes": [ 688 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 688 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2737", "outputIndexes": [ 689 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 689, 688 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2738", "outputIndexes": [ 690 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 673, 690, 676 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_109_output_0", "outputIndexes": [ 691 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 691, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_225_output_0", "outputIndexes": [ 692 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 684, 692, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_91_output_0", "outputIndexes": [ 693 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 672, 693 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_64_output_0", "outputIndexes": [ 694 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 694, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_56_output_0", "outputIndexes": [ 695 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 694 ], "main_type": "NONE", "name": "/Shape_100_output_0", "outputIndexes": [ 696 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 696 ], "main_type": "NONE", "name": "Shape2839", "outputIndexes": [ 697 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 696 ], "main_type": "NONE", "name": "Rank2841", "outputIndexes": [ 698 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 698, 698 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2843", "outputIndexes": [ 699 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 699 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2844", "outputIndexes": [ 700 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 699, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2846", "outputIndexes": [ 701 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 701 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2847", "outputIndexes": [ 702 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 697, 700, 702, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2850", "outputIndexes": [ 703 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 703 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2851", "outputIndexes": [ 704 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 704 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2852", "outputIndexes": [ 705 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 705, 704 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2853", "outputIndexes": [ 706 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 696, 706, 699 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_110_output_0", "outputIndexes": [ 707 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 707, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_21_output_0", "outputIndexes": [ 708 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 708, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_231_output_0", "outputIndexes": [ 709 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 694, 709, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_29_output_0", "outputIndexes": [ 710 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 710 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_14_output_0", "outputIndexes": [ 711 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 708, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_230_output_0", "outputIndexes": [ 712 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 694, 20, 712, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_28_output_0", "outputIndexes": [ 713 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 711, 713 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_94_output_0", "outputIndexes": [ 714 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 714, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_57_output_0", "outputIndexes": [ 715 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 695, 715 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_35_output_0", "outputIndexes": [ 716 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 671 ], "outputIndexes": [ 3455 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3455 ], "outputIndexes": [ 3456 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear", "inputIndexes": [ 3456 ], "outputIndexes": [ 3457 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 707880012, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3457 ], "outputIndexes": [ 3458 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3458 ], "outputIndexes": [ 717 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 683, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_226_output_0", "outputIndexes": [ 718 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 691, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_227_output_0", "outputIndexes": [ 719 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 718, 719, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_92_output_0", "outputIndexes": [ 720 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 717, 720 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_65_output_0", "outputIndexes": [ 721 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 721, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_58_output_0", "outputIndexes": [ 722 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 721 ], "main_type": "NONE", "name": "/Shape_101_output_0", "outputIndexes": [ 723 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 723 ], "main_type": "NONE", "name": "Shape2745", "outputIndexes": [ 724 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 723 ], "main_type": "NONE", "name": "Rank2747", "outputIndexes": [ 725 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 725, 725 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2749", "outputIndexes": [ 726 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 726 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2750", "outputIndexes": [ 727 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 726, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2752", "outputIndexes": [ 728 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 728 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2753", "outputIndexes": [ 729 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 724, 727, 729, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2756", "outputIndexes": [ 730 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 730 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2757", "outputIndexes": [ 731 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 731 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2758", "outputIndexes": [ 732 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 732, 731 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2759", "outputIndexes": [ 733 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 723, 733, 726 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_111_output_0", "outputIndexes": [ 734 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 734, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_22_output_0", "outputIndexes": [ 735 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 735, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_233_output_0", "outputIndexes": [ 736 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 721, 736, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_31_output_0", "outputIndexes": [ 737 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 737 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_15_output_0", "outputIndexes": [ 738 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 735, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_232_output_0", "outputIndexes": [ 739 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 721, 20, 739, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_30_output_0", "outputIndexes": [ 740 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 738, 740 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_95_output_0", "outputIndexes": [ 741 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 741, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_59_output_0", "outputIndexes": [ 742 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 722, 742 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_36_output_0", "outputIndexes": [ 743 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 671 ], "outputIndexes": [ 3459 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3459 ], "outputIndexes": [ 3460 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear", "inputIndexes": [ 3460 ], "outputIndexes": [ 3461 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 710243426, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3461 ], "outputIndexes": [ 3462 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3462 ], "outputIndexes": [ 744 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 683, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_228_output_0", "outputIndexes": [ 745 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 691, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_229_output_0", "outputIndexes": [ 746 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 745, 746, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_93_output_0", "outputIndexes": [ 747 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 744, 747 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_66_output_0", "outputIndexes": [ 748 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 716, 743, 748, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_71_output_0", "outputIndexes": [ 749 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 749 ], "outputIndexes": [ 3463 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3463 ], "outputIndexes": [ 3464 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear", "inputIndexes": [ 3464 ], "outputIndexes": [ 3465 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 712606840, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3465 ], "outputIndexes": [ 3466 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3466 ], "outputIndexes": [ 750 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 670, 750 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_38_output_0", "outputIndexes": [ 751 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_7/Mul_1_output_0", "inputIndexes": [ 751 ], "outputIndexes": [ 752 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 722044046, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 752 ], "outputIndexes": [ 3467 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3467 ], "outputIndexes": [ 3468 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear", "inputIndexes": [ 3468 ], "outputIndexes": [ 3469 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 722076814, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3469 ], "outputIndexes": [ 3470 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3470 ], "outputIndexes": [ 753 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 753 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_7/Mul_output_0", "outputIndexes": [ 754 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 752 ], "outputIndexes": [ 3471 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3471 ], "outputIndexes": [ 3472 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear", "inputIndexes": [ 3472 ], "outputIndexes": [ 3473 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 747439268, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3473 ], "outputIndexes": [ 3474 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3474 ], "outputIndexes": [ 755 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 754, 755 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_7/Mul_output_0", "outputIndexes": [ 756 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 756 ], "outputIndexes": [ 3475 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3475 ], "outputIndexes": [ 3476 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear", "inputIndexes": [ 3476 ], "outputIndexes": [ 3477 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 772801722, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3477 ], "outputIndexes": [ 3478 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3478 ], "outputIndexes": [ 757 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 751, 757 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_39_output_0", "outputIndexes": [ 758 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 758, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_72_output_0", "outputIndexes": [ 759 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_8/Mul_1_output_0", "inputIndexes": [ 759 ], "outputIndexes": [ 760 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 798164176, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 760 ], "outputIndexes": [ 3479 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3479 ], "outputIndexes": [ 3480 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear", "inputIndexes": [ 3480 ], "outputIndexes": [ 3481 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 798196944, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3481 ], "outputIndexes": [ 3482 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3482 ], "outputIndexes": [ 761 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 760 ], "main_type": "NONE", "name": "/Shape_112_output_0", "outputIndexes": [ 762 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 762 ], "main_type": "NONE", "name": "Shape3078", "outputIndexes": [ 763 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 762 ], "main_type": "NONE", "name": "Rank3080", "outputIndexes": [ 764 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 764, 764 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3082", "outputIndexes": [ 765 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 765 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3083", "outputIndexes": [ 766 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 765, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3085", "outputIndexes": [ 767 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 767 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3086", "outputIndexes": [ 768 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 763, 766, 768, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3089", "outputIndexes": [ 769 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 769 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3090", "outputIndexes": [ 770 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 770, 770 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3092", "outputIndexes": [ 771 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 762, 771, 765 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_123_output_0", "outputIndexes": [ 772 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 772, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_256_output_0", "outputIndexes": [ 773 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 765 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3100", "outputIndexes": [ 774 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 767 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3103", "outputIndexes": [ 775 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 763, 774, 775, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3106", "outputIndexes": [ 776 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 776 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3107", "outputIndexes": [ 777 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 777 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3108", "outputIndexes": [ 778 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 778, 777 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3109", "outputIndexes": [ 779 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 762, 779, 765 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_124_output_0", "outputIndexes": [ 780 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 780, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_257_output_0", "outputIndexes": [ 781 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 773, 781, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_104_output_0", "outputIndexes": [ 782 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 761, 782 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_73_output_0", "outputIndexes": [ 783 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 783, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_64_output_0", "outputIndexes": [ 784 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 783 ], "main_type": "NONE", "name": "/Shape_114_output_0", "outputIndexes": [ 785 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 785 ], "main_type": "NONE", "name": "Shape3210", "outputIndexes": [ 786 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 785 ], "main_type": "NONE", "name": "Rank3212", "outputIndexes": [ 787 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 787, 787 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3214", "outputIndexes": [ 788 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 788 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3215", "outputIndexes": [ 789 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 788, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3217", "outputIndexes": [ 790 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 790 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3218", "outputIndexes": [ 791 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 786, 789, 791, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3221", "outputIndexes": [ 792 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 792 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3222", "outputIndexes": [ 793 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 793 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3223", "outputIndexes": [ 794 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 794, 793 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3224", "outputIndexes": [ 795 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 785, 795, 788 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_125_output_0", "outputIndexes": [ 796 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 796, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_24_output_0", "outputIndexes": [ 797 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 797, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_263_output_0", "outputIndexes": [ 798 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 783, 798, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_33_output_0", "outputIndexes": [ 799 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 799 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_16_output_0", "outputIndexes": [ 800 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 797, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_262_output_0", "outputIndexes": [ 801 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 783, 20, 801, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_32_output_0", "outputIndexes": [ 802 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 800, 802 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_107_output_0", "outputIndexes": [ 803 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 803, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_65_output_0", "outputIndexes": [ 804 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 784, 804 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_40_output_0", "outputIndexes": [ 805 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 760 ], "outputIndexes": [ 3483 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3483 ], "outputIndexes": [ 3484 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear", "inputIndexes": [ 3484 ], "outputIndexes": [ 3485 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 807650534, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3485 ], "outputIndexes": [ 3486 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3486 ], "outputIndexes": [ 806 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 772, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_258_output_0", "outputIndexes": [ 807 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 780, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_259_output_0", "outputIndexes": [ 808 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 807, 808, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_105_output_0", "outputIndexes": [ 809 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 806, 809 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_74_output_0", "outputIndexes": [ 810 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 810, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_66_output_0", "outputIndexes": [ 811 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 810 ], "main_type": "NONE", "name": "/Shape_115_output_0", "outputIndexes": [ 812 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 812 ], "main_type": "NONE", "name": "Shape3116", "outputIndexes": [ 813 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 812 ], "main_type": "NONE", "name": "Rank3118", "outputIndexes": [ 814 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 814, 814 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3120", "outputIndexes": [ 815 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 815 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3121", "outputIndexes": [ 816 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 815, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3123", "outputIndexes": [ 817 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 817 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3124", "outputIndexes": [ 818 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 813, 816, 818, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3127", "outputIndexes": [ 819 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 819 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3128", "outputIndexes": [ 820 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 820 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3129", "outputIndexes": [ 821 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 821, 820 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3130", "outputIndexes": [ 822 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 812, 822, 815 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_126_output_0", "outputIndexes": [ 823 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 823, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_25_output_0", "outputIndexes": [ 824 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 824, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_265_output_0", "outputIndexes": [ 825 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 810, 825, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_35_output_0", "outputIndexes": [ 826 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 826 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_17_output_0", "outputIndexes": [ 827 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 824, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_264_output_0", "outputIndexes": [ 828 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 810, 20, 828, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_34_output_0", "outputIndexes": [ 829 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 827, 829 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_108_output_0", "outputIndexes": [ 830 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 830, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_67_output_0", "outputIndexes": [ 831 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 811, 831 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_41_output_0", "outputIndexes": [ 832 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 760 ], "outputIndexes": [ 3487 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3487 ], "outputIndexes": [ 3488 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear", "inputIndexes": [ 3488 ], "outputIndexes": [ 3489 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 810013948, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3489 ], "outputIndexes": [ 3490 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3490 ], "outputIndexes": [ 833 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 772, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_260_output_0", "outputIndexes": [ 834 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 780, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_261_output_0", "outputIndexes": [ 835 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 834, 835, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_106_output_0", "outputIndexes": [ 836 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 833, 836 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_75_output_0", "outputIndexes": [ 837 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 805, 832, 837, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_80_output_0", "outputIndexes": [ 838 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 838 ], "outputIndexes": [ 3491 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3491 ], "outputIndexes": [ 3492 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear", "inputIndexes": [ 3492 ], "outputIndexes": [ 3493 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 812377362, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3493 ], "outputIndexes": [ 3494 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3494 ], "outputIndexes": [ 839 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 759, 839 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_43_output_0", "outputIndexes": [ 840 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_8/Mul_1_output_0", "inputIndexes": [ 840 ], "outputIndexes": [ 841 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 821814568, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 841 ], "outputIndexes": [ 3495 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3495 ], "outputIndexes": [ 3496 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear", "inputIndexes": [ 3496 ], "outputIndexes": [ 3497 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 821847336, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3497 ], "outputIndexes": [ 3498 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3498 ], "outputIndexes": [ 842 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 842 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_8/Mul_output_0", "outputIndexes": [ 843 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 841 ], "outputIndexes": [ 3499 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3499 ], "outputIndexes": [ 3500 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear", "inputIndexes": [ 3500 ], "outputIndexes": [ 3501 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 847209790, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3501 ], "outputIndexes": [ 3502 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3502 ], "outputIndexes": [ 844 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 843, 844 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_8/Mul_output_0", "outputIndexes": [ 845 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 845 ], "outputIndexes": [ 3503 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3503 ], "outputIndexes": [ 3504 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear", "inputIndexes": [ 3504 ], "outputIndexes": [ 3505 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 872572244, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3505 ], "outputIndexes": [ 3506 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3506 ], "outputIndexes": [ 846 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 840, 846 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_44_output_0", "outputIndexes": [ 847 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 847, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_81_output_0", "outputIndexes": [ 848 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_9/Mul_1_output_0", "inputIndexes": [ 848 ], "outputIndexes": [ 849 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 897934698, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 849 ], "outputIndexes": [ 3507 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3507 ], "outputIndexes": [ 3508 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear", "inputIndexes": [ 3508 ], "outputIndexes": [ 3509 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 897967466, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3509 ], "outputIndexes": [ 3510 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3510 ], "outputIndexes": [ 850 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 849 ], "main_type": "NONE", "name": "/Shape_126_output_0", "outputIndexes": [ 851 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 851 ], "main_type": "NONE", "name": "Shape3449", "outputIndexes": [ 852 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 851 ], "main_type": "NONE", "name": "Rank3451", "outputIndexes": [ 853 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 853, 853 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3453", "outputIndexes": [ 854 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 854 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3454", "outputIndexes": [ 855 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 854, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3456", "outputIndexes": [ 856 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 856 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3457", "outputIndexes": [ 857 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 852, 855, 857, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3460", "outputIndexes": [ 858 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 858 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3461", "outputIndexes": [ 859 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 859, 859 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3463", "outputIndexes": [ 860 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 851, 860, 854 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_138_output_0", "outputIndexes": [ 861 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 861, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_288_output_0", "outputIndexes": [ 862 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 854 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3471", "outputIndexes": [ 863 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 856 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3474", "outputIndexes": [ 864 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 852, 863, 864, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3477", "outputIndexes": [ 865 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 865 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3478", "outputIndexes": [ 866 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 866 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3479", "outputIndexes": [ 867 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 867, 866 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3480", "outputIndexes": [ 868 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 851, 868, 854 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_139_output_0", "outputIndexes": [ 869 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 869, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_289_output_0", "outputIndexes": [ 870 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 862, 870, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_117_output_0", "outputIndexes": [ 871 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 850, 871 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_82_output_0", "outputIndexes": [ 872 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 872, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_72_output_0", "outputIndexes": [ 873 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 872 ], "main_type": "NONE", "name": "/Shape_128_output_0", "outputIndexes": [ 874 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 874 ], "main_type": "NONE", "name": "Shape3581", "outputIndexes": [ 875 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 874 ], "main_type": "NONE", "name": "Rank3583", "outputIndexes": [ 876 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 876, 876 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3585", "outputIndexes": [ 877 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 877 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3586", "outputIndexes": [ 878 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 877, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3588", "outputIndexes": [ 879 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 879 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3589", "outputIndexes": [ 880 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 875, 878, 880, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3592", "outputIndexes": [ 881 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 881 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3593", "outputIndexes": [ 882 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 882 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3594", "outputIndexes": [ 883 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 883, 882 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3595", "outputIndexes": [ 884 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 874, 884, 877 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_140_output_0", "outputIndexes": [ 885 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 885, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_27_output_0", "outputIndexes": [ 886 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 886, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_295_output_0", "outputIndexes": [ 887 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 872, 887, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_37_output_0", "outputIndexes": [ 888 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 888 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_18_output_0", "outputIndexes": [ 889 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 886, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_294_output_0", "outputIndexes": [ 890 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 872, 20, 890, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_36_output_0", "outputIndexes": [ 891 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 889, 891 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_120_output_0", "outputIndexes": [ 892 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 892, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_73_output_0", "outputIndexes": [ 893 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 873, 893 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_45_output_0", "outputIndexes": [ 894 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 849 ], "outputIndexes": [ 3511 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3511 ], "outputIndexes": [ 3512 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear", "inputIndexes": [ 3512 ], "outputIndexes": [ 3513 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 907421056, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3513 ], "outputIndexes": [ 3514 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3514 ], "outputIndexes": [ 895 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 861, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_290_output_0", "outputIndexes": [ 896 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 869, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_291_output_0", "outputIndexes": [ 897 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 896, 897, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_118_output_0", "outputIndexes": [ 898 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 895, 898 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_83_output_0", "outputIndexes": [ 899 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 899, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_74_output_0", "outputIndexes": [ 900 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 899 ], "main_type": "NONE", "name": "/Shape_129_output_0", "outputIndexes": [ 901 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 901 ], "main_type": "NONE", "name": "Shape3487", "outputIndexes": [ 902 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 901 ], "main_type": "NONE", "name": "Rank3489", "outputIndexes": [ 903 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 903, 903 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3491", "outputIndexes": [ 904 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 904 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3492", "outputIndexes": [ 905 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 904, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3494", "outputIndexes": [ 906 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 906 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3495", "outputIndexes": [ 907 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 902, 905, 907, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3498", "outputIndexes": [ 908 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 908 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3499", "outputIndexes": [ 909 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 909 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3500", "outputIndexes": [ 910 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 910, 909 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3501", "outputIndexes": [ 911 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 901, 911, 904 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_141_output_0", "outputIndexes": [ 912 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 912, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_28_output_0", "outputIndexes": [ 913 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 913, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_297_output_0", "outputIndexes": [ 914 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 899, 914, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_39_output_0", "outputIndexes": [ 915 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 915 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_19_output_0", "outputIndexes": [ 916 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 913, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_296_output_0", "outputIndexes": [ 917 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 899, 20, 917, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_38_output_0", "outputIndexes": [ 918 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 916, 918 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_121_output_0", "outputIndexes": [ 919 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 919, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_75_output_0", "outputIndexes": [ 920 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 900, 920 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_46_output_0", "outputIndexes": [ 921 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 849 ], "outputIndexes": [ 3515 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3515 ], "outputIndexes": [ 3516 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear", "inputIndexes": [ 3516 ], "outputIndexes": [ 3517 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 909784470, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3517 ], "outputIndexes": [ 3518 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3518 ], "outputIndexes": [ 922 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 861, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_292_output_0", "outputIndexes": [ 923 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 869, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_293_output_0", "outputIndexes": [ 924 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 923, 924, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_119_output_0", "outputIndexes": [ 925 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 922, 925 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_84_output_0", "outputIndexes": [ 926 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 894, 921, 926, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_89_output_0", "outputIndexes": [ 927 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 927 ], "outputIndexes": [ 3519 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3519 ], "outputIndexes": [ 3520 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear", "inputIndexes": [ 3520 ], "outputIndexes": [ 3521 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 912147884, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3521 ], "outputIndexes": [ 3522 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3522 ], "outputIndexes": [ 928 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 848, 928 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_48_output_0", "outputIndexes": [ 929 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_9/Mul_1_output_0", "inputIndexes": [ 929 ], "outputIndexes": [ 930 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 921585090, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 930 ], "outputIndexes": [ 3523 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3523 ], "outputIndexes": [ 3524 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear", "inputIndexes": [ 3524 ], "outputIndexes": [ 3525 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 921617858, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3525 ], "outputIndexes": [ 3526 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3526 ], "outputIndexes": [ 931 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 931 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_9/Mul_output_0", "outputIndexes": [ 932 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 930 ], "outputIndexes": [ 3527 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3527 ], "outputIndexes": [ 3528 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear", "inputIndexes": [ 3528 ], "outputIndexes": [ 3529 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 946980312, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3529 ], "outputIndexes": [ 3530 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3530 ], "outputIndexes": [ 933 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 932, 933 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_9/Mul_output_0", "outputIndexes": [ 934 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 934 ], "outputIndexes": [ 3531 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3531 ], "outputIndexes": [ 3532 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear", "inputIndexes": [ 3532 ], "outputIndexes": [ 3533 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 972342766, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3533 ], "outputIndexes": [ 3534 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3534 ], "outputIndexes": [ 935 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 929, 935 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_49_output_0", "outputIndexes": [ 936 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 936, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_90_output_0", "outputIndexes": [ 937 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_10/Mul_1_output_0", "inputIndexes": [ 937 ], "outputIndexes": [ 938 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 997705220, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 938 ], "outputIndexes": [ 3535 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3535 ], "outputIndexes": [ 3536 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear", "inputIndexes": [ 3536 ], "outputIndexes": [ 3537 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 997737988, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3537 ], "outputIndexes": [ 3538 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3538 ], "outputIndexes": [ 939 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 938 ], "main_type": "NONE", "name": "/Shape_140_output_0", "outputIndexes": [ 940 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 940 ], "main_type": "NONE", "name": "Shape3820", "outputIndexes": [ 941 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 940 ], "main_type": "NONE", "name": "Rank3822", "outputIndexes": [ 942 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 942, 942 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3824", "outputIndexes": [ 943 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 943 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3825", "outputIndexes": [ 944 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 943, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3827", "outputIndexes": [ 945 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 945 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3828", "outputIndexes": [ 946 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 941, 944, 946, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3831", "outputIndexes": [ 947 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 947 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3832", "outputIndexes": [ 948 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 948, 948 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3834", "outputIndexes": [ 949 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 940, 949, 943 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_153_output_0", "outputIndexes": [ 950 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 950, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_320_output_0", "outputIndexes": [ 951 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 943 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3842", "outputIndexes": [ 952 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 945 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3845", "outputIndexes": [ 953 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 941, 952, 953, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3848", "outputIndexes": [ 954 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 954 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3849", "outputIndexes": [ 955 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 955 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3850", "outputIndexes": [ 956 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 956, 955 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3851", "outputIndexes": [ 957 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 940, 957, 943 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_154_output_0", "outputIndexes": [ 958 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 958, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_321_output_0", "outputIndexes": [ 959 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 951, 959, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_130_output_0", "outputIndexes": [ 960 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 939, 960 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_91_output_0", "outputIndexes": [ 961 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 961, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_80_output_0", "outputIndexes": [ 962 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 961 ], "main_type": "NONE", "name": "/Shape_142_output_0", "outputIndexes": [ 963 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 963 ], "main_type": "NONE", "name": "Shape3952", "outputIndexes": [ 964 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 963 ], "main_type": "NONE", "name": "Rank3954", "outputIndexes": [ 965 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 965, 965 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3956", "outputIndexes": [ 966 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 966 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3957", "outputIndexes": [ 967 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 966, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3959", "outputIndexes": [ 968 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 968 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3960", "outputIndexes": [ 969 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 964, 967, 969, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3963", "outputIndexes": [ 970 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 970 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3964", "outputIndexes": [ 971 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 971 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3965", "outputIndexes": [ 972 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 972, 971 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3966", "outputIndexes": [ 973 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 963, 973, 966 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_155_output_0", "outputIndexes": [ 974 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 974, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_30_output_0", "outputIndexes": [ 975 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 975, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_327_output_0", "outputIndexes": [ 976 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 961, 976, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_41_output_0", "outputIndexes": [ 977 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 977 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_20_output_0", "outputIndexes": [ 978 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 975, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_326_output_0", "outputIndexes": [ 979 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 961, 20, 979, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_40_output_0", "outputIndexes": [ 980 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 978, 980 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_133_output_0", "outputIndexes": [ 981 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 981, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_81_output_0", "outputIndexes": [ 982 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 962, 982 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_50_output_0", "outputIndexes": [ 983 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 938 ], "outputIndexes": [ 3539 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3539 ], "outputIndexes": [ 3540 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear", "inputIndexes": [ 3540 ], "outputIndexes": [ 3541 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1007191578, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3541 ], "outputIndexes": [ 3542 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3542 ], "outputIndexes": [ 984 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 950, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_322_output_0", "outputIndexes": [ 985 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 958, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_323_output_0", "outputIndexes": [ 986 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 985, 986, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_131_output_0", "outputIndexes": [ 987 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 984, 987 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_92_output_0", "outputIndexes": [ 988 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 988, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_82_output_0", "outputIndexes": [ 989 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 988 ], "main_type": "NONE", "name": "/Shape_143_output_0", "outputIndexes": [ 990 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 990 ], "main_type": "NONE", "name": "Shape3858", "outputIndexes": [ 991 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 990 ], "main_type": "NONE", "name": "Rank3860", "outputIndexes": [ 992 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 992, 992 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3862", "outputIndexes": [ 993 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 993 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3863", "outputIndexes": [ 994 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 993, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3865", "outputIndexes": [ 995 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 995 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3866", "outputIndexes": [ 996 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 991, 994, 996, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3869", "outputIndexes": [ 997 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 997 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3870", "outputIndexes": [ 998 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 998 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3871", "outputIndexes": [ 999 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 999, 998 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3872", "outputIndexes": [ 1000 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 990, 1000, 993 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_156_output_0", "outputIndexes": [ 1001 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1001, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_31_output_0", "outputIndexes": [ 1002 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1002, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_329_output_0", "outputIndexes": [ 1003 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 988, 1003, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_43_output_0", "outputIndexes": [ 1004 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1004 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_21_output_0", "outputIndexes": [ 1005 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1002, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_328_output_0", "outputIndexes": [ 1006 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 988, 20, 1006, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_42_output_0", "outputIndexes": [ 1007 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1005, 1007 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_134_output_0", "outputIndexes": [ 1008 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1008, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_83_output_0", "outputIndexes": [ 1009 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 989, 1009 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_51_output_0", "outputIndexes": [ 1010 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 938 ], "outputIndexes": [ 3543 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3543 ], "outputIndexes": [ 3544 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear", "inputIndexes": [ 3544 ], "outputIndexes": [ 3545 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1009554992, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3545 ], "outputIndexes": [ 3546 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3546 ], "outputIndexes": [ 1011 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 950, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_324_output_0", "outputIndexes": [ 1012 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 958, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_325_output_0", "outputIndexes": [ 1013 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1012, 1013, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_132_output_0", "outputIndexes": [ 1014 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1011, 1014 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_93_output_0", "outputIndexes": [ 1015 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 983, 1010, 1015, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_98_output_0", "outputIndexes": [ 1016 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1016 ], "outputIndexes": [ 3547 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3547 ], "outputIndexes": [ 3548 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear", "inputIndexes": [ 3548 ], "outputIndexes": [ 3549 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1011918406, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3549 ], "outputIndexes": [ 3550 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3550 ], "outputIndexes": [ 1017 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 937, 1017 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_53_output_0", "outputIndexes": [ 1018 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_10/Mul_1_output_0", "inputIndexes": [ 1018 ], "outputIndexes": [ 1019 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1021355612, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1019 ], "outputIndexes": [ 3551 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3551 ], "outputIndexes": [ 3552 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear", "inputIndexes": [ 3552 ], "outputIndexes": [ 3553 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1021388380, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3553 ], "outputIndexes": [ 3554 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3554 ], "outputIndexes": [ 1020 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1020 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_10/Mul_output_0", "outputIndexes": [ 1021 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1019 ], "outputIndexes": [ 3555 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3555 ], "outputIndexes": [ 3556 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear", "inputIndexes": [ 3556 ], "outputIndexes": [ 3557 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1046750834, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3557 ], "outputIndexes": [ 3558 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3558 ], "outputIndexes": [ 1022 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1021, 1022 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_10/Mul_output_0", "outputIndexes": [ 1023 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1023 ], "outputIndexes": [ 3559 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3559 ], "outputIndexes": [ 3560 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear", "inputIndexes": [ 3560 ], "outputIndexes": [ 3561 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1072113288, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3561 ], "outputIndexes": [ 3562 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3562 ], "outputIndexes": [ 1024 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1018, 1024 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_54_output_0", "outputIndexes": [ 1025 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1025, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_99_output_0", "outputIndexes": [ 1026 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_11/Mul_1_output_0", "inputIndexes": [ 1026 ], "outputIndexes": [ 1027 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1097475742, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1027 ], "outputIndexes": [ 3563 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3563 ], "outputIndexes": [ 3564 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear", "inputIndexes": [ 3564 ], "outputIndexes": [ 3565 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1097508510, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3565 ], "outputIndexes": [ 3566 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3566 ], "outputIndexes": [ 1028 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1027 ], "main_type": "NONE", "name": "/Shape_154_output_0", "outputIndexes": [ 1029 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1029 ], "main_type": "NONE", "name": "Shape4191", "outputIndexes": [ 1030 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1029 ], "main_type": "NONE", "name": "Rank4193", "outputIndexes": [ 1031 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1031, 1031 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4195", "outputIndexes": [ 1032 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1032 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4196", "outputIndexes": [ 1033 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1032, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4198", "outputIndexes": [ 1034 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1034 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4199", "outputIndexes": [ 1035 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1030, 1033, 1035, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4202", "outputIndexes": [ 1036 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1036 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4203", "outputIndexes": [ 1037 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1037, 1037 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4205", "outputIndexes": [ 1038 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1029, 1038, 1032 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_168_output_0", "outputIndexes": [ 1039 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1039, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_352_output_0", "outputIndexes": [ 1040 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1032 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4213", "outputIndexes": [ 1041 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1034 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4216", "outputIndexes": [ 1042 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1030, 1041, 1042, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4219", "outputIndexes": [ 1043 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1043 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4220", "outputIndexes": [ 1044 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1044 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4221", "outputIndexes": [ 1045 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1045, 1044 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4222", "outputIndexes": [ 1046 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1029, 1046, 1032 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_169_output_0", "outputIndexes": [ 1047 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1047, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_353_output_0", "outputIndexes": [ 1048 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1040, 1048, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_143_output_0", "outputIndexes": [ 1049 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1028, 1049 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_100_output_0", "outputIndexes": [ 1050 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1050, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_88_output_0", "outputIndexes": [ 1051 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1050 ], "main_type": "NONE", "name": "/Shape_156_output_0", "outputIndexes": [ 1052 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1052 ], "main_type": "NONE", "name": "Shape4323", "outputIndexes": [ 1053 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1052 ], "main_type": "NONE", "name": "Rank4325", "outputIndexes": [ 1054 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1054, 1054 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4327", "outputIndexes": [ 1055 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1055 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4328", "outputIndexes": [ 1056 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1055, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4330", "outputIndexes": [ 1057 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1057 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4331", "outputIndexes": [ 1058 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1053, 1056, 1058, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4334", "outputIndexes": [ 1059 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1059 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4335", "outputIndexes": [ 1060 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1060 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4336", "outputIndexes": [ 1061 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1061, 1060 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4337", "outputIndexes": [ 1062 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1052, 1062, 1055 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_170_output_0", "outputIndexes": [ 1063 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1063, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_33_output_0", "outputIndexes": [ 1064 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1064, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_359_output_0", "outputIndexes": [ 1065 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1050, 1065, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_45_output_0", "outputIndexes": [ 1066 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1066 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_22_output_0", "outputIndexes": [ 1067 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1064, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_358_output_0", "outputIndexes": [ 1068 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1050, 20, 1068, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_44_output_0", "outputIndexes": [ 1069 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1067, 1069 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_146_output_0", "outputIndexes": [ 1070 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1070, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_89_output_0", "outputIndexes": [ 1071 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1051, 1071 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_55_output_0", "outputIndexes": [ 1072 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1027 ], "outputIndexes": [ 3567 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3567 ], "outputIndexes": [ 3568 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear", "inputIndexes": [ 3568 ], "outputIndexes": [ 3569 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1106962100, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3569 ], "outputIndexes": [ 3570 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3570 ], "outputIndexes": [ 1073 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1039, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_354_output_0", "outputIndexes": [ 1074 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1047, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_355_output_0", "outputIndexes": [ 1075 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1074, 1075, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_144_output_0", "outputIndexes": [ 1076 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1073, 1076 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_101_output_0", "outputIndexes": [ 1077 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1077, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_90_output_0", "outputIndexes": [ 1078 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1077 ], "main_type": "NONE", "name": "/Shape_157_output_0", "outputIndexes": [ 1079 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1079 ], "main_type": "NONE", "name": "Shape4229", "outputIndexes": [ 1080 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1079 ], "main_type": "NONE", "name": "Rank4231", "outputIndexes": [ 1081 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1081, 1081 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4233", "outputIndexes": [ 1082 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1082 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4234", "outputIndexes": [ 1083 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1082, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4236", "outputIndexes": [ 1084 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1084 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4237", "outputIndexes": [ 1085 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1080, 1083, 1085, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4240", "outputIndexes": [ 1086 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1086 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4241", "outputIndexes": [ 1087 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1087 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4242", "outputIndexes": [ 1088 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1088, 1087 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4243", "outputIndexes": [ 1089 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1079, 1089, 1082 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_171_output_0", "outputIndexes": [ 1090 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1090, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_34_output_0", "outputIndexes": [ 1091 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1091, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_361_output_0", "outputIndexes": [ 1092 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1077, 1092, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_47_output_0", "outputIndexes": [ 1093 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1093 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_23_output_0", "outputIndexes": [ 1094 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1091, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_360_output_0", "outputIndexes": [ 1095 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1077, 20, 1095, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_46_output_0", "outputIndexes": [ 1096 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1094, 1096 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_147_output_0", "outputIndexes": [ 1097 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1097, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_91_output_0", "outputIndexes": [ 1098 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1078, 1098 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_56_output_0", "outputIndexes": [ 1099 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1027 ], "outputIndexes": [ 3571 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3571 ], "outputIndexes": [ 3572 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear", "inputIndexes": [ 3572 ], "outputIndexes": [ 3573 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1109325514, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3573 ], "outputIndexes": [ 3574 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3574 ], "outputIndexes": [ 1100 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1039, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_356_output_0", "outputIndexes": [ 1101 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1047, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_357_output_0", "outputIndexes": [ 1102 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1101, 1102, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_145_output_0", "outputIndexes": [ 1103 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1100, 1103 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_102_output_0", "outputIndexes": [ 1104 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1072, 1099, 1104, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_107_output_0", "outputIndexes": [ 1105 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1105 ], "outputIndexes": [ 3575 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3575 ], "outputIndexes": [ 3576 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear", "inputIndexes": [ 3576 ], "outputIndexes": [ 3577 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1111688928, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3577 ], "outputIndexes": [ 3578 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3578 ], "outputIndexes": [ 1106 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1026, 1106 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_58_output_0", "outputIndexes": [ 1107 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_11/Mul_1_output_0", "inputIndexes": [ 1107 ], "outputIndexes": [ 1108 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1121126134, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1108 ], "outputIndexes": [ 3579 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3579 ], "outputIndexes": [ 3580 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear", "inputIndexes": [ 3580 ], "outputIndexes": [ 3581 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1121158902, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3581 ], "outputIndexes": [ 3582 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3582 ], "outputIndexes": [ 1109 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1109 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_11/Mul_output_0", "outputIndexes": [ 1110 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1108 ], "outputIndexes": [ 3583 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3583 ], "outputIndexes": [ 3584 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear", "inputIndexes": [ 3584 ], "outputIndexes": [ 3585 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1146521356, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3585 ], "outputIndexes": [ 3586 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3586 ], "outputIndexes": [ 1111 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1110, 1111 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_11/Mul_output_0", "outputIndexes": [ 1112 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1112 ], "outputIndexes": [ 3587 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3587 ], "outputIndexes": [ 3588 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear", "inputIndexes": [ 3588 ], "outputIndexes": [ 3589 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1171883810, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3589 ], "outputIndexes": [ 3590 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3590 ], "outputIndexes": [ 1113 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1107, 1113 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_59_output_0", "outputIndexes": [ 1114 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1114, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_108_output_0", "outputIndexes": [ 1115 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_12/Mul_1_output_0", "inputIndexes": [ 1115 ], "outputIndexes": [ 1116 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1197246264, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1116 ], "outputIndexes": [ 3591 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3591 ], "outputIndexes": [ 3592 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear", "inputIndexes": [ 3592 ], "outputIndexes": [ 3593 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1197279032, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3593 ], "outputIndexes": [ 3594 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3594 ], "outputIndexes": [ 1117 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1116 ], "main_type": "NONE", "name": "/Shape_168_output_0", "outputIndexes": [ 1118 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1118 ], "main_type": "NONE", "name": "Shape4562", "outputIndexes": [ 1119 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1118 ], "main_type": "NONE", "name": "Rank4564", "outputIndexes": [ 1120 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1120, 1120 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4566", "outputIndexes": [ 1121 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1121 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4567", "outputIndexes": [ 1122 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1121, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4569", "outputIndexes": [ 1123 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1123 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4570", "outputIndexes": [ 1124 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1119, 1122, 1124, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4573", "outputIndexes": [ 1125 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1125 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4574", "outputIndexes": [ 1126 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1126, 1126 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4576", "outputIndexes": [ 1127 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1118, 1127, 1121 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_183_output_0", "outputIndexes": [ 1128 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1128, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_384_output_0", "outputIndexes": [ 1129 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1121 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4584", "outputIndexes": [ 1130 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1123 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4587", "outputIndexes": [ 1131 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1119, 1130, 1131, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4590", "outputIndexes": [ 1132 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1132 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4591", "outputIndexes": [ 1133 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1133 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4592", "outputIndexes": [ 1134 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1134, 1133 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4593", "outputIndexes": [ 1135 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1118, 1135, 1121 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_184_output_0", "outputIndexes": [ 1136 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1136, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_385_output_0", "outputIndexes": [ 1137 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1129, 1137, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_156_output_0", "outputIndexes": [ 1138 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1117, 1138 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_109_output_0", "outputIndexes": [ 1139 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1139, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_96_output_0", "outputIndexes": [ 1140 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1139 ], "main_type": "NONE", "name": "/Shape_170_output_0", "outputIndexes": [ 1141 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1141 ], "main_type": "NONE", "name": "Shape4694", "outputIndexes": [ 1142 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1141 ], "main_type": "NONE", "name": "Rank4696", "outputIndexes": [ 1143 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1143, 1143 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4698", "outputIndexes": [ 1144 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1144 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4699", "outputIndexes": [ 1145 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1144, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4701", "outputIndexes": [ 1146 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1146 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4702", "outputIndexes": [ 1147 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1142, 1145, 1147, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4705", "outputIndexes": [ 1148 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1148 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4706", "outputIndexes": [ 1149 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1149 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4707", "outputIndexes": [ 1150 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1150, 1149 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4708", "outputIndexes": [ 1151 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1141, 1151, 1144 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_185_output_0", "outputIndexes": [ 1152 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1152, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_36_output_0", "outputIndexes": [ 1153 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1153, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_391_output_0", "outputIndexes": [ 1154 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1139, 1154, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_49_output_0", "outputIndexes": [ 1155 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1155 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_24_output_0", "outputIndexes": [ 1156 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1153, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_390_output_0", "outputIndexes": [ 1157 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1139, 20, 1157, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_48_output_0", "outputIndexes": [ 1158 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1156, 1158 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_159_output_0", "outputIndexes": [ 1159 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1159, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_97_output_0", "outputIndexes": [ 1160 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1140, 1160 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_60_output_0", "outputIndexes": [ 1161 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1116 ], "outputIndexes": [ 3595 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3595 ], "outputIndexes": [ 3596 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear", "inputIndexes": [ 3596 ], "outputIndexes": [ 3597 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1206732622, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3597 ], "outputIndexes": [ 3598 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3598 ], "outputIndexes": [ 1162 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1128, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_386_output_0", "outputIndexes": [ 1163 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1136, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_387_output_0", "outputIndexes": [ 1164 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1163, 1164, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_157_output_0", "outputIndexes": [ 1165 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1162, 1165 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_110_output_0", "outputIndexes": [ 1166 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1166, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_98_output_0", "outputIndexes": [ 1167 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1166 ], "main_type": "NONE", "name": "/Shape_171_output_0", "outputIndexes": [ 1168 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1168 ], "main_type": "NONE", "name": "Shape4600", "outputIndexes": [ 1169 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1168 ], "main_type": "NONE", "name": "Rank4602", "outputIndexes": [ 1170 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1170, 1170 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4604", "outputIndexes": [ 1171 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1171 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4605", "outputIndexes": [ 1172 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1171, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4607", "outputIndexes": [ 1173 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1173 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4608", "outputIndexes": [ 1174 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1169, 1172, 1174, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4611", "outputIndexes": [ 1175 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1175 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4612", "outputIndexes": [ 1176 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1176 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4613", "outputIndexes": [ 1177 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1177, 1176 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4614", "outputIndexes": [ 1178 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1168, 1178, 1171 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_186_output_0", "outputIndexes": [ 1179 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1179, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_37_output_0", "outputIndexes": [ 1180 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1180, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_393_output_0", "outputIndexes": [ 1181 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1166, 1181, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_51_output_0", "outputIndexes": [ 1182 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1182 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_25_output_0", "outputIndexes": [ 1183 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1180, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_392_output_0", "outputIndexes": [ 1184 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1166, 20, 1184, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_50_output_0", "outputIndexes": [ 1185 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1183, 1185 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_160_output_0", "outputIndexes": [ 1186 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1186, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_99_output_0", "outputIndexes": [ 1187 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1167, 1187 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_61_output_0", "outputIndexes": [ 1188 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1116 ], "outputIndexes": [ 3599 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3599 ], "outputIndexes": [ 3600 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear", "inputIndexes": [ 3600 ], "outputIndexes": [ 3601 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1209096036, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3601 ], "outputIndexes": [ 3602 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3602 ], "outputIndexes": [ 1189 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1128, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_388_output_0", "outputIndexes": [ 1190 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1136, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_389_output_0", "outputIndexes": [ 1191 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1190, 1191, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_158_output_0", "outputIndexes": [ 1192 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1189, 1192 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_111_output_0", "outputIndexes": [ 1193 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1161, 1188, 1193, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_116_output_0", "outputIndexes": [ 1194 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1194 ], "outputIndexes": [ 3603 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3603 ], "outputIndexes": [ 3604 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear", "inputIndexes": [ 3604 ], "outputIndexes": [ 3605 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1211459450, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3605 ], "outputIndexes": [ 3606 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3606 ], "outputIndexes": [ 1195 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1115, 1195 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_63_output_0", "outputIndexes": [ 1196 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_12/Mul_1_output_0", "inputIndexes": [ 1196 ], "outputIndexes": [ 1197 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1220896656, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1197 ], "outputIndexes": [ 3607 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3607 ], "outputIndexes": [ 3608 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear", "inputIndexes": [ 3608 ], "outputIndexes": [ 3609 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1220929424, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3609 ], "outputIndexes": [ 3610 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3610 ], "outputIndexes": [ 1198 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1198 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_12/Mul_output_0", "outputIndexes": [ 1199 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1197 ], "outputIndexes": [ 3611 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3611 ], "outputIndexes": [ 3612 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear", "inputIndexes": [ 3612 ], "outputIndexes": [ 3613 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1246291878, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3613 ], "outputIndexes": [ 3614 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3614 ], "outputIndexes": [ 1200 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1199, 1200 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_12/Mul_output_0", "outputIndexes": [ 1201 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1201 ], "outputIndexes": [ 3615 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3615 ], "outputIndexes": [ 3616 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear", "inputIndexes": [ 3616 ], "outputIndexes": [ 3617 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1271654332, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3617 ], "outputIndexes": [ 3618 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3618 ], "outputIndexes": [ 1202 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1196, 1202 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_64_output_0", "outputIndexes": [ 1203 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1203, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_117_output_0", "outputIndexes": [ 1204 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_13/Mul_1_output_0", "inputIndexes": [ 1204 ], "outputIndexes": [ 1205 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1297016786, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1205 ], "outputIndexes": [ 3619 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3619 ], "outputIndexes": [ 3620 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear", "inputIndexes": [ 3620 ], "outputIndexes": [ 3621 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1297049554, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3621 ], "outputIndexes": [ 3622 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3622 ], "outputIndexes": [ 1206 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1205 ], "main_type": "NONE", "name": "/Shape_182_output_0", "outputIndexes": [ 1207 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1207 ], "main_type": "NONE", "name": "Shape4933", "outputIndexes": [ 1208 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1207 ], "main_type": "NONE", "name": "Rank4935", "outputIndexes": [ 1209 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1209, 1209 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4937", "outputIndexes": [ 1210 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1210 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4938", "outputIndexes": [ 1211 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1210, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4940", "outputIndexes": [ 1212 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1212 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4941", "outputIndexes": [ 1213 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1208, 1211, 1213, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4944", "outputIndexes": [ 1214 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1214 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4945", "outputIndexes": [ 1215 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1215, 1215 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4947", "outputIndexes": [ 1216 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1207, 1216, 1210 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_198_output_0", "outputIndexes": [ 1217 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1217, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_416_output_0", "outputIndexes": [ 1218 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1210 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4955", "outputIndexes": [ 1219 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1212 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4958", "outputIndexes": [ 1220 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1208, 1219, 1220, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4961", "outputIndexes": [ 1221 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1221 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4962", "outputIndexes": [ 1222 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1222 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4963", "outputIndexes": [ 1223 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1223, 1222 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4964", "outputIndexes": [ 1224 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1207, 1224, 1210 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_199_output_0", "outputIndexes": [ 1225 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1225, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_417_output_0", "outputIndexes": [ 1226 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1218, 1226, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_169_output_0", "outputIndexes": [ 1227 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1206, 1227 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_118_output_0", "outputIndexes": [ 1228 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1228, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_104_output_0", "outputIndexes": [ 1229 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1228 ], "main_type": "NONE", "name": "/Shape_184_output_0", "outputIndexes": [ 1230 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1230 ], "main_type": "NONE", "name": "Shape5065", "outputIndexes": [ 1231 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1230 ], "main_type": "NONE", "name": "Rank5067", "outputIndexes": [ 1232 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1232, 1232 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5069", "outputIndexes": [ 1233 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1233 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5070", "outputIndexes": [ 1234 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1233, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5072", "outputIndexes": [ 1235 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1235 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5073", "outputIndexes": [ 1236 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1231, 1234, 1236, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5076", "outputIndexes": [ 1237 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1237 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5077", "outputIndexes": [ 1238 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1238 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5078", "outputIndexes": [ 1239 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1239, 1238 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5079", "outputIndexes": [ 1240 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1230, 1240, 1233 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_200_output_0", "outputIndexes": [ 1241 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1241, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_39_output_0", "outputIndexes": [ 1242 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1242, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_423_output_0", "outputIndexes": [ 1243 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1228, 1243, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_53_output_0", "outputIndexes": [ 1244 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1244 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_26_output_0", "outputIndexes": [ 1245 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1242, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_422_output_0", "outputIndexes": [ 1246 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1228, 20, 1246, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_52_output_0", "outputIndexes": [ 1247 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1245, 1247 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_172_output_0", "outputIndexes": [ 1248 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1248, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_105_output_0", "outputIndexes": [ 1249 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1229, 1249 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_65_output_0", "outputIndexes": [ 1250 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1205 ], "outputIndexes": [ 3623 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3623 ], "outputIndexes": [ 3624 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear", "inputIndexes": [ 3624 ], "outputIndexes": [ 3625 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1306503144, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3625 ], "outputIndexes": [ 3626 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3626 ], "outputIndexes": [ 1251 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1217, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_418_output_0", "outputIndexes": [ 1252 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1225, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_419_output_0", "outputIndexes": [ 1253 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1252, 1253, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_170_output_0", "outputIndexes": [ 1254 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1251, 1254 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_119_output_0", "outputIndexes": [ 1255 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1255, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_106_output_0", "outputIndexes": [ 1256 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1255 ], "main_type": "NONE", "name": "/Shape_185_output_0", "outputIndexes": [ 1257 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1257 ], "main_type": "NONE", "name": "Shape4971", "outputIndexes": [ 1258 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1257 ], "main_type": "NONE", "name": "Rank4973", "outputIndexes": [ 1259 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1259, 1259 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4975", "outputIndexes": [ 1260 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1260 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4976", "outputIndexes": [ 1261 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1260, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4978", "outputIndexes": [ 1262 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1262 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4979", "outputIndexes": [ 1263 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1258, 1261, 1263, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4982", "outputIndexes": [ 1264 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1264 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4983", "outputIndexes": [ 1265 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1265 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4984", "outputIndexes": [ 1266 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1266, 1265 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4985", "outputIndexes": [ 1267 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1257, 1267, 1260 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_201_output_0", "outputIndexes": [ 1268 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1268, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_40_output_0", "outputIndexes": [ 1269 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1269, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_425_output_0", "outputIndexes": [ 1270 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1255, 1270, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_55_output_0", "outputIndexes": [ 1271 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1271 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_27_output_0", "outputIndexes": [ 1272 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1269, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_424_output_0", "outputIndexes": [ 1273 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1255, 20, 1273, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_54_output_0", "outputIndexes": [ 1274 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1272, 1274 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_173_output_0", "outputIndexes": [ 1275 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1275, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_107_output_0", "outputIndexes": [ 1276 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1256, 1276 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_66_output_0", "outputIndexes": [ 1277 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1205 ], "outputIndexes": [ 3627 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3627 ], "outputIndexes": [ 3628 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear", "inputIndexes": [ 3628 ], "outputIndexes": [ 3629 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1308866558, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3629 ], "outputIndexes": [ 3630 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3630 ], "outputIndexes": [ 1278 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1217, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_420_output_0", "outputIndexes": [ 1279 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1225, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_421_output_0", "outputIndexes": [ 1280 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1279, 1280, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_171_output_0", "outputIndexes": [ 1281 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1278, 1281 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_120_output_0", "outputIndexes": [ 1282 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1250, 1277, 1282, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_125_output_0", "outputIndexes": [ 1283 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1283 ], "outputIndexes": [ 3631 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3631 ], "outputIndexes": [ 3632 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear", "inputIndexes": [ 3632 ], "outputIndexes": [ 3633 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1311229972, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3633 ], "outputIndexes": [ 3634 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3634 ], "outputIndexes": [ 1284 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1204, 1284 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_68_output_0", "outputIndexes": [ 1285 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_13/Mul_1_output_0", "inputIndexes": [ 1285 ], "outputIndexes": [ 1286 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1320667178, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1286 ], "outputIndexes": [ 3635 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3635 ], "outputIndexes": [ 3636 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear", "inputIndexes": [ 3636 ], "outputIndexes": [ 3637 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1320699946, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3637 ], "outputIndexes": [ 3638 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3638 ], "outputIndexes": [ 1287 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1287 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_13/Mul_output_0", "outputIndexes": [ 1288 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1286 ], "outputIndexes": [ 3639 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3639 ], "outputIndexes": [ 3640 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear", "inputIndexes": [ 3640 ], "outputIndexes": [ 3641 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1346062400, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3641 ], "outputIndexes": [ 3642 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3642 ], "outputIndexes": [ 1289 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1288, 1289 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_13/Mul_output_0", "outputIndexes": [ 1290 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1290 ], "outputIndexes": [ 3643 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3643 ], "outputIndexes": [ 3644 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear", "inputIndexes": [ 3644 ], "outputIndexes": [ 3645 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1371424854, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3645 ], "outputIndexes": [ 3646 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3646 ], "outputIndexes": [ 1291 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1285, 1291 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_69_output_0", "outputIndexes": [ 1292 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1292, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_126_output_0", "outputIndexes": [ 1293 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_14/Mul_1_output_0", "inputIndexes": [ 1293 ], "outputIndexes": [ 1294 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1396787308, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1294 ], "outputIndexes": [ 3647 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3647 ], "outputIndexes": [ 3648 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear", "inputIndexes": [ 3648 ], "outputIndexes": [ 3649 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1396820076, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3649 ], "outputIndexes": [ 3650 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3650 ], "outputIndexes": [ 1295 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1294 ], "main_type": "NONE", "name": "/Shape_196_output_0", "outputIndexes": [ 1296 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1296 ], "main_type": "NONE", "name": "Shape5304", "outputIndexes": [ 1297 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1296 ], "main_type": "NONE", "name": "Rank5306", "outputIndexes": [ 1298 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1298, 1298 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5308", "outputIndexes": [ 1299 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1299 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5309", "outputIndexes": [ 1300 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1299, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5311", "outputIndexes": [ 1301 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1301 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5312", "outputIndexes": [ 1302 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1297, 1300, 1302, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5315", "outputIndexes": [ 1303 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1303 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5316", "outputIndexes": [ 1304 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1304, 1304 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5318", "outputIndexes": [ 1305 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1296, 1305, 1299 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_213_output_0", "outputIndexes": [ 1306 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1306, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_448_output_0", "outputIndexes": [ 1307 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1299 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5326", "outputIndexes": [ 1308 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1301 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5329", "outputIndexes": [ 1309 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1297, 1308, 1309, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5332", "outputIndexes": [ 1310 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1310 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5333", "outputIndexes": [ 1311 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1311 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5334", "outputIndexes": [ 1312 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1312, 1311 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5335", "outputIndexes": [ 1313 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1296, 1313, 1299 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_214_output_0", "outputIndexes": [ 1314 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1314, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_449_output_0", "outputIndexes": [ 1315 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1307, 1315, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_182_output_0", "outputIndexes": [ 1316 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1295, 1316 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_127_output_0", "outputIndexes": [ 1317 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1317, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_112_output_0", "outputIndexes": [ 1318 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1317 ], "main_type": "NONE", "name": "/Shape_198_output_0", "outputIndexes": [ 1319 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1319 ], "main_type": "NONE", "name": "Shape5436", "outputIndexes": [ 1320 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1319 ], "main_type": "NONE", "name": "Rank5438", "outputIndexes": [ 1321 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1321, 1321 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5440", "outputIndexes": [ 1322 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1322 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5441", "outputIndexes": [ 1323 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1322, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5443", "outputIndexes": [ 1324 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1324 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5444", "outputIndexes": [ 1325 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1320, 1323, 1325, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5447", "outputIndexes": [ 1326 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1326 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5448", "outputIndexes": [ 1327 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1327 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5449", "outputIndexes": [ 1328 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1328, 1327 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5450", "outputIndexes": [ 1329 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1319, 1329, 1322 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_215_output_0", "outputIndexes": [ 1330 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1330, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_42_output_0", "outputIndexes": [ 1331 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1331, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_455_output_0", "outputIndexes": [ 1332 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1317, 1332, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_57_output_0", "outputIndexes": [ 1333 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1333 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_28_output_0", "outputIndexes": [ 1334 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1331, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_454_output_0", "outputIndexes": [ 1335 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1317, 20, 1335, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_56_output_0", "outputIndexes": [ 1336 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1334, 1336 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_185_output_0", "outputIndexes": [ 1337 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1337, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_113_output_0", "outputIndexes": [ 1338 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1318, 1338 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_70_output_0", "outputIndexes": [ 1339 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1294 ], "outputIndexes": [ 3651 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3651 ], "outputIndexes": [ 3652 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear", "inputIndexes": [ 3652 ], "outputIndexes": [ 3653 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1406273666, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3653 ], "outputIndexes": [ 3654 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3654 ], "outputIndexes": [ 1340 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1306, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_450_output_0", "outputIndexes": [ 1341 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1314, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_451_output_0", "outputIndexes": [ 1342 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1341, 1342, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_183_output_0", "outputIndexes": [ 1343 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1340, 1343 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_128_output_0", "outputIndexes": [ 1344 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1344, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_114_output_0", "outputIndexes": [ 1345 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1344 ], "main_type": "NONE", "name": "/Shape_199_output_0", "outputIndexes": [ 1346 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1346 ], "main_type": "NONE", "name": "Shape5342", "outputIndexes": [ 1347 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1346 ], "main_type": "NONE", "name": "Rank5344", "outputIndexes": [ 1348 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1348, 1348 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5346", "outputIndexes": [ 1349 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1349 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5347", "outputIndexes": [ 1350 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1349, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5349", "outputIndexes": [ 1351 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1351 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5350", "outputIndexes": [ 1352 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1347, 1350, 1352, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5353", "outputIndexes": [ 1353 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1353 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5354", "outputIndexes": [ 1354 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1354 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5355", "outputIndexes": [ 1355 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1355, 1354 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5356", "outputIndexes": [ 1356 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1346, 1356, 1349 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_216_output_0", "outputIndexes": [ 1357 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1357, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_43_output_0", "outputIndexes": [ 1358 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1358, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_457_output_0", "outputIndexes": [ 1359 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1344, 1359, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_59_output_0", "outputIndexes": [ 1360 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1360 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_29_output_0", "outputIndexes": [ 1361 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1358, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_456_output_0", "outputIndexes": [ 1362 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1344, 20, 1362, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_58_output_0", "outputIndexes": [ 1363 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1361, 1363 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_186_output_0", "outputIndexes": [ 1364 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1364, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_115_output_0", "outputIndexes": [ 1365 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1345, 1365 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_71_output_0", "outputIndexes": [ 1366 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1294 ], "outputIndexes": [ 3655 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3655 ], "outputIndexes": [ 3656 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear", "inputIndexes": [ 3656 ], "outputIndexes": [ 3657 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1408637080, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3657 ], "outputIndexes": [ 3658 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3658 ], "outputIndexes": [ 1367 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1306, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_452_output_0", "outputIndexes": [ 1368 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1314, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_453_output_0", "outputIndexes": [ 1369 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1368, 1369, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_184_output_0", "outputIndexes": [ 1370 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1367, 1370 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_129_output_0", "outputIndexes": [ 1371 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1339, 1366, 1371, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_134_output_0", "outputIndexes": [ 1372 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1372 ], "outputIndexes": [ 3659 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3659 ], "outputIndexes": [ 3660 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear", "inputIndexes": [ 3660 ], "outputIndexes": [ 3661 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1411000494, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3661 ], "outputIndexes": [ 3662 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3662 ], "outputIndexes": [ 1373 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1293, 1373 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_73_output_0", "outputIndexes": [ 1374 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_14/Mul_1_output_0", "inputIndexes": [ 1374 ], "outputIndexes": [ 1375 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1420437700, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1375 ], "outputIndexes": [ 3663 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3663 ], "outputIndexes": [ 3664 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear", "inputIndexes": [ 3664 ], "outputIndexes": [ 3665 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1420470468, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3665 ], "outputIndexes": [ 3666 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3666 ], "outputIndexes": [ 1376 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1376 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_14/Mul_output_0", "outputIndexes": [ 1377 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1375 ], "outputIndexes": [ 3667 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3667 ], "outputIndexes": [ 3668 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear", "inputIndexes": [ 3668 ], "outputIndexes": [ 3669 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1445832922, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3669 ], "outputIndexes": [ 3670 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3670 ], "outputIndexes": [ 1378 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1377, 1378 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_14/Mul_output_0", "outputIndexes": [ 1379 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1379 ], "outputIndexes": [ 3671 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3671 ], "outputIndexes": [ 3672 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear", "inputIndexes": [ 3672 ], "outputIndexes": [ 3673 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1471195376, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3673 ], "outputIndexes": [ 3674 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3674 ], "outputIndexes": [ 1380 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1374, 1380 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_74_output_0", "outputIndexes": [ 1381 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1381, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_135_output_0", "outputIndexes": [ 1382 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_15/Mul_1_output_0", "inputIndexes": [ 1382 ], "outputIndexes": [ 1383 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1496557830, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1383 ], "outputIndexes": [ 3675 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3675 ], "outputIndexes": [ 3676 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear", "inputIndexes": [ 3676 ], "outputIndexes": [ 3677 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1496590598, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3677 ], "outputIndexes": [ 3678 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3678 ], "outputIndexes": [ 1384 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1383 ], "main_type": "NONE", "name": "/Shape_210_output_0", "outputIndexes": [ 1385 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1385 ], "main_type": "NONE", "name": "Shape5675", "outputIndexes": [ 1386 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1385 ], "main_type": "NONE", "name": "Rank5677", "outputIndexes": [ 1387 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1387, 1387 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5679", "outputIndexes": [ 1388 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1388 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5680", "outputIndexes": [ 1389 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1388, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5682", "outputIndexes": [ 1390 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1390 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5683", "outputIndexes": [ 1391 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1386, 1389, 1391, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5686", "outputIndexes": [ 1392 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1392 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5687", "outputIndexes": [ 1393 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1393, 1393 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5689", "outputIndexes": [ 1394 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1385, 1394, 1388 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_228_output_0", "outputIndexes": [ 1395 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1395, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_480_output_0", "outputIndexes": [ 1396 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1388 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5697", "outputIndexes": [ 1397 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1390 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5700", "outputIndexes": [ 1398 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1386, 1397, 1398, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5703", "outputIndexes": [ 1399 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1399 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5704", "outputIndexes": [ 1400 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1400 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5705", "outputIndexes": [ 1401 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1401, 1400 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5706", "outputIndexes": [ 1402 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1385, 1402, 1388 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_229_output_0", "outputIndexes": [ 1403 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1403, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_481_output_0", "outputIndexes": [ 1404 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1396, 1404, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_195_output_0", "outputIndexes": [ 1405 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1384, 1405 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_136_output_0", "outputIndexes": [ 1406 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1406, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_120_output_0", "outputIndexes": [ 1407 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1406 ], "main_type": "NONE", "name": "/Shape_212_output_0", "outputIndexes": [ 1408 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1408 ], "main_type": "NONE", "name": "Shape5807", "outputIndexes": [ 1409 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1408 ], "main_type": "NONE", "name": "Rank5809", "outputIndexes": [ 1410 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1410, 1410 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5811", "outputIndexes": [ 1411 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1411 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5812", "outputIndexes": [ 1412 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1411, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5814", "outputIndexes": [ 1413 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1413 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5815", "outputIndexes": [ 1414 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1409, 1412, 1414, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5818", "outputIndexes": [ 1415 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1415 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5819", "outputIndexes": [ 1416 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1416 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5820", "outputIndexes": [ 1417 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1417, 1416 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5821", "outputIndexes": [ 1418 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1408, 1418, 1411 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_230_output_0", "outputIndexes": [ 1419 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1419, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_45_output_0", "outputIndexes": [ 1420 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1420, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_487_output_0", "outputIndexes": [ 1421 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1406, 1421, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_61_output_0", "outputIndexes": [ 1422 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1422 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_30_output_0", "outputIndexes": [ 1423 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1420, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_486_output_0", "outputIndexes": [ 1424 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1406, 20, 1424, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_60_output_0", "outputIndexes": [ 1425 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1423, 1425 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_198_output_0", "outputIndexes": [ 1426 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1426, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_121_output_0", "outputIndexes": [ 1427 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1407, 1427 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_75_output_0", "outputIndexes": [ 1428 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1383 ], "outputIndexes": [ 3679 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3679 ], "outputIndexes": [ 3680 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear", "inputIndexes": [ 3680 ], "outputIndexes": [ 3681 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1506044188, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3681 ], "outputIndexes": [ 3682 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3682 ], "outputIndexes": [ 1429 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1395, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_482_output_0", "outputIndexes": [ 1430 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1403, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_483_output_0", "outputIndexes": [ 1431 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1430, 1431, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_196_output_0", "outputIndexes": [ 1432 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1429, 1432 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_137_output_0", "outputIndexes": [ 1433 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1433, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_122_output_0", "outputIndexes": [ 1434 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1433 ], "main_type": "NONE", "name": "/Shape_213_output_0", "outputIndexes": [ 1435 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1435 ], "main_type": "NONE", "name": "Shape5713", "outputIndexes": [ 1436 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1435 ], "main_type": "NONE", "name": "Rank5715", "outputIndexes": [ 1437 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1437, 1437 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5717", "outputIndexes": [ 1438 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1438 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5718", "outputIndexes": [ 1439 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1438, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5720", "outputIndexes": [ 1440 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1440 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5721", "outputIndexes": [ 1441 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1436, 1439, 1441, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5724", "outputIndexes": [ 1442 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1442 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5725", "outputIndexes": [ 1443 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1443 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5726", "outputIndexes": [ 1444 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1444, 1443 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5727", "outputIndexes": [ 1445 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1435, 1445, 1438 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_231_output_0", "outputIndexes": [ 1446 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1446, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_46_output_0", "outputIndexes": [ 1447 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1447, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_489_output_0", "outputIndexes": [ 1448 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1433, 1448, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_63_output_0", "outputIndexes": [ 1449 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1449 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_31_output_0", "outputIndexes": [ 1450 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1447, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_488_output_0", "outputIndexes": [ 1451 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1433, 20, 1451, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_62_output_0", "outputIndexes": [ 1452 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1450, 1452 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_199_output_0", "outputIndexes": [ 1453 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1453, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_123_output_0", "outputIndexes": [ 1454 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1434, 1454 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_76_output_0", "outputIndexes": [ 1455 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1383 ], "outputIndexes": [ 3683 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3683 ], "outputIndexes": [ 3684 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear", "inputIndexes": [ 3684 ], "outputIndexes": [ 3685 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1508407602, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3685 ], "outputIndexes": [ 3686 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3686 ], "outputIndexes": [ 1456 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1395, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_484_output_0", "outputIndexes": [ 1457 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1403, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_485_output_0", "outputIndexes": [ 1458 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1457, 1458, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_197_output_0", "outputIndexes": [ 1459 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1456, 1459 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_138_output_0", "outputIndexes": [ 1460 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1428, 1455, 1460, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_143_output_0", "outputIndexes": [ 1461 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1461 ], "outputIndexes": [ 3687 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3687 ], "outputIndexes": [ 3688 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear", "inputIndexes": [ 3688 ], "outputIndexes": [ 3689 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1510771016, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3689 ], "outputIndexes": [ 3690 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3690 ], "outputIndexes": [ 1462 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1382, 1462 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_78_output_0", "outputIndexes": [ 1463 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_15/Mul_1_output_0", "inputIndexes": [ 1463 ], "outputIndexes": [ 1464 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1520208222, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1464 ], "outputIndexes": [ 3691 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3691 ], "outputIndexes": [ 3692 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear", "inputIndexes": [ 3692 ], "outputIndexes": [ 3693 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1520240990, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3693 ], "outputIndexes": [ 3694 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3694 ], "outputIndexes": [ 1465 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1465 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_15/Mul_output_0", "outputIndexes": [ 1466 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1464 ], "outputIndexes": [ 3695 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3695 ], "outputIndexes": [ 3696 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear", "inputIndexes": [ 3696 ], "outputIndexes": [ 3697 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1545603444, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3697 ], "outputIndexes": [ 3698 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3698 ], "outputIndexes": [ 1467 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1466, 1467 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_15/Mul_output_0", "outputIndexes": [ 1468 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1468 ], "outputIndexes": [ 3699 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3699 ], "outputIndexes": [ 3700 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear", "inputIndexes": [ 3700 ], "outputIndexes": [ 3701 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1570965898, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3701 ], "outputIndexes": [ 3702 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3702 ], "outputIndexes": [ 1469 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1463, 1469 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_79_output_0", "outputIndexes": [ 1470 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1470, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_144_output_0", "outputIndexes": [ 1471 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_16/Mul_1_output_0", "inputIndexes": [ 1471 ], "outputIndexes": [ 1472 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1596328352, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1472 ], "outputIndexes": [ 3703 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3703 ], "outputIndexes": [ 3704 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear", "inputIndexes": [ 3704 ], "outputIndexes": [ 3705 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1596361120, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3705 ], "outputIndexes": [ 3706 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3706 ], "outputIndexes": [ 1473 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1472 ], "main_type": "NONE", "name": "/Shape_224_output_0", "outputIndexes": [ 1474 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1474 ], "main_type": "NONE", "name": "Shape6046", "outputIndexes": [ 1475 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1474 ], "main_type": "NONE", "name": "Rank6048", "outputIndexes": [ 1476 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1476, 1476 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6050", "outputIndexes": [ 1477 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1477 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6051", "outputIndexes": [ 1478 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1477, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6053", "outputIndexes": [ 1479 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1479 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6054", "outputIndexes": [ 1480 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1475, 1478, 1480, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6057", "outputIndexes": [ 1481 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1481 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6058", "outputIndexes": [ 1482 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1482, 1482 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6060", "outputIndexes": [ 1483 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1474, 1483, 1477 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_243_output_0", "outputIndexes": [ 1484 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1484, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_512_output_0", "outputIndexes": [ 1485 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1477 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6068", "outputIndexes": [ 1486 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1479 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6071", "outputIndexes": [ 1487 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1475, 1486, 1487, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6074", "outputIndexes": [ 1488 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1488 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6075", "outputIndexes": [ 1489 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1489 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6076", "outputIndexes": [ 1490 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1490, 1489 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6077", "outputIndexes": [ 1491 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1474, 1491, 1477 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_244_output_0", "outputIndexes": [ 1492 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1492, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_513_output_0", "outputIndexes": [ 1493 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1485, 1493, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_208_output_0", "outputIndexes": [ 1494 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1473, 1494 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_145_output_0", "outputIndexes": [ 1495 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1495, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_128_output_0", "outputIndexes": [ 1496 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1495 ], "main_type": "NONE", "name": "/Shape_226_output_0", "outputIndexes": [ 1497 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1497 ], "main_type": "NONE", "name": "Shape6178", "outputIndexes": [ 1498 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1497 ], "main_type": "NONE", "name": "Rank6180", "outputIndexes": [ 1499 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1499, 1499 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6182", "outputIndexes": [ 1500 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1500 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6183", "outputIndexes": [ 1501 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1500, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6185", "outputIndexes": [ 1502 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1502 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6186", "outputIndexes": [ 1503 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1498, 1501, 1503, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6189", "outputIndexes": [ 1504 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1504 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6190", "outputIndexes": [ 1505 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1505 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6191", "outputIndexes": [ 1506 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1506, 1505 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6192", "outputIndexes": [ 1507 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1497, 1507, 1500 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_245_output_0", "outputIndexes": [ 1508 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1508, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_48_output_0", "outputIndexes": [ 1509 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1509, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_519_output_0", "outputIndexes": [ 1510 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1495, 1510, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_65_output_0", "outputIndexes": [ 1511 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1511 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_32_output_0", "outputIndexes": [ 1512 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1509, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_518_output_0", "outputIndexes": [ 1513 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1495, 20, 1513, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_64_output_0", "outputIndexes": [ 1514 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1512, 1514 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_211_output_0", "outputIndexes": [ 1515 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1515, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_129_output_0", "outputIndexes": [ 1516 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1496, 1516 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_80_output_0", "outputIndexes": [ 1517 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1472 ], "outputIndexes": [ 3707 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3707 ], "outputIndexes": [ 3708 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear", "inputIndexes": [ 3708 ], "outputIndexes": [ 3709 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1605814710, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3709 ], "outputIndexes": [ 3710 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3710 ], "outputIndexes": [ 1518 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1484, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_514_output_0", "outputIndexes": [ 1519 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1492, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_515_output_0", "outputIndexes": [ 1520 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1519, 1520, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_209_output_0", "outputIndexes": [ 1521 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1518, 1521 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_146_output_0", "outputIndexes": [ 1522 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1522, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_130_output_0", "outputIndexes": [ 1523 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1522 ], "main_type": "NONE", "name": "/Shape_227_output_0", "outputIndexes": [ 1524 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1524 ], "main_type": "NONE", "name": "Shape6084", "outputIndexes": [ 1525 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1524 ], "main_type": "NONE", "name": "Rank6086", "outputIndexes": [ 1526 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1526, 1526 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6088", "outputIndexes": [ 1527 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1527 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6089", "outputIndexes": [ 1528 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1527, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6091", "outputIndexes": [ 1529 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1529 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6092", "outputIndexes": [ 1530 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1525, 1528, 1530, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6095", "outputIndexes": [ 1531 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1531 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6096", "outputIndexes": [ 1532 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1532 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6097", "outputIndexes": [ 1533 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1533, 1532 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6098", "outputIndexes": [ 1534 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1524, 1534, 1527 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_246_output_0", "outputIndexes": [ 1535 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1535, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_49_output_0", "outputIndexes": [ 1536 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1536, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_521_output_0", "outputIndexes": [ 1537 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1522, 1537, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_67_output_0", "outputIndexes": [ 1538 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1538 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_33_output_0", "outputIndexes": [ 1539 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1536, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_520_output_0", "outputIndexes": [ 1540 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1522, 20, 1540, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_66_output_0", "outputIndexes": [ 1541 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1539, 1541 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_212_output_0", "outputIndexes": [ 1542 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1542, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_131_output_0", "outputIndexes": [ 1543 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1523, 1543 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_81_output_0", "outputIndexes": [ 1544 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1472 ], "outputIndexes": [ 3711 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3711 ], "outputIndexes": [ 3712 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear", "inputIndexes": [ 3712 ], "outputIndexes": [ 3713 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1608178124, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3713 ], "outputIndexes": [ 3714 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3714 ], "outputIndexes": [ 1545 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1484, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_516_output_0", "outputIndexes": [ 1546 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1492, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_517_output_0", "outputIndexes": [ 1547 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1546, 1547, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_210_output_0", "outputIndexes": [ 1548 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1545, 1548 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_147_output_0", "outputIndexes": [ 1549 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1517, 1544, 1549, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_152_output_0", "outputIndexes": [ 1550 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1550 ], "outputIndexes": [ 3715 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3715 ], "outputIndexes": [ 3716 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear", "inputIndexes": [ 3716 ], "outputIndexes": [ 3717 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1610541538, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3717 ], "outputIndexes": [ 3718 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3718 ], "outputIndexes": [ 1551 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1471, 1551 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_83_output_0", "outputIndexes": [ 1552 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_16/Mul_1_output_0", "inputIndexes": [ 1552 ], "outputIndexes": [ 1553 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1619978744, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1553 ], "outputIndexes": [ 3719 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3719 ], "outputIndexes": [ 3720 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear", "inputIndexes": [ 3720 ], "outputIndexes": [ 3721 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1620011512, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3721 ], "outputIndexes": [ 3722 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3722 ], "outputIndexes": [ 1554 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1554 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_16/Mul_output_0", "outputIndexes": [ 1555 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1553 ], "outputIndexes": [ 3723 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3723 ], "outputIndexes": [ 3724 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear", "inputIndexes": [ 3724 ], "outputIndexes": [ 3725 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1645373966, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3725 ], "outputIndexes": [ 3726 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3726 ], "outputIndexes": [ 1556 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1555, 1556 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_16/Mul_output_0", "outputIndexes": [ 1557 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1557 ], "outputIndexes": [ 3727 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3727 ], "outputIndexes": [ 3728 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear", "inputIndexes": [ 3728 ], "outputIndexes": [ 3729 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1670736420, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3729 ], "outputIndexes": [ 3730 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3730 ], "outputIndexes": [ 1558 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1552, 1558 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_84_output_0", "outputIndexes": [ 1559 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1559, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_153_output_0", "outputIndexes": [ 1560 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_17/Mul_1_output_0", "inputIndexes": [ 1560 ], "outputIndexes": [ 1561 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1696098874, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1561 ], "outputIndexes": [ 3731 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3731 ], "outputIndexes": [ 3732 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear", "inputIndexes": [ 3732 ], "outputIndexes": [ 3733 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1696131642, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3733 ], "outputIndexes": [ 3734 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3734 ], "outputIndexes": [ 1562 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1561 ], "main_type": "NONE", "name": "/Shape_238_output_0", "outputIndexes": [ 1563 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1563 ], "main_type": "NONE", "name": "Shape6417", "outputIndexes": [ 1564 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1563 ], "main_type": "NONE", "name": "Rank6419", "outputIndexes": [ 1565 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1565, 1565 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6421", "outputIndexes": [ 1566 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1566 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6422", "outputIndexes": [ 1567 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1566, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6424", "outputIndexes": [ 1568 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1568 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6425", "outputIndexes": [ 1569 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1564, 1567, 1569, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6428", "outputIndexes": [ 1570 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1570 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6429", "outputIndexes": [ 1571 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1571, 1571 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6431", "outputIndexes": [ 1572 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1563, 1572, 1566 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_258_output_0", "outputIndexes": [ 1573 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1573, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_544_output_0", "outputIndexes": [ 1574 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1566 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6439", "outputIndexes": [ 1575 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1568 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6442", "outputIndexes": [ 1576 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1564, 1575, 1576, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6445", "outputIndexes": [ 1577 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1577 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6446", "outputIndexes": [ 1578 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1578 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6447", "outputIndexes": [ 1579 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1579, 1578 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6448", "outputIndexes": [ 1580 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1563, 1580, 1566 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_259_output_0", "outputIndexes": [ 1581 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1581, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_545_output_0", "outputIndexes": [ 1582 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1574, 1582, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_221_output_0", "outputIndexes": [ 1583 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1562, 1583 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_154_output_0", "outputIndexes": [ 1584 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1584, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_136_output_0", "outputIndexes": [ 1585 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1584 ], "main_type": "NONE", "name": "/Shape_240_output_0", "outputIndexes": [ 1586 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1586 ], "main_type": "NONE", "name": "Shape6549", "outputIndexes": [ 1587 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1586 ], "main_type": "NONE", "name": "Rank6551", "outputIndexes": [ 1588 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1588, 1588 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6553", "outputIndexes": [ 1589 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1589 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6554", "outputIndexes": [ 1590 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1589, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6556", "outputIndexes": [ 1591 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1591 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6557", "outputIndexes": [ 1592 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1587, 1590, 1592, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6560", "outputIndexes": [ 1593 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1593 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6561", "outputIndexes": [ 1594 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1594 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6562", "outputIndexes": [ 1595 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1595, 1594 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6563", "outputIndexes": [ 1596 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1586, 1596, 1589 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_260_output_0", "outputIndexes": [ 1597 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1597, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_51_output_0", "outputIndexes": [ 1598 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1598, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_551_output_0", "outputIndexes": [ 1599 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1584, 1599, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_69_output_0", "outputIndexes": [ 1600 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1600 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_34_output_0", "outputIndexes": [ 1601 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1598, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_550_output_0", "outputIndexes": [ 1602 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1584, 20, 1602, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_68_output_0", "outputIndexes": [ 1603 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1601, 1603 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_224_output_0", "outputIndexes": [ 1604 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1604, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_137_output_0", "outputIndexes": [ 1605 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1585, 1605 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_85_output_0", "outputIndexes": [ 1606 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1561 ], "outputIndexes": [ 3735 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3735 ], "outputIndexes": [ 3736 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear", "inputIndexes": [ 3736 ], "outputIndexes": [ 3737 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1705585232, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3737 ], "outputIndexes": [ 3738 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3738 ], "outputIndexes": [ 1607 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1573, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_546_output_0", "outputIndexes": [ 1608 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1581, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_547_output_0", "outputIndexes": [ 1609 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1608, 1609, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_222_output_0", "outputIndexes": [ 1610 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1607, 1610 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_155_output_0", "outputIndexes": [ 1611 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1611, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_138_output_0", "outputIndexes": [ 1612 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1611 ], "main_type": "NONE", "name": "/Shape_241_output_0", "outputIndexes": [ 1613 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1613 ], "main_type": "NONE", "name": "Shape6455", "outputIndexes": [ 1614 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1613 ], "main_type": "NONE", "name": "Rank6457", "outputIndexes": [ 1615 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1615, 1615 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6459", "outputIndexes": [ 1616 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1616 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6460", "outputIndexes": [ 1617 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1616, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6462", "outputIndexes": [ 1618 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1618 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6463", "outputIndexes": [ 1619 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1614, 1617, 1619, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6466", "outputIndexes": [ 1620 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1620 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6467", "outputIndexes": [ 1621 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1621 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6468", "outputIndexes": [ 1622 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1622, 1621 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6469", "outputIndexes": [ 1623 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1613, 1623, 1616 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_261_output_0", "outputIndexes": [ 1624 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1624, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_52_output_0", "outputIndexes": [ 1625 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1625, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_553_output_0", "outputIndexes": [ 1626 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1611, 1626, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_71_output_0", "outputIndexes": [ 1627 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1627 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_35_output_0", "outputIndexes": [ 1628 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1625, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_552_output_0", "outputIndexes": [ 1629 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1611, 20, 1629, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_70_output_0", "outputIndexes": [ 1630 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1628, 1630 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_225_output_0", "outputIndexes": [ 1631 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1631, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_139_output_0", "outputIndexes": [ 1632 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1612, 1632 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_86_output_0", "outputIndexes": [ 1633 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1561 ], "outputIndexes": [ 3739 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3739 ], "outputIndexes": [ 3740 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear", "inputIndexes": [ 3740 ], "outputIndexes": [ 3741 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1707948646, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3741 ], "outputIndexes": [ 3742 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3742 ], "outputIndexes": [ 1634 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1573, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_548_output_0", "outputIndexes": [ 1635 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1581, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_549_output_0", "outputIndexes": [ 1636 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1635, 1636, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_223_output_0", "outputIndexes": [ 1637 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1634, 1637 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_156_output_0", "outputIndexes": [ 1638 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1606, 1633, 1638, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_161_output_0", "outputIndexes": [ 1639 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1639 ], "outputIndexes": [ 3743 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3743 ], "outputIndexes": [ 3744 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear", "inputIndexes": [ 3744 ], "outputIndexes": [ 3745 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1710312060, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3745 ], "outputIndexes": [ 3746 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3746 ], "outputIndexes": [ 1640 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1560, 1640 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_88_output_0", "outputIndexes": [ 1641 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_17/Mul_1_output_0", "inputIndexes": [ 1641 ], "outputIndexes": [ 1642 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1719749266, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1642 ], "outputIndexes": [ 3747 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3747 ], "outputIndexes": [ 3748 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear", "inputIndexes": [ 3748 ], "outputIndexes": [ 3749 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1719782034, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3749 ], "outputIndexes": [ 3750 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3750 ], "outputIndexes": [ 1643 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1643 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_17/Mul_output_0", "outputIndexes": [ 1644 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1642 ], "outputIndexes": [ 3751 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3751 ], "outputIndexes": [ 3752 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear", "inputIndexes": [ 3752 ], "outputIndexes": [ 3753 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1745144488, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3753 ], "outputIndexes": [ 3754 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3754 ], "outputIndexes": [ 1645 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1644, 1645 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_17/Mul_output_0", "outputIndexes": [ 1646 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1646 ], "outputIndexes": [ 3755 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3755 ], "outputIndexes": [ 3756 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear", "inputIndexes": [ 3756 ], "outputIndexes": [ 3757 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1770506942, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3757 ], "outputIndexes": [ 3758 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3758 ], "outputIndexes": [ 1647 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1641, 1647 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_89_output_0", "outputIndexes": [ 1648 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1648, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_162_output_0", "outputIndexes": [ 1649 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_18/Mul_1_output_0", "inputIndexes": [ 1649 ], "outputIndexes": [ 1650 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1795869396, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1650 ], "outputIndexes": [ 3759 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3759 ], "outputIndexes": [ 3760 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear", "inputIndexes": [ 3760 ], "outputIndexes": [ 3761 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1795902164, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3761 ], "outputIndexes": [ 3762 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3762 ], "outputIndexes": [ 1651 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1650 ], "main_type": "NONE", "name": "/Shape_252_output_0", "outputIndexes": [ 1652 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1652 ], "main_type": "NONE", "name": "Shape6788", "outputIndexes": [ 1653 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1652 ], "main_type": "NONE", "name": "Rank6790", "outputIndexes": [ 1654 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1654, 1654 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6792", "outputIndexes": [ 1655 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1655 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6793", "outputIndexes": [ 1656 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1655, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6795", "outputIndexes": [ 1657 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1657 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6796", "outputIndexes": [ 1658 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1653, 1656, 1658, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6799", "outputIndexes": [ 1659 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1659 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6800", "outputIndexes": [ 1660 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1660, 1660 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6802", "outputIndexes": [ 1661 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1652, 1661, 1655 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_273_output_0", "outputIndexes": [ 1662 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1662, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_576_output_0", "outputIndexes": [ 1663 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1655 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6810", "outputIndexes": [ 1664 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1657 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6813", "outputIndexes": [ 1665 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1653, 1664, 1665, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6816", "outputIndexes": [ 1666 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1666 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6817", "outputIndexes": [ 1667 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1667 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6818", "outputIndexes": [ 1668 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1668, 1667 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6819", "outputIndexes": [ 1669 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1652, 1669, 1655 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_274_output_0", "outputIndexes": [ 1670 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1670, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_577_output_0", "outputIndexes": [ 1671 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1663, 1671, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_234_output_0", "outputIndexes": [ 1672 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1651, 1672 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_163_output_0", "outputIndexes": [ 1673 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1673, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_144_output_0", "outputIndexes": [ 1674 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1673 ], "main_type": "NONE", "name": "/Shape_254_output_0", "outputIndexes": [ 1675 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1675 ], "main_type": "NONE", "name": "Shape6920", "outputIndexes": [ 1676 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1675 ], "main_type": "NONE", "name": "Rank6922", "outputIndexes": [ 1677 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1677, 1677 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6924", "outputIndexes": [ 1678 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1678 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6925", "outputIndexes": [ 1679 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1678, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6927", "outputIndexes": [ 1680 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1680 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6928", "outputIndexes": [ 1681 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1676, 1679, 1681, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6931", "outputIndexes": [ 1682 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1682 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6932", "outputIndexes": [ 1683 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1683 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6933", "outputIndexes": [ 1684 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1684, 1683 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6934", "outputIndexes": [ 1685 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1675, 1685, 1678 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_275_output_0", "outputIndexes": [ 1686 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1686, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_54_output_0", "outputIndexes": [ 1687 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1687, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_583_output_0", "outputIndexes": [ 1688 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1673, 1688, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_73_output_0", "outputIndexes": [ 1689 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1689 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_36_output_0", "outputIndexes": [ 1690 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1687, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_582_output_0", "outputIndexes": [ 1691 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1673, 20, 1691, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_72_output_0", "outputIndexes": [ 1692 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1690, 1692 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_237_output_0", "outputIndexes": [ 1693 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1693, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_145_output_0", "outputIndexes": [ 1694 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1674, 1694 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_90_output_0", "outputIndexes": [ 1695 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1650 ], "outputIndexes": [ 3763 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3763 ], "outputIndexes": [ 3764 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear", "inputIndexes": [ 3764 ], "outputIndexes": [ 3765 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1805355754, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3765 ], "outputIndexes": [ 3766 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3766 ], "outputIndexes": [ 1696 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1662, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_578_output_0", "outputIndexes": [ 1697 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1670, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_579_output_0", "outputIndexes": [ 1698 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1697, 1698, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_235_output_0", "outputIndexes": [ 1699 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1696, 1699 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_164_output_0", "outputIndexes": [ 1700 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1700, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_146_output_0", "outputIndexes": [ 1701 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1700 ], "main_type": "NONE", "name": "/Shape_255_output_0", "outputIndexes": [ 1702 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1702 ], "main_type": "NONE", "name": "Shape6826", "outputIndexes": [ 1703 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1702 ], "main_type": "NONE", "name": "Rank6828", "outputIndexes": [ 1704 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1704, 1704 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6830", "outputIndexes": [ 1705 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1705 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6831", "outputIndexes": [ 1706 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1705, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6833", "outputIndexes": [ 1707 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1707 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6834", "outputIndexes": [ 1708 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1703, 1706, 1708, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6837", "outputIndexes": [ 1709 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1709 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6838", "outputIndexes": [ 1710 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1710 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6839", "outputIndexes": [ 1711 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1711, 1710 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6840", "outputIndexes": [ 1712 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1702, 1712, 1705 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_276_output_0", "outputIndexes": [ 1713 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1713, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_55_output_0", "outputIndexes": [ 1714 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1714, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_585_output_0", "outputIndexes": [ 1715 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1700, 1715, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_75_output_0", "outputIndexes": [ 1716 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1716 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_37_output_0", "outputIndexes": [ 1717 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1714, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_584_output_0", "outputIndexes": [ 1718 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1700, 20, 1718, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_74_output_0", "outputIndexes": [ 1719 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1717, 1719 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_238_output_0", "outputIndexes": [ 1720 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1720, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_147_output_0", "outputIndexes": [ 1721 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1701, 1721 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_91_output_0", "outputIndexes": [ 1722 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1650 ], "outputIndexes": [ 3767 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3767 ], "outputIndexes": [ 3768 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear", "inputIndexes": [ 3768 ], "outputIndexes": [ 3769 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1807719168, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3769 ], "outputIndexes": [ 3770 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3770 ], "outputIndexes": [ 1723 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1662, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_580_output_0", "outputIndexes": [ 1724 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1670, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_581_output_0", "outputIndexes": [ 1725 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1724, 1725, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_236_output_0", "outputIndexes": [ 1726 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1723, 1726 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_165_output_0", "outputIndexes": [ 1727 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1695, 1722, 1727, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_170_output_0", "outputIndexes": [ 1728 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1728 ], "outputIndexes": [ 3771 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3771 ], "outputIndexes": [ 3772 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear", "inputIndexes": [ 3772 ], "outputIndexes": [ 3773 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1810082582, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3773 ], "outputIndexes": [ 3774 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3774 ], "outputIndexes": [ 1729 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1649, 1729 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_93_output_0", "outputIndexes": [ 1730 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_18/Mul_1_output_0", "inputIndexes": [ 1730 ], "outputIndexes": [ 1731 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1819519788, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1731 ], "outputIndexes": [ 3775 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3775 ], "outputIndexes": [ 3776 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear", "inputIndexes": [ 3776 ], "outputIndexes": [ 3777 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1819552556, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3777 ], "outputIndexes": [ 3778 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3778 ], "outputIndexes": [ 1732 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1732 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_18/Mul_output_0", "outputIndexes": [ 1733 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1731 ], "outputIndexes": [ 3779 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3779 ], "outputIndexes": [ 3780 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear", "inputIndexes": [ 3780 ], "outputIndexes": [ 3781 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1844915010, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3781 ], "outputIndexes": [ 3782 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3782 ], "outputIndexes": [ 1734 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1733, 1734 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_18/Mul_output_0", "outputIndexes": [ 1735 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1735 ], "outputIndexes": [ 3783 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3783 ], "outputIndexes": [ 3784 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear", "inputIndexes": [ 3784 ], "outputIndexes": [ 3785 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1870277464, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3785 ], "outputIndexes": [ 3786 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3786 ], "outputIndexes": [ 1736 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1730, 1736 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_94_output_0", "outputIndexes": [ 1737 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1737, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_171_output_0", "outputIndexes": [ 1738 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_19/Mul_1_output_0", "inputIndexes": [ 1738 ], "outputIndexes": [ 1739 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1895639918, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1739 ], "outputIndexes": [ 3787 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3787 ], "outputIndexes": [ 3788 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear", "inputIndexes": [ 3788 ], "outputIndexes": [ 3789 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1895672686, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3789 ], "outputIndexes": [ 3790 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3790 ], "outputIndexes": [ 1740 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1739 ], "main_type": "NONE", "name": "/Shape_266_output_0", "outputIndexes": [ 1741 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1741 ], "main_type": "NONE", "name": "Shape7159", "outputIndexes": [ 1742 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1741 ], "main_type": "NONE", "name": "Rank7161", "outputIndexes": [ 1743 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1743, 1743 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7163", "outputIndexes": [ 1744 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1744 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7164", "outputIndexes": [ 1745 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1744, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7166", "outputIndexes": [ 1746 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1746 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7167", "outputIndexes": [ 1747 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1742, 1745, 1747, 15 ], "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": [ 1748 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1748 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7171", "outputIndexes": [ 1749 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1749, 1749 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7173", "outputIndexes": [ 1750 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1741, 1750, 1744 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_288_output_0", "outputIndexes": [ 1751 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1751, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_608_output_0", "outputIndexes": [ 1752 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1744 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7181", "outputIndexes": [ 1753 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1746 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7184", "outputIndexes": [ 1754 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1742, 1753, 1754, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7187", "outputIndexes": [ 1755 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1755 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7188", "outputIndexes": [ 1756 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1756 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7189", "outputIndexes": [ 1757 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1757, 1756 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7190", "outputIndexes": [ 1758 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1741, 1758, 1744 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_289_output_0", "outputIndexes": [ 1759 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1759, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_609_output_0", "outputIndexes": [ 1760 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1752, 1760, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_247_output_0", "outputIndexes": [ 1761 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1740, 1761 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_172_output_0", "outputIndexes": [ 1762 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1762, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_152_output_0", "outputIndexes": [ 1763 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1762 ], "main_type": "NONE", "name": "/Shape_268_output_0", "outputIndexes": [ 1764 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1764 ], "main_type": "NONE", "name": "Shape7291", "outputIndexes": [ 1765 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1764 ], "main_type": "NONE", "name": "Rank7293", "outputIndexes": [ 1766 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1766, 1766 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7295", "outputIndexes": [ 1767 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1767 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7296", "outputIndexes": [ 1768 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1767, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7298", "outputIndexes": [ 1769 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1769 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7299", "outputIndexes": [ 1770 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1765, 1768, 1770, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7302", "outputIndexes": [ 1771 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1771 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7303", "outputIndexes": [ 1772 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1772 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7304", "outputIndexes": [ 1773 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1773, 1772 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7305", "outputIndexes": [ 1774 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1764, 1774, 1767 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_290_output_0", "outputIndexes": [ 1775 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1775, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_57_output_0", "outputIndexes": [ 1776 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1776, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_615_output_0", "outputIndexes": [ 1777 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1762, 1777, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_77_output_0", "outputIndexes": [ 1778 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1778 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_38_output_0", "outputIndexes": [ 1779 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1776, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_614_output_0", "outputIndexes": [ 1780 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1762, 20, 1780, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_76_output_0", "outputIndexes": [ 1781 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1779, 1781 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_250_output_0", "outputIndexes": [ 1782 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1782, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_153_output_0", "outputIndexes": [ 1783 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1763, 1783 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_95_output_0", "outputIndexes": [ 1784 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1739 ], "outputIndexes": [ 3791 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3791 ], "outputIndexes": [ 3792 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear", "inputIndexes": [ 3792 ], "outputIndexes": [ 3793 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1905126276, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3793 ], "outputIndexes": [ 3794 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3794 ], "outputIndexes": [ 1785 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1751, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_610_output_0", "outputIndexes": [ 1786 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1759, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_611_output_0", "outputIndexes": [ 1787 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1786, 1787, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_248_output_0", "outputIndexes": [ 1788 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1785, 1788 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_173_output_0", "outputIndexes": [ 1789 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1789, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_154_output_0", "outputIndexes": [ 1790 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1789 ], "main_type": "NONE", "name": "/Shape_269_output_0", "outputIndexes": [ 1791 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1791 ], "main_type": "NONE", "name": "Shape7197", "outputIndexes": [ 1792 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1791 ], "main_type": "NONE", "name": "Rank7199", "outputIndexes": [ 1793 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1793, 1793 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7201", "outputIndexes": [ 1794 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1794 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7202", "outputIndexes": [ 1795 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1794, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7204", "outputIndexes": [ 1796 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1796 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7205", "outputIndexes": [ 1797 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1792, 1795, 1797, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7208", "outputIndexes": [ 1798 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1798 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7209", "outputIndexes": [ 1799 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1799 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7210", "outputIndexes": [ 1800 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1800, 1799 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7211", "outputIndexes": [ 1801 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1791, 1801, 1794 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_291_output_0", "outputIndexes": [ 1802 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1802, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_58_output_0", "outputIndexes": [ 1803 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1803, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_617_output_0", "outputIndexes": [ 1804 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1789, 1804, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_79_output_0", "outputIndexes": [ 1805 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1805 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_39_output_0", "outputIndexes": [ 1806 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1803, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_616_output_0", "outputIndexes": [ 1807 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1789, 20, 1807, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_78_output_0", "outputIndexes": [ 1808 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1806, 1808 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_251_output_0", "outputIndexes": [ 1809 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1809, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_155_output_0", "outputIndexes": [ 1810 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1790, 1810 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_96_output_0", "outputIndexes": [ 1811 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1739 ], "outputIndexes": [ 3795 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3795 ], "outputIndexes": [ 3796 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear", "inputIndexes": [ 3796 ], "outputIndexes": [ 3797 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 1907489690, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3797 ], "outputIndexes": [ 3798 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3798 ], "outputIndexes": [ 1812 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1751, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_612_output_0", "outputIndexes": [ 1813 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1759, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_613_output_0", "outputIndexes": [ 1814 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1813, 1814, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_249_output_0", "outputIndexes": [ 1815 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1812, 1815 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_174_output_0", "outputIndexes": [ 1816 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1784, 1811, 1816, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_179_output_0", "outputIndexes": [ 1817 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1817 ], "outputIndexes": [ 3799 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3799 ], "outputIndexes": [ 3800 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear", "inputIndexes": [ 3800 ], "outputIndexes": [ 3801 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1909853104, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3801 ], "outputIndexes": [ 3802 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3802 ], "outputIndexes": [ 1818 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1738, 1818 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_98_output_0", "outputIndexes": [ 1819 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_19/Mul_1_output_0", "inputIndexes": [ 1819 ], "outputIndexes": [ 1820 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1919290310, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1820 ], "outputIndexes": [ 3803 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3803 ], "outputIndexes": [ 3804 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear", "inputIndexes": [ 3804 ], "outputIndexes": [ 3805 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1919323078, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3805 ], "outputIndexes": [ 3806 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3806 ], "outputIndexes": [ 1821 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1821 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_19/Mul_output_0", "outputIndexes": [ 1822 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1820 ], "outputIndexes": [ 3807 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3807 ], "outputIndexes": [ 3808 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear", "inputIndexes": [ 3808 ], "outputIndexes": [ 3809 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1944685532, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3809 ], "outputIndexes": [ 3810 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3810 ], "outputIndexes": [ 1823 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1822, 1823 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_19/Mul_output_0", "outputIndexes": [ 1824 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1824 ], "outputIndexes": [ 3811 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3811 ], "outputIndexes": [ 3812 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear", "inputIndexes": [ 3812 ], "outputIndexes": [ 3813 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 1970047986, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3813 ], "outputIndexes": [ 3814 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3814 ], "outputIndexes": [ 1825 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1819, 1825 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_99_output_0", "outputIndexes": [ 1826 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1826, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_180_output_0", "outputIndexes": [ 1827 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_20/Mul_1_output_0", "inputIndexes": [ 1827 ], "outputIndexes": [ 1828 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 1995410440, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1828 ], "outputIndexes": [ 3815 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3815 ], "outputIndexes": [ 3816 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear", "inputIndexes": [ 3816 ], "outputIndexes": [ 3817 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 1995443208, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3817 ], "outputIndexes": [ 3818 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3818 ], "outputIndexes": [ 1829 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1828 ], "main_type": "NONE", "name": "/Shape_280_output_0", "outputIndexes": [ 1830 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1830 ], "main_type": "NONE", "name": "Shape7530", "outputIndexes": [ 1831 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1830 ], "main_type": "NONE", "name": "Rank7532", "outputIndexes": [ 1832 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1832, 1832 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7534", "outputIndexes": [ 1833 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1833 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7535", "outputIndexes": [ 1834 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1833, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7537", "outputIndexes": [ 1835 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1835 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7538", "outputIndexes": [ 1836 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1831, 1834, 1836, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7541", "outputIndexes": [ 1837 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1837 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7542", "outputIndexes": [ 1838 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1838, 1838 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7544", "outputIndexes": [ 1839 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1830, 1839, 1833 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_303_output_0", "outputIndexes": [ 1840 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1840, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_640_output_0", "outputIndexes": [ 1841 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1833 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7552", "outputIndexes": [ 1842 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1835 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7555", "outputIndexes": [ 1843 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1831, 1842, 1843, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7558", "outputIndexes": [ 1844 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1844 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7559", "outputIndexes": [ 1845 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1845 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7560", "outputIndexes": [ 1846 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1846, 1845 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7561", "outputIndexes": [ 1847 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1830, 1847, 1833 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_304_output_0", "outputIndexes": [ 1848 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1848, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_641_output_0", "outputIndexes": [ 1849 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1841, 1849, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_260_output_0", "outputIndexes": [ 1850 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1829, 1850 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_181_output_0", "outputIndexes": [ 1851 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1851, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_160_output_0", "outputIndexes": [ 1852 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1851 ], "main_type": "NONE", "name": "/Shape_282_output_0", "outputIndexes": [ 1853 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1853 ], "main_type": "NONE", "name": "Shape7662", "outputIndexes": [ 1854 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1853 ], "main_type": "NONE", "name": "Rank7664", "outputIndexes": [ 1855 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1855, 1855 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7666", "outputIndexes": [ 1856 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1856 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7667", "outputIndexes": [ 1857 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1856, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7669", "outputIndexes": [ 1858 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1858 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7670", "outputIndexes": [ 1859 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1854, 1857, 1859, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7673", "outputIndexes": [ 1860 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1860 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7674", "outputIndexes": [ 1861 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1861 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7675", "outputIndexes": [ 1862 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1862, 1861 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7676", "outputIndexes": [ 1863 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1853, 1863, 1856 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_305_output_0", "outputIndexes": [ 1864 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1864, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_60_output_0", "outputIndexes": [ 1865 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1865, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_647_output_0", "outputIndexes": [ 1866 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1851, 1866, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_81_output_0", "outputIndexes": [ 1867 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1867 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_40_output_0", "outputIndexes": [ 1868 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1865, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_646_output_0", "outputIndexes": [ 1869 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1851, 20, 1869, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_80_output_0", "outputIndexes": [ 1870 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1868, 1870 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_263_output_0", "outputIndexes": [ 1871 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1871, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_161_output_0", "outputIndexes": [ 1872 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1852, 1872 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_100_output_0", "outputIndexes": [ 1873 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1828 ], "outputIndexes": [ 3819 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3819 ], "outputIndexes": [ 3820 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear", "inputIndexes": [ 3820 ], "outputIndexes": [ 3821 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2004896798, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3821 ], "outputIndexes": [ 3822 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3822 ], "outputIndexes": [ 1874 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1840, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_642_output_0", "outputIndexes": [ 1875 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1848, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_643_output_0", "outputIndexes": [ 1876 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1875, 1876, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_261_output_0", "outputIndexes": [ 1877 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1874, 1877 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_182_output_0", "outputIndexes": [ 1878 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1878, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_162_output_0", "outputIndexes": [ 1879 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1878 ], "main_type": "NONE", "name": "/Shape_283_output_0", "outputIndexes": [ 1880 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1880 ], "main_type": "NONE", "name": "Shape7568", "outputIndexes": [ 1881 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1880 ], "main_type": "NONE", "name": "Rank7570", "outputIndexes": [ 1882 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1882, 1882 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7572", "outputIndexes": [ 1883 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1883 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7573", "outputIndexes": [ 1884 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1883, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7575", "outputIndexes": [ 1885 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1885 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7576", "outputIndexes": [ 1886 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1881, 1884, 1886, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7579", "outputIndexes": [ 1887 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1887 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7580", "outputIndexes": [ 1888 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1888 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7581", "outputIndexes": [ 1889 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1889, 1888 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7582", "outputIndexes": [ 1890 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1880, 1890, 1883 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_306_output_0", "outputIndexes": [ 1891 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1891, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_61_output_0", "outputIndexes": [ 1892 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1892, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_649_output_0", "outputIndexes": [ 1893 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1878, 1893, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_83_output_0", "outputIndexes": [ 1894 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1894 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_41_output_0", "outputIndexes": [ 1895 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1892, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_648_output_0", "outputIndexes": [ 1896 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1878, 20, 1896, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_82_output_0", "outputIndexes": [ 1897 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1895, 1897 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_264_output_0", "outputIndexes": [ 1898 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1898, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_163_output_0", "outputIndexes": [ 1899 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1879, 1899 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_101_output_0", "outputIndexes": [ 1900 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1828 ], "outputIndexes": [ 3823 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3823 ], "outputIndexes": [ 3824 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear", "inputIndexes": [ 3824 ], "outputIndexes": [ 3825 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2007260212, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3825 ], "outputIndexes": [ 3826 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3826 ], "outputIndexes": [ 1901 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1840, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_644_output_0", "outputIndexes": [ 1902 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1848, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_645_output_0", "outputIndexes": [ 1903 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1902, 1903, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_262_output_0", "outputIndexes": [ 1904 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1901, 1904 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_183_output_0", "outputIndexes": [ 1905 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1873, 1900, 1905, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_188_output_0", "outputIndexes": [ 1906 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1906 ], "outputIndexes": [ 3827 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3827 ], "outputIndexes": [ 3828 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear", "inputIndexes": [ 3828 ], "outputIndexes": [ 3829 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2009623626, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3829 ], "outputIndexes": [ 3830 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3830 ], "outputIndexes": [ 1907 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1827, 1907 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_103_output_0", "outputIndexes": [ 1908 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_20/Mul_1_output_0", "inputIndexes": [ 1908 ], "outputIndexes": [ 1909 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2019060832, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1909 ], "outputIndexes": [ 3831 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3831 ], "outputIndexes": [ 3832 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear", "inputIndexes": [ 3832 ], "outputIndexes": [ 3833 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2019093600, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3833 ], "outputIndexes": [ 3834 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3834 ], "outputIndexes": [ 1910 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1910 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_20/Mul_output_0", "outputIndexes": [ 1911 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1909 ], "outputIndexes": [ 3835 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3835 ], "outputIndexes": [ 3836 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear", "inputIndexes": [ 3836 ], "outputIndexes": [ 3837 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2044456054, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3837 ], "outputIndexes": [ 3838 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3838 ], "outputIndexes": [ 1912 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1911, 1912 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_20/Mul_output_0", "outputIndexes": [ 1913 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1913 ], "outputIndexes": [ 3839 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3839 ], "outputIndexes": [ 3840 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear", "inputIndexes": [ 3840 ], "outputIndexes": [ 3841 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2069818508, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3841 ], "outputIndexes": [ 3842 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3842 ], "outputIndexes": [ 1914 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1908, 1914 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_104_output_0", "outputIndexes": [ 1915 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1915, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_189_output_0", "outputIndexes": [ 1916 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_21/Mul_1_output_0", "inputIndexes": [ 1916 ], "outputIndexes": [ 1917 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2095180962, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1917 ], "outputIndexes": [ 3843 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3843 ], "outputIndexes": [ 3844 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear", "inputIndexes": [ 3844 ], "outputIndexes": [ 3845 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2095213730, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3845 ], "outputIndexes": [ 3846 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3846 ], "outputIndexes": [ 1918 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1917 ], "main_type": "NONE", "name": "/Shape_294_output_0", "outputIndexes": [ 1919 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1919 ], "main_type": "NONE", "name": "Shape7901", "outputIndexes": [ 1920 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1919 ], "main_type": "NONE", "name": "Rank7903", "outputIndexes": [ 1921 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1921, 1921 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7905", "outputIndexes": [ 1922 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1922 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7906", "outputIndexes": [ 1923 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1922, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7908", "outputIndexes": [ 1924 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1924 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7909", "outputIndexes": [ 1925 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1920, 1923, 1925, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7912", "outputIndexes": [ 1926 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1926 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7913", "outputIndexes": [ 1927 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1927, 1927 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7915", "outputIndexes": [ 1928 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1919, 1928, 1922 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_318_output_0", "outputIndexes": [ 1929 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1929, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_672_output_0", "outputIndexes": [ 1930 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1922 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7923", "outputIndexes": [ 1931 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1924 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7926", "outputIndexes": [ 1932 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1920, 1931, 1932, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7929", "outputIndexes": [ 1933 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1933 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7930", "outputIndexes": [ 1934 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1934 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7931", "outputIndexes": [ 1935 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1935, 1934 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7932", "outputIndexes": [ 1936 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1919, 1936, 1922 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_319_output_0", "outputIndexes": [ 1937 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1937, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_673_output_0", "outputIndexes": [ 1938 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1930, 1938, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_273_output_0", "outputIndexes": [ 1939 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1918, 1939 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_190_output_0", "outputIndexes": [ 1940 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1940, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_168_output_0", "outputIndexes": [ 1941 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1940 ], "main_type": "NONE", "name": "/Shape_296_output_0", "outputIndexes": [ 1942 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1942 ], "main_type": "NONE", "name": "Shape8033", "outputIndexes": [ 1943 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1942 ], "main_type": "NONE", "name": "Rank8035", "outputIndexes": [ 1944 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1944, 1944 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8037", "outputIndexes": [ 1945 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1945 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8038", "outputIndexes": [ 1946 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1945, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8040", "outputIndexes": [ 1947 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1947 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8041", "outputIndexes": [ 1948 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1943, 1946, 1948, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8044", "outputIndexes": [ 1949 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1949 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8045", "outputIndexes": [ 1950 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1950 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8046", "outputIndexes": [ 1951 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1951, 1950 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8047", "outputIndexes": [ 1952 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1942, 1952, 1945 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_320_output_0", "outputIndexes": [ 1953 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1953, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_63_output_0", "outputIndexes": [ 1954 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1954, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_679_output_0", "outputIndexes": [ 1955 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1940, 1955, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_85_output_0", "outputIndexes": [ 1956 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1956 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_42_output_0", "outputIndexes": [ 1957 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1954, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_678_output_0", "outputIndexes": [ 1958 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1940, 20, 1958, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_84_output_0", "outputIndexes": [ 1959 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1957, 1959 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_276_output_0", "outputIndexes": [ 1960 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1960, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_169_output_0", "outputIndexes": [ 1961 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1941, 1961 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_105_output_0", "outputIndexes": [ 1962 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1917 ], "outputIndexes": [ 3847 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3847 ], "outputIndexes": [ 3848 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear", "inputIndexes": [ 3848 ], "outputIndexes": [ 3849 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2104667320, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3849 ], "outputIndexes": [ 3850 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3850 ], "outputIndexes": [ 1963 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1929, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_674_output_0", "outputIndexes": [ 1964 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1937, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_675_output_0", "outputIndexes": [ 1965 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1964, 1965, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_274_output_0", "outputIndexes": [ 1966 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1963, 1966 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_191_output_0", "outputIndexes": [ 1967 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1967, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_170_output_0", "outputIndexes": [ 1968 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1967 ], "main_type": "NONE", "name": "/Shape_297_output_0", "outputIndexes": [ 1969 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1969 ], "main_type": "NONE", "name": "Shape7939", "outputIndexes": [ 1970 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1969 ], "main_type": "NONE", "name": "Rank7941", "outputIndexes": [ 1971 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1971, 1971 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7943", "outputIndexes": [ 1972 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1972 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7944", "outputIndexes": [ 1973 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1972, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7946", "outputIndexes": [ 1974 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1974 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7947", "outputIndexes": [ 1975 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1970, 1973, 1975, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7950", "outputIndexes": [ 1976 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1976 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7951", "outputIndexes": [ 1977 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 1977 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7952", "outputIndexes": [ 1978 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1978, 1977 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7953", "outputIndexes": [ 1979 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1969, 1979, 1972 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_321_output_0", "outputIndexes": [ 1980 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1980, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_64_output_0", "outputIndexes": [ 1981 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1981, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_681_output_0", "outputIndexes": [ 1982 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1967, 1982, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_87_output_0", "outputIndexes": [ 1983 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1983 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_43_output_0", "outputIndexes": [ 1984 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1981, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_680_output_0", "outputIndexes": [ 1985 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1967, 20, 1985, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_86_output_0", "outputIndexes": [ 1986 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1984, 1986 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_277_output_0", "outputIndexes": [ 1987 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1987, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_171_output_0", "outputIndexes": [ 1988 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1968, 1988 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_106_output_0", "outputIndexes": [ 1989 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1917 ], "outputIndexes": [ 3851 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3851 ], "outputIndexes": [ 3852 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear", "inputIndexes": [ 3852 ], "outputIndexes": [ 3853 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2107030734, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3853 ], "outputIndexes": [ 3854 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3854 ], "outputIndexes": [ 1990 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1929, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_676_output_0", "outputIndexes": [ 1991 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1937, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_677_output_0", "outputIndexes": [ 1992 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1991, 1992, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_275_output_0", "outputIndexes": [ 1993 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1990, 1993 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_192_output_0", "outputIndexes": [ 1994 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1962, 1989, 1994, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_197_output_0", "outputIndexes": [ 1995 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1995 ], "outputIndexes": [ 3855 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3855 ], "outputIndexes": [ 3856 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear", "inputIndexes": [ 3856 ], "outputIndexes": [ 3857 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2109394148, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3857 ], "outputIndexes": [ 3858 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3858 ], "outputIndexes": [ 1996 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1916, 1996 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_108_output_0", "outputIndexes": [ 1997 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_21/Mul_1_output_0", "inputIndexes": [ 1997 ], "outputIndexes": [ 1998 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2118831354, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1998 ], "outputIndexes": [ 3859 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3859 ], "outputIndexes": [ 3860 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear", "inputIndexes": [ 3860 ], "outputIndexes": [ 3861 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2118864122, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3861 ], "outputIndexes": [ 3862 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3862 ], "outputIndexes": [ 1999 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1999 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_21/Mul_output_0", "outputIndexes": [ 2000 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1998 ], "outputIndexes": [ 3863 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3863 ], "outputIndexes": [ 3864 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear", "inputIndexes": [ 3864 ], "outputIndexes": [ 3865 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2144226576, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3865 ], "outputIndexes": [ 3866 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3866 ], "outputIndexes": [ 2001 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2000, 2001 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_21/Mul_output_0", "outputIndexes": [ 2002 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2002 ], "outputIndexes": [ 3867 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3867 ], "outputIndexes": [ 3868 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear", "inputIndexes": [ 3868 ], "outputIndexes": [ 3869 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2169589030, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3869 ], "outputIndexes": [ 3870 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3870 ], "outputIndexes": [ 2003 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1997, 2003 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_109_output_0", "outputIndexes": [ 2004 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2004, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_198_output_0", "outputIndexes": [ 2005 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_22/Mul_1_output_0", "inputIndexes": [ 2005 ], "outputIndexes": [ 2006 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2194951484, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2006 ], "outputIndexes": [ 3871 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3871 ], "outputIndexes": [ 3872 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear", "inputIndexes": [ 3872 ], "outputIndexes": [ 3873 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2194984252, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3873 ], "outputIndexes": [ 3874 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3874 ], "outputIndexes": [ 2007 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2006 ], "main_type": "NONE", "name": "/Shape_308_output_0", "outputIndexes": [ 2008 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2008 ], "main_type": "NONE", "name": "Shape8272", "outputIndexes": [ 2009 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2008 ], "main_type": "NONE", "name": "Rank8274", "outputIndexes": [ 2010 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2010, 2010 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8276", "outputIndexes": [ 2011 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2011 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8277", "outputIndexes": [ 2012 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2011, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8279", "outputIndexes": [ 2013 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2013 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8280", "outputIndexes": [ 2014 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2009, 2012, 2014, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8283", "outputIndexes": [ 2015 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2015 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8284", "outputIndexes": [ 2016 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2016, 2016 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8286", "outputIndexes": [ 2017 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2008, 2017, 2011 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_333_output_0", "outputIndexes": [ 2018 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2018, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_704_output_0", "outputIndexes": [ 2019 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2011 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8294", "outputIndexes": [ 2020 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2013 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8297", "outputIndexes": [ 2021 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2009, 2020, 2021, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8300", "outputIndexes": [ 2022 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2022 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8301", "outputIndexes": [ 2023 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2023 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8302", "outputIndexes": [ 2024 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2024, 2023 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8303", "outputIndexes": [ 2025 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2008, 2025, 2011 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_334_output_0", "outputIndexes": [ 2026 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2026, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_705_output_0", "outputIndexes": [ 2027 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2019, 2027, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_286_output_0", "outputIndexes": [ 2028 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2007, 2028 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_199_output_0", "outputIndexes": [ 2029 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2029, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_176_output_0", "outputIndexes": [ 2030 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2029 ], "main_type": "NONE", "name": "/Shape_310_output_0", "outputIndexes": [ 2031 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2031 ], "main_type": "NONE", "name": "Shape8404", "outputIndexes": [ 2032 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2031 ], "main_type": "NONE", "name": "Rank8406", "outputIndexes": [ 2033 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2033, 2033 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8408", "outputIndexes": [ 2034 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2034 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8409", "outputIndexes": [ 2035 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2034, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8411", "outputIndexes": [ 2036 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2036 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8412", "outputIndexes": [ 2037 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2032, 2035, 2037, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8415", "outputIndexes": [ 2038 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2038 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8416", "outputIndexes": [ 2039 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2039 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8417", "outputIndexes": [ 2040 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2040, 2039 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8418", "outputIndexes": [ 2041 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2031, 2041, 2034 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_335_output_0", "outputIndexes": [ 2042 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2042, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_66_output_0", "outputIndexes": [ 2043 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2043, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_711_output_0", "outputIndexes": [ 2044 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2029, 2044, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_89_output_0", "outputIndexes": [ 2045 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2045 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_44_output_0", "outputIndexes": [ 2046 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2043, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_710_output_0", "outputIndexes": [ 2047 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2029, 20, 2047, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_88_output_0", "outputIndexes": [ 2048 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2046, 2048 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_289_output_0", "outputIndexes": [ 2049 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2049, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_177_output_0", "outputIndexes": [ 2050 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2030, 2050 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_110_output_0", "outputIndexes": [ 2051 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2006 ], "outputIndexes": [ 3875 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3875 ], "outputIndexes": [ 3876 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear", "inputIndexes": [ 3876 ], "outputIndexes": [ 3877 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2204437842, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3877 ], "outputIndexes": [ 3878 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3878 ], "outputIndexes": [ 2052 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2018, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_706_output_0", "outputIndexes": [ 2053 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2026, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_707_output_0", "outputIndexes": [ 2054 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2053, 2054, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_287_output_0", "outputIndexes": [ 2055 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2052, 2055 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_200_output_0", "outputIndexes": [ 2056 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2056, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_178_output_0", "outputIndexes": [ 2057 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2056 ], "main_type": "NONE", "name": "/Shape_311_output_0", "outputIndexes": [ 2058 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2058 ], "main_type": "NONE", "name": "Shape8310", "outputIndexes": [ 2059 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2058 ], "main_type": "NONE", "name": "Rank8312", "outputIndexes": [ 2060 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2060, 2060 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8314", "outputIndexes": [ 2061 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2061 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8315", "outputIndexes": [ 2062 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2061, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8317", "outputIndexes": [ 2063 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2063 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8318", "outputIndexes": [ 2064 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2059, 2062, 2064, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8321", "outputIndexes": [ 2065 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2065 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8322", "outputIndexes": [ 2066 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2066 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8323", "outputIndexes": [ 2067 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2067, 2066 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8324", "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_336_output_0", "outputIndexes": [ 2069 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2069, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_67_output_0", "outputIndexes": [ 2070 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2070, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_713_output_0", "outputIndexes": [ 2071 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2056, 2071, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_91_output_0", "outputIndexes": [ 2072 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2072 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_45_output_0", "outputIndexes": [ 2073 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2070, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_712_output_0", "outputIndexes": [ 2074 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2056, 20, 2074, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_90_output_0", "outputIndexes": [ 2075 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2073, 2075 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_290_output_0", "outputIndexes": [ 2076 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2076, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_179_output_0", "outputIndexes": [ 2077 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2057, 2077 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_111_output_0", "outputIndexes": [ 2078 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2006 ], "outputIndexes": [ 3879 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3879 ], "outputIndexes": [ 3880 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear", "inputIndexes": [ 3880 ], "outputIndexes": [ 3881 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2206801256, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3881 ], "outputIndexes": [ 3882 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3882 ], "outputIndexes": [ 2079 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2018, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_708_output_0", "outputIndexes": [ 2080 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2026, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_709_output_0", "outputIndexes": [ 2081 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2080, 2081, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_288_output_0", "outputIndexes": [ 2082 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2079, 2082 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_201_output_0", "outputIndexes": [ 2083 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2051, 2078, 2083, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_206_output_0", "outputIndexes": [ 2084 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2084 ], "outputIndexes": [ 3883 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3883 ], "outputIndexes": [ 3884 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear", "inputIndexes": [ 3884 ], "outputIndexes": [ 3885 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2209164670, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3885 ], "outputIndexes": [ 3886 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3886 ], "outputIndexes": [ 2085 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2005, 2085 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_113_output_0", "outputIndexes": [ 2086 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_22/Mul_1_output_0", "inputIndexes": [ 2086 ], "outputIndexes": [ 2087 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2218601876, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2087 ], "outputIndexes": [ 3887 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3887 ], "outputIndexes": [ 3888 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear", "inputIndexes": [ 3888 ], "outputIndexes": [ 3889 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2218634644, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3889 ], "outputIndexes": [ 3890 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3890 ], "outputIndexes": [ 2088 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2088 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_22/Mul_output_0", "outputIndexes": [ 2089 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2087 ], "outputIndexes": [ 3891 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3891 ], "outputIndexes": [ 3892 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear", "inputIndexes": [ 3892 ], "outputIndexes": [ 3893 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2243997098, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3893 ], "outputIndexes": [ 3894 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3894 ], "outputIndexes": [ 2090 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2089, 2090 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_22/Mul_output_0", "outputIndexes": [ 2091 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2091 ], "outputIndexes": [ 3895 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3895 ], "outputIndexes": [ 3896 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear", "inputIndexes": [ 3896 ], "outputIndexes": [ 3897 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2269359552, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3897 ], "outputIndexes": [ 3898 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3898 ], "outputIndexes": [ 2092 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2086, 2092 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_114_output_0", "outputIndexes": [ 2093 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2093, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_207_output_0", "outputIndexes": [ 2094 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_23/Mul_1_output_0", "inputIndexes": [ 2094 ], "outputIndexes": [ 2095 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2294722006, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2095 ], "outputIndexes": [ 3899 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3899 ], "outputIndexes": [ 3900 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear", "inputIndexes": [ 3900 ], "outputIndexes": [ 3901 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2294754774, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3901 ], "outputIndexes": [ 3902 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3902 ], "outputIndexes": [ 2096 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2095 ], "main_type": "NONE", "name": "/Shape_322_output_0", "outputIndexes": [ 2097 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2097 ], "main_type": "NONE", "name": "Shape8643", "outputIndexes": [ 2098 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2097 ], "main_type": "NONE", "name": "Rank8645", "outputIndexes": [ 2099 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2099, 2099 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8647", "outputIndexes": [ 2100 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2100 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8648", "outputIndexes": [ 2101 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2100, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8650", "outputIndexes": [ 2102 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2102 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8651", "outputIndexes": [ 2103 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2098, 2101, 2103, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8654", "outputIndexes": [ 2104 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2104 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8655", "outputIndexes": [ 2105 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2105, 2105 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8657", "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_348_output_0", "outputIndexes": [ 2107 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2107, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_736_output_0", "outputIndexes": [ 2108 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2100 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8665", "outputIndexes": [ 2109 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2102 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8668", "outputIndexes": [ 2110 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2098, 2109, 2110, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8671", "outputIndexes": [ 2111 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2111 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8672", "outputIndexes": [ 2112 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2112 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8673", "outputIndexes": [ 2113 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2113, 2112 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8674", "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_349_output_0", "outputIndexes": [ 2115 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2115, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_737_output_0", "outputIndexes": [ 2116 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2108, 2116, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_299_output_0", "outputIndexes": [ 2117 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2096, 2117 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_208_output_0", "outputIndexes": [ 2118 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2118, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_184_output_0", "outputIndexes": [ 2119 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2118 ], "main_type": "NONE", "name": "/Shape_324_output_0", "outputIndexes": [ 2120 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2120 ], "main_type": "NONE", "name": "Shape8775", "outputIndexes": [ 2121 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2120 ], "main_type": "NONE", "name": "Rank8777", "outputIndexes": [ 2122 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2122, 2122 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8779", "outputIndexes": [ 2123 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2123 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8780", "outputIndexes": [ 2124 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2123, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8782", "outputIndexes": [ 2125 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2125 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8783", "outputIndexes": [ 2126 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2121, 2124, 2126, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8786", "outputIndexes": [ 2127 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2127 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8787", "outputIndexes": [ 2128 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2128 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8788", "outputIndexes": [ 2129 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2129, 2128 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8789", "outputIndexes": [ 2130 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2120, 2130, 2123 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_350_output_0", "outputIndexes": [ 2131 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2131, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_69_output_0", "outputIndexes": [ 2132 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2132, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_743_output_0", "outputIndexes": [ 2133 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2118, 2133, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_93_output_0", "outputIndexes": [ 2134 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2134 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_46_output_0", "outputIndexes": [ 2135 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2132, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_742_output_0", "outputIndexes": [ 2136 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2118, 20, 2136, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_92_output_0", "outputIndexes": [ 2137 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2135, 2137 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_302_output_0", "outputIndexes": [ 2138 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2138, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_185_output_0", "outputIndexes": [ 2139 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2119, 2139 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_115_output_0", "outputIndexes": [ 2140 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2095 ], "outputIndexes": [ 3903 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3903 ], "outputIndexes": [ 3904 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear", "inputIndexes": [ 3904 ], "outputIndexes": [ 3905 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2304208364, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3905 ], "outputIndexes": [ 3906 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3906 ], "outputIndexes": [ 2141 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2107, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_738_output_0", "outputIndexes": [ 2142 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2115, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_739_output_0", "outputIndexes": [ 2143 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2142, 2143, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_300_output_0", "outputIndexes": [ 2144 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2141, 2144 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_209_output_0", "outputIndexes": [ 2145 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2145, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_186_output_0", "outputIndexes": [ 2146 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2145 ], "main_type": "NONE", "name": "/Shape_325_output_0", "outputIndexes": [ 2147 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2147 ], "main_type": "NONE", "name": "Shape8681", "outputIndexes": [ 2148 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2147 ], "main_type": "NONE", "name": "Rank8683", "outputIndexes": [ 2149 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2149, 2149 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8685", "outputIndexes": [ 2150 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2150 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8686", "outputIndexes": [ 2151 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2150, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8688", "outputIndexes": [ 2152 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2152 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8689", "outputIndexes": [ 2153 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2148, 2151, 2153, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8692", "outputIndexes": [ 2154 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2154 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8693", "outputIndexes": [ 2155 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2155 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8694", "outputIndexes": [ 2156 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2156, 2155 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8695", "outputIndexes": [ 2157 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2147, 2157, 2150 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_351_output_0", "outputIndexes": [ 2158 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2158, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_70_output_0", "outputIndexes": [ 2159 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2159, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_745_output_0", "outputIndexes": [ 2160 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2145, 2160, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_95_output_0", "outputIndexes": [ 2161 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2161 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_47_output_0", "outputIndexes": [ 2162 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2159, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_744_output_0", "outputIndexes": [ 2163 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2145, 20, 2163, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_94_output_0", "outputIndexes": [ 2164 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2162, 2164 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_303_output_0", "outputIndexes": [ 2165 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2165, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_187_output_0", "outputIndexes": [ 2166 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2146, 2166 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_116_output_0", "outputIndexes": [ 2167 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2095 ], "outputIndexes": [ 3907 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3907 ], "outputIndexes": [ 3908 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear", "inputIndexes": [ 3908 ], "outputIndexes": [ 3909 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2306571778, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3909 ], "outputIndexes": [ 3910 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3910 ], "outputIndexes": [ 2168 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2107, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_740_output_0", "outputIndexes": [ 2169 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2115, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_741_output_0", "outputIndexes": [ 2170 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2169, 2170, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_301_output_0", "outputIndexes": [ 2171 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2168, 2171 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_210_output_0", "outputIndexes": [ 2172 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2140, 2167, 2172, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_215_output_0", "outputIndexes": [ 2173 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2173 ], "outputIndexes": [ 3911 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3911 ], "outputIndexes": [ 3912 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear", "inputIndexes": [ 3912 ], "outputIndexes": [ 3913 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2308935192, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3913 ], "outputIndexes": [ 3914 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3914 ], "outputIndexes": [ 2174 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2094, 2174 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_118_output_0", "outputIndexes": [ 2175 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_23/Mul_1_output_0", "inputIndexes": [ 2175 ], "outputIndexes": [ 2176 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2318372398, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2176 ], "outputIndexes": [ 3915 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3915 ], "outputIndexes": [ 3916 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear", "inputIndexes": [ 3916 ], "outputIndexes": [ 3917 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2318405166, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3917 ], "outputIndexes": [ 3918 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3918 ], "outputIndexes": [ 2177 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2177 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_23/Mul_output_0", "outputIndexes": [ 2178 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2176 ], "outputIndexes": [ 3919 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3919 ], "outputIndexes": [ 3920 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear", "inputIndexes": [ 3920 ], "outputIndexes": [ 3921 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2343767620, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3921 ], "outputIndexes": [ 3922 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3922 ], "outputIndexes": [ 2179 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2178, 2179 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_23/Mul_output_0", "outputIndexes": [ 2180 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2180 ], "outputIndexes": [ 3923 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3923 ], "outputIndexes": [ 3924 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear", "inputIndexes": [ 3924 ], "outputIndexes": [ 3925 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2369130074, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3925 ], "outputIndexes": [ 3926 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3926 ], "outputIndexes": [ 2181 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2175, 2181 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_119_output_0", "outputIndexes": [ 2182 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2182, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_216_output_0", "outputIndexes": [ 2183 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_24/Mul_1_output_0", "inputIndexes": [ 2183 ], "outputIndexes": [ 2184 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2394492528, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2184 ], "outputIndexes": [ 3927 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3927 ], "outputIndexes": [ 3928 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear", "inputIndexes": [ 3928 ], "outputIndexes": [ 3929 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2394525296, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3929 ], "outputIndexes": [ 3930 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3930 ], "outputIndexes": [ 2185 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2184 ], "main_type": "NONE", "name": "/Shape_336_output_0", "outputIndexes": [ 2186 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2186 ], "main_type": "NONE", "name": "Shape9014", "outputIndexes": [ 2187 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2186 ], "main_type": "NONE", "name": "Rank9016", "outputIndexes": [ 2188 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2188, 2188 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9018", "outputIndexes": [ 2189 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2189 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9019", "outputIndexes": [ 2190 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2189, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9021", "outputIndexes": [ 2191 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2191 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9022", "outputIndexes": [ 2192 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2187, 2190, 2192, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9025", "outputIndexes": [ 2193 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2193 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9026", "outputIndexes": [ 2194 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2194, 2194 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9028", "outputIndexes": [ 2195 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2186, 2195, 2189 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_363_output_0", "outputIndexes": [ 2196 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2196, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_768_output_0", "outputIndexes": [ 2197 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2189 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9036", "outputIndexes": [ 2198 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2191 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9039", "outputIndexes": [ 2199 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2187, 2198, 2199, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9042", "outputIndexes": [ 2200 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2200 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9043", "outputIndexes": [ 2201 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2201 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9044", "outputIndexes": [ 2202 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2202, 2201 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9045", "outputIndexes": [ 2203 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2186, 2203, 2189 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_364_output_0", "outputIndexes": [ 2204 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2204, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_769_output_0", "outputIndexes": [ 2205 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2197, 2205, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_312_output_0", "outputIndexes": [ 2206 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2185, 2206 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_217_output_0", "outputIndexes": [ 2207 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2207, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_192_output_0", "outputIndexes": [ 2208 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2207 ], "main_type": "NONE", "name": "/Shape_338_output_0", "outputIndexes": [ 2209 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2209 ], "main_type": "NONE", "name": "Shape9146", "outputIndexes": [ 2210 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2209 ], "main_type": "NONE", "name": "Rank9148", "outputIndexes": [ 2211 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2211, 2211 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9150", "outputIndexes": [ 2212 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2212 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9151", "outputIndexes": [ 2213 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2212, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9153", "outputIndexes": [ 2214 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2214 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9154", "outputIndexes": [ 2215 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2210, 2213, 2215, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9157", "outputIndexes": [ 2216 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2216 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9158", "outputIndexes": [ 2217 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2217 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9159", "outputIndexes": [ 2218 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2218, 2217 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9160", "outputIndexes": [ 2219 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2209, 2219, 2212 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_365_output_0", "outputIndexes": [ 2220 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2220, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_72_output_0", "outputIndexes": [ 2221 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2221, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_775_output_0", "outputIndexes": [ 2222 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2207, 2222, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_97_output_0", "outputIndexes": [ 2223 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2223 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_48_output_0", "outputIndexes": [ 2224 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2221, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_774_output_0", "outputIndexes": [ 2225 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2207, 20, 2225, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_96_output_0", "outputIndexes": [ 2226 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2224, 2226 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_315_output_0", "outputIndexes": [ 2227 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2227, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_193_output_0", "outputIndexes": [ 2228 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2208, 2228 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_120_output_0", "outputIndexes": [ 2229 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2184 ], "outputIndexes": [ 3931 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3931 ], "outputIndexes": [ 3932 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear", "inputIndexes": [ 3932 ], "outputIndexes": [ 3933 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2403978886, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3933 ], "outputIndexes": [ 3934 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3934 ], "outputIndexes": [ 2230 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2196, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_770_output_0", "outputIndexes": [ 2231 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2204, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_771_output_0", "outputIndexes": [ 2232 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2231, 2232, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_313_output_0", "outputIndexes": [ 2233 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2230, 2233 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_218_output_0", "outputIndexes": [ 2234 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2234, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_194_output_0", "outputIndexes": [ 2235 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2234 ], "main_type": "NONE", "name": "/Shape_339_output_0", "outputIndexes": [ 2236 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2236 ], "main_type": "NONE", "name": "Shape9052", "outputIndexes": [ 2237 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2236 ], "main_type": "NONE", "name": "Rank9054", "outputIndexes": [ 2238 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2238, 2238 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9056", "outputIndexes": [ 2239 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2239 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9057", "outputIndexes": [ 2240 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2239, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9059", "outputIndexes": [ 2241 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2241 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9060", "outputIndexes": [ 2242 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2237, 2240, 2242, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9063", "outputIndexes": [ 2243 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2243 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9064", "outputIndexes": [ 2244 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2244 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9065", "outputIndexes": [ 2245 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2245, 2244 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9066", "outputIndexes": [ 2246 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2236, 2246, 2239 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_366_output_0", "outputIndexes": [ 2247 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2247, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_73_output_0", "outputIndexes": [ 2248 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2248, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_777_output_0", "outputIndexes": [ 2249 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2234, 2249, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_99_output_0", "outputIndexes": [ 2250 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2250 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_49_output_0", "outputIndexes": [ 2251 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2248, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_776_output_0", "outputIndexes": [ 2252 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2234, 20, 2252, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_98_output_0", "outputIndexes": [ 2253 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2251, 2253 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_316_output_0", "outputIndexes": [ 2254 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2254, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_195_output_0", "outputIndexes": [ 2255 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2235, 2255 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_121_output_0", "outputIndexes": [ 2256 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2184 ], "outputIndexes": [ 3935 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3935 ], "outputIndexes": [ 3936 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear", "inputIndexes": [ 3936 ], "outputIndexes": [ 3937 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2406342300, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3937 ], "outputIndexes": [ 3938 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3938 ], "outputIndexes": [ 2257 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2196, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_772_output_0", "outputIndexes": [ 2258 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2204, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_773_output_0", "outputIndexes": [ 2259 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2258, 2259, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_314_output_0", "outputIndexes": [ 2260 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2257, 2260 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_219_output_0", "outputIndexes": [ 2261 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2229, 2256, 2261, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_224_output_0", "outputIndexes": [ 2262 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2262 ], "outputIndexes": [ 3939 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3939 ], "outputIndexes": [ 3940 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear", "inputIndexes": [ 3940 ], "outputIndexes": [ 3941 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2408705714, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3941 ], "outputIndexes": [ 3942 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3942 ], "outputIndexes": [ 2263 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2183, 2263 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_123_output_0", "outputIndexes": [ 2264 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_24/Mul_1_output_0", "inputIndexes": [ 2264 ], "outputIndexes": [ 2265 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2418142920, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2265 ], "outputIndexes": [ 3943 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3943 ], "outputIndexes": [ 3944 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear", "inputIndexes": [ 3944 ], "outputIndexes": [ 3945 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2418175688, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3945 ], "outputIndexes": [ 3946 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3946 ], "outputIndexes": [ 2266 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2266 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_24/Mul_output_0", "outputIndexes": [ 2267 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2265 ], "outputIndexes": [ 3947 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3947 ], "outputIndexes": [ 3948 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear", "inputIndexes": [ 3948 ], "outputIndexes": [ 3949 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2443538142, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3949 ], "outputIndexes": [ 3950 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3950 ], "outputIndexes": [ 2268 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2267, 2268 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_24/Mul_output_0", "outputIndexes": [ 2269 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2269 ], "outputIndexes": [ 3951 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3951 ], "outputIndexes": [ 3952 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear", "inputIndexes": [ 3952 ], "outputIndexes": [ 3953 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2468900596, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3953 ], "outputIndexes": [ 3954 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3954 ], "outputIndexes": [ 2270 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2264, 2270 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_124_output_0", "outputIndexes": [ 2271 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2271, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_225_output_0", "outputIndexes": [ 2272 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_25/Mul_1_output_0", "inputIndexes": [ 2272 ], "outputIndexes": [ 2273 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2494263050, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2273 ], "outputIndexes": [ 3955 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3955 ], "outputIndexes": [ 3956 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear", "inputIndexes": [ 3956 ], "outputIndexes": [ 3957 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2494295818, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3957 ], "outputIndexes": [ 3958 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3958 ], "outputIndexes": [ 2274 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2273 ], "main_type": "NONE", "name": "/Shape_350_output_0", "outputIndexes": [ 2275 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2275 ], "main_type": "NONE", "name": "Shape9385", "outputIndexes": [ 2276 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2275 ], "main_type": "NONE", "name": "Rank9387", "outputIndexes": [ 2277 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2277, 2277 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9389", "outputIndexes": [ 2278 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2278 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9390", "outputIndexes": [ 2279 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2278, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9392", "outputIndexes": [ 2280 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2280 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9393", "outputIndexes": [ 2281 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2276, 2279, 2281, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9396", "outputIndexes": [ 2282 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2282 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9397", "outputIndexes": [ 2283 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2283, 2283 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9399", "outputIndexes": [ 2284 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2275, 2284, 2278 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_378_output_0", "outputIndexes": [ 2285 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2285, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_800_output_0", "outputIndexes": [ 2286 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2278 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9407", "outputIndexes": [ 2287 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2280 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9410", "outputIndexes": [ 2288 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2276, 2287, 2288, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9413", "outputIndexes": [ 2289 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2289 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9414", "outputIndexes": [ 2290 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2290 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9415", "outputIndexes": [ 2291 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2291, 2290 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9416", "outputIndexes": [ 2292 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2275, 2292, 2278 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_379_output_0", "outputIndexes": [ 2293 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2293, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_801_output_0", "outputIndexes": [ 2294 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2286, 2294, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_325_output_0", "outputIndexes": [ 2295 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2274, 2295 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_226_output_0", "outputIndexes": [ 2296 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2296, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_200_output_0", "outputIndexes": [ 2297 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2296 ], "main_type": "NONE", "name": "/Shape_352_output_0", "outputIndexes": [ 2298 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2298 ], "main_type": "NONE", "name": "Shape9517", "outputIndexes": [ 2299 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2298 ], "main_type": "NONE", "name": "Rank9519", "outputIndexes": [ 2300 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2300, 2300 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9521", "outputIndexes": [ 2301 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2301 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9522", "outputIndexes": [ 2302 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2301, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9524", "outputIndexes": [ 2303 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2303 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9525", "outputIndexes": [ 2304 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2299, 2302, 2304, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9528", "outputIndexes": [ 2305 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2305 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9529", "outputIndexes": [ 2306 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2306 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9530", "outputIndexes": [ 2307 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2307, 2306 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9531", "outputIndexes": [ 2308 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2298, 2308, 2301 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_380_output_0", "outputIndexes": [ 2309 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2309, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_75_output_0", "outputIndexes": [ 2310 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2310, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_807_output_0", "outputIndexes": [ 2311 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2296, 2311, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_101_output_0", "outputIndexes": [ 2312 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2312 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_50_output_0", "outputIndexes": [ 2313 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2310, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_806_output_0", "outputIndexes": [ 2314 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2296, 20, 2314, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_100_output_0", "outputIndexes": [ 2315 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2313, 2315 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_328_output_0", "outputIndexes": [ 2316 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2316, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_201_output_0", "outputIndexes": [ 2317 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2297, 2317 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_125_output_0", "outputIndexes": [ 2318 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2273 ], "outputIndexes": [ 3959 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3959 ], "outputIndexes": [ 3960 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear", "inputIndexes": [ 3960 ], "outputIndexes": [ 3961 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2503749408, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3961 ], "outputIndexes": [ 3962 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3962 ], "outputIndexes": [ 2319 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2285, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_802_output_0", "outputIndexes": [ 2320 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2293, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_803_output_0", "outputIndexes": [ 2321 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2320, 2321, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_326_output_0", "outputIndexes": [ 2322 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2319, 2322 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_227_output_0", "outputIndexes": [ 2323 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2323, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_202_output_0", "outputIndexes": [ 2324 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2323 ], "main_type": "NONE", "name": "/Shape_353_output_0", "outputIndexes": [ 2325 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2325 ], "main_type": "NONE", "name": "Shape9423", "outputIndexes": [ 2326 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2325 ], "main_type": "NONE", "name": "Rank9425", "outputIndexes": [ 2327 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2327, 2327 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9427", "outputIndexes": [ 2328 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2328 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9428", "outputIndexes": [ 2329 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2328, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9430", "outputIndexes": [ 2330 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2330 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9431", "outputIndexes": [ 2331 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2326, 2329, 2331, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9434", "outputIndexes": [ 2332 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2332 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9435", "outputIndexes": [ 2333 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2333 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9436", "outputIndexes": [ 2334 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2334, 2333 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9437", "outputIndexes": [ 2335 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2325, 2335, 2328 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_381_output_0", "outputIndexes": [ 2336 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2336, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_76_output_0", "outputIndexes": [ 2337 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2337, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_809_output_0", "outputIndexes": [ 2338 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2323, 2338, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_103_output_0", "outputIndexes": [ 2339 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2339 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_51_output_0", "outputIndexes": [ 2340 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2337, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_808_output_0", "outputIndexes": [ 2341 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2323, 20, 2341, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_102_output_0", "outputIndexes": [ 2342 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2340, 2342 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_329_output_0", "outputIndexes": [ 2343 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2343, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_203_output_0", "outputIndexes": [ 2344 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2324, 2344 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_126_output_0", "outputIndexes": [ 2345 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2273 ], "outputIndexes": [ 3963 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3963 ], "outputIndexes": [ 3964 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear", "inputIndexes": [ 3964 ], "outputIndexes": [ 3965 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2506112822, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3965 ], "outputIndexes": [ 3966 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3966 ], "outputIndexes": [ 2346 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2285, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_804_output_0", "outputIndexes": [ 2347 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2293, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_805_output_0", "outputIndexes": [ 2348 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2347, 2348, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_327_output_0", "outputIndexes": [ 2349 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2346, 2349 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_228_output_0", "outputIndexes": [ 2350 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2318, 2345, 2350, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_233_output_0", "outputIndexes": [ 2351 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2351 ], "outputIndexes": [ 3967 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3967 ], "outputIndexes": [ 3968 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear", "inputIndexes": [ 3968 ], "outputIndexes": [ 3969 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2508476236, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3969 ], "outputIndexes": [ 3970 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3970 ], "outputIndexes": [ 2352 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2272, 2352 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_128_output_0", "outputIndexes": [ 2353 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_25/Mul_1_output_0", "inputIndexes": [ 2353 ], "outputIndexes": [ 2354 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2517913442, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2354 ], "outputIndexes": [ 3971 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3971 ], "outputIndexes": [ 3972 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear", "inputIndexes": [ 3972 ], "outputIndexes": [ 3973 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2517946210, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 3973 ], "outputIndexes": [ 3974 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3974 ], "outputIndexes": [ 2355 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2355 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_25/Mul_output_0", "outputIndexes": [ 2356 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2354 ], "outputIndexes": [ 3975 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 3975 ], "outputIndexes": [ 3976 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear", "inputIndexes": [ 3976 ], "outputIndexes": [ 3977 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2543308664, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 3977 ], "outputIndexes": [ 3978 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3978 ], "outputIndexes": [ 2357 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2356, 2357 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_25/Mul_output_0", "outputIndexes": [ 2358 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2358 ], "outputIndexes": [ 3979 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 3979 ], "outputIndexes": [ 3980 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear", "inputIndexes": [ 3980 ], "outputIndexes": [ 3981 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2568671118, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 3981 ], "outputIndexes": [ 3982 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3982 ], "outputIndexes": [ 2359 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2353, 2359 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_129_output_0", "outputIndexes": [ 2360 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2360, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_234_output_0", "outputIndexes": [ 2361 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_26/Mul_1_output_0", "inputIndexes": [ 2361 ], "outputIndexes": [ 2362 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2594033572, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2362 ], "outputIndexes": [ 3983 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 3983 ], "outputIndexes": [ 3984 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear", "inputIndexes": [ 3984 ], "outputIndexes": [ 3985 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2594066340, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 3985 ], "outputIndexes": [ 3986 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3986 ], "outputIndexes": [ 2363 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2362 ], "main_type": "NONE", "name": "/Shape_364_output_0", "outputIndexes": [ 2364 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2364 ], "main_type": "NONE", "name": "Shape9756", "outputIndexes": [ 2365 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2364 ], "main_type": "NONE", "name": "Rank9758", "outputIndexes": [ 2366 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2366, 2366 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9760", "outputIndexes": [ 2367 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2367 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9761", "outputIndexes": [ 2368 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2367, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9763", "outputIndexes": [ 2369 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2369 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9764", "outputIndexes": [ 2370 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2365, 2368, 2370, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9767", "outputIndexes": [ 2371 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2371 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9768", "outputIndexes": [ 2372 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2372, 2372 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9770", "outputIndexes": [ 2373 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2364, 2373, 2367 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_393_output_0", "outputIndexes": [ 2374 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2374, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_832_output_0", "outputIndexes": [ 2375 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2367 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9778", "outputIndexes": [ 2376 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2369 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9781", "outputIndexes": [ 2377 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2365, 2376, 2377, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9784", "outputIndexes": [ 2378 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2378 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9785", "outputIndexes": [ 2379 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2379 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9786", "outputIndexes": [ 2380 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2380, 2379 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9787", "outputIndexes": [ 2381 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2364, 2381, 2367 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_394_output_0", "outputIndexes": [ 2382 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2382, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_833_output_0", "outputIndexes": [ 2383 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2375, 2383, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_338_output_0", "outputIndexes": [ 2384 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2363, 2384 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_235_output_0", "outputIndexes": [ 2385 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2385, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_208_output_0", "outputIndexes": [ 2386 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2385 ], "main_type": "NONE", "name": "/Shape_366_output_0", "outputIndexes": [ 2387 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2387 ], "main_type": "NONE", "name": "Shape9888", "outputIndexes": [ 2388 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2387 ], "main_type": "NONE", "name": "Rank9890", "outputIndexes": [ 2389 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2389, 2389 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9892", "outputIndexes": [ 2390 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2390 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9893", "outputIndexes": [ 2391 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2390, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9895", "outputIndexes": [ 2392 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2392 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9896", "outputIndexes": [ 2393 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2388, 2391, 2393, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9899", "outputIndexes": [ 2394 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2394 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9900", "outputIndexes": [ 2395 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2395 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9901", "outputIndexes": [ 2396 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2396, 2395 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9902", "outputIndexes": [ 2397 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2387, 2397, 2390 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_395_output_0", "outputIndexes": [ 2398 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2398, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_78_output_0", "outputIndexes": [ 2399 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2399, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_839_output_0", "outputIndexes": [ 2400 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2385, 2400, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_105_output_0", "outputIndexes": [ 2401 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2401 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_52_output_0", "outputIndexes": [ 2402 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2399, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_838_output_0", "outputIndexes": [ 2403 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2385, 20, 2403, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_104_output_0", "outputIndexes": [ 2404 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2402, 2404 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_341_output_0", "outputIndexes": [ 2405 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2405, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_209_output_0", "outputIndexes": [ 2406 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2386, 2406 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_130_output_0", "outputIndexes": [ 2407 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2362 ], "outputIndexes": [ 3987 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 3987 ], "outputIndexes": [ 3988 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear", "inputIndexes": [ 3988 ], "outputIndexes": [ 3989 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2603519930, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 3989 ], "outputIndexes": [ 3990 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3990 ], "outputIndexes": [ 2408 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2374, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_834_output_0", "outputIndexes": [ 2409 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2382, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_835_output_0", "outputIndexes": [ 2410 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2409, 2410, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_339_output_0", "outputIndexes": [ 2411 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2408, 2411 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_236_output_0", "outputIndexes": [ 2412 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2412, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_210_output_0", "outputIndexes": [ 2413 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2412 ], "main_type": "NONE", "name": "/Shape_367_output_0", "outputIndexes": [ 2414 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2414 ], "main_type": "NONE", "name": "Shape9794", "outputIndexes": [ 2415 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2414 ], "main_type": "NONE", "name": "Rank9796", "outputIndexes": [ 2416 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2416, 2416 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9798", "outputIndexes": [ 2417 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2417 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9799", "outputIndexes": [ 2418 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2417, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9801", "outputIndexes": [ 2419 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2419 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9802", "outputIndexes": [ 2420 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2415, 2418, 2420, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9805", "outputIndexes": [ 2421 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2421 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9806", "outputIndexes": [ 2422 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2422 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9807", "outputIndexes": [ 2423 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2423, 2422 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9808", "outputIndexes": [ 2424 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2414, 2424, 2417 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_396_output_0", "outputIndexes": [ 2425 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2425, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_79_output_0", "outputIndexes": [ 2426 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2426, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_841_output_0", "outputIndexes": [ 2427 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2412, 2427, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_107_output_0", "outputIndexes": [ 2428 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2428 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_53_output_0", "outputIndexes": [ 2429 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2426, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_840_output_0", "outputIndexes": [ 2430 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2412, 20, 2430, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_106_output_0", "outputIndexes": [ 2431 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2429, 2431 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_342_output_0", "outputIndexes": [ 2432 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2432, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_211_output_0", "outputIndexes": [ 2433 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2413, 2433 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_131_output_0", "outputIndexes": [ 2434 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2362 ], "outputIndexes": [ 3991 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 3991 ], "outputIndexes": [ 3992 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear", "inputIndexes": [ 3992 ], "outputIndexes": [ 3993 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2605883344, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 3993 ], "outputIndexes": [ 3994 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3994 ], "outputIndexes": [ 2435 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2374, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_836_output_0", "outputIndexes": [ 2436 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2382, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_837_output_0", "outputIndexes": [ 2437 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2436, 2437, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_340_output_0", "outputIndexes": [ 2438 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2435, 2438 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_237_output_0", "outputIndexes": [ 2439 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2407, 2434, 2439, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_242_output_0", "outputIndexes": [ 2440 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2440 ], "outputIndexes": [ 3995 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 3995 ], "outputIndexes": [ 3996 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear", "inputIndexes": [ 3996 ], "outputIndexes": [ 3997 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2608246758, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 3997 ], "outputIndexes": [ 3998 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 3998 ], "outputIndexes": [ 2441 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2361, 2441 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_133_output_0", "outputIndexes": [ 2442 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_26/Mul_1_output_0", "inputIndexes": [ 2442 ], "outputIndexes": [ 2443 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2617683964, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2443 ], "outputIndexes": [ 3999 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 3999 ], "outputIndexes": [ 4000 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear", "inputIndexes": [ 4000 ], "outputIndexes": [ 4001 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2617716732, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4001 ], "outputIndexes": [ 4002 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4002 ], "outputIndexes": [ 2444 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2444 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_26/Mul_output_0", "outputIndexes": [ 2445 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2443 ], "outputIndexes": [ 4003 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4003 ], "outputIndexes": [ 4004 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear", "inputIndexes": [ 4004 ], "outputIndexes": [ 4005 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2643079186, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4005 ], "outputIndexes": [ 4006 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4006 ], "outputIndexes": [ 2446 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2445, 2446 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_26/Mul_output_0", "outputIndexes": [ 2447 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2447 ], "outputIndexes": [ 4007 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4007 ], "outputIndexes": [ 4008 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear", "inputIndexes": [ 4008 ], "outputIndexes": [ 4009 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2668441640, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4009 ], "outputIndexes": [ 4010 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4010 ], "outputIndexes": [ 2448 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2442, 2448 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_134_output_0", "outputIndexes": [ 2449 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2449, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_243_output_0", "outputIndexes": [ 2450 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_27/Mul_1_output_0", "inputIndexes": [ 2450 ], "outputIndexes": [ 2451 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2693804094, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2451 ], "outputIndexes": [ 4011 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4011 ], "outputIndexes": [ 4012 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear", "inputIndexes": [ 4012 ], "outputIndexes": [ 4013 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2693836862, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4013 ], "outputIndexes": [ 4014 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4014 ], "outputIndexes": [ 2452 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2451 ], "main_type": "NONE", "name": "/Shape_378_output_0", "outputIndexes": [ 2453 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2453 ], "main_type": "NONE", "name": "Shape10127", "outputIndexes": [ 2454 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2453 ], "main_type": "NONE", "name": "Rank10129", "outputIndexes": [ 2455 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2455, 2455 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10131", "outputIndexes": [ 2456 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2456 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10132", "outputIndexes": [ 2457 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2456, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10134", "outputIndexes": [ 2458 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2458 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10135", "outputIndexes": [ 2459 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2454, 2457, 2459, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10138", "outputIndexes": [ 2460 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2460 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10139", "outputIndexes": [ 2461 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2461, 2461 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10141", "outputIndexes": [ 2462 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2453, 2462, 2456 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_408_output_0", "outputIndexes": [ 2463 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2463, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_864_output_0", "outputIndexes": [ 2464 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2456 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10149", "outputIndexes": [ 2465 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2458 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10152", "outputIndexes": [ 2466 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2454, 2465, 2466, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10155", "outputIndexes": [ 2467 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2467 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10156", "outputIndexes": [ 2468 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2468 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10157", "outputIndexes": [ 2469 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2469, 2468 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10158", "outputIndexes": [ 2470 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2453, 2470, 2456 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_409_output_0", "outputIndexes": [ 2471 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2471, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_865_output_0", "outputIndexes": [ 2472 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2464, 2472, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_351_output_0", "outputIndexes": [ 2473 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2452, 2473 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_244_output_0", "outputIndexes": [ 2474 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2474, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_216_output_0", "outputIndexes": [ 2475 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2474 ], "main_type": "NONE", "name": "/Shape_380_output_0", "outputIndexes": [ 2476 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2476 ], "main_type": "NONE", "name": "Shape10259", "outputIndexes": [ 2477 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2476 ], "main_type": "NONE", "name": "Rank10261", "outputIndexes": [ 2478 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2478, 2478 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10263", "outputIndexes": [ 2479 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2479 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10264", "outputIndexes": [ 2480 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2479, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10266", "outputIndexes": [ 2481 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2481 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10267", "outputIndexes": [ 2482 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2477, 2480, 2482, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10270", "outputIndexes": [ 2483 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2483 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10271", "outputIndexes": [ 2484 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2484 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10272", "outputIndexes": [ 2485 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2485, 2484 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10273", "outputIndexes": [ 2486 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2476, 2486, 2479 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_410_output_0", "outputIndexes": [ 2487 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2487, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_81_output_0", "outputIndexes": [ 2488 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2488, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_871_output_0", "outputIndexes": [ 2489 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2474, 2489, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_109_output_0", "outputIndexes": [ 2490 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2490 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_54_output_0", "outputIndexes": [ 2491 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2488, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_870_output_0", "outputIndexes": [ 2492 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2474, 20, 2492, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_108_output_0", "outputIndexes": [ 2493 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2491, 2493 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_354_output_0", "outputIndexes": [ 2494 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2494, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_217_output_0", "outputIndexes": [ 2495 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2475, 2495 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_135_output_0", "outputIndexes": [ 2496 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2451 ], "outputIndexes": [ 4015 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4015 ], "outputIndexes": [ 4016 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear", "inputIndexes": [ 4016 ], "outputIndexes": [ 4017 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2703290452, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4017 ], "outputIndexes": [ 4018 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4018 ], "outputIndexes": [ 2497 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2463, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_866_output_0", "outputIndexes": [ 2498 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2471, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_867_output_0", "outputIndexes": [ 2499 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2498, 2499, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_352_output_0", "outputIndexes": [ 2500 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2497, 2500 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_245_output_0", "outputIndexes": [ 2501 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2501, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_218_output_0", "outputIndexes": [ 2502 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2501 ], "main_type": "NONE", "name": "/Shape_381_output_0", "outputIndexes": [ 2503 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2503 ], "main_type": "NONE", "name": "Shape10165", "outputIndexes": [ 2504 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2503 ], "main_type": "NONE", "name": "Rank10167", "outputIndexes": [ 2505 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2505, 2505 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10169", "outputIndexes": [ 2506 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2506 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10170", "outputIndexes": [ 2507 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2506, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10172", "outputIndexes": [ 2508 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2508 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10173", "outputIndexes": [ 2509 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2504, 2507, 2509, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10176", "outputIndexes": [ 2510 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2510 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10177", "outputIndexes": [ 2511 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2511 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10178", "outputIndexes": [ 2512 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2512, 2511 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10179", "outputIndexes": [ 2513 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2503, 2513, 2506 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_411_output_0", "outputIndexes": [ 2514 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2514, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_82_output_0", "outputIndexes": [ 2515 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2515, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_873_output_0", "outputIndexes": [ 2516 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2501, 2516, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_111_output_0", "outputIndexes": [ 2517 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2517 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_55_output_0", "outputIndexes": [ 2518 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2515, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_872_output_0", "outputIndexes": [ 2519 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2501, 20, 2519, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_110_output_0", "outputIndexes": [ 2520 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2518, 2520 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_355_output_0", "outputIndexes": [ 2521 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2521, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_219_output_0", "outputIndexes": [ 2522 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2502, 2522 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_136_output_0", "outputIndexes": [ 2523 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2451 ], "outputIndexes": [ 4019 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4019 ], "outputIndexes": [ 4020 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear", "inputIndexes": [ 4020 ], "outputIndexes": [ 4021 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2705653866, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4021 ], "outputIndexes": [ 4022 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4022 ], "outputIndexes": [ 2524 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2463, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_868_output_0", "outputIndexes": [ 2525 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2471, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_869_output_0", "outputIndexes": [ 2526 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2525, 2526, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_353_output_0", "outputIndexes": [ 2527 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2524, 2527 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_246_output_0", "outputIndexes": [ 2528 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2496, 2523, 2528, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_251_output_0", "outputIndexes": [ 2529 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2529 ], "outputIndexes": [ 4023 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4023 ], "outputIndexes": [ 4024 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear", "inputIndexes": [ 4024 ], "outputIndexes": [ 4025 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2708017280, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4025 ], "outputIndexes": [ 4026 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4026 ], "outputIndexes": [ 2530 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2450, 2530 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_138_output_0", "outputIndexes": [ 2531 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_27/Mul_1_output_0", "inputIndexes": [ 2531 ], "outputIndexes": [ 2532 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2717454486, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2532 ], "outputIndexes": [ 4027 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4027 ], "outputIndexes": [ 4028 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear", "inputIndexes": [ 4028 ], "outputIndexes": [ 4029 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2717487254, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4029 ], "outputIndexes": [ 4030 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4030 ], "outputIndexes": [ 2533 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2533 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_27/Mul_output_0", "outputIndexes": [ 2534 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2532 ], "outputIndexes": [ 4031 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4031 ], "outputIndexes": [ 4032 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear", "inputIndexes": [ 4032 ], "outputIndexes": [ 4033 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2742849708, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4033 ], "outputIndexes": [ 4034 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4034 ], "outputIndexes": [ 2535 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2534, 2535 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_27/Mul_output_0", "outputIndexes": [ 2536 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2536 ], "outputIndexes": [ 4035 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4035 ], "outputIndexes": [ 4036 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear", "inputIndexes": [ 4036 ], "outputIndexes": [ 4037 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2768212162, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4037 ], "outputIndexes": [ 4038 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4038 ], "outputIndexes": [ 2537 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2531, 2537 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_139_output_0", "outputIndexes": [ 2538 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2538, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_252_output_0", "outputIndexes": [ 2539 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_28/Mul_1_output_0", "inputIndexes": [ 2539 ], "outputIndexes": [ 2540 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2793574616, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2540 ], "outputIndexes": [ 4039 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4039 ], "outputIndexes": [ 4040 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear", "inputIndexes": [ 4040 ], "outputIndexes": [ 4041 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2793607384, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4041 ], "outputIndexes": [ 4042 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4042 ], "outputIndexes": [ 2541 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2540 ], "main_type": "NONE", "name": "/Shape_392_output_0", "outputIndexes": [ 2542 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2542 ], "main_type": "NONE", "name": "Shape10498", "outputIndexes": [ 2543 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2542 ], "main_type": "NONE", "name": "Rank10500", "outputIndexes": [ 2544 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2544, 2544 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10502", "outputIndexes": [ 2545 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2545 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10503", "outputIndexes": [ 2546 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2545, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10505", "outputIndexes": [ 2547 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2547 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10506", "outputIndexes": [ 2548 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2543, 2546, 2548, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10509", "outputIndexes": [ 2549 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2549 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10510", "outputIndexes": [ 2550 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2550, 2550 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10512", "outputIndexes": [ 2551 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2542, 2551, 2545 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_423_output_0", "outputIndexes": [ 2552 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2552, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_896_output_0", "outputIndexes": [ 2553 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2545 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10520", "outputIndexes": [ 2554 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2547 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10523", "outputIndexes": [ 2555 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2543, 2554, 2555, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10526", "outputIndexes": [ 2556 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2556 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10527", "outputIndexes": [ 2557 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2557 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10528", "outputIndexes": [ 2558 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2558, 2557 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10529", "outputIndexes": [ 2559 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2542, 2559, 2545 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_424_output_0", "outputIndexes": [ 2560 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2560, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_897_output_0", "outputIndexes": [ 2561 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2553, 2561, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_364_output_0", "outputIndexes": [ 2562 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2541, 2562 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_253_output_0", "outputIndexes": [ 2563 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2563, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_224_output_0", "outputIndexes": [ 2564 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2563 ], "main_type": "NONE", "name": "/Shape_394_output_0", "outputIndexes": [ 2565 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2565 ], "main_type": "NONE", "name": "Shape10630", "outputIndexes": [ 2566 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2565 ], "main_type": "NONE", "name": "Rank10632", "outputIndexes": [ 2567 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2567, 2567 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10634", "outputIndexes": [ 2568 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2568 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10635", "outputIndexes": [ 2569 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2568, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10637", "outputIndexes": [ 2570 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2570 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10638", "outputIndexes": [ 2571 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2566, 2569, 2571, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10641", "outputIndexes": [ 2572 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2572 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10642", "outputIndexes": [ 2573 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2573 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10643", "outputIndexes": [ 2574 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2574, 2573 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10644", "outputIndexes": [ 2575 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2565, 2575, 2568 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_425_output_0", "outputIndexes": [ 2576 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2576, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_84_output_0", "outputIndexes": [ 2577 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2577, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_903_output_0", "outputIndexes": [ 2578 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2563, 2578, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_113_output_0", "outputIndexes": [ 2579 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2579 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_56_output_0", "outputIndexes": [ 2580 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2577, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_902_output_0", "outputIndexes": [ 2581 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2563, 20, 2581, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_112_output_0", "outputIndexes": [ 2582 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2580, 2582 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_367_output_0", "outputIndexes": [ 2583 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2583, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_225_output_0", "outputIndexes": [ 2584 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2564, 2584 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_140_output_0", "outputIndexes": [ 2585 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2540 ], "outputIndexes": [ 4043 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4043 ], "outputIndexes": [ 4044 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear", "inputIndexes": [ 4044 ], "outputIndexes": [ 4045 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2803060974, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4045 ], "outputIndexes": [ 4046 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4046 ], "outputIndexes": [ 2586 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2552, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_898_output_0", "outputIndexes": [ 2587 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2560, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_899_output_0", "outputIndexes": [ 2588 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2587, 2588, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_365_output_0", "outputIndexes": [ 2589 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2586, 2589 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_254_output_0", "outputIndexes": [ 2590 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2590, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_226_output_0", "outputIndexes": [ 2591 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2590 ], "main_type": "NONE", "name": "/Shape_395_output_0", "outputIndexes": [ 2592 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2592 ], "main_type": "NONE", "name": "Shape10536", "outputIndexes": [ 2593 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2592 ], "main_type": "NONE", "name": "Rank10538", "outputIndexes": [ 2594 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2594, 2594 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10540", "outputIndexes": [ 2595 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2595 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10541", "outputIndexes": [ 2596 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2595, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10543", "outputIndexes": [ 2597 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2597 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10544", "outputIndexes": [ 2598 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2593, 2596, 2598, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10547", "outputIndexes": [ 2599 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2599 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10548", "outputIndexes": [ 2600 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2600 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10549", "outputIndexes": [ 2601 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2601, 2600 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10550", "outputIndexes": [ 2602 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2592, 2602, 2595 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_426_output_0", "outputIndexes": [ 2603 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2603, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_85_output_0", "outputIndexes": [ 2604 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2604, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_905_output_0", "outputIndexes": [ 2605 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2590, 2605, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_115_output_0", "outputIndexes": [ 2606 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2606 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_57_output_0", "outputIndexes": [ 2607 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2604, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_904_output_0", "outputIndexes": [ 2608 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2590, 20, 2608, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_114_output_0", "outputIndexes": [ 2609 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2607, 2609 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_368_output_0", "outputIndexes": [ 2610 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2610, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_227_output_0", "outputIndexes": [ 2611 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2591, 2611 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_141_output_0", "outputIndexes": [ 2612 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2540 ], "outputIndexes": [ 4047 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4047 ], "outputIndexes": [ 4048 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear", "inputIndexes": [ 4048 ], "outputIndexes": [ 4049 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2805424388, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4049 ], "outputIndexes": [ 4050 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4050 ], "outputIndexes": [ 2613 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2552, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_900_output_0", "outputIndexes": [ 2614 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2560, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_901_output_0", "outputIndexes": [ 2615 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2614, 2615, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_366_output_0", "outputIndexes": [ 2616 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2613, 2616 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_255_output_0", "outputIndexes": [ 2617 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2585, 2612, 2617, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_260_output_0", "outputIndexes": [ 2618 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2618 ], "outputIndexes": [ 4051 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4051 ], "outputIndexes": [ 4052 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear", "inputIndexes": [ 4052 ], "outputIndexes": [ 4053 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2807787802, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4053 ], "outputIndexes": [ 4054 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4054 ], "outputIndexes": [ 2619 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2539, 2619 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_143_output_0", "outputIndexes": [ 2620 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_28/Mul_1_output_0", "inputIndexes": [ 2620 ], "outputIndexes": [ 2621 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2817225008, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2621 ], "outputIndexes": [ 4055 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4055 ], "outputIndexes": [ 4056 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear", "inputIndexes": [ 4056 ], "outputIndexes": [ 4057 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2817257776, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4057 ], "outputIndexes": [ 4058 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4058 ], "outputIndexes": [ 2622 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2622 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_28/Mul_output_0", "outputIndexes": [ 2623 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2621 ], "outputIndexes": [ 4059 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4059 ], "outputIndexes": [ 4060 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear", "inputIndexes": [ 4060 ], "outputIndexes": [ 4061 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2842620230, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4061 ], "outputIndexes": [ 4062 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4062 ], "outputIndexes": [ 2624 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2623, 2624 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_28/Mul_output_0", "outputIndexes": [ 2625 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2625 ], "outputIndexes": [ 4063 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4063 ], "outputIndexes": [ 4064 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear", "inputIndexes": [ 4064 ], "outputIndexes": [ 4065 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2867982684, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4065 ], "outputIndexes": [ 4066 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4066 ], "outputIndexes": [ 2626 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2620, 2626 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_144_output_0", "outputIndexes": [ 2627 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2627, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_261_output_0", "outputIndexes": [ 2628 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_29/Mul_1_output_0", "inputIndexes": [ 2628 ], "outputIndexes": [ 2629 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2893345138, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2629 ], "outputIndexes": [ 4067 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4067 ], "outputIndexes": [ 4068 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear", "inputIndexes": [ 4068 ], "outputIndexes": [ 4069 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2893377906, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4069 ], "outputIndexes": [ 4070 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4070 ], "outputIndexes": [ 2630 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2629 ], "main_type": "NONE", "name": "/Shape_406_output_0", "outputIndexes": [ 2631 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2631 ], "main_type": "NONE", "name": "Shape10869", "outputIndexes": [ 2632 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2631 ], "main_type": "NONE", "name": "Rank10871", "outputIndexes": [ 2633 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2633, 2633 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10873", "outputIndexes": [ 2634 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2634 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10874", "outputIndexes": [ 2635 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2634, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10876", "outputIndexes": [ 2636 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2636 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10877", "outputIndexes": [ 2637 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2632, 2635, 2637, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10880", "outputIndexes": [ 2638 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2638 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10881", "outputIndexes": [ 2639 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2639, 2639 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10883", "outputIndexes": [ 2640 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2631, 2640, 2634 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_438_output_0", "outputIndexes": [ 2641 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2641, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_928_output_0", "outputIndexes": [ 2642 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2634 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10891", "outputIndexes": [ 2643 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2636 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10894", "outputIndexes": [ 2644 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2632, 2643, 2644, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10897", "outputIndexes": [ 2645 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2645 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10898", "outputIndexes": [ 2646 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2646 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10899", "outputIndexes": [ 2647 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2647, 2646 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10900", "outputIndexes": [ 2648 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2631, 2648, 2634 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_439_output_0", "outputIndexes": [ 2649 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2649, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_929_output_0", "outputIndexes": [ 2650 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2642, 2650, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_377_output_0", "outputIndexes": [ 2651 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2630, 2651 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_262_output_0", "outputIndexes": [ 2652 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2652, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_232_output_0", "outputIndexes": [ 2653 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2652 ], "main_type": "NONE", "name": "/Shape_408_output_0", "outputIndexes": [ 2654 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2654 ], "main_type": "NONE", "name": "Shape11001", "outputIndexes": [ 2655 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2654 ], "main_type": "NONE", "name": "Rank11003", "outputIndexes": [ 2656 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2656, 2656 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11005", "outputIndexes": [ 2657 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2657 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11006", "outputIndexes": [ 2658 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2657, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11008", "outputIndexes": [ 2659 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2659 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11009", "outputIndexes": [ 2660 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2655, 2658, 2660, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11012", "outputIndexes": [ 2661 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2661 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11013", "outputIndexes": [ 2662 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2662 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11014", "outputIndexes": [ 2663 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2663, 2662 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11015", "outputIndexes": [ 2664 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2654, 2664, 2657 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_440_output_0", "outputIndexes": [ 2665 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2665, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_87_output_0", "outputIndexes": [ 2666 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2666, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_935_output_0", "outputIndexes": [ 2667 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2652, 2667, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_117_output_0", "outputIndexes": [ 2668 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2668 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_58_output_0", "outputIndexes": [ 2669 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2666, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_934_output_0", "outputIndexes": [ 2670 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2652, 20, 2670, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_116_output_0", "outputIndexes": [ 2671 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2669, 2671 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_380_output_0", "outputIndexes": [ 2672 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2672, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_233_output_0", "outputIndexes": [ 2673 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2653, 2673 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_145_output_0", "outputIndexes": [ 2674 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2629 ], "outputIndexes": [ 4071 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4071 ], "outputIndexes": [ 4072 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear", "inputIndexes": [ 4072 ], "outputIndexes": [ 4073 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2902831496, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4073 ], "outputIndexes": [ 4074 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4074 ], "outputIndexes": [ 2675 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2641, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_930_output_0", "outputIndexes": [ 2676 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2649, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_931_output_0", "outputIndexes": [ 2677 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2676, 2677, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_378_output_0", "outputIndexes": [ 2678 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2675, 2678 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_263_output_0", "outputIndexes": [ 2679 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2679, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_234_output_0", "outputIndexes": [ 2680 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2679 ], "main_type": "NONE", "name": "/Shape_409_output_0", "outputIndexes": [ 2681 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2681 ], "main_type": "NONE", "name": "Shape10907", "outputIndexes": [ 2682 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2681 ], "main_type": "NONE", "name": "Rank10909", "outputIndexes": [ 2683 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2683, 2683 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10911", "outputIndexes": [ 2684 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2684 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10912", "outputIndexes": [ 2685 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2684, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10914", "outputIndexes": [ 2686 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2686 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10915", "outputIndexes": [ 2687 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2682, 2685, 2687, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10918", "outputIndexes": [ 2688 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2688 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10919", "outputIndexes": [ 2689 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2689 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10920", "outputIndexes": [ 2690 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2690, 2689 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10921", "outputIndexes": [ 2691 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2681, 2691, 2684 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_441_output_0", "outputIndexes": [ 2692 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2692, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_88_output_0", "outputIndexes": [ 2693 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2693, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_937_output_0", "outputIndexes": [ 2694 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2679, 2694, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_119_output_0", "outputIndexes": [ 2695 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2695 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_59_output_0", "outputIndexes": [ 2696 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2693, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_936_output_0", "outputIndexes": [ 2697 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2679, 20, 2697, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_118_output_0", "outputIndexes": [ 2698 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2696, 2698 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_381_output_0", "outputIndexes": [ 2699 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2699, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_235_output_0", "outputIndexes": [ 2700 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2680, 2700 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_146_output_0", "outputIndexes": [ 2701 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2629 ], "outputIndexes": [ 4075 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4075 ], "outputIndexes": [ 4076 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear", "inputIndexes": [ 4076 ], "outputIndexes": [ 4077 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 2905194910, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4077 ], "outputIndexes": [ 4078 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4078 ], "outputIndexes": [ 2702 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2641, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_932_output_0", "outputIndexes": [ 2703 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2649, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_933_output_0", "outputIndexes": [ 2704 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2703, 2704, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_379_output_0", "outputIndexes": [ 2705 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2702, 2705 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_264_output_0", "outputIndexes": [ 2706 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2674, 2701, 2706, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_269_output_0", "outputIndexes": [ 2707 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2707 ], "outputIndexes": [ 4079 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4079 ], "outputIndexes": [ 4080 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear", "inputIndexes": [ 4080 ], "outputIndexes": [ 4081 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2907558324, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4081 ], "outputIndexes": [ 4082 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4082 ], "outputIndexes": [ 2708 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2628, 2708 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_148_output_0", "outputIndexes": [ 2709 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_29/Mul_1_output_0", "inputIndexes": [ 2709 ], "outputIndexes": [ 2710 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2916995530, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2710 ], "outputIndexes": [ 4083 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4083 ], "outputIndexes": [ 4084 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear", "inputIndexes": [ 4084 ], "outputIndexes": [ 4085 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2917028298, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4085 ], "outputIndexes": [ 4086 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4086 ], "outputIndexes": [ 2711 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2711 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_29/Mul_output_0", "outputIndexes": [ 2712 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2710 ], "outputIndexes": [ 4087 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4087 ], "outputIndexes": [ 4088 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear", "inputIndexes": [ 4088 ], "outputIndexes": [ 4089 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2942390752, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4089 ], "outputIndexes": [ 4090 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4090 ], "outputIndexes": [ 2713 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2712, 2713 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_29/Mul_output_0", "outputIndexes": [ 2714 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2714 ], "outputIndexes": [ 4091 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4091 ], "outputIndexes": [ 4092 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear", "inputIndexes": [ 4092 ], "outputIndexes": [ 4093 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 2967753206, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4093 ], "outputIndexes": [ 4094 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4094 ], "outputIndexes": [ 2715 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2709, 2715 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_149_output_0", "outputIndexes": [ 2716 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2716, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_270_output_0", "outputIndexes": [ 2717 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_30/Mul_1_output_0", "inputIndexes": [ 2717 ], "outputIndexes": [ 2718 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 2993115660, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2718 ], "outputIndexes": [ 4095 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4095 ], "outputIndexes": [ 4096 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear", "inputIndexes": [ 4096 ], "outputIndexes": [ 4097 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 2993148428, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4097 ], "outputIndexes": [ 4098 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4098 ], "outputIndexes": [ 2719 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2718 ], "main_type": "NONE", "name": "/Shape_420_output_0", "outputIndexes": [ 2720 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2720 ], "main_type": "NONE", "name": "Shape11240", "outputIndexes": [ 2721 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2720 ], "main_type": "NONE", "name": "Rank11242", "outputIndexes": [ 2722 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2722, 2722 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11244", "outputIndexes": [ 2723 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2723 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11245", "outputIndexes": [ 2724 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2723, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11247", "outputIndexes": [ 2725 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2725 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11248", "outputIndexes": [ 2726 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2721, 2724, 2726, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11251", "outputIndexes": [ 2727 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2727 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11252", "outputIndexes": [ 2728 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2728, 2728 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11254", "outputIndexes": [ 2729 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2720, 2729, 2723 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_453_output_0", "outputIndexes": [ 2730 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2730, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_960_output_0", "outputIndexes": [ 2731 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2723 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11262", "outputIndexes": [ 2732 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2725 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11265", "outputIndexes": [ 2733 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2721, 2732, 2733, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11268", "outputIndexes": [ 2734 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2734 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11269", "outputIndexes": [ 2735 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2735 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11270", "outputIndexes": [ 2736 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2736, 2735 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11271", "outputIndexes": [ 2737 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2720, 2737, 2723 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_454_output_0", "outputIndexes": [ 2738 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2738, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_961_output_0", "outputIndexes": [ 2739 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2731, 2739, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_390_output_0", "outputIndexes": [ 2740 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2719, 2740 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_271_output_0", "outputIndexes": [ 2741 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2741, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_240_output_0", "outputIndexes": [ 2742 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2741 ], "main_type": "NONE", "name": "/Shape_422_output_0", "outputIndexes": [ 2743 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2743 ], "main_type": "NONE", "name": "Shape11372", "outputIndexes": [ 2744 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2743 ], "main_type": "NONE", "name": "Rank11374", "outputIndexes": [ 2745 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2745, 2745 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11376", "outputIndexes": [ 2746 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2746 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11377", "outputIndexes": [ 2747 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2746, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11379", "outputIndexes": [ 2748 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2748 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11380", "outputIndexes": [ 2749 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2744, 2747, 2749, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11383", "outputIndexes": [ 2750 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2750 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11384", "outputIndexes": [ 2751 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2751 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11385", "outputIndexes": [ 2752 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2752, 2751 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11386", "outputIndexes": [ 2753 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2743, 2753, 2746 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_455_output_0", "outputIndexes": [ 2754 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2754, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_90_output_0", "outputIndexes": [ 2755 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2755, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_967_output_0", "outputIndexes": [ 2756 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2741, 2756, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_121_output_0", "outputIndexes": [ 2757 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2757 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_60_output_0", "outputIndexes": [ 2758 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2755, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_966_output_0", "outputIndexes": [ 2759 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2741, 20, 2759, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_120_output_0", "outputIndexes": [ 2760 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2758, 2760 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_393_output_0", "outputIndexes": [ 2761 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2761, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_241_output_0", "outputIndexes": [ 2762 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2742, 2762 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_150_output_0", "outputIndexes": [ 2763 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2718 ], "outputIndexes": [ 4099 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4099 ], "outputIndexes": [ 4100 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear", "inputIndexes": [ 4100 ], "outputIndexes": [ 4101 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3002602018, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4101 ], "outputIndexes": [ 4102 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4102 ], "outputIndexes": [ 2764 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2730, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_962_output_0", "outputIndexes": [ 2765 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2738, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_963_output_0", "outputIndexes": [ 2766 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2765, 2766, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_391_output_0", "outputIndexes": [ 2767 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2764, 2767 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_272_output_0", "outputIndexes": [ 2768 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2768, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_242_output_0", "outputIndexes": [ 2769 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2768 ], "main_type": "NONE", "name": "/Shape_423_output_0", "outputIndexes": [ 2770 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2770 ], "main_type": "NONE", "name": "Shape11278", "outputIndexes": [ 2771 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2770 ], "main_type": "NONE", "name": "Rank11280", "outputIndexes": [ 2772 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2772, 2772 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11282", "outputIndexes": [ 2773 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2773 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11283", "outputIndexes": [ 2774 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2773, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11285", "outputIndexes": [ 2775 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2775 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11286", "outputIndexes": [ 2776 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2771, 2774, 2776, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11289", "outputIndexes": [ 2777 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2777 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11290", "outputIndexes": [ 2778 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2778 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11291", "outputIndexes": [ 2779 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2779, 2778 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11292", "outputIndexes": [ 2780 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2770, 2780, 2773 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_456_output_0", "outputIndexes": [ 2781 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2781, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_91_output_0", "outputIndexes": [ 2782 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2782, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_969_output_0", "outputIndexes": [ 2783 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2768, 2783, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_123_output_0", "outputIndexes": [ 2784 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2784 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_61_output_0", "outputIndexes": [ 2785 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2782, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_968_output_0", "outputIndexes": [ 2786 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2768, 20, 2786, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_122_output_0", "outputIndexes": [ 2787 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2785, 2787 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_394_output_0", "outputIndexes": [ 2788 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2788, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_243_output_0", "outputIndexes": [ 2789 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2769, 2789 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_151_output_0", "outputIndexes": [ 2790 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2718 ], "outputIndexes": [ 4103 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4103 ], "outputIndexes": [ 4104 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear", "inputIndexes": [ 4104 ], "outputIndexes": [ 4105 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3004965432, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4105 ], "outputIndexes": [ 4106 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4106 ], "outputIndexes": [ 2791 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2730, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_964_output_0", "outputIndexes": [ 2792 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2738, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_965_output_0", "outputIndexes": [ 2793 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2792, 2793, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_392_output_0", "outputIndexes": [ 2794 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2791, 2794 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_273_output_0", "outputIndexes": [ 2795 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2763, 2790, 2795, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_278_output_0", "outputIndexes": [ 2796 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2796 ], "outputIndexes": [ 4107 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4107 ], "outputIndexes": [ 4108 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear", "inputIndexes": [ 4108 ], "outputIndexes": [ 4109 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3007328846, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4109 ], "outputIndexes": [ 4110 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4110 ], "outputIndexes": [ 2797 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2717, 2797 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_153_output_0", "outputIndexes": [ 2798 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_30/Mul_1_output_0", "inputIndexes": [ 2798 ], "outputIndexes": [ 2799 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3016766052, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2799 ], "outputIndexes": [ 4111 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4111 ], "outputIndexes": [ 4112 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear", "inputIndexes": [ 4112 ], "outputIndexes": [ 4113 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3016798820, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4113 ], "outputIndexes": [ 4114 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4114 ], "outputIndexes": [ 2800 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2800 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_30/Mul_output_0", "outputIndexes": [ 2801 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2799 ], "outputIndexes": [ 4115 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4115 ], "outputIndexes": [ 4116 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear", "inputIndexes": [ 4116 ], "outputIndexes": [ 4117 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3042161274, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4117 ], "outputIndexes": [ 4118 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4118 ], "outputIndexes": [ 2802 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2801, 2802 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_30/Mul_output_0", "outputIndexes": [ 2803 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2803 ], "outputIndexes": [ 4119 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4119 ], "outputIndexes": [ 4120 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear", "inputIndexes": [ 4120 ], "outputIndexes": [ 4121 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3067523728, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4121 ], "outputIndexes": [ 4122 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4122 ], "outputIndexes": [ 2804 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2798, 2804 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_154_output_0", "outputIndexes": [ 2805 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2805, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_279_output_0", "outputIndexes": [ 2806 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_31/Mul_1_output_0", "inputIndexes": [ 2806 ], "outputIndexes": [ 2807 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3092886182, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2807 ], "outputIndexes": [ 4123 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4123 ], "outputIndexes": [ 4124 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear", "inputIndexes": [ 4124 ], "outputIndexes": [ 4125 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3092918950, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4125 ], "outputIndexes": [ 4126 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4126 ], "outputIndexes": [ 2808 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2807 ], "main_type": "NONE", "name": "/Shape_434_output_0", "outputIndexes": [ 2809 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2809 ], "main_type": "NONE", "name": "Shape11611", "outputIndexes": [ 2810 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2809 ], "main_type": "NONE", "name": "Rank11613", "outputIndexes": [ 2811 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2811, 2811 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11615", "outputIndexes": [ 2812 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2812 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11616", "outputIndexes": [ 2813 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2812, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11618", "outputIndexes": [ 2814 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2814 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11619", "outputIndexes": [ 2815 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2810, 2813, 2815, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11622", "outputIndexes": [ 2816 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2816 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11623", "outputIndexes": [ 2817 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2817, 2817 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11625", "outputIndexes": [ 2818 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2809, 2818, 2812 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_468_output_0", "outputIndexes": [ 2819 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2819, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_992_output_0", "outputIndexes": [ 2820 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2812 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11633", "outputIndexes": [ 2821 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2814 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11636", "outputIndexes": [ 2822 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2810, 2821, 2822, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11639", "outputIndexes": [ 2823 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2823 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11640", "outputIndexes": [ 2824 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2824 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11641", "outputIndexes": [ 2825 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2825, 2824 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11642", "outputIndexes": [ 2826 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2809, 2826, 2812 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_469_output_0", "outputIndexes": [ 2827 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2827, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_993_output_0", "outputIndexes": [ 2828 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2820, 2828, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_403_output_0", "outputIndexes": [ 2829 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2808, 2829 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_280_output_0", "outputIndexes": [ 2830 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2830, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_248_output_0", "outputIndexes": [ 2831 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2830 ], "main_type": "NONE", "name": "/Shape_436_output_0", "outputIndexes": [ 2832 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2832 ], "main_type": "NONE", "name": "Shape11743", "outputIndexes": [ 2833 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2832 ], "main_type": "NONE", "name": "Rank11745", "outputIndexes": [ 2834 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2834, 2834 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11747", "outputIndexes": [ 2835 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2835 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11748", "outputIndexes": [ 2836 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2835, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11750", "outputIndexes": [ 2837 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2837 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11751", "outputIndexes": [ 2838 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2833, 2836, 2838, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11754", "outputIndexes": [ 2839 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2839 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11755", "outputIndexes": [ 2840 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2840 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11756", "outputIndexes": [ 2841 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2841, 2840 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11757", "outputIndexes": [ 2842 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2832, 2842, 2835 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_470_output_0", "outputIndexes": [ 2843 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2843, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_93_output_0", "outputIndexes": [ 2844 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2844, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_999_output_0", "outputIndexes": [ 2845 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2830, 2845, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_125_output_0", "outputIndexes": [ 2846 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2846 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_62_output_0", "outputIndexes": [ 2847 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2844, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_998_output_0", "outputIndexes": [ 2848 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2830, 20, 2848, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_124_output_0", "outputIndexes": [ 2849 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2847, 2849 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_406_output_0", "outputIndexes": [ 2850 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2850, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_249_output_0", "outputIndexes": [ 2851 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2831, 2851 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_155_output_0", "outputIndexes": [ 2852 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2807 ], "outputIndexes": [ 4127 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4127 ], "outputIndexes": [ 4128 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear", "inputIndexes": [ 4128 ], "outputIndexes": [ 4129 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3102372540, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4129 ], "outputIndexes": [ 4130 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4130 ], "outputIndexes": [ 2853 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2819, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_994_output_0", "outputIndexes": [ 2854 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2827, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_995_output_0", "outputIndexes": [ 2855 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2854, 2855, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_404_output_0", "outputIndexes": [ 2856 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2853, 2856 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_281_output_0", "outputIndexes": [ 2857 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2857, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_250_output_0", "outputIndexes": [ 2858 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2857 ], "main_type": "NONE", "name": "/Shape_437_output_0", "outputIndexes": [ 2859 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2859 ], "main_type": "NONE", "name": "Shape11649", "outputIndexes": [ 2860 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2859 ], "main_type": "NONE", "name": "Rank11651", "outputIndexes": [ 2861 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2861, 2861 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11653", "outputIndexes": [ 2862 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2862 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11654", "outputIndexes": [ 2863 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2862, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11656", "outputIndexes": [ 2864 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2864 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11657", "outputIndexes": [ 2865 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2860, 2863, 2865, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11660", "outputIndexes": [ 2866 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2866 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11661", "outputIndexes": [ 2867 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2867 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11662", "outputIndexes": [ 2868 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2868, 2867 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11663", "outputIndexes": [ 2869 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2859, 2869, 2862 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_471_output_0", "outputIndexes": [ 2870 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2870, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_94_output_0", "outputIndexes": [ 2871 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2871, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1001_output_0", "outputIndexes": [ 2872 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2857, 2872, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_127_output_0", "outputIndexes": [ 2873 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2873 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_63_output_0", "outputIndexes": [ 2874 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2871, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1000_output_0", "outputIndexes": [ 2875 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2857, 20, 2875, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_126_output_0", "outputIndexes": [ 2876 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2874, 2876 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_407_output_0", "outputIndexes": [ 2877 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2877, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_251_output_0", "outputIndexes": [ 2878 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2858, 2878 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_156_output_0", "outputIndexes": [ 2879 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2807 ], "outputIndexes": [ 4131 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4131 ], "outputIndexes": [ 4132 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear", "inputIndexes": [ 4132 ], "outputIndexes": [ 4133 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3104735954, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4133 ], "outputIndexes": [ 4134 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4134 ], "outputIndexes": [ 2880 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2819, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_996_output_0", "outputIndexes": [ 2881 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2827, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_997_output_0", "outputIndexes": [ 2882 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2881, 2882, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_405_output_0", "outputIndexes": [ 2883 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2880, 2883 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_282_output_0", "outputIndexes": [ 2884 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2852, 2879, 2884, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_287_output_0", "outputIndexes": [ 2885 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2885 ], "outputIndexes": [ 4135 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4135 ], "outputIndexes": [ 4136 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear", "inputIndexes": [ 4136 ], "outputIndexes": [ 4137 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3107099368, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4137 ], "outputIndexes": [ 4138 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4138 ], "outputIndexes": [ 2886 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2806, 2886 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_158_output_0", "outputIndexes": [ 2887 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_31/Mul_1_output_0", "inputIndexes": [ 2887 ], "outputIndexes": [ 2888 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3116536574, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2888 ], "outputIndexes": [ 4139 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4139 ], "outputIndexes": [ 4140 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear", "inputIndexes": [ 4140 ], "outputIndexes": [ 4141 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3116569342, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4141 ], "outputIndexes": [ 4142 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4142 ], "outputIndexes": [ 2889 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2889 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_31/Mul_output_0", "outputIndexes": [ 2890 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2888 ], "outputIndexes": [ 4143 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4143 ], "outputIndexes": [ 4144 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear", "inputIndexes": [ 4144 ], "outputIndexes": [ 4145 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3141931796, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4145 ], "outputIndexes": [ 4146 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4146 ], "outputIndexes": [ 2891 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2890, 2891 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_31/Mul_output_0", "outputIndexes": [ 2892 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2892 ], "outputIndexes": [ 4147 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4147 ], "outputIndexes": [ 4148 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear", "inputIndexes": [ 4148 ], "outputIndexes": [ 4149 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3167294250, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4149 ], "outputIndexes": [ 4150 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4150 ], "outputIndexes": [ 2893 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2887, 2893 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_159_output_0", "outputIndexes": [ 2894 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2894, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_288_output_0", "outputIndexes": [ 2895 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_32/Mul_1_output_0", "inputIndexes": [ 2895 ], "outputIndexes": [ 2896 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3192656704, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2896 ], "outputIndexes": [ 4151 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4151 ], "outputIndexes": [ 4152 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear", "inputIndexes": [ 4152 ], "outputIndexes": [ 4153 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3192689472, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4153 ], "outputIndexes": [ 4154 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4154 ], "outputIndexes": [ 2897 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2896 ], "main_type": "NONE", "name": "/Shape_448_output_0", "outputIndexes": [ 2898 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2898 ], "main_type": "NONE", "name": "Shape11982", "outputIndexes": [ 2899 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2898 ], "main_type": "NONE", "name": "Rank11984", "outputIndexes": [ 2900 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2900, 2900 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11986", "outputIndexes": [ 2901 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2901 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11987", "outputIndexes": [ 2902 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2901, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11989", "outputIndexes": [ 2903 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2903 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11990", "outputIndexes": [ 2904 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2899, 2902, 2904, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11993", "outputIndexes": [ 2905 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2905 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11994", "outputIndexes": [ 2906 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2906, 2906 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11996", "outputIndexes": [ 2907 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2898, 2907, 2901 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_483_output_0", "outputIndexes": [ 2908 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2908, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1024_output_0", "outputIndexes": [ 2909 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2901 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12004", "outputIndexes": [ 2910 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2903 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12007", "outputIndexes": [ 2911 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2899, 2910, 2911, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12010", "outputIndexes": [ 2912 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2912 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12011", "outputIndexes": [ 2913 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2913 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12012", "outputIndexes": [ 2914 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2914, 2913 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12013", "outputIndexes": [ 2915 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2898, 2915, 2901 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_484_output_0", "outputIndexes": [ 2916 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2916, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1025_output_0", "outputIndexes": [ 2917 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2909, 2917, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_416_output_0", "outputIndexes": [ 2918 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2897, 2918 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_289_output_0", "outputIndexes": [ 2919 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2919, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_256_output_0", "outputIndexes": [ 2920 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2919 ], "main_type": "NONE", "name": "/Shape_450_output_0", "outputIndexes": [ 2921 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2921 ], "main_type": "NONE", "name": "Shape12114", "outputIndexes": [ 2922 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2921 ], "main_type": "NONE", "name": "Rank12116", "outputIndexes": [ 2923 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2923, 2923 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12118", "outputIndexes": [ 2924 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2924 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12119", "outputIndexes": [ 2925 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2924, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12121", "outputIndexes": [ 2926 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2926 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12122", "outputIndexes": [ 2927 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2922, 2925, 2927, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12125", "outputIndexes": [ 2928 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2928 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12126", "outputIndexes": [ 2929 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2929 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12127", "outputIndexes": [ 2930 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2930, 2929 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12128", "outputIndexes": [ 2931 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2921, 2931, 2924 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_485_output_0", "outputIndexes": [ 2932 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2932, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_96_output_0", "outputIndexes": [ 2933 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2933, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1031_output_0", "outputIndexes": [ 2934 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2919, 2934, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_129_output_0", "outputIndexes": [ 2935 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2935 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_64_output_0", "outputIndexes": [ 2936 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2933, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1030_output_0", "outputIndexes": [ 2937 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2919, 20, 2937, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_128_output_0", "outputIndexes": [ 2938 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2936, 2938 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_419_output_0", "outputIndexes": [ 2939 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2939, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_257_output_0", "outputIndexes": [ 2940 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2920, 2940 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_160_output_0", "outputIndexes": [ 2941 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2896 ], "outputIndexes": [ 4155 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4155 ], "outputIndexes": [ 4156 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear", "inputIndexes": [ 4156 ], "outputIndexes": [ 4157 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3202143062, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4157 ], "outputIndexes": [ 4158 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4158 ], "outputIndexes": [ 2942 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2908, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1026_output_0", "outputIndexes": [ 2943 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2916, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1027_output_0", "outputIndexes": [ 2944 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2943, 2944, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_417_output_0", "outputIndexes": [ 2945 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2942, 2945 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_290_output_0", "outputIndexes": [ 2946 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2946, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_258_output_0", "outputIndexes": [ 2947 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2946 ], "main_type": "NONE", "name": "/Shape_451_output_0", "outputIndexes": [ 2948 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2948 ], "main_type": "NONE", "name": "Shape12020", "outputIndexes": [ 2949 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2948 ], "main_type": "NONE", "name": "Rank12022", "outputIndexes": [ 2950 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2950, 2950 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12024", "outputIndexes": [ 2951 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2951 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12025", "outputIndexes": [ 2952 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2951, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12027", "outputIndexes": [ 2953 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2953 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12028", "outputIndexes": [ 2954 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2949, 2952, 2954, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12031", "outputIndexes": [ 2955 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2955 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12032", "outputIndexes": [ 2956 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 2956 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12033", "outputIndexes": [ 2957 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2957, 2956 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12034", "outputIndexes": [ 2958 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2948, 2958, 2951 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_486_output_0", "outputIndexes": [ 2959 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2959, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_97_output_0", "outputIndexes": [ 2960 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2960, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1033_output_0", "outputIndexes": [ 2961 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2946, 2961, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_131_output_0", "outputIndexes": [ 2962 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2962 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_65_output_0", "outputIndexes": [ 2963 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2960, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1032_output_0", "outputIndexes": [ 2964 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2946, 20, 2964, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_130_output_0", "outputIndexes": [ 2965 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2963, 2965 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_420_output_0", "outputIndexes": [ 2966 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2966, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_259_output_0", "outputIndexes": [ 2967 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2947, 2967 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_161_output_0", "outputIndexes": [ 2968 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2896 ], "outputIndexes": [ 4159 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4159 ], "outputIndexes": [ 4160 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear", "inputIndexes": [ 4160 ], "outputIndexes": [ 4161 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3204506476, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4161 ], "outputIndexes": [ 4162 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4162 ], "outputIndexes": [ 2969 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2908, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1028_output_0", "outputIndexes": [ 2970 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2916, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1029_output_0", "outputIndexes": [ 2971 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2970, 2971, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_418_output_0", "outputIndexes": [ 2972 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2969, 2972 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_291_output_0", "outputIndexes": [ 2973 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2941, 2968, 2973, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_296_output_0", "outputIndexes": [ 2974 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2974 ], "outputIndexes": [ 4163 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4163 ], "outputIndexes": [ 4164 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear", "inputIndexes": [ 4164 ], "outputIndexes": [ 4165 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3206869890, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4165 ], "outputIndexes": [ 4166 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4166 ], "outputIndexes": [ 2975 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2895, 2975 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_163_output_0", "outputIndexes": [ 2976 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_32/Mul_1_output_0", "inputIndexes": [ 2976 ], "outputIndexes": [ 2977 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3216307096, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2977 ], "outputIndexes": [ 4167 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4167 ], "outputIndexes": [ 4168 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear", "inputIndexes": [ 4168 ], "outputIndexes": [ 4169 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3216339864, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4169 ], "outputIndexes": [ 4170 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4170 ], "outputIndexes": [ 2978 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2978 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_32/Mul_output_0", "outputIndexes": [ 2979 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2977 ], "outputIndexes": [ 4171 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4171 ], "outputIndexes": [ 4172 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear", "inputIndexes": [ 4172 ], "outputIndexes": [ 4173 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3241702318, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4173 ], "outputIndexes": [ 4174 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4174 ], "outputIndexes": [ 2980 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2979, 2980 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_32/Mul_output_0", "outputIndexes": [ 2981 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2981 ], "outputIndexes": [ 4175 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4175 ], "outputIndexes": [ 4176 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear", "inputIndexes": [ 4176 ], "outputIndexes": [ 4177 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3267064772, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4177 ], "outputIndexes": [ 4178 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4178 ], "outputIndexes": [ 2982 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2976, 2982 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_164_output_0", "outputIndexes": [ 2983 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2983, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_297_output_0", "outputIndexes": [ 2984 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_33/Mul_1_output_0", "inputIndexes": [ 2984 ], "outputIndexes": [ 2985 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3292427226, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2985 ], "outputIndexes": [ 4179 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4179 ], "outputIndexes": [ 4180 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear", "inputIndexes": [ 4180 ], "outputIndexes": [ 4181 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3292459994, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4181 ], "outputIndexes": [ 4182 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4182 ], "outputIndexes": [ 2986 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2985 ], "main_type": "NONE", "name": "/Shape_462_output_0", "outputIndexes": [ 2987 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2987 ], "main_type": "NONE", "name": "Shape12353", "outputIndexes": [ 2988 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2987 ], "main_type": "NONE", "name": "Rank12355", "outputIndexes": [ 2989 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2989, 2989 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12357", "outputIndexes": [ 2990 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2990 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12358", "outputIndexes": [ 2991 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2990, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12360", "outputIndexes": [ 2992 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2992 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12361", "outputIndexes": [ 2993 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2988, 2991, 2993, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12364", "outputIndexes": [ 2994 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2994 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12365", "outputIndexes": [ 2995 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2995, 2995 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12367", "outputIndexes": [ 2996 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2987, 2996, 2990 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_498_output_0", "outputIndexes": [ 2997 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2997, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1056_output_0", "outputIndexes": [ 2998 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2990 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12375", "outputIndexes": [ 2999 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2992 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12378", "outputIndexes": [ 3000 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2988, 2999, 3000, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12381", "outputIndexes": [ 3001 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3001 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12382", "outputIndexes": [ 3002 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3002 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12383", "outputIndexes": [ 3003 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3003, 3002 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12384", "outputIndexes": [ 3004 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2987, 3004, 2990 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_499_output_0", "outputIndexes": [ 3005 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3005, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1057_output_0", "outputIndexes": [ 3006 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2998, 3006, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_429_output_0", "outputIndexes": [ 3007 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2986, 3007 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_298_output_0", "outputIndexes": [ 3008 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3008, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_264_output_0", "outputIndexes": [ 3009 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3008 ], "main_type": "NONE", "name": "/Shape_464_output_0", "outputIndexes": [ 3010 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3010 ], "main_type": "NONE", "name": "Shape12485", "outputIndexes": [ 3011 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3010 ], "main_type": "NONE", "name": "Rank12487", "outputIndexes": [ 3012 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3012, 3012 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12489", "outputIndexes": [ 3013 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3013 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12490", "outputIndexes": [ 3014 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3013, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12492", "outputIndexes": [ 3015 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3015 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12493", "outputIndexes": [ 3016 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3011, 3014, 3016, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12496", "outputIndexes": [ 3017 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3017 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12497", "outputIndexes": [ 3018 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 3018 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12498", "outputIndexes": [ 3019 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3019, 3018 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12499", "outputIndexes": [ 3020 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3010, 3020, 3013 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_500_output_0", "outputIndexes": [ 3021 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3021, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_99_output_0", "outputIndexes": [ 3022 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3022, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1063_output_0", "outputIndexes": [ 3023 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3008, 3023, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_133_output_0", "outputIndexes": [ 3024 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3024 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_66_output_0", "outputIndexes": [ 3025 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3022, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1062_output_0", "outputIndexes": [ 3026 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3008, 20, 3026, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_132_output_0", "outputIndexes": [ 3027 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3025, 3027 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_432_output_0", "outputIndexes": [ 3028 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3028, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_265_output_0", "outputIndexes": [ 3029 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3009, 3029 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_165_output_0", "outputIndexes": [ 3030 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2985 ], "outputIndexes": [ 4183 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4183 ], "outputIndexes": [ 4184 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear", "inputIndexes": [ 4184 ], "outputIndexes": [ 4185 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3301913584, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4185 ], "outputIndexes": [ 4186 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4186 ], "outputIndexes": [ 3031 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2997, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1058_output_0", "outputIndexes": [ 3032 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3005, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1059_output_0", "outputIndexes": [ 3033 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3032, 3033, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_430_output_0", "outputIndexes": [ 3034 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3031, 3034 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_299_output_0", "outputIndexes": [ 3035 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3035, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_266_output_0", "outputIndexes": [ 3036 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3035 ], "main_type": "NONE", "name": "/Shape_465_output_0", "outputIndexes": [ 3037 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3037 ], "main_type": "NONE", "name": "Shape12391", "outputIndexes": [ 3038 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3037 ], "main_type": "NONE", "name": "Rank12393", "outputIndexes": [ 3039 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3039, 3039 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12395", "outputIndexes": [ 3040 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3040 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12396", "outputIndexes": [ 3041 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3040, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12398", "outputIndexes": [ 3042 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3042 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12399", "outputIndexes": [ 3043 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3038, 3041, 3043, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12402", "outputIndexes": [ 3044 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3044 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12403", "outputIndexes": [ 3045 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 3045 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12404", "outputIndexes": [ 3046 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3046, 3045 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12405", "outputIndexes": [ 3047 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3037, 3047, 3040 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_501_output_0", "outputIndexes": [ 3048 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3048, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_100_output_0", "outputIndexes": [ 3049 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3049, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1065_output_0", "outputIndexes": [ 3050 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3035, 3050, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_135_output_0", "outputIndexes": [ 3051 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3051 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_67_output_0", "outputIndexes": [ 3052 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3049, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1064_output_0", "outputIndexes": [ 3053 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3035, 20, 3053, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_134_output_0", "outputIndexes": [ 3054 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3052, 3054 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_433_output_0", "outputIndexes": [ 3055 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3055, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_267_output_0", "outputIndexes": [ 3056 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3036, 3056 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_166_output_0", "outputIndexes": [ 3057 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2985 ], "outputIndexes": [ 4187 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4187 ], "outputIndexes": [ 4188 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear", "inputIndexes": [ 4188 ], "outputIndexes": [ 4189 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3304276998, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4189 ], "outputIndexes": [ 4190 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4190 ], "outputIndexes": [ 3058 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2997, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1060_output_0", "outputIndexes": [ 3059 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3005, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1061_output_0", "outputIndexes": [ 3060 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3059, 3060, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_431_output_0", "outputIndexes": [ 3061 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3058, 3061 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_300_output_0", "outputIndexes": [ 3062 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3030, 3057, 3062, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_305_output_0", "outputIndexes": [ 3063 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3063 ], "outputIndexes": [ 4191 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4191 ], "outputIndexes": [ 4192 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear", "inputIndexes": [ 4192 ], "outputIndexes": [ 4193 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3306640412, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4193 ], "outputIndexes": [ 4194 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4194 ], "outputIndexes": [ 3064 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2984, 3064 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_168_output_0", "outputIndexes": [ 3065 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_33/Mul_1_output_0", "inputIndexes": [ 3065 ], "outputIndexes": [ 3066 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3316077618, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3066 ], "outputIndexes": [ 4195 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4195 ], "outputIndexes": [ 4196 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear", "inputIndexes": [ 4196 ], "outputIndexes": [ 4197 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3316110386, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4197 ], "outputIndexes": [ 4198 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4198 ], "outputIndexes": [ 3067 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3067 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_33/Mul_output_0", "outputIndexes": [ 3068 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3066 ], "outputIndexes": [ 4199 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4199 ], "outputIndexes": [ 4200 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear", "inputIndexes": [ 4200 ], "outputIndexes": [ 4201 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3341472840, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4201 ], "outputIndexes": [ 4202 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4202 ], "outputIndexes": [ 3069 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3068, 3069 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_33/Mul_output_0", "outputIndexes": [ 3070 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3070 ], "outputIndexes": [ 4203 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4203 ], "outputIndexes": [ 4204 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear", "inputIndexes": [ 4204 ], "outputIndexes": [ 4205 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3366835294, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4205 ], "outputIndexes": [ 4206 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4206 ], "outputIndexes": [ 3071 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3065, 3071 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_169_output_0", "outputIndexes": [ 3072 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3072, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_306_output_0", "outputIndexes": [ 3073 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_34/Mul_1_output_0", "inputIndexes": [ 3073 ], "outputIndexes": [ 3074 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3392197748, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3074 ], "outputIndexes": [ 4207 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4207 ], "outputIndexes": [ 4208 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear", "inputIndexes": [ 4208 ], "outputIndexes": [ 4209 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3392230516, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4209 ], "outputIndexes": [ 4210 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4210 ], "outputIndexes": [ 3075 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3074 ], "main_type": "NONE", "name": "/Shape_476_output_0", "outputIndexes": [ 3076 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3076 ], "main_type": "NONE", "name": "Shape12724", "outputIndexes": [ 3077 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3076 ], "main_type": "NONE", "name": "Rank12726", "outputIndexes": [ 3078 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3078, 3078 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12728", "outputIndexes": [ 3079 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3079 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12729", "outputIndexes": [ 3080 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3079, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12731", "outputIndexes": [ 3081 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3081 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12732", "outputIndexes": [ 3082 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3077, 3080, 3082, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12735", "outputIndexes": [ 3083 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3083 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12736", "outputIndexes": [ 3084 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3084, 3084 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12738", "outputIndexes": [ 3085 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3076, 3085, 3079 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_513_output_0", "outputIndexes": [ 3086 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3086, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1088_output_0", "outputIndexes": [ 3087 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3079 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12746", "outputIndexes": [ 3088 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3081 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12749", "outputIndexes": [ 3089 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3077, 3088, 3089, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12752", "outputIndexes": [ 3090 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3090 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12753", "outputIndexes": [ 3091 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3091 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12754", "outputIndexes": [ 3092 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3092, 3091 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12755", "outputIndexes": [ 3093 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3076, 3093, 3079 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_514_output_0", "outputIndexes": [ 3094 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3094, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1089_output_0", "outputIndexes": [ 3095 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3087, 3095, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_442_output_0", "outputIndexes": [ 3096 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3075, 3096 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_307_output_0", "outputIndexes": [ 3097 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3097, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_272_output_0", "outputIndexes": [ 3098 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3097 ], "main_type": "NONE", "name": "/Shape_478_output_0", "outputIndexes": [ 3099 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3099 ], "main_type": "NONE", "name": "Shape12856", "outputIndexes": [ 3100 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3099 ], "main_type": "NONE", "name": "Rank12858", "outputIndexes": [ 3101 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3101, 3101 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12860", "outputIndexes": [ 3102 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3102 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12861", "outputIndexes": [ 3103 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3102, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12863", "outputIndexes": [ 3104 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3104 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12864", "outputIndexes": [ 3105 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3100, 3103, 3105, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12867", "outputIndexes": [ 3106 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3106 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12868", "outputIndexes": [ 3107 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 3107 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12869", "outputIndexes": [ 3108 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3108, 3107 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12870", "outputIndexes": [ 3109 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3099, 3109, 3102 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_515_output_0", "outputIndexes": [ 3110 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3110, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_102_output_0", "outputIndexes": [ 3111 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3111, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1095_output_0", "outputIndexes": [ 3112 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3097, 3112, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_137_output_0", "outputIndexes": [ 3113 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3113 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_68_output_0", "outputIndexes": [ 3114 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3111, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1094_output_0", "outputIndexes": [ 3115 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3097, 20, 3115, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_136_output_0", "outputIndexes": [ 3116 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3114, 3116 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_445_output_0", "outputIndexes": [ 3117 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3117, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_273_output_0", "outputIndexes": [ 3118 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3098, 3118 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_170_output_0", "outputIndexes": [ 3119 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3074 ], "outputIndexes": [ 4211 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4211 ], "outputIndexes": [ 4212 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear", "inputIndexes": [ 4212 ], "outputIndexes": [ 4213 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3401684106, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4213 ], "outputIndexes": [ 4214 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4214 ], "outputIndexes": [ 3120 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3086, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1090_output_0", "outputIndexes": [ 3121 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3094, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1091_output_0", "outputIndexes": [ 3122 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3121, 3122, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_443_output_0", "outputIndexes": [ 3123 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3120, 3123 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_308_output_0", "outputIndexes": [ 3124 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3124, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_274_output_0", "outputIndexes": [ 3125 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3124 ], "main_type": "NONE", "name": "/Shape_479_output_0", "outputIndexes": [ 3126 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3126 ], "main_type": "NONE", "name": "Shape12762", "outputIndexes": [ 3127 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3126 ], "main_type": "NONE", "name": "Rank12764", "outputIndexes": [ 3128 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3128, 3128 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12766", "outputIndexes": [ 3129 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3129 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12767", "outputIndexes": [ 3130 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3129, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12769", "outputIndexes": [ 3131 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3131 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12770", "outputIndexes": [ 3132 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3127, 3130, 3132, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12773", "outputIndexes": [ 3133 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3133 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12774", "outputIndexes": [ 3134 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 3134 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12775", "outputIndexes": [ 3135 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3135, 3134 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12776", "outputIndexes": [ 3136 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3126, 3136, 3129 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_516_output_0", "outputIndexes": [ 3137 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3137, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_103_output_0", "outputIndexes": [ 3138 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3138, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1097_output_0", "outputIndexes": [ 3139 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3124, 3139, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_139_output_0", "outputIndexes": [ 3140 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3140 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_69_output_0", "outputIndexes": [ 3141 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3138, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1096_output_0", "outputIndexes": [ 3142 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3124, 20, 3142, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_138_output_0", "outputIndexes": [ 3143 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3141, 3143 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_446_output_0", "outputIndexes": [ 3144 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3144, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_275_output_0", "outputIndexes": [ 3145 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3125, 3145 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_171_output_0", "outputIndexes": [ 3146 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3074 ], "outputIndexes": [ 4215 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4215 ], "outputIndexes": [ 4216 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear", "inputIndexes": [ 4216 ], "outputIndexes": [ 4217 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3404047520, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4217 ], "outputIndexes": [ 4218 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4218 ], "outputIndexes": [ 3147 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3086, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1092_output_0", "outputIndexes": [ 3148 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3094, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1093_output_0", "outputIndexes": [ 3149 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3148, 3149, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_444_output_0", "outputIndexes": [ 3150 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3147, 3150 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_309_output_0", "outputIndexes": [ 3151 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3119, 3146, 3151, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_314_output_0", "outputIndexes": [ 3152 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3152 ], "outputIndexes": [ 4219 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4219 ], "outputIndexes": [ 4220 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear", "inputIndexes": [ 4220 ], "outputIndexes": [ 4221 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3406410934, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4221 ], "outputIndexes": [ 4222 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4222 ], "outputIndexes": [ 3153 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3073, 3153 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_173_output_0", "outputIndexes": [ 3154 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_34/Mul_1_output_0", "inputIndexes": [ 3154 ], "outputIndexes": [ 3155 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3415848140, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3155 ], "outputIndexes": [ 4223 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4223 ], "outputIndexes": [ 4224 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear", "inputIndexes": [ 4224 ], "outputIndexes": [ 4225 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3415880908, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4225 ], "outputIndexes": [ 4226 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4226 ], "outputIndexes": [ 3156 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3156 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_34/Mul_output_0", "outputIndexes": [ 3157 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3155 ], "outputIndexes": [ 4227 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4227 ], "outputIndexes": [ 4228 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear", "inputIndexes": [ 4228 ], "outputIndexes": [ 4229 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3441243362, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4229 ], "outputIndexes": [ 4230 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4230 ], "outputIndexes": [ 3158 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3157, 3158 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_34/Mul_output_0", "outputIndexes": [ 3159 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3159 ], "outputIndexes": [ 4231 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4231 ], "outputIndexes": [ 4232 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear", "inputIndexes": [ 4232 ], "outputIndexes": [ 4233 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3466605816, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4233 ], "outputIndexes": [ 4234 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4234 ], "outputIndexes": [ 3160 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3154, 3160 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_174_output_0", "outputIndexes": [ 3161 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3161, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_315_output_0", "outputIndexes": [ 3162 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_35/Mul_1_output_0", "inputIndexes": [ 3162 ], "outputIndexes": [ 3163 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3491968270, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3163 ], "outputIndexes": [ 4235 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 4235 ], "outputIndexes": [ 4236 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear", "inputIndexes": [ 4236 ], "outputIndexes": [ 4237 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3492001038, 8388630, 1048576, 16384, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 4237 ], "outputIndexes": [ 4238 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4238 ], "outputIndexes": [ 3164 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3163 ], "main_type": "NONE", "name": "/Shape_490_output_0", "outputIndexes": [ 3165 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3165 ], "main_type": "NONE", "name": "Shape13095", "outputIndexes": [ 3166 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3165 ], "main_type": "NONE", "name": "Rank13097", "outputIndexes": [ 3167 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3167, 3167 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13099", "outputIndexes": [ 3168 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3168 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13100", "outputIndexes": [ 3169 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3168, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13102", "outputIndexes": [ 3170 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3170 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13103", "outputIndexes": [ 3171 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3166, 3169, 3171, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13106", "outputIndexes": [ 3172 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3172 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13107", "outputIndexes": [ 3173 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3173, 3173 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13109", "outputIndexes": [ 3174 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3165, 3174, 3168 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_528_output_0", "outputIndexes": [ 3175 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3175, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1120_output_0", "outputIndexes": [ 3176 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3168 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13117", "outputIndexes": [ 3177 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3170 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13120", "outputIndexes": [ 3178 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3166, 3177, 3178, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13123", "outputIndexes": [ 3179 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3179 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13124", "outputIndexes": [ 3180 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3180 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13125", "outputIndexes": [ 3181 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3181, 3180 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13126", "outputIndexes": [ 3182 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3165, 3182, 3168 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_529_output_0", "outputIndexes": [ 3183 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3183, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1121_output_0", "outputIndexes": [ 3184 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3176, 3184, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_455_output_0", "outputIndexes": [ 3185 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3164, 3185 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_316_output_0", "outputIndexes": [ 3186 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3186, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_280_output_0", "outputIndexes": [ 3187 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3186 ], "main_type": "NONE", "name": "/Shape_492_output_0", "outputIndexes": [ 3188 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3188 ], "main_type": "NONE", "name": "Shape13195", "outputIndexes": [ 3189 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3188 ], "main_type": "NONE", "name": "Rank13197", "outputIndexes": [ 3190 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3190, 3190 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13199", "outputIndexes": [ 3191 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3191 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13200", "outputIndexes": [ 3192 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3191, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13202", "outputIndexes": [ 3193 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3193 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13203", "outputIndexes": [ 3194 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3189, 3192, 3194, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13206", "outputIndexes": [ 3195 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3195 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13207", "outputIndexes": [ 3196 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 3196 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13208", "outputIndexes": [ 3197 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3197, 3196 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13209", "outputIndexes": [ 3198 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3188, 3198, 3191 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_530_output_0", "outputIndexes": [ 3199 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3199, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_105_output_0", "outputIndexes": [ 3200 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3200, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1127_output_0", "outputIndexes": [ 3201 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3186, 3201, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_141_output_0", "outputIndexes": [ 3202 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3202 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_70_output_0", "outputIndexes": [ 3203 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3200, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1126_output_0", "outputIndexes": [ 3204 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3186, 20, 3204, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_140_output_0", "outputIndexes": [ 3205 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3203, 3205 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_458_output_0", "outputIndexes": [ 3206 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3206, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_281_output_0", "outputIndexes": [ 3207 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3187, 3207 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_175_output_0", "outputIndexes": [ 3208 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3163 ], "outputIndexes": [ 4239 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 4239 ], "outputIndexes": [ 4240 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear", "inputIndexes": [ 4240 ], "outputIndexes": [ 4241 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3501454628, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 4241 ], "outputIndexes": [ 4242 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4242 ], "outputIndexes": [ 3209 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3175, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1122_output_0", "outputIndexes": [ 3210 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3183, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1123_output_0", "outputIndexes": [ 3211 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3210, 3211, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_456_output_0", "outputIndexes": [ 3212 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3209, 3212 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_317_output_0", "outputIndexes": [ 3213 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3213, 58 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_282_output_0", "outputIndexes": [ 3214 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3213 ], "main_type": "NONE", "name": "/Shape_493_output_0", "outputIndexes": [ 3215 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3215 ], "main_type": "NONE", "name": "Shape13133", "outputIndexes": [ 3216 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3215 ], "main_type": "NONE", "name": "Rank13135", "outputIndexes": [ 3217 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3217, 3217 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13137", "outputIndexes": [ 3218 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3218 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13138", "outputIndexes": [ 3219 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3218, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13140", "outputIndexes": [ 3220 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3220 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13141", "outputIndexes": [ 3221 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3216, 3219, 3221, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13144", "outputIndexes": [ 3222 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3222 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13145", "outputIndexes": [ 3223 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 3223 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13146", "outputIndexes": [ 3224 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3224, 3223 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13147", "outputIndexes": [ 3225 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3215, 3225, 3218 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_531_output_0", "outputIndexes": [ 3226 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3226, 73 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_106_output_0", "outputIndexes": [ 3227 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3227, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1129_output_0", "outputIndexes": [ 3228 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3213, 3228, 76, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_143_output_0", "outputIndexes": [ 3229 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3229 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_71_output_0", "outputIndexes": [ 3230 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3227, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1128_output_0", "outputIndexes": [ 3231 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3213, 20, 3231, 77, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_142_output_0", "outputIndexes": [ 3232 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3230, 3232 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_459_output_0", "outputIndexes": [ 3233 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3233, 89 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_283_output_0", "outputIndexes": [ 3234 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3214, 3234 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_176_output_0", "outputIndexes": [ 3235 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3163 ], "outputIndexes": [ 4243 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 4243 ], "outputIndexes": [ 4244 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear", "inputIndexes": [ 4244 ], "outputIndexes": [ 4245 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 32768, "weightSize": 0 }, "external": [ 3503818042, 2097174, 262144, 4096, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 4245 ], "outputIndexes": [ 4246 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4246 ], "outputIndexes": [ 3236 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3175, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1124_output_0", "outputIndexes": [ 3237 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3183, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1125_output_0", "outputIndexes": [ 3238 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3237, 3238, 95, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_457_output_0", "outputIndexes": [ 3239 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3236, 3239 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_318_output_0", "outputIndexes": [ 3240 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3208, 3235, 3240, 125 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_323_output_0", "outputIndexes": [ 3241 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3241 ], "outputIndexes": [ 4247 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 4247 ], "outputIndexes": [ 4248 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear", "inputIndexes": [ 4248 ], "outputIndexes": [ 4249 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 131072, "weightSize": 0 }, "external": [ 3506181456, 8388630, 1048576, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 4249 ], "outputIndexes": [ 4250 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4250 ], "outputIndexes": [ 3242 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3162, 3242 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_178_output_0", "outputIndexes": [ 3243 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_35/Mul_1_output_0", "inputIndexes": [ 3243 ], "outputIndexes": [ 3244 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3515618662, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3244 ], "outputIndexes": [ 4251 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 4251 ], "outputIndexes": [ 4252 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear", "inputIndexes": [ 4252 ], "outputIndexes": [ 4253 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3515651430, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 4253 ], "outputIndexes": [ 4254 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4254 ], "outputIndexes": [ 3245 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3245 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_35/Mul_output_0", "outputIndexes": [ 3246 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3244 ], "outputIndexes": [ 4255 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 4255 ], "outputIndexes": [ 4256 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear", "inputIndexes": [ 4256 ], "outputIndexes": [ 4257 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 11008, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3541013884, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 4257 ], "outputIndexes": [ 4258 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4258 ], "outputIndexes": [ 3247 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 11008 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3246, 3247 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_35/Mul_output_0", "outputIndexes": [ 3248 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3248 ], "outputIndexes": [ 4259 ], "main_type": "Reshape", "main": { "dims": [ -1, 11008, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 4259 ], "outputIndexes": [ 4260 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear", "inputIndexes": [ 4260 ], "outputIndexes": [ 4261 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 4096, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 11008, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 352256, "weightSize": 0 }, "external": [ 3566376338, 22544406, 2818048, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 4261 ], "outputIndexes": [ 4262 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4262 ], "outputIndexes": [ 3249 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 4096 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3243, 3249 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_179_output_0", "outputIndexes": [ 3250 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3250, 3251, 76, 15, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/lm/Slice_output_0", "outputIndexes": [ 3252 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "name": "/lm/norm/Mul_1_output_0", "inputIndexes": [ 3252 ], "outputIndexes": [ 3253 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-05, "group": 1, "useRMSNorm": true, "external": [ 3591738792, 16384, 16384 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3253 ], "outputIndexes": [ 4263 ], "main_type": "Reshape", "main": { "dims": [ -1, 4096, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/pre_convert", "inputIndexes": [ 4263 ], "outputIndexes": [ 4264 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear", "inputIndexes": [ 4264 ], "outputIndexes": [ 4265 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 151680, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 4096, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": true, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 4853760, "weightSize": 0 }, "external": [ 3591771560, 310640666, 38830080, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/post_convert", "inputIndexes": [ 4265 ], "outputIndexes": [ 4266 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 4266 ], "outputIndexes": [ 3254 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 151680 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" } ], "outputName": [ "logits", "presents" ], "preferForwardType": "CPU", "sourceType": "ONNX", "tensorName": [ "past_key_values", "presents", "input_ids", "/Constant_output_0", "/Reshape_output_0", "/input_layernorm/Mul_1_output_0", "/FakeLinear_output_0", "/Shape_output_0", "Shape51", "BinaryOp54", "BinaryOp55", "Unsqueeze56", "Const9", "BinaryOp58", "Unsqueeze59", "Unsqueeze61", "StridedSlice62", "BinaryOp64", "BinaryOp65", "/Gather_1_output_0", "/rotary/Constant_4_output_0", "/Unsqueeze_output_0", "Unsqueeze75", "Unsqueeze78", "StridedSlice81", "Squeeze82", "BinaryOp83", "BinaryOp84", "/Gather_2_output_0", "/Unsqueeze_1_output_0", "/Constant_4_output_0", "/Constant_5_output_0", "/Concat_output_0", "/Reshape_1_output_0", "position_ids", "/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", "Shape107", "BinaryOp110", "BinaryOp111", "Unsqueeze112", "BinaryOp114", "Unsqueeze115", "StridedSlice118", "BinaryOp120", "BinaryOp121", "/Gather_3_output_0", "/Mul_output_0", "/Shape_2_output_0", "/Constant_11_output_0", "Shape239", "BinaryOp242", "BinaryOp243", "Unsqueeze244", "BinaryOp246", "Unsqueeze247", "StridedSlice250", "Squeeze251", "BinaryOp252", "BinaryOp253", "/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", "Unsqueeze158", "Unsqueeze161", "StridedSlice164", "Squeeze165", "BinaryOp166", "BinaryOp167", "/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", "/Mul_2_output_0", "/Shape_3_output_0", "Shape126", "BinaryOp129", "BinaryOp130", "Unsqueeze131", "BinaryOp133", "Unsqueeze134", "StridedSlice137", "Squeeze138", "BinaryOp139", "BinaryOp140", "/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", "Shape484", "BinaryOp487", "BinaryOp488", "Unsqueeze489", "BinaryOp491", "Unsqueeze492", "StridedSlice495", "BinaryOp497", "BinaryOp498", "/Gather_18_output_0", "/Unsqueeze_32_output_0", "Unsqueeze506", "Unsqueeze509", "StridedSlice512", "Squeeze513", "BinaryOp514", "BinaryOp515", "/Gather_19_output_0", "/Unsqueeze_33_output_0", "/Concat_13_output_0", "/Reshape_10_output_0", "/Mul_8_output_0", "/Shape_16_output_0", "Shape615", "BinaryOp618", "BinaryOp619", "Unsqueeze620", "BinaryOp622", "Unsqueeze623", "StridedSlice626", "Squeeze627", "BinaryOp628", "BinaryOp629", "/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", "/Mul_10_output_0", "/Shape_17_output_0", "Shape522", "BinaryOp525", "BinaryOp526", "Unsqueeze527", "BinaryOp529", "Unsqueeze530", "StridedSlice533", "Squeeze534", "BinaryOp535", "BinaryOp536", "/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", "Shape854", "BinaryOp857", "BinaryOp858", "Unsqueeze859", "BinaryOp861", "Unsqueeze862", "StridedSlice865", "BinaryOp867", "BinaryOp868", "/Gather_33_output_0", "/Unsqueeze_64_output_0", "Unsqueeze876", "Unsqueeze879", "StridedSlice882", "Squeeze883", "BinaryOp884", "BinaryOp885", "/Gather_34_output_0", "/Unsqueeze_65_output_0", "/Concat_26_output_0", "/Reshape_19_output_0", "/Mul_16_output_0", "/Shape_30_output_0", "Shape985", "BinaryOp988", "BinaryOp989", "Unsqueeze990", "BinaryOp992", "Unsqueeze993", "StridedSlice996", "Squeeze997", "BinaryOp998", "BinaryOp999", "/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", "/Mul_18_output_0", "/Shape_31_output_0", "Shape892", "BinaryOp895", "BinaryOp896", "Unsqueeze897", "BinaryOp899", "Unsqueeze900", "StridedSlice903", "Squeeze904", "BinaryOp905", "BinaryOp906", "/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", "Shape1224", "BinaryOp1227", "BinaryOp1228", "Unsqueeze1229", "BinaryOp1231", "Unsqueeze1232", "StridedSlice1235", "BinaryOp1237", "BinaryOp1238", "/Gather_48_output_0", "/Unsqueeze_96_output_0", "Unsqueeze1246", "Unsqueeze1249", "StridedSlice1252", "Squeeze1253", "BinaryOp1254", "BinaryOp1255", "/Gather_49_output_0", "/Unsqueeze_97_output_0", "/Concat_39_output_0", "/Reshape_28_output_0", "/Mul_24_output_0", "/Shape_44_output_0", "Shape1355", "BinaryOp1358", "BinaryOp1359", "Unsqueeze1360", "BinaryOp1362", "Unsqueeze1363", "StridedSlice1366", "Squeeze1367", "BinaryOp1368", "BinaryOp1369", "/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", "/Mul_26_output_0", "/Shape_45_output_0", "Shape1262", "BinaryOp1265", "BinaryOp1266", "Unsqueeze1267", "BinaryOp1269", "Unsqueeze1270", "StridedSlice1273", "Squeeze1274", "BinaryOp1275", "BinaryOp1276", "/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", "Shape1594", "BinaryOp1597", "BinaryOp1598", "Unsqueeze1599", "BinaryOp1601", "Unsqueeze1602", "StridedSlice1605", "BinaryOp1607", "BinaryOp1608", "/Gather_63_output_0", "/Unsqueeze_128_output_0", "Unsqueeze1616", "Unsqueeze1619", "StridedSlice1622", "Squeeze1623", "BinaryOp1624", "BinaryOp1625", "/Gather_64_output_0", "/Unsqueeze_129_output_0", "/Concat_52_output_0", "/Reshape_37_output_0", "/Mul_32_output_0", "/Shape_58_output_0", "Shape1726", "BinaryOp1729", "BinaryOp1730", "Unsqueeze1731", "BinaryOp1733", "Unsqueeze1734", "StridedSlice1737", "Squeeze1738", "BinaryOp1739", "BinaryOp1740", "/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", "/Mul_34_output_0", "/Shape_59_output_0", "Shape1632", "BinaryOp1635", "BinaryOp1636", "Unsqueeze1637", "BinaryOp1639", "Unsqueeze1640", "StridedSlice1643", "Squeeze1644", "BinaryOp1645", "BinaryOp1646", "/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", "Shape1965", "BinaryOp1968", "BinaryOp1969", "Unsqueeze1970", "BinaryOp1972", "Unsqueeze1973", "StridedSlice1976", "BinaryOp1978", "BinaryOp1979", "/Gather_78_output_0", "/Unsqueeze_160_output_0", "Unsqueeze1987", "Unsqueeze1990", "StridedSlice1993", "Squeeze1994", "BinaryOp1995", "BinaryOp1996", "/Gather_79_output_0", "/Unsqueeze_161_output_0", "/Concat_65_output_0", "/Reshape_46_output_0", "/Mul_40_output_0", "/Shape_72_output_0", "Shape2097", "BinaryOp2100", "BinaryOp2101", "Unsqueeze2102", "BinaryOp2104", "Unsqueeze2105", "StridedSlice2108", "Squeeze2109", "BinaryOp2110", "BinaryOp2111", "/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", "/Mul_42_output_0", "/Shape_73_output_0", "Shape2003", "BinaryOp2006", "BinaryOp2007", "Unsqueeze2008", "BinaryOp2010", "Unsqueeze2011", "StridedSlice2014", "Squeeze2015", "BinaryOp2016", "BinaryOp2017", "/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", "Shape2336", "BinaryOp2339", "BinaryOp2340", "Unsqueeze2341", "BinaryOp2343", "Unsqueeze2344", "StridedSlice2347", "BinaryOp2349", "BinaryOp2350", "/Gather_93_output_0", "/Unsqueeze_192_output_0", "Unsqueeze2358", "Unsqueeze2361", "StridedSlice2364", "Squeeze2365", "BinaryOp2366", "BinaryOp2367", "/Gather_94_output_0", "/Unsqueeze_193_output_0", "/Concat_78_output_0", "/Reshape_55_output_0", "/Mul_48_output_0", "/Shape_86_output_0", "Shape2468", "BinaryOp2471", "BinaryOp2472", "Unsqueeze2473", "BinaryOp2475", "Unsqueeze2476", "StridedSlice2479", "Squeeze2480", "BinaryOp2481", "BinaryOp2482", "/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", "/Mul_50_output_0", "/Shape_87_output_0", "Shape2374", "BinaryOp2377", "BinaryOp2378", "Unsqueeze2379", "BinaryOp2381", "Unsqueeze2382", "StridedSlice2385", "Squeeze2386", "BinaryOp2387", "BinaryOp2388", "/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", "Shape2707", "BinaryOp2710", "BinaryOp2711", "Unsqueeze2712", "BinaryOp2714", "Unsqueeze2715", "StridedSlice2718", "BinaryOp2720", "BinaryOp2721", "/Gather_108_output_0", "/Unsqueeze_224_output_0", "Unsqueeze2729", "Unsqueeze2732", "StridedSlice2735", "Squeeze2736", "BinaryOp2737", "BinaryOp2738", "/Gather_109_output_0", "/Unsqueeze_225_output_0", "/Concat_91_output_0", "/Reshape_64_output_0", "/Mul_56_output_0", "/Shape_100_output_0", "Shape2839", "BinaryOp2842", "BinaryOp2843", "Unsqueeze2844", "BinaryOp2846", "Unsqueeze2847", "StridedSlice2850", "Squeeze2851", "BinaryOp2852", "BinaryOp2853", "/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", "/Mul_58_output_0", "/Shape_101_output_0", "Shape2745", "BinaryOp2748", "BinaryOp2749", "Unsqueeze2750", "BinaryOp2752", "Unsqueeze2753", "StridedSlice2756", "Squeeze2757", "BinaryOp2758", "BinaryOp2759", "/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", "Shape3078", "BinaryOp3081", "BinaryOp3082", "Unsqueeze3083", "BinaryOp3085", "Unsqueeze3086", "StridedSlice3089", "BinaryOp3091", "BinaryOp3092", "/Gather_123_output_0", "/Unsqueeze_256_output_0", "Unsqueeze3100", "Unsqueeze3103", "StridedSlice3106", "Squeeze3107", "BinaryOp3108", "BinaryOp3109", "/Gather_124_output_0", "/Unsqueeze_257_output_0", "/Concat_104_output_0", "/Reshape_73_output_0", "/Mul_64_output_0", "/Shape_114_output_0", "Shape3210", "BinaryOp3213", "BinaryOp3214", "Unsqueeze3215", "BinaryOp3217", "Unsqueeze3218", "StridedSlice3221", "Squeeze3222", "BinaryOp3223", "BinaryOp3224", "/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", "/Mul_66_output_0", "/Shape_115_output_0", "Shape3116", "BinaryOp3119", "BinaryOp3120", "Unsqueeze3121", "BinaryOp3123", "Unsqueeze3124", "StridedSlice3127", "Squeeze3128", "BinaryOp3129", "BinaryOp3130", "/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", "Shape3449", "BinaryOp3452", "BinaryOp3453", "Unsqueeze3454", "BinaryOp3456", "Unsqueeze3457", "StridedSlice3460", "BinaryOp3462", "BinaryOp3463", "/Gather_138_output_0", "/Unsqueeze_288_output_0", "Unsqueeze3471", "Unsqueeze3474", "StridedSlice3477", "Squeeze3478", "BinaryOp3479", "BinaryOp3480", "/Gather_139_output_0", "/Unsqueeze_289_output_0", "/Concat_117_output_0", "/Reshape_82_output_0", "/Mul_72_output_0", "/Shape_128_output_0", "Shape3581", "BinaryOp3584", "BinaryOp3585", "Unsqueeze3586", "BinaryOp3588", "Unsqueeze3589", "StridedSlice3592", "Squeeze3593", "BinaryOp3594", "BinaryOp3595", "/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", "/Mul_74_output_0", "/Shape_129_output_0", "Shape3487", "BinaryOp3490", "BinaryOp3491", "Unsqueeze3492", "BinaryOp3494", "Unsqueeze3495", "StridedSlice3498", "Squeeze3499", "BinaryOp3500", "BinaryOp3501", "/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", "Shape3820", "BinaryOp3823", "BinaryOp3824", "Unsqueeze3825", "BinaryOp3827", "Unsqueeze3828", "StridedSlice3831", "BinaryOp3833", "BinaryOp3834", "/Gather_153_output_0", "/Unsqueeze_320_output_0", "Unsqueeze3842", "Unsqueeze3845", "StridedSlice3848", "Squeeze3849", "BinaryOp3850", "BinaryOp3851", "/Gather_154_output_0", "/Unsqueeze_321_output_0", "/Concat_130_output_0", "/Reshape_91_output_0", "/Mul_80_output_0", "/Shape_142_output_0", "Shape3952", "BinaryOp3955", "BinaryOp3956", "Unsqueeze3957", "BinaryOp3959", "Unsqueeze3960", "StridedSlice3963", "Squeeze3964", "BinaryOp3965", "BinaryOp3966", "/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", "/Mul_82_output_0", "/Shape_143_output_0", "Shape3858", "BinaryOp3861", "BinaryOp3862", "Unsqueeze3863", "BinaryOp3865", "Unsqueeze3866", "StridedSlice3869", "Squeeze3870", "BinaryOp3871", "BinaryOp3872", "/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", "Shape4191", "BinaryOp4194", "BinaryOp4195", "Unsqueeze4196", "BinaryOp4198", "Unsqueeze4199", "StridedSlice4202", "BinaryOp4204", "BinaryOp4205", "/Gather_168_output_0", "/Unsqueeze_352_output_0", "Unsqueeze4213", "Unsqueeze4216", "StridedSlice4219", "Squeeze4220", "BinaryOp4221", "BinaryOp4222", "/Gather_169_output_0", "/Unsqueeze_353_output_0", "/Concat_143_output_0", "/Reshape_100_output_0", "/Mul_88_output_0", "/Shape_156_output_0", "Shape4323", "BinaryOp4326", "BinaryOp4327", "Unsqueeze4328", "BinaryOp4330", "Unsqueeze4331", "StridedSlice4334", "Squeeze4335", "BinaryOp4336", "BinaryOp4337", "/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", "/Mul_90_output_0", "/Shape_157_output_0", "Shape4229", "BinaryOp4232", "BinaryOp4233", "Unsqueeze4234", "BinaryOp4236", "Unsqueeze4237", "StridedSlice4240", "Squeeze4241", "BinaryOp4242", "BinaryOp4243", "/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", "Shape4562", "BinaryOp4565", "BinaryOp4566", "Unsqueeze4567", "BinaryOp4569", "Unsqueeze4570", "StridedSlice4573", "BinaryOp4575", "BinaryOp4576", "/Gather_183_output_0", "/Unsqueeze_384_output_0", "Unsqueeze4584", "Unsqueeze4587", "StridedSlice4590", "Squeeze4591", "BinaryOp4592", "BinaryOp4593", "/Gather_184_output_0", "/Unsqueeze_385_output_0", "/Concat_156_output_0", "/Reshape_109_output_0", "/Mul_96_output_0", "/Shape_170_output_0", "Shape4694", "BinaryOp4697", "BinaryOp4698", "Unsqueeze4699", "BinaryOp4701", "Unsqueeze4702", "StridedSlice4705", "Squeeze4706", "BinaryOp4707", "BinaryOp4708", "/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", "/Mul_98_output_0", "/Shape_171_output_0", "Shape4600", "BinaryOp4603", "BinaryOp4604", "Unsqueeze4605", "BinaryOp4607", "Unsqueeze4608", "StridedSlice4611", "Squeeze4612", "BinaryOp4613", "BinaryOp4614", "/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", "Shape4933", "BinaryOp4936", "BinaryOp4937", "Unsqueeze4938", "BinaryOp4940", "Unsqueeze4941", "StridedSlice4944", "BinaryOp4946", "BinaryOp4947", "/Gather_198_output_0", "/Unsqueeze_416_output_0", "Unsqueeze4955", "Unsqueeze4958", "StridedSlice4961", "Squeeze4962", "BinaryOp4963", "BinaryOp4964", "/Gather_199_output_0", "/Unsqueeze_417_output_0", "/Concat_169_output_0", "/Reshape_118_output_0", "/Mul_104_output_0", "/Shape_184_output_0", "Shape5065", "BinaryOp5068", "BinaryOp5069", "Unsqueeze5070", "BinaryOp5072", "Unsqueeze5073", "StridedSlice5076", "Squeeze5077", "BinaryOp5078", "BinaryOp5079", "/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", "/Mul_106_output_0", "/Shape_185_output_0", "Shape4971", "BinaryOp4974", "BinaryOp4975", "Unsqueeze4976", "BinaryOp4978", "Unsqueeze4979", "StridedSlice4982", "Squeeze4983", "BinaryOp4984", "BinaryOp4985", "/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", "Shape5304", "BinaryOp5307", "BinaryOp5308", "Unsqueeze5309", "BinaryOp5311", "Unsqueeze5312", "StridedSlice5315", "BinaryOp5317", "BinaryOp5318", "/Gather_213_output_0", "/Unsqueeze_448_output_0", "Unsqueeze5326", "Unsqueeze5329", "StridedSlice5332", "Squeeze5333", "BinaryOp5334", "BinaryOp5335", "/Gather_214_output_0", "/Unsqueeze_449_output_0", "/Concat_182_output_0", "/Reshape_127_output_0", "/Mul_112_output_0", "/Shape_198_output_0", "Shape5436", "BinaryOp5439", "BinaryOp5440", "Unsqueeze5441", "BinaryOp5443", "Unsqueeze5444", "StridedSlice5447", "Squeeze5448", "BinaryOp5449", "BinaryOp5450", "/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", "/Mul_114_output_0", "/Shape_199_output_0", "Shape5342", "BinaryOp5345", "BinaryOp5346", "Unsqueeze5347", "BinaryOp5349", "Unsqueeze5350", "StridedSlice5353", "Squeeze5354", "BinaryOp5355", "BinaryOp5356", "/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", "Shape5675", "BinaryOp5678", "BinaryOp5679", "Unsqueeze5680", "BinaryOp5682", "Unsqueeze5683", "StridedSlice5686", "BinaryOp5688", "BinaryOp5689", "/Gather_228_output_0", "/Unsqueeze_480_output_0", "Unsqueeze5697", "Unsqueeze5700", "StridedSlice5703", "Squeeze5704", "BinaryOp5705", "BinaryOp5706", "/Gather_229_output_0", "/Unsqueeze_481_output_0", "/Concat_195_output_0", "/Reshape_136_output_0", "/Mul_120_output_0", "/Shape_212_output_0", "Shape5807", "BinaryOp5810", "BinaryOp5811", "Unsqueeze5812", "BinaryOp5814", "Unsqueeze5815", "StridedSlice5818", "Squeeze5819", "BinaryOp5820", "BinaryOp5821", "/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", "/Mul_122_output_0", "/Shape_213_output_0", "Shape5713", "BinaryOp5716", "BinaryOp5717", "Unsqueeze5718", "BinaryOp5720", "Unsqueeze5721", "StridedSlice5724", "Squeeze5725", "BinaryOp5726", "BinaryOp5727", "/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", "Shape6046", "BinaryOp6049", "BinaryOp6050", "Unsqueeze6051", "BinaryOp6053", "Unsqueeze6054", "StridedSlice6057", "BinaryOp6059", "BinaryOp6060", "/Gather_243_output_0", "/Unsqueeze_512_output_0", "Unsqueeze6068", "Unsqueeze6071", "StridedSlice6074", "Squeeze6075", "BinaryOp6076", "BinaryOp6077", "/Gather_244_output_0", "/Unsqueeze_513_output_0", "/Concat_208_output_0", "/Reshape_145_output_0", "/Mul_128_output_0", "/Shape_226_output_0", "Shape6178", "BinaryOp6181", "BinaryOp6182", "Unsqueeze6183", "BinaryOp6185", "Unsqueeze6186", "StridedSlice6189", "Squeeze6190", "BinaryOp6191", "BinaryOp6192", "/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", "/Mul_130_output_0", "/Shape_227_output_0", "Shape6084", "BinaryOp6087", "BinaryOp6088", "Unsqueeze6089", "BinaryOp6091", "Unsqueeze6092", "StridedSlice6095", "Squeeze6096", "BinaryOp6097", "BinaryOp6098", "/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", "Shape6417", "BinaryOp6420", "BinaryOp6421", "Unsqueeze6422", "BinaryOp6424", "Unsqueeze6425", "StridedSlice6428", "BinaryOp6430", "BinaryOp6431", "/Gather_258_output_0", "/Unsqueeze_544_output_0", "Unsqueeze6439", "Unsqueeze6442", "StridedSlice6445", "Squeeze6446", "BinaryOp6447", "BinaryOp6448", "/Gather_259_output_0", "/Unsqueeze_545_output_0", "/Concat_221_output_0", "/Reshape_154_output_0", "/Mul_136_output_0", "/Shape_240_output_0", "Shape6549", "BinaryOp6552", "BinaryOp6553", "Unsqueeze6554", "BinaryOp6556", "Unsqueeze6557", "StridedSlice6560", "Squeeze6561", "BinaryOp6562", "BinaryOp6563", "/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", "/Mul_138_output_0", "/Shape_241_output_0", "Shape6455", "BinaryOp6458", "BinaryOp6459", "Unsqueeze6460", "BinaryOp6462", "Unsqueeze6463", "StridedSlice6466", "Squeeze6467", "BinaryOp6468", "BinaryOp6469", "/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", "Shape6788", "BinaryOp6791", "BinaryOp6792", "Unsqueeze6793", "BinaryOp6795", "Unsqueeze6796", "StridedSlice6799", "BinaryOp6801", "BinaryOp6802", "/Gather_273_output_0", "/Unsqueeze_576_output_0", "Unsqueeze6810", "Unsqueeze6813", "StridedSlice6816", "Squeeze6817", "BinaryOp6818", "BinaryOp6819", "/Gather_274_output_0", "/Unsqueeze_577_output_0", "/Concat_234_output_0", "/Reshape_163_output_0", "/Mul_144_output_0", "/Shape_254_output_0", "Shape6920", "BinaryOp6923", "BinaryOp6924", "Unsqueeze6925", "BinaryOp6927", "Unsqueeze6928", "StridedSlice6931", "Squeeze6932", "BinaryOp6933", "BinaryOp6934", "/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", "/Mul_146_output_0", "/Shape_255_output_0", "Shape6826", "BinaryOp6829", "BinaryOp6830", "Unsqueeze6831", "BinaryOp6833", "Unsqueeze6834", "StridedSlice6837", "Squeeze6838", "BinaryOp6839", "BinaryOp6840", "/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", "Shape7159", "BinaryOp7162", "BinaryOp7163", "Unsqueeze7164", "BinaryOp7166", "Unsqueeze7167", "StridedSlice7170", "BinaryOp7172", "BinaryOp7173", "/Gather_288_output_0", "/Unsqueeze_608_output_0", "Unsqueeze7181", "Unsqueeze7184", "StridedSlice7187", "Squeeze7188", "BinaryOp7189", "BinaryOp7190", "/Gather_289_output_0", "/Unsqueeze_609_output_0", "/Concat_247_output_0", "/Reshape_172_output_0", "/Mul_152_output_0", "/Shape_268_output_0", "Shape7291", "BinaryOp7294", "BinaryOp7295", "Unsqueeze7296", "BinaryOp7298", "Unsqueeze7299", "StridedSlice7302", "Squeeze7303", "BinaryOp7304", "BinaryOp7305", "/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", "/Mul_154_output_0", "/Shape_269_output_0", "Shape7197", "BinaryOp7200", "BinaryOp7201", "Unsqueeze7202", "BinaryOp7204", "Unsqueeze7205", "StridedSlice7208", "Squeeze7209", "BinaryOp7210", "BinaryOp7211", "/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", "Shape7530", "BinaryOp7533", "BinaryOp7534", "Unsqueeze7535", "BinaryOp7537", "Unsqueeze7538", "StridedSlice7541", "BinaryOp7543", "BinaryOp7544", "/Gather_303_output_0", "/Unsqueeze_640_output_0", "Unsqueeze7552", "Unsqueeze7555", "StridedSlice7558", "Squeeze7559", "BinaryOp7560", "BinaryOp7561", "/Gather_304_output_0", "/Unsqueeze_641_output_0", "/Concat_260_output_0", "/Reshape_181_output_0", "/Mul_160_output_0", "/Shape_282_output_0", "Shape7662", "BinaryOp7665", "BinaryOp7666", "Unsqueeze7667", "BinaryOp7669", "Unsqueeze7670", "StridedSlice7673", "Squeeze7674", "BinaryOp7675", "BinaryOp7676", "/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", "/Mul_162_output_0", "/Shape_283_output_0", "Shape7568", "BinaryOp7571", "BinaryOp7572", "Unsqueeze7573", "BinaryOp7575", "Unsqueeze7576", "StridedSlice7579", "Squeeze7580", "BinaryOp7581", "BinaryOp7582", "/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", "Shape7901", "BinaryOp7904", "BinaryOp7905", "Unsqueeze7906", "BinaryOp7908", "Unsqueeze7909", "StridedSlice7912", "BinaryOp7914", "BinaryOp7915", "/Gather_318_output_0", "/Unsqueeze_672_output_0", "Unsqueeze7923", "Unsqueeze7926", "StridedSlice7929", "Squeeze7930", "BinaryOp7931", "BinaryOp7932", "/Gather_319_output_0", "/Unsqueeze_673_output_0", "/Concat_273_output_0", "/Reshape_190_output_0", "/Mul_168_output_0", "/Shape_296_output_0", "Shape8033", "BinaryOp8036", "BinaryOp8037", "Unsqueeze8038", "BinaryOp8040", "Unsqueeze8041", "StridedSlice8044", "Squeeze8045", "BinaryOp8046", "BinaryOp8047", "/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", "/Mul_170_output_0", "/Shape_297_output_0", "Shape7939", "BinaryOp7942", "BinaryOp7943", "Unsqueeze7944", "BinaryOp7946", "Unsqueeze7947", "StridedSlice7950", "Squeeze7951", "BinaryOp7952", "BinaryOp7953", "/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", "Shape8272", "BinaryOp8275", "BinaryOp8276", "Unsqueeze8277", "BinaryOp8279", "Unsqueeze8280", "StridedSlice8283", "BinaryOp8285", "BinaryOp8286", "/Gather_333_output_0", "/Unsqueeze_704_output_0", "Unsqueeze8294", "Unsqueeze8297", "StridedSlice8300", "Squeeze8301", "BinaryOp8302", "BinaryOp8303", "/Gather_334_output_0", "/Unsqueeze_705_output_0", "/Concat_286_output_0", "/Reshape_199_output_0", "/Mul_176_output_0", "/Shape_310_output_0", "Shape8404", "BinaryOp8407", "BinaryOp8408", "Unsqueeze8409", "BinaryOp8411", "Unsqueeze8412", "StridedSlice8415", "Squeeze8416", "BinaryOp8417", "BinaryOp8418", "/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", "/Mul_178_output_0", "/Shape_311_output_0", "Shape8310", "BinaryOp8313", "BinaryOp8314", "Unsqueeze8315", "BinaryOp8317", "Unsqueeze8318", "StridedSlice8321", "Squeeze8322", "BinaryOp8323", "BinaryOp8324", "/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", "Shape8643", "BinaryOp8646", "BinaryOp8647", "Unsqueeze8648", "BinaryOp8650", "Unsqueeze8651", "StridedSlice8654", "BinaryOp8656", "BinaryOp8657", "/Gather_348_output_0", "/Unsqueeze_736_output_0", "Unsqueeze8665", "Unsqueeze8668", "StridedSlice8671", "Squeeze8672", "BinaryOp8673", "BinaryOp8674", "/Gather_349_output_0", "/Unsqueeze_737_output_0", "/Concat_299_output_0", "/Reshape_208_output_0", "/Mul_184_output_0", "/Shape_324_output_0", "Shape8775", "BinaryOp8778", "BinaryOp8779", "Unsqueeze8780", "BinaryOp8782", "Unsqueeze8783", "StridedSlice8786", "Squeeze8787", "BinaryOp8788", "BinaryOp8789", "/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", "/Mul_186_output_0", "/Shape_325_output_0", "Shape8681", "BinaryOp8684", "BinaryOp8685", "Unsqueeze8686", "BinaryOp8688", "Unsqueeze8689", "StridedSlice8692", "Squeeze8693", "BinaryOp8694", "BinaryOp8695", "/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", "Shape9014", "BinaryOp9017", "BinaryOp9018", "Unsqueeze9019", "BinaryOp9021", "Unsqueeze9022", "StridedSlice9025", "BinaryOp9027", "BinaryOp9028", "/Gather_363_output_0", "/Unsqueeze_768_output_0", "Unsqueeze9036", "Unsqueeze9039", "StridedSlice9042", "Squeeze9043", "BinaryOp9044", "BinaryOp9045", "/Gather_364_output_0", "/Unsqueeze_769_output_0", "/Concat_312_output_0", "/Reshape_217_output_0", "/Mul_192_output_0", "/Shape_338_output_0", "Shape9146", "BinaryOp9149", "BinaryOp9150", "Unsqueeze9151", "BinaryOp9153", "Unsqueeze9154", "StridedSlice9157", "Squeeze9158", "BinaryOp9159", "BinaryOp9160", "/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", "/Mul_194_output_0", "/Shape_339_output_0", "Shape9052", "BinaryOp9055", "BinaryOp9056", "Unsqueeze9057", "BinaryOp9059", "Unsqueeze9060", "StridedSlice9063", "Squeeze9064", "BinaryOp9065", "BinaryOp9066", "/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", "Shape9385", "BinaryOp9388", "BinaryOp9389", "Unsqueeze9390", "BinaryOp9392", "Unsqueeze9393", "StridedSlice9396", "BinaryOp9398", "BinaryOp9399", "/Gather_378_output_0", "/Unsqueeze_800_output_0", "Unsqueeze9407", "Unsqueeze9410", "StridedSlice9413", "Squeeze9414", "BinaryOp9415", "BinaryOp9416", "/Gather_379_output_0", "/Unsqueeze_801_output_0", "/Concat_325_output_0", "/Reshape_226_output_0", "/Mul_200_output_0", "/Shape_352_output_0", "Shape9517", "BinaryOp9520", "BinaryOp9521", "Unsqueeze9522", "BinaryOp9524", "Unsqueeze9525", "StridedSlice9528", "Squeeze9529", "BinaryOp9530", "BinaryOp9531", "/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", "/Mul_202_output_0", "/Shape_353_output_0", "Shape9423", "BinaryOp9426", "BinaryOp9427", "Unsqueeze9428", "BinaryOp9430", "Unsqueeze9431", "StridedSlice9434", "Squeeze9435", "BinaryOp9436", "BinaryOp9437", "/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", "Shape9756", "BinaryOp9759", "BinaryOp9760", "Unsqueeze9761", "BinaryOp9763", "Unsqueeze9764", "StridedSlice9767", "BinaryOp9769", "BinaryOp9770", "/Gather_393_output_0", "/Unsqueeze_832_output_0", "Unsqueeze9778", "Unsqueeze9781", "StridedSlice9784", "Squeeze9785", "BinaryOp9786", "BinaryOp9787", "/Gather_394_output_0", "/Unsqueeze_833_output_0", "/Concat_338_output_0", "/Reshape_235_output_0", "/Mul_208_output_0", "/Shape_366_output_0", "Shape9888", "BinaryOp9891", "BinaryOp9892", "Unsqueeze9893", "BinaryOp9895", "Unsqueeze9896", "StridedSlice9899", "Squeeze9900", "BinaryOp9901", "BinaryOp9902", "/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", "/Mul_210_output_0", "/Shape_367_output_0", "Shape9794", "BinaryOp9797", "BinaryOp9798", "Unsqueeze9799", "BinaryOp9801", "Unsqueeze9802", "StridedSlice9805", "Squeeze9806", "BinaryOp9807", "BinaryOp9808", "/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", "Shape10127", "BinaryOp10130", "BinaryOp10131", "Unsqueeze10132", "BinaryOp10134", "Unsqueeze10135", "StridedSlice10138", "BinaryOp10140", "BinaryOp10141", "/Gather_408_output_0", "/Unsqueeze_864_output_0", "Unsqueeze10149", "Unsqueeze10152", "StridedSlice10155", "Squeeze10156", "BinaryOp10157", "BinaryOp10158", "/Gather_409_output_0", "/Unsqueeze_865_output_0", "/Concat_351_output_0", "/Reshape_244_output_0", "/Mul_216_output_0", "/Shape_380_output_0", "Shape10259", "BinaryOp10262", "BinaryOp10263", "Unsqueeze10264", "BinaryOp10266", "Unsqueeze10267", "StridedSlice10270", "Squeeze10271", "BinaryOp10272", "BinaryOp10273", "/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", "/Mul_218_output_0", "/Shape_381_output_0", "Shape10165", "BinaryOp10168", "BinaryOp10169", "Unsqueeze10170", "BinaryOp10172", "Unsqueeze10173", "StridedSlice10176", "Squeeze10177", "BinaryOp10178", "BinaryOp10179", "/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", "Shape10498", "BinaryOp10501", "BinaryOp10502", "Unsqueeze10503", "BinaryOp10505", "Unsqueeze10506", "StridedSlice10509", "BinaryOp10511", "BinaryOp10512", "/Gather_423_output_0", "/Unsqueeze_896_output_0", "Unsqueeze10520", "Unsqueeze10523", "StridedSlice10526", "Squeeze10527", "BinaryOp10528", "BinaryOp10529", "/Gather_424_output_0", "/Unsqueeze_897_output_0", "/Concat_364_output_0", "/Reshape_253_output_0", "/Mul_224_output_0", "/Shape_394_output_0", "Shape10630", "BinaryOp10633", "BinaryOp10634", "Unsqueeze10635", "BinaryOp10637", "Unsqueeze10638", "StridedSlice10641", "Squeeze10642", "BinaryOp10643", "BinaryOp10644", "/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", "/Mul_226_output_0", "/Shape_395_output_0", "Shape10536", "BinaryOp10539", "BinaryOp10540", "Unsqueeze10541", "BinaryOp10543", "Unsqueeze10544", "StridedSlice10547", "Squeeze10548", "BinaryOp10549", "BinaryOp10550", "/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", "Shape10869", "BinaryOp10872", "BinaryOp10873", "Unsqueeze10874", "BinaryOp10876", "Unsqueeze10877", "StridedSlice10880", "BinaryOp10882", "BinaryOp10883", "/Gather_438_output_0", "/Unsqueeze_928_output_0", "Unsqueeze10891", "Unsqueeze10894", "StridedSlice10897", "Squeeze10898", "BinaryOp10899", "BinaryOp10900", "/Gather_439_output_0", "/Unsqueeze_929_output_0", "/Concat_377_output_0", "/Reshape_262_output_0", "/Mul_232_output_0", "/Shape_408_output_0", "Shape11001", "BinaryOp11004", "BinaryOp11005", "Unsqueeze11006", "BinaryOp11008", "Unsqueeze11009", "StridedSlice11012", "Squeeze11013", "BinaryOp11014", "BinaryOp11015", "/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", "/Mul_234_output_0", "/Shape_409_output_0", "Shape10907", "BinaryOp10910", "BinaryOp10911", "Unsqueeze10912", "BinaryOp10914", "Unsqueeze10915", "StridedSlice10918", "Squeeze10919", "BinaryOp10920", "BinaryOp10921", "/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", "Shape11240", "BinaryOp11243", "BinaryOp11244", "Unsqueeze11245", "BinaryOp11247", "Unsqueeze11248", "StridedSlice11251", "BinaryOp11253", "BinaryOp11254", "/Gather_453_output_0", "/Unsqueeze_960_output_0", "Unsqueeze11262", "Unsqueeze11265", "StridedSlice11268", "Squeeze11269", "BinaryOp11270", "BinaryOp11271", "/Gather_454_output_0", "/Unsqueeze_961_output_0", "/Concat_390_output_0", "/Reshape_271_output_0", "/Mul_240_output_0", "/Shape_422_output_0", "Shape11372", "BinaryOp11375", "BinaryOp11376", "Unsqueeze11377", "BinaryOp11379", "Unsqueeze11380", "StridedSlice11383", "Squeeze11384", "BinaryOp11385", "BinaryOp11386", "/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", "/Mul_242_output_0", "/Shape_423_output_0", "Shape11278", "BinaryOp11281", "BinaryOp11282", "Unsqueeze11283", "BinaryOp11285", "Unsqueeze11286", "StridedSlice11289", "Squeeze11290", "BinaryOp11291", "BinaryOp11292", "/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", "Shape11611", "BinaryOp11614", "BinaryOp11615", "Unsqueeze11616", "BinaryOp11618", "Unsqueeze11619", "StridedSlice11622", "BinaryOp11624", "BinaryOp11625", "/Gather_468_output_0", "/Unsqueeze_992_output_0", "Unsqueeze11633", "Unsqueeze11636", "StridedSlice11639", "Squeeze11640", "BinaryOp11641", "BinaryOp11642", "/Gather_469_output_0", "/Unsqueeze_993_output_0", "/Concat_403_output_0", "/Reshape_280_output_0", "/Mul_248_output_0", "/Shape_436_output_0", "Shape11743", "BinaryOp11746", "BinaryOp11747", "Unsqueeze11748", "BinaryOp11750", "Unsqueeze11751", "StridedSlice11754", "Squeeze11755", "BinaryOp11756", "BinaryOp11757", "/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", "/Mul_250_output_0", "/Shape_437_output_0", "Shape11649", "BinaryOp11652", "BinaryOp11653", "Unsqueeze11654", "BinaryOp11656", "Unsqueeze11657", "StridedSlice11660", "Squeeze11661", "BinaryOp11662", "BinaryOp11663", "/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", "Shape11982", "BinaryOp11985", "BinaryOp11986", "Unsqueeze11987", "BinaryOp11989", "Unsqueeze11990", "StridedSlice11993", "BinaryOp11995", "BinaryOp11996", "/Gather_483_output_0", "/Unsqueeze_1024_output_0", "Unsqueeze12004", "Unsqueeze12007", "StridedSlice12010", "Squeeze12011", "BinaryOp12012", "BinaryOp12013", "/Gather_484_output_0", "/Unsqueeze_1025_output_0", "/Concat_416_output_0", "/Reshape_289_output_0", "/Mul_256_output_0", "/Shape_450_output_0", "Shape12114", "BinaryOp12117", "BinaryOp12118", "Unsqueeze12119", "BinaryOp12121", "Unsqueeze12122", "StridedSlice12125", "Squeeze12126", "BinaryOp12127", "BinaryOp12128", "/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", "/Mul_258_output_0", "/Shape_451_output_0", "Shape12020", "BinaryOp12023", "BinaryOp12024", "Unsqueeze12025", "BinaryOp12027", "Unsqueeze12028", "StridedSlice12031", "Squeeze12032", "BinaryOp12033", "BinaryOp12034", "/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", "Shape12353", "BinaryOp12356", "BinaryOp12357", "Unsqueeze12358", "BinaryOp12360", "Unsqueeze12361", "StridedSlice12364", "BinaryOp12366", "BinaryOp12367", "/Gather_498_output_0", "/Unsqueeze_1056_output_0", "Unsqueeze12375", "Unsqueeze12378", "StridedSlice12381", "Squeeze12382", "BinaryOp12383", "BinaryOp12384", "/Gather_499_output_0", "/Unsqueeze_1057_output_0", "/Concat_429_output_0", "/Reshape_298_output_0", "/Mul_264_output_0", "/Shape_464_output_0", "Shape12485", "BinaryOp12488", "BinaryOp12489", "Unsqueeze12490", "BinaryOp12492", "Unsqueeze12493", "StridedSlice12496", "Squeeze12497", "BinaryOp12498", "BinaryOp12499", "/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", "/Mul_266_output_0", "/Shape_465_output_0", "Shape12391", "BinaryOp12394", "BinaryOp12395", "Unsqueeze12396", "BinaryOp12398", "Unsqueeze12399", "StridedSlice12402", "Squeeze12403", "BinaryOp12404", "BinaryOp12405", "/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", "Shape12724", "BinaryOp12727", "BinaryOp12728", "Unsqueeze12729", "BinaryOp12731", "Unsqueeze12732", "StridedSlice12735", "BinaryOp12737", "BinaryOp12738", "/Gather_513_output_0", "/Unsqueeze_1088_output_0", "Unsqueeze12746", "Unsqueeze12749", "StridedSlice12752", "Squeeze12753", "BinaryOp12754", "BinaryOp12755", "/Gather_514_output_0", "/Unsqueeze_1089_output_0", "/Concat_442_output_0", "/Reshape_307_output_0", "/Mul_272_output_0", "/Shape_478_output_0", "Shape12856", "BinaryOp12859", "BinaryOp12860", "Unsqueeze12861", "BinaryOp12863", "Unsqueeze12864", "StridedSlice12867", "Squeeze12868", "BinaryOp12869", "BinaryOp12870", "/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", "/Mul_274_output_0", "/Shape_479_output_0", "Shape12762", "BinaryOp12765", "BinaryOp12766", "Unsqueeze12767", "BinaryOp12769", "Unsqueeze12770", "StridedSlice12773", "Squeeze12774", "BinaryOp12775", "BinaryOp12776", "/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", "Shape13095", "BinaryOp13098", "BinaryOp13099", "Unsqueeze13100", "BinaryOp13102", "Unsqueeze13103", "StridedSlice13106", "BinaryOp13108", "BinaryOp13109", "/Gather_528_output_0", "/Unsqueeze_1120_output_0", "Unsqueeze13117", "Unsqueeze13120", "StridedSlice13123", "Squeeze13124", "BinaryOp13125", "BinaryOp13126", "/Gather_529_output_0", "/Unsqueeze_1121_output_0", "/Concat_455_output_0", "/Reshape_316_output_0", "/Mul_280_output_0", "/Shape_492_output_0", "Shape13195", "BinaryOp13198", "BinaryOp13199", "Unsqueeze13200", "BinaryOp13202", "Unsqueeze13203", "StridedSlice13206", "Squeeze13207", "BinaryOp13208", "BinaryOp13209", "/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", "/Mul_282_output_0", "/Shape_493_output_0", "Shape13133", "BinaryOp13136", "BinaryOp13137", "Unsqueeze13138", "BinaryOp13140", "Unsqueeze13141", "StridedSlice13144", "Squeeze13145", "BinaryOp13146", "BinaryOp13147", "/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", "logits_index", "/lm/Slice_output_0", "/lm/norm/Mul_1_output_0", "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": "88269394-9aef-4cf2-a5c8-ae90520bc008" }