NPU运行ConvNeXt多标签分类模型输出结果错误

[happyme531]

RKNN-Toolkit2版本: 2.3.0

模型: https://huggingface.co/SmilingWolf/wd-convnext-tagger-v3/tree/main

转换脚本:

#!/usr/bin/env python
# coding: utf-8

import datetime
from rknn.api import RKNN
from sys import exit

ONNX_MODEL="model.onnx"
RKNN_MODEL=ONNX_MODEL.replace(".onnx",".rknn")
DATASET="/home/zt/rk3588-nn/rknn_model_zoo/datasets/COCO/coco_subset_20.txt"
QUANTIZE=False
detailed_performance_log = True

timedate_iso = datetime.datetime.now().isoformat()

rknn = RKNN(verbose=True)
rknn.config(
    # mean_values=[x * 255 for x in [0.485, 0.456, 0.406]],
    # std_values=[x * 255 for x in [0.229, 0.224, 0.225]],
    quantized_dtype='w8a8',
    quantized_algorithm='normal',
    quantized_method='channel',
    quantized_hybrid_level=0,
    target_platform='rk3588',
    quant_img_RGB2BGR = False,
    float_dtype='float16',
    optimization_level=3,
    custom_string=f"converted at {timedate_iso}",
    remove_weight=False,
    compress_weight=False,
    inputs_yuv_fmt=None,
    single_core_mode=False,
    dynamic_input=None,
    model_pruning=False,
    op_target=None,
    quantize_weight=False,
    remove_reshape=False,
    sparse_infer=False,
    enable_flash_attention=False,
    )

ret = rknn.load_onnx(model=ONNX_MODEL, inputs=["input"], input_size_list=[[1,448,448,3]])
ret = rknn.build(do_quantization=QUANTIZE, dataset=DATASET, rknn_batch_size=None)
ret = rknn.export_rknn(RKNN_MODEL)

# ret = rknn.init_runtime(target='rk3588',device_id='cbb956772bf5dac9',core_mask=RKNN.NPU_CORE_0,perf_debug=detailed_performance_log)
# rknn.eval_perf()
ret = rknn.accuracy_analysis(inputs=['img.npy'], target='rk3588')

NPU运行代码:

import os
import sys
import cv2
import numpy as np
import pandas as pd
from rknnlite.api import RKNNLite

pd.set_option("display.max_rows", 1000)

dim = 448
thresh = 0.4

# 初始化RKNNLite
rknn_lite = RKNNLite(verbose=False)

# 加载RKNN模型
ret = rknn_lite.load_rknn("model.rknn")
if ret != 0:
    print('加载RKNN模型失败')
    exit(ret)

# 初始化运行时环境
ret = rknn_lite.init_runtime()
if ret != 0:
    print('初始化运行时环境失败')
    exit(ret)

label_names = pd.read_csv("selected_tags.csv")

target_img = "input.jpg" if len(sys.argv) < 2 else sys.argv[1]

try:
    # 图像预处理
    # 1. 读取图像并转RGB
    img = cv2.imread(target_img)
    if img is None:
        print(f"无法读取图像: {target_img}")
        exit(1)
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

    # 2. 处理透明通道
    if img.shape[2] == 4:  # RGBA图像
        img = img[:, :, :3]  # 只保留RGB通道
    elif len(img.shape) == 2:  # 灰度图
        img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)

    # 3. 填充为正方形
    h, w = img.shape[:2]
    if h > w:
        diff = h - w
        pad_left = diff // 2
        pad_right = diff - pad_left
        img = cv2.copyMakeBorder(img, 0, 0, pad_left, pad_right, 
                                cv2.BORDER_CONSTANT, value=(255, 255, 255))
    elif w > h:
        diff = w - h
        pad_top = diff // 2
        pad_bottom = diff - pad_top
        img = cv2.copyMakeBorder(img, pad_top, pad_bottom, 0, 0, 
                                cv2.BORDER_CONSTANT, value=(255, 255, 255))

    # 4. resize到目标尺寸
    img = cv2.resize(img, (dim, dim), interpolation=cv2.INTER_AREA)

    # 5. 转换为float32并添加batch维度
    img = img.astype(np.float32)
    img = np.expand_dims(img, 0)
    print(img.shape)
    # 执行推理
    np.save("img.npy",img)
    outputs = rknn_lite.inference(inputs=[img], data_format="nhwc")
    probs = outputs[0]  # 获取第一个输出
    print(probs.shape)

    # 后处理
    label_names["probs"] = probs[0]
    found_tags = label_names[label_names["probs"] > thresh][["tag_id", "name", "probs"]]
    print(found_tags)

finally:
    # 释放资源
    rknn_lite.release()

CPU运行代码:

import os
import sys
import cv2
import numpy as np
import pandas as pd
import onnxruntime as ort

pd.set_option("display.max_rows", 1000)

dim = 448
thresh = 0.4

# 加载模型
session = ort.InferenceSession("model.onnx")

label_names = pd.read_csv("selected_tags.csv")

target_img = "input.jpg" if len(sys.argv) < 2 else sys.argv[1]

try:
    # 图像预处理
    # 1. 读取图像并转RGB
    img = cv2.imread(target_img)
    if img is None:
        print(f"无法读取图像: {target_img}")
        exit(1)
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

    # 2. 处理透明通道
    if img.shape[2] == 4:  # RGBA图像
        img = img[:, :, :3]  # 只保留RGB通道
    elif len(img.shape) == 2:  # 灰度图
        img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)

    # 3. 填充为正方形
    h, w = img.shape[:2]
    if h > w:
        diff = h - w
        pad_left = diff // 2
        pad_right = diff - pad_left
        img = cv2.copyMakeBorder(img, 0, 0, pad_left, pad_right, 
                                cv2.BORDER_CONSTANT, value=(255, 255, 255))
    elif w > h:
        diff = w - h
        pad_top = diff // 2
        pad_bottom = diff - pad_top
        img = cv2.copyMakeBorder(img, pad_top, pad_bottom, 0, 0, 
                                cv2.BORDER_CONSTANT, value=(255, 255, 255))

    # 4. resize到目标尺寸
    img = cv2.resize(img, (dim, dim), interpolation=cv2.INTER_AREA)

    # 5. 转换为float32并添加batch维度
    img = img.astype(np.float32)
    img = np.expand_dims(img, 0)
    print(img.shape)
    # 执行推理
    input_name = session.get_inputs()[0].name
    label_name = session.get_outputs()[0].name
    probs = session.run([label_name], {input_name: img})[0]
    print(probs.shape)

    # 后处理
    label_names["probs"] = probs[0]
    found_tags = label_names[label_names["probs"] > thresh][["tag_id", "name", "probs"]]
    print(found_tags)

finally:
    # 释放资源
    pass

输入图片:

ONNX推理结果(正确):

       tag_id           name     probs
0     9999999        general  0.538390
1     9999998      sensitive  0.484545
5      212816           solo  0.932813
11      11906     open_mouth  0.405107
12      15080     short_hair  0.527191
22     383159   long_sleeves  0.523863
25     540830           1boy  0.945991
40      16613        jewelry  0.558022
47      15675       standing  0.452569
72    1300281     male_focus  0.913289
130     10926          pants  0.834378
230      3477        sweater  0.402990
346   1094664   colored_skin  0.603011
373      4009     turtleneck  0.544890
1532  1314823  black_sweater  0.499709

RKNN推理结果(错误):

       tag_id                  name     probs
1     9999998             sensitive  0.518066
4      470575                 1girl  0.444580
5      212816                  solo  0.435547
6       13197             long_hair  0.740723
19     566835        multiple_girls  0.547852
38       1821                2girls  0.459961
44       1709             twintails  0.514160
65      11429             pink_hair  0.695801
337      2508                 blood  0.424316
1403   720637  chromatic_aberration  0.526855
8110   385430          hatsune_miku  0.790039

精度分析结果:

# simulator_error: calculate the output error of each layer of the simulator (compared to the 'golden' value).
#              entire: output error of each layer between 'golden' and 'simulator', these errors will accumulate layer by layer.
#              single: single-layer output error between 'golden' and 'simulator', can better reflect the single-layer accuracy of the simulator.
# runtime_error: calculate the output error of each layer of the runtime.
#              entire: output error of each layer between 'golden' and 'runtime', these errors will accumulate layer by layer.
#              single_sim: single-layer output error between 'simulator' and 'runtime', can better reflect the single-layer accuracy of runtime.

layer_name                                                                                                 simulator_error                             runtime_error                      
                                                                                                       entire              single                  entire           single_sim            
                                                                                                    cos      euc        cos      euc            cos      euc        cos      euc          
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
[Input] input                                                                                     1.00000 | 0.0       1.00000 | 0.0           1.00000 | 0.0       1.00000 | 0.0           
[BatchNormalization] /Div_output_0_sw                                                             1.00000 | 0.0520    1.00000 | 0.0520        1.00000 | 0.0536    1.00000 | 0.0376        
[Transpose] /Sub_output_0                                                                         1.00000 | 0.0520    1.00000 | 0.0520        1.00000 | 0.0536    1.00000 | 0.0           
[Conv] /core_model/stem/stem.0/Conv_output_0                                                      1.00000 | 0.4337    1.00000 | 0.4337        1.00000 | 0.4339    1.00000 | 0.0022        
[exNorm] /core_model/stem/stem.1/Transpose_output_0_tp_rs_sw                                      1.00000 | 0.2652    1.00000 | 0.1549        1.00000 | 0.7212    1.00000 | 0.6860        
[Conv] /core_model/stages/stages.0/blocks/blocks.0/conv_dw/Conv_output_0                          1.00000 | 0.1781    1.00000 | 0.0713        1.00000 | 0.3850    1.00000 | 0.0008        
[exNorm] /core_model/stages/stages.0/blocks/blocks.0/Transpose_output_0_tp_rs_sw                  0.99999 | 8.7693    1.00000 | 0.6680        0.99997 | 15.510    1.00000 | 2.0370        
[Conv] /core_model/stages/stages.0/blocks/blocks.0/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 26.967    1.00000 | 2.6462        
[exGelu] /core_model/stages/stages.0/blocks/blocks.0/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 10.242    1.00000 | 0.8090        0.99997 | 26.137    1.00000 | 5.7634        
[Conv] /core_model/stages/stages.0/blocks/blocks.0/Mul                                            1.00000 | 5.8694    1.00000 | 1.2471        
[Add] /core_model/stages/stages.0/blocks/blocks.0/Add_tp                                          1.00000 | 6.0621    1.00000 | 1.6632        1.00000 | 15.546    1.00000 | 1.5519        
[Conv] /core_model/stages/stages.0/blocks/blocks.1/conv_dw/Conv_output_0                          1.00000 | 4.3364    1.00000 | 0.8735        0.99999 | 9.4880    1.00000 | 0.0104        
[exNorm] /core_model/stages/stages.0/blocks/blocks.1/Transpose_output_0_tp_rs_sw                  1.00000 | 2.0800    1.00000 | 0.3138        0.99999 | 3.7449    1.00000 | 0.7789        
[Conv] /core_model/stages/stages.0/blocks/blocks.1/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 8.2362    1.00000 | 1.5194        
[exGelu] /core_model/stages/stages.0/blocks/blocks.1/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 1.6299    1.00000 | 0.1053        0.99992 | 4.2616    0.99995 | 3.1520        
[Conv] /core_model/stages/stages.0/blocks/blocks.1/Mul                                            1.00000 | 13.150    1.00000 | 3.2147        
[Add] /core_model/stages/stages.0/blocks/blocks.1/Add_tp                                          1.00000 | 15.498    1.00000 | 4.7431        0.99999 | 97.716    1.00000 | 3.8536        
[Conv] /core_model/stages/stages.0/blocks/blocks.2/conv_dw/Conv_output_0                          1.00000 | 5.3657    1.00000 | 0.7520        0.99985 | 31.952    1.00000 | 0.0110        
[exNorm] /core_model/stages/stages.0/blocks/blocks.2/Transpose_output_0_tp_rs_sw                  0.99999 | 2.8247    1.00000 | 0.2477        0.99977 | 14.736    1.00000 | 0.6807        
[Conv] /core_model/stages/stages.0/blocks/blocks.2/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 9.1141    1.00000 | 1.1492        
[exGelu] /core_model/stages/stages.0/blocks/blocks.2/mlp/fc1/MatMul_output_0_sw_sw                0.99998 | 2.4339    1.00000 | 0.1316        0.99952 | 13.106    0.99999 | 2.2258        
[Conv] /core_model/stages/stages.0/blocks/blocks.2/Mul                                            1.00000 | 34.713    1.00000 | 3.6944        
[Add] /core_model/stages/stages.0/blocks/blocks.2/Add                                             1.00000 | 37.332    1.00000 | 7.3145        0.99996 | 265.54    1.00000 | 11.449        
[exNorm] /core_model/stages/stages.0/blocks/blocks.2/Add_tp_rs_sw                                 0.99999 | 0.8700    1.00000 | 0.0715        0.99970 | 4.3177    1.00000 | 0.2157        
[Conv] /core_model/stages/stages.1/downsample/downsample.1/Conv_output_0                          0.99999 | 2.7481    1.00000 | 0.2071        0.99985 | 13.890    1.00000 | 0.0075        
[Conv] /core_model/stages/stages.1/blocks/blocks.0/conv_dw/Conv_output_0                          0.99999 | 1.6479    1.00000 | 0.1787        0.99985 | 9.2222    1.00000 | 0.0039        
[exNorm] /core_model/stages/stages.1/blocks/blocks.0/Transpose_output_0_tp_rs_sw                  0.99999 | 1.1727    1.00000 | 0.1207        0.99984 | 5.6794    1.00000 | 0.3214        
[Conv] /core_model/stages/stages.1/blocks/blocks.0/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 5.6439    1.00000 | 0.7624        
[exGelu] /core_model/stages/stages.1/blocks/blocks.0/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 1.4437    1.00000 | 0.0936        0.99973 | 6.9007    0.99998 | 1.6650        
[Conv] /core_model/stages/stages.1/blocks/blocks.0/Mul                                            1.00000 | 6.6549    1.00000 | 2.0843        
[Add] /core_model/stages/stages.1/blocks/blocks.0/Add_output_0_tp_tp                              1.00000 | 7.6756    1.00000 | 2.8409        0.99999 | 76.236    1.00000 | 3.7374        
[Conv] /core_model/stages/stages.1/blocks/blocks.1/conv_dw/Conv_output_0                          0.99999 | 6.9869    1.00000 | 0.6112        0.99982 | 39.508    1.00000 | 0.0122        
[exNorm] /core_model/stages/stages.1/blocks/blocks.1/Transpose_output_0_tp_rs_sw                  0.99999 | 1.3834    1.00000 | 0.1432        0.99987 | 6.8398    1.00000 | 0.3849        
[Conv] /core_model/stages/stages.1/blocks/blocks.1/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 7.2545    1.00000 | 0.9486        
[exGelu] /core_model/stages/stages.1/blocks/blocks.1/mlp/fc1/MatMul_output_0_sw_sw                0.99998 | 1.5757    1.00000 | 0.0924        0.99961 | 7.9394    0.99998 | 1.9830        
[Conv] /core_model/stages/stages.1/blocks/blocks.1/Mul                                            1.00000 | 15.006    1.00000 | 1.9839        
[Add] /core_model/stages/stages.1/blocks/blocks.1/Add_output_0_tp_tp                              1.00000 | 16.957    1.00000 | 5.5882        0.99999 | 85.386    1.00000 | 6.4523        
[Conv] /core_model/stages/stages.1/blocks/blocks.2/conv_dw/Conv_output_0                          0.99999 | 9.6383    1.00000 | 2.5898        0.99982 | 50.067    1.00000 | 0.0136        
[exNorm] /core_model/stages/stages.1/blocks/blocks.2/Transpose_output_0_tp_rs_sw                  0.99999 | 1.5785    1.00000 | 0.1567        0.99985 | 8.0117    1.00000 | 0.4384        
[Conv] /core_model/stages/stages.1/blocks/blocks.2/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 8.2731    1.00000 | 1.1041        
[exGelu] /core_model/stages/stages.1/blocks/blocks.2/mlp/fc1/MatMul_output_0_sw_sw                0.99998 | 1.8875    1.00000 | 0.1062        0.99958 | 9.5626    0.99998 | 2.1530        
[Conv] /core_model/stages/stages.1/blocks/blocks.2/Mul                                            1.00000 | 50.664    1.00000 | 6.3097        
[Add] /core_model/stages/stages.1/blocks/blocks.2/Add_output_0                                    1.00000 | 57.117    1.00000 | 12.456        0.99998 | 371.59    1.00000 | 11.375        
[exNorm] /core_model/stages/stages.1/blocks/blocks.2/Add_output_0_tp_rs_sw                        0.99999 | 0.0565    1.00000 | 0.0050        0.99974 | 0.2851    1.00000 | 0.0147        
[Conv] /core_model/stages/stages.2/downsample/downsample.1/Conv_output_0                          0.99999 | 0.2799    1.00000 | 0.0221        0.99987 | 1.5164    1.00000 | 0.0009        
[Conv] /core_model/stages/stages.2/blocks/blocks.0/conv_dw/Conv_output_0                          1.00000 | 0.2692    1.00000 | 0.0388        0.99994 | 1.6870    1.00000 | 0.0007        
[exNorm] /core_model/stages/stages.2/blocks/blocks.0/Transpose_output_0_tp_rs_sw                  1.00000 | 0.4613    1.00000 | 0.0797        0.99993 | 2.3538    1.00000 | 0.2400        
[Conv] /core_model/stages/stages.2/blocks/blocks.0/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 3.2392    1.00000 | 0.5924        
[exGelu] /core_model/stages/stages.2/blocks/blocks.0/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.8842    1.00000 | 0.0693        0.99977 | 4.8253    0.99999 | 1.0671        
[Conv] /core_model/stages/stages.2/blocks/blocks.0/Mul                                            1.00000 | 0.2157    1.00000 | 0.0226        
[Add] /core_model/stages/stages.2/blocks/blocks.0/Add_output_0_tp_tp                              1.00000 | 0.3688    1.00000 | 0.0374        0.99990 | 1.9785    1.00000 | 0.0385        
[Conv] /core_model/stages/stages.2/blocks/blocks.1/conv_dw/Conv_output_0                          1.00000 | 0.4518    1.00000 | 0.0458        0.99990 | 2.3457    1.00000 | 0.0010        
[exNorm] /core_model/stages/stages.2/blocks/blocks.1/Transpose_output_0_tp_rs_sw                  1.00000 | 0.4876    1.00000 | 0.0757        0.99992 | 2.5248    1.00000 | 0.2189        
[Conv] /core_model/stages/stages.2/blocks/blocks.1/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 3.4631    1.00000 | 0.7081        
[exGelu] /core_model/stages/stages.2/blocks/blocks.1/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.8244    1.00000 | 0.0616        0.99974 | 4.3593    0.99998 | 1.3470        
[Conv] /core_model/stages/stages.2/blocks/blocks.1/Mul                                            0.99999 | 0.2148    1.00000 | 0.0179        
[Add] /core_model/stages/stages.2/blocks/blocks.1/Add_output_0_tp_tp                              1.00000 | 0.4318    1.00000 | 0.0482        0.99991 | 2.3433    1.00000 | 0.0416        
[Conv] /core_model/stages/stages.2/blocks/blocks.2/conv_dw/Conv_output_0                          1.00000 | 0.3404    1.00000 | 0.0467        0.99994 | 1.8587    1.00000 | 0.0005        
[exNorm] /core_model/stages/stages.2/blocks/blocks.2/Transpose_output_0_tp_rs_sw                  1.00000 | 0.5217    1.00000 | 0.0643        0.99990 | 2.6697    1.00000 | 0.2002        
[Conv] /core_model/stages/stages.2/blocks/blocks.2/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 3.6632    1.00000 | 0.5986        
[exGelu] /core_model/stages/stages.2/blocks/blocks.2/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.9429    1.00000 | 0.0682        0.99974 | 4.9279    0.99998 | 1.2612        
[Conv] /core_model/stages/stages.2/blocks/blocks.2/Mul                                            0.99999 | 0.3791    1.00000 | 0.0309        
[Add] /core_model/stages/stages.2/blocks/blocks.2/Add_output_0_tp_tp                              1.00000 | 0.5892    1.00000 | 0.0684        0.99991 | 3.1163    1.00000 | 0.0671        
[Conv] /core_model/stages/stages.2/blocks/blocks.3/conv_dw/Conv_output_0                          1.00000 | 0.6233    1.00000 | 0.0957        0.99995 | 3.7812    1.00000 | 0.0016        
[exNorm] /core_model/stages/stages.2/blocks/blocks.3/Transpose_output_0_tp_rs_sw                  1.00000 | 0.4507    1.00000 | 0.0628        0.99989 | 2.4686    1.00000 | 0.1678        
[Conv] /core_model/stages/stages.2/blocks/blocks.3/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 3.3606    1.00000 | 0.5857        
[exGelu] /core_model/stages/stages.2/blocks/blocks.3/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.8769    1.00000 | 0.0674        0.99974 | 4.8338    0.99998 | 1.3066        
[Conv] /core_model/stages/stages.2/blocks/blocks.3/Mul                                            0.99999 | 0.5393    1.00000 | 0.0443        
[Add] /core_model/stages/stages.2/blocks/blocks.3/Add_output_0_tp_tp                              1.00000 | 0.8402    1.00000 | 0.0915        0.99991 | 4.4720    1.00000 | 0.0875        
[Conv] /core_model/stages/stages.2/blocks/blocks.4/conv_dw/Conv_output_0                          1.00000 | 0.6862    1.00000 | 0.0912        0.99994 | 4.0351    1.00000 | 0.0015        
[exNorm] /core_model/stages/stages.2/blocks/blocks.4/Transpose_output_0_tp_rs_sw                  1.00000 | 0.5082    1.00000 | 0.0674        0.99989 | 2.6855    1.00000 | 0.1909        
[Conv] /core_model/stages/stages.2/blocks/blocks.4/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 3.7791    1.00000 | 0.6579        
[exGelu] /core_model/stages/stages.2/blocks/blocks.4/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 1.0072    1.00000 | 0.0678        0.99969 | 5.1696    0.99998 | 1.4190        
[Conv] /core_model/stages/stages.2/blocks/blocks.4/Mul                                            0.99999 | 0.8188    1.00000 | 0.0577        
[Add] /core_model/stages/stages.2/blocks/blocks.4/Add_output_0_tp_tp                              1.00000 | 1.2235    1.00000 | 0.1188        0.99989 | 6.3236    1.00000 | 0.1074        
[Conv] /core_model/stages/stages.2/blocks/blocks.5/conv_dw/Conv_output_0                          1.00000 | 0.9030    1.00000 | 0.1252        0.99994 | 5.2962    1.00000 | 0.0017        
[exNorm] /core_model/stages/stages.2/blocks/blocks.5/Transpose_output_0_tp_rs_sw                  1.00000 | 0.5777    1.00000 | 0.0730        0.99991 | 3.0166    1.00000 | 0.2249        
[Conv] /core_model/stages/stages.2/blocks/blocks.5/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 4.4417    1.00000 | 0.8116        
[exGelu] /core_model/stages/stages.2/blocks/blocks.5/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 1.0031    1.00000 | 0.0663        0.99967 | 5.2905    0.99997 | 1.6362        
[Conv] /core_model/stages/stages.2/blocks/blocks.5/Mul                                            0.99999 | 1.2782    1.00000 | 0.0937        
[Add] /core_model/stages/stages.2/blocks/blocks.5/Add_output_0_tp_tp                              0.99999 | 1.8889    1.00000 | 0.1682        0.99986 | 9.9420    1.00000 | 0.1609        
[Conv] /core_model/stages/stages.2/blocks/blocks.6/conv_dw/Conv_output_0                          1.00000 | 1.3570    1.00000 | 0.1637        0.99992 | 7.8443    1.00000 | 0.0083        
[exNorm] /core_model/stages/stages.2/blocks/blocks.6/Transpose_output_0_tp_rs_sw                  1.00000 | 0.6362    1.00000 | 0.0864        0.99991 | 3.3521    1.00000 | 0.2395        
[Conv] /core_model/stages/stages.2/blocks/blocks.6/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 4.8108    1.00000 | 0.8959        
[exGelu] /core_model/stages/stages.2/blocks/blocks.6/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 1.0179    1.00000 | 0.0621        0.99958 | 5.4758    0.99996 | 1.7060        
[Conv] /core_model/stages/stages.2/blocks/blocks.6/Mul                                            0.99999 | 1.6284    1.00000 | 0.1057        
[Add] /core_model/stages/stages.2/blocks/blocks.6/Add_output_0_tp_tp                              0.99999 | 2.6560    1.00000 | 0.2178        0.99983 | 13.954    1.00000 | 0.1870        
[Conv] /core_model/stages/stages.2/blocks/blocks.7/conv_dw/Conv_output_0                          1.00000 | 2.6708    1.00000 | 0.2898        0.99990 | 16.431    1.00000 | 0.0039        
[exNorm] /core_model/stages/stages.2/blocks/blocks.7/Transpose_output_0_tp_rs_sw                  1.00000 | 0.6435    1.00000 | 0.0796        0.99989 | 3.5828    1.00000 | 0.2167        
[Conv] /core_model/stages/stages.2/blocks/blocks.7/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 4.8424    1.00000 | 0.9562        
[exGelu] /core_model/stages/stages.2/blocks/blocks.7/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.9794    1.00000 | 0.0594        0.99958 | 5.3082    0.99995 | 1.7802        
[Conv] /core_model/stages/stages.2/blocks/blocks.7/Mul                                            0.99999 | 1.6477    1.00000 | 0.1123        
[Add] /core_model/stages/stages.2/blocks/blocks.7/Add_output_0_tp_tp                              0.99999 | 3.1878    1.00000 | 0.2573        0.99981 | 17.404    1.00000 | 0.2067        
[Conv] /core_model/stages/stages.2/blocks/blocks.8/conv_dw/Conv_output_0                          1.00000 | 3.3814    1.00000 | 0.3804        0.99989 | 22.417    1.00000 | 0.0175        
[exNorm] /core_model/stages/stages.2/blocks/blocks.8/Transpose_output_0_tp_rs_sw                  1.00000 | 0.6081    1.00000 | 0.0794        0.99987 | 3.6687    1.00000 | 0.1942        
[Conv] /core_model/stages/stages.2/blocks/blocks.8/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 4.6344    1.00000 | 0.9102        
[exGelu] /core_model/stages/stages.2/blocks/blocks.8/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 1.0736    1.00000 | 0.0658        0.99953 | 6.2395    0.99996 | 1.7144        
[Conv] /core_model/stages/stages.2/blocks/blocks.8/Mul                                            0.99999 | 1.9659    1.00000 | 0.1339        
[Add] /core_model/stages/stages.2/blocks/blocks.8/Add_output_0_tp_tp                              0.99999 | 3.8854    1.00000 | 0.3079        0.99980 | 21.835    1.00000 | 0.2582        
[Conv] /core_model/stages/stages.2/blocks/blocks.9/conv_dw/Conv_output_0                          1.00000 | 3.0943    1.00000 | 0.3607        0.99989 | 19.920    1.00000 | 0.0193        
[exNorm] /core_model/stages/stages.2/blocks/blocks.9/Transpose_output_0_tp_rs_sw                  1.00000 | 0.7018    1.00000 | 0.0905        0.99986 | 4.2144    1.00000 | 0.2403        
[Conv] /core_model/stages/stages.2/blocks/blocks.9/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 5.1151    1.00000 | 0.9788        
[exGelu] /core_model/stages/stages.2/blocks/blocks.9/mlp/fc1/MatMul_output_0_sw_sw                0.99998 | 1.0199    1.00000 | 0.0555        0.99944 | 5.7620    0.99994 | 1.8168        
[Conv] /core_model/stages/stages.2/blocks/blocks.9/Mul                                            0.99998 | 2.6935    1.00000 | 0.1629        
[Add] /core_model/stages/stages.2/blocks/blocks.9/Add_output_0_tp_tp                              0.99999 | 5.0023    1.00000 | 0.3772        0.99977 | 28.179    1.00000 | 0.3101        
[Conv] /core_model/stages/stages.2/blocks/blocks.10/conv_dw/Conv_output_0                         1.00000 | 8.7919    1.00000 | 0.7930        0.99985 | 55.061    1.00000 | 0.0208        
[exNorm] /core_model/stages/stages.2/blocks/blocks.10/Transpose_output_0_tp_rs_sw                 1.00000 | 0.5855    1.00000 | 0.0824        0.99988 | 3.8776    1.00000 | 0.2273        
[Conv] /core_model/stages/stages.2/blocks/blocks.10/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 4.3600    1.00000 | 0.9054        
[exGelu] /core_model/stages/stages.2/blocks/blocks.10/mlp/fc1/MatMul_output_0_sw_sw               0.99999 | 0.9389    1.00000 | 0.0630        0.99945 | 6.3456    0.99996 | 1.6959        
[Conv] /core_model/stages/stages.2/blocks/blocks.10/Mul                                           0.99999 | 2.0694    1.00000 | 0.1503        
[Add] /core_model/stages/stages.2/blocks/blocks.10/Add_output_0_tp_tp                             0.99999 | 5.4775    1.00000 | 0.4259        0.99976 | 31.675    1.00000 | 0.2968        
[Conv] /core_model/stages/stages.2/blocks/blocks.11/conv_dw/Conv_output_0                         1.00000 | 5.1577    1.00000 | 0.5321        0.99986 | 34.599    1.00000 | 0.0109        
[exNorm] /core_model/stages/stages.2/blocks/blocks.11/Transpose_output_0_tp_rs_sw                 1.00000 | 0.7717    1.00000 | 0.0947        0.99985 | 4.7655    1.00000 | 0.2408        
[Conv] /core_model/stages/stages.2/blocks/blocks.11/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 5.8779    1.00000 | 1.2118        
[exGelu] /core_model/stages/stages.2/blocks/blocks.11/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 0.7395    1.00000 | 0.0420        0.99929 | 4.7998    0.99991 | 1.6672        
[Conv] /core_model/stages/stages.2/blocks/blocks.11/Mul                                           0.99999 | 3.1887    1.00000 | 0.2001        
[Add] /core_model/stages/stages.2/blocks/blocks.11/Add_output_0_tp_tp                             0.99999 | 6.4642    1.00000 | 0.4905        0.99973 | 39.713    1.00000 | 0.3775        
[Conv] /core_model/stages/stages.2/blocks/blocks.12/conv_dw/Conv_output_0                         1.00000 | 6.2979    1.00000 | 0.7168        0.99986 | 48.280    1.00000 | 0.0260        
[exNorm] /core_model/stages/stages.2/blocks/blocks.12/Transpose_output_0_tp_rs_sw                 1.00000 | 0.6945    1.00000 | 0.0886        0.99984 | 4.5649    1.00000 | 0.2556        
[Conv] /core_model/stages/stages.2/blocks/blocks.12/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 5.3192    1.00000 | 1.0274        
[exGelu] /core_model/stages/stages.2/blocks/blocks.12/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 0.9953    1.00000 | 0.0576        0.99935 | 6.4515    0.99995 | 1.7743        
[Conv] /core_model/stages/stages.2/blocks/blocks.12/Mul                                           0.99999 | 4.1482    1.00000 | 0.2756        
[Add] /core_model/stages/stages.2/blocks/blocks.12/Add_output_0_tp_tp                             0.99999 | 7.9987    1.00000 | 0.6117        0.99971 | 50.162    1.00000 | 0.5109        
[Conv] /core_model/stages/stages.2/blocks/blocks.13/conv_dw/Conv_output_0                         1.00000 | 7.2780    1.00000 | 0.8903        0.99989 | 55.593    1.00000 | 0.0181        
[exNorm] /core_model/stages/stages.2/blocks/blocks.13/Transpose_output_0_tp_rs_sw                 1.00000 | 0.7272    1.00000 | 0.0883        0.99982 | 4.5997    1.00000 | 0.2679        
[Conv] /core_model/stages/stages.2/blocks/blocks.13/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 5.4471    1.00000 | 0.9776        
[exGelu] /core_model/stages/stages.2/blocks/blocks.13/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 1.1115    1.00000 | 0.0594        0.99929 | 6.9709    0.99995 | 1.8141        
[Conv] /core_model/stages/stages.2/blocks/blocks.13/Mul                                           0.99999 | 5.7149    1.00000 | 0.3629        
[Add] /core_model/stages/stages.2/blocks/blocks.13/Add_output_0_tp_tp                             0.99999 | 10.423    1.00000 | 0.7736        0.99969 | 66.497    1.00000 | 0.6889        
[Conv] /core_model/stages/stages.2/blocks/blocks.14/conv_dw/Conv_output_0                         1.00000 | 9.4366    1.00000 | 1.3369        0.99992 | 68.933    1.00000 | 0.1296        
[exNorm] /core_model/stages/stages.2/blocks/blocks.14/Transpose_output_0_tp_rs_sw                 0.99999 | 0.8293    1.00000 | 0.0848        0.99979 | 4.9586    1.00000 | 0.2607        
[Conv] /core_model/stages/stages.2/blocks/blocks.14/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 6.0089    1.00000 | 0.9250        
[exGelu] /core_model/stages/stages.2/blocks/blocks.14/mlp/fc1/MatMul_output_0_sw_sw               0.99997 | 1.5339    1.00000 | 0.0609        0.99913 | 8.2408    0.99996 | 1.7623        
[Conv] /core_model/stages/stages.2/blocks/blocks.14/Mul                                           0.99999 | 8.0104    1.00000 | 0.4646        
[Add] /core_model/stages/stages.2/blocks/blocks.14/Add_output_0_tp_tp                             0.99999 | 13.925    1.00000 | 0.9812        0.99969 | 85.462    1.00000 | 0.8454        
[Conv] /core_model/stages/stages.2/blocks/blocks.15/conv_dw/Conv_output_0                         1.00000 | 16.742    1.00000 | 1.7782        0.99987 | 125.06    1.00000 | 0.0242        
[exNorm] /core_model/stages/stages.2/blocks/blocks.15/Transpose_output_0_tp_rs_sw                 1.00000 | 0.7294    1.00000 | 0.0922        0.99982 | 4.4425    1.00000 | 0.2382        
[Conv] /core_model/stages/stages.2/blocks/blocks.15/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 5.4620    1.00000 | 0.8868        
[exGelu] /core_model/stages/stages.2/blocks/blocks.15/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 1.3080    1.00000 | 0.0646        0.99930 | 7.3326    0.99995 | 1.8906        
[Conv] /core_model/stages/stages.2/blocks/blocks.15/Mul                                           0.99999 | 8.5323    1.00000 | 0.5060        
[Add] /core_model/stages/stages.2/blocks/blocks.15/Add_output_0_tp_tp                             0.99999 | 16.900    1.00000 | 1.2867        0.99975 | 104.73    1.00000 | 1.1232        
[Conv] /core_model/stages/stages.2/blocks/blocks.16/conv_dw/Conv_output_0                         1.00000 | 17.464    1.00000 | 1.7820        0.99985 | 124.10    1.00000 | 0.0376        
[exNorm] /core_model/stages/stages.2/blocks/blocks.16/Transpose_output_0_tp_rs_sw                 0.99999 | 0.9050    1.00000 | 0.0971        0.99981 | 5.3032    1.00000 | 0.2699        
[Conv] /core_model/stages/stages.2/blocks/blocks.16/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 6.5533    1.00000 | 1.0245        
[exGelu] /core_model/stages/stages.2/blocks/blocks.16/mlp/fc1/MatMul_output_0_sw_sw               0.99997 | 1.1433    1.00000 | 0.0473        0.99896 | 6.4331    0.99990 | 1.9672        
[Conv] /core_model/stages/stages.2/blocks/blocks.16/Mul                                           0.99998 | 9.6403    1.00000 | 0.5067        
[Add] /core_model/stages/stages.2/blocks/blocks.16/Add_output_0_tp_tp                             0.99999 | 20.782    1.00000 | 1.5976        0.99976 | 131.81    1.00000 | 1.3443        
[Conv] /core_model/stages/stages.2/blocks/blocks.17/conv_dw/Conv_output_0                         1.00000 | 26.555    1.00000 | 2.4370        0.99982 | 191.69    1.00000 | 0.0619        
[exNorm] /core_model/stages/stages.2/blocks/blocks.17/Transpose_output_0_tp_rs_sw                 0.99999 | 1.0619    1.00000 | 0.1007        0.99978 | 5.8734    1.00000 | 0.2716        
[Conv] /core_model/stages/stages.2/blocks/blocks.17/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 8.3130    1.00000 | 1.1316        
[exGelu] /core_model/stages/stages.2/blocks/blocks.17/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 1.1258    1.00000 | 0.0518        0.99913 | 6.7021    0.99994 | 1.8047        
[Conv] /core_model/stages/stages.2/blocks/blocks.17/Mul                                           0.99999 | 9.2450    1.00000 | 0.5250        
[Add] /core_model/stages/stages.2/blocks/blocks.17/Add_output_0_tp_tp                             0.99999 | 24.116    1.00000 | 1.8857        0.99977 | 154.80    1.00000 | 1.4521        
[Conv] /core_model/stages/stages.2/blocks/blocks.18/conv_dw/Conv_output_0                         1.00000 | 35.221    1.00000 | 3.1955        0.99977 | 288.18    1.00000 | 0.1117        
[exNorm] /core_model/stages/stages.2/blocks/blocks.18/Transpose_output_0_tp_rs_sw                 0.99999 | 1.1296    1.00000 | 0.1223        0.99979 | 6.9383    1.00000 | 0.3205        
[Conv] /core_model/stages/stages.2/blocks/blocks.18/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 8.6446    1.00000 | 1.6140        
[exGelu] /core_model/stages/stages.2/blocks/blocks.18/mlp/fc1/MatMul_output_0_sw_sw               0.99997 | 0.2919    1.00000 | 0.0136        0.99751 | 2.5379    0.99955 | 1.0759        
[Conv] /core_model/stages/stages.2/blocks/blocks.18/Mul                                           0.99999 | 5.3299    1.00000 | 0.3148        
[Add] /core_model/stages/stages.2/blocks/blocks.18/Add_output_0_tp_tp                             0.99999 | 25.106    1.00000 | 2.1030        0.99975 | 191.56    1.00000 | 1.2286        
[Conv] /core_model/stages/stages.2/blocks/blocks.19/conv_dw/Conv_output_0                         1.00000 | 35.548    1.00000 | 3.3633        0.99976 | 312.59    1.00000 | 0.0670        
[exNorm] /core_model/stages/stages.2/blocks/blocks.19/Transpose_output_0_tp_rs_sw                 0.99999 | 1.2167    1.00000 | 0.1289        0.99975 | 7.5931    1.00000 | 0.3336        
[Conv] /core_model/stages/stages.2/blocks/blocks.19/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 9.9176    1.00000 | 1.4606        
[exGelu] /core_model/stages/stages.2/blocks/blocks.19/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 0.8619    1.00000 | 0.0388        0.99856 | 6.9386    0.99995 | 1.2801        
[Conv] /core_model/stages/stages.2/blocks/blocks.19/Mul                                           0.99999 | 9.1184    1.00000 | 0.5657        
[Add] /core_model/stages/stages.2/blocks/blocks.19/Add_output_0_tp_tp                             0.99999 | 27.806    1.00000 | 2.4645        0.99975 | 210.29    1.00000 | 1.6738        
[Conv] /core_model/stages/stages.2/blocks/blocks.20/conv_dw/Conv_output_0                         1.00000 | 39.498    1.00000 | 4.2072        0.99980 | 337.58    1.00000 | 0.0586        
[exNorm] /core_model/stages/stages.2/blocks/blocks.20/Transpose_output_0_tp_rs_sw                 0.99999 | 1.2402    1.00000 | 0.1320        0.99976 | 7.7935    1.00000 | 0.3367        
[Conv] /core_model/stages/stages.2/blocks/blocks.20/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 11.100    1.00000 | 1.5904        
[exGelu] /core_model/stages/stages.2/blocks/blocks.20/mlp/fc1/MatMul_output_0_sw_sw               0.99996 | 0.8821    1.00000 | 0.0327        0.99796 | 6.5160    0.99994 | 1.1170        
[Conv] /core_model/stages/stages.2/blocks/blocks.20/Mul                                           0.99999 | 15.295    1.00000 | 0.8990        
[Add] /core_model/stages/stages.2/blocks/blocks.20/Add_output_0_tp_tp                             1.00000 | 33.080    1.00000 | 3.0517        0.99980 | 231.84    1.00000 | 2.4472        
[Conv] /core_model/stages/stages.2/blocks/blocks.21/conv_dw/Conv_output_0                         1.00000 | 43.275    1.00000 | 4.7738        0.99982 | 354.33    1.00000 | 0.1131        
[exNorm] /core_model/stages/stages.2/blocks/blocks.21/Transpose_output_0_tp_rs_sw                 0.99999 | 1.1445    1.00000 | 0.1154        0.99977 | 6.9322    1.00000 | 0.3022        
[Conv] /core_model/stages/stages.2/blocks/blocks.21/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 9.9430    1.00000 | 1.3724        
[exGelu] /core_model/stages/stages.2/blocks/blocks.21/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 1.0118    1.00000 | 0.0509        0.99927 | 6.6085    0.99997 | 1.4451        
[Conv] /core_model/stages/stages.2/blocks/blocks.21/Mul                                           1.00000 | 70.174    1.00000 | 10.938        
[Add] /core_model/stages/stages.2/blocks/blocks.21/Add_output_0_tp_tp                             1.00000 | 80.642    1.00000 | 11.138        0.99996 | 433.11    1.00000 | 13.219        
[Conv] /core_model/stages/stages.2/blocks/blocks.22/conv_dw/Conv_output_0                         1.00000 | 40.203    1.00000 | 4.1400        0.99980 | 334.79    1.00000 | 0.0673        
[exNorm] /core_model/stages/stages.2/blocks/blocks.22/Transpose_output_0_tp_rs_sw                 0.99999 | 1.0973    1.00000 | 0.1141        0.99975 | 6.9852    1.00000 | 0.3096        
[Conv] /core_model/stages/stages.2/blocks/blocks.22/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 9.5003    1.00000 | 1.4080        
[exGelu] /core_model/stages/stages.2/blocks/blocks.22/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 0.9515    1.00000 | 0.0456        0.99927 | 5.9943    0.99996 | 1.4071        
[Conv] /core_model/stages/stages.2/blocks/blocks.22/Mul                                           1.00000 | 76.234    1.00000 | 12.482        
[Add] /core_model/stages/stages.2/blocks/blocks.22/Add_output_0_tp_tp                             1.00000 | 142.22    1.00000 | 24.626        0.99997 | 821.79    1.00000 | 27.483        
[Conv] /core_model/stages/stages.2/blocks/blocks.23/conv_dw/Conv_output_0                         1.00000 | 40.943    1.00000 | 4.5496        0.99980 | 297.74    1.00000 | 0.1471        
[exNorm] /core_model/stages/stages.2/blocks/blocks.23/Transpose_output_0_tp_rs_sw                 0.99999 | 0.9653    1.00000 | 0.1013        0.99975 | 6.0447    1.00000 | 0.2972        
[Conv] /core_model/stages/stages.2/blocks/blocks.23/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 8.0764    1.00000 | 1.1749        
[exGelu] /core_model/stages/stages.2/blocks/blocks.23/mlp/fc1/MatMul_output_0_sw_sw               0.99998 | 1.2145    1.00000 | 0.0520        0.99921 | 7.0187    0.99995 | 1.7389        
[Conv] /core_model/stages/stages.2/blocks/blocks.23/Mul                                           1.00000 | 86.669    1.00000 | 16.266        
[Add] /core_model/stages/stages.2/blocks/blocks.23/Add_output_0_tp_tp                             1.00000 | 210.87    1.00000 | 35.719        0.99997 | 1124.2    1.00000 | 35.783        
[Conv] /core_model/stages/stages.2/blocks/blocks.24/conv_dw/Conv_output_0                         1.00000 | 58.969    1.00000 | 6.2639        0.99981 | 394.87    1.00000 | 0.7103        
[exNorm] /core_model/stages/stages.2/blocks/blocks.24/Transpose_output_0_tp_rs_sw                 0.99999 | 0.8831    1.00000 | 0.0938        0.99977 | 5.7076    1.00000 | 0.2652        
[Conv] /core_model/stages/stages.2/blocks/blocks.24/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 7.6605    1.00000 | 1.2004        
[exGelu] /core_model/stages/stages.2/blocks/blocks.24/mlp/fc1/MatMul_output_0_sw_sw               0.99999 | 1.1554    1.00000 | 0.0605        0.99935 | 7.6976    0.99997 | 1.6360        
[Conv] /core_model/stages/stages.2/blocks/blocks.24/Mul                                           1.00000 | 114.05    1.00000 | 23.232        
[Add] /core_model/stages/stages.2/blocks/blocks.24/Add_output_0_tp_tp                             1.00000 | 292.12    1.00000 | 60.307        0.99998 | 1715.0    1.00000 | 71.288        
[Conv] /core_model/stages/stages.2/blocks/blocks.25/conv_dw/Conv_output_0                         1.00000 | 88.267    1.00000 | 10.777        0.99987 | 569.37    1.00000 | 1.0375        
[exNorm] /core_model/stages/stages.2/blocks/blocks.25/Transpose_output_0_tp_rs_sw                 0.99999 | 1.0343    1.00000 | 0.0916        0.99981 | 5.2693    1.00000 | 0.2724        
[Conv] /core_model/stages/stages.2/blocks/blocks.25/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 9.5084    1.00000 | 1.2177        
[exGelu] /core_model/stages/stages.2/blocks/blocks.25/mlp/fc1/MatMul_output_0_sw_sw               0.99997 | 1.5533    1.00000 | 0.0571        0.99936 | 7.3450    0.99997 | 1.5172        
[Conv] /core_model/stages/stages.2/blocks/blocks.25/Mul                                           1.00000 | 179.44    1.00000 | 23.912        
[Add] /core_model/stages/stages.2/blocks/blocks.25/Add_output_0_tp_tp                             1.00000 | 407.01    1.00000 | 86.773        0.99997 | 2365.5    1.00000 | 63.555        
[Conv] /core_model/stages/stages.2/blocks/blocks.26/conv_dw/Conv_output_0                         1.00000 | 139.21    1.00000 | 27.404        0.99994 | 785.43    1.00000 | 1.0042        
[exNorm] /core_model/stages/stages.2/blocks/blocks.26/Transpose_output_0_tp_rs_sw                 1.00000 | 0.8196    1.00000 | 0.1137        0.99987 | 5.2440    1.00000 | 0.3433        
[Conv] /core_model/stages/stages.2/blocks/blocks.26/norm/LayerNormalization_output_0_tp_rs_sw     1.00000 | 6.9916    1.00000 | 1.1787        
[exGelu] /core_model/stages/stages.2/blocks/blocks.26/mlp/fc1/MatMul_output_0_sw_sw               0.99999 | 1.2665    1.00000 | 0.0648        0.99940 | 8.2429    0.99998 | 1.6506        
[Conv] /core_model/stages/stages.2/blocks/blocks.26/Mul                                           1.00000 | 158.99    1.00000 | 33.201        
[Add] /core_model/stages/stages.2/blocks/blocks.26/Add_output_0                                   1.00000 | 460.17    1.00000 | 113.40        0.99998 | 2734.2    1.00000 | 112.83        
[exNorm] /core_model/stages/stages.2/blocks/blocks.26/Add_output_0_tp_rs_sw                       0.99999 | 0.1386    1.00000 | 0.0105        0.99921 | 1.0342    1.00000 | 0.0305        
[Conv] /core_model/stages/stages.3/downsample/downsample.1/Conv_output_0                          0.99999 | 0.8730    1.00000 | 0.0604        0.99959 | 6.3997    1.00000 | 0.0013        
[Conv] /core_model/stages/stages.3/blocks/blocks.0/conv_dw/Conv_output_0                          1.00000 | 1.3354    1.00000 | 0.1569        0.99984 | 11.542    1.00000 | 0.0036        
[exNorm] /core_model/stages/stages.3/blocks/blocks.0/Transpose_output_0_tp_rs_sw                  1.00000 | 0.1808    1.00000 | 0.0689        0.99999 | 1.2067    1.00000 | 0.1993        
[Conv] /core_model/stages/stages.3/blocks/blocks.0/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 2.2010    1.00000 | 1.4116        
[exGelu] /core_model/stages/stages.3/blocks/blocks.0/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.0119    1.00000 | 0.0023        0.99232 | 0.4313    0.99305 | 0.4120        
[Conv] /core_model/stages/stages.3/blocks/blocks.0/Mul_output_0                                   0.99999 | 0.1740    1.00000 | 0.0160        
[Add] /core_model/stages/stages.3/blocks/blocks.0/Add_output_0                                    0.99999 | 0.8897    1.00000 | 0.0663        0.99860 | 11.872    1.00000 | 0.0366        
[Conv] /core_model/stages/stages.3/blocks/blocks.1/conv_dw/Conv_output_0                          1.00000 | 1.2923    1.00000 | 0.1496        0.99933 | 22.159    1.00000 | 0.0019        
[exNorm] /core_model/stages/stages.3/blocks/blocks.1/Transpose_output_0_tp_rs_sw                  1.00000 | 0.2367    1.00000 | 0.1103        0.99989 | 4.7703    1.00000 | 0.2830        
[Conv] /core_model/stages/stages.3/blocks/blocks.1/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 2.8117    1.00000 | 1.7015        
[exGelu] /core_model/stages/stages.3/blocks/blocks.1/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.0049    1.00000 | 0.0009        0.98598 | 0.2878    0.99058 | 0.2126        
[Conv] /core_model/stages/stages.3/blocks/blocks.1/Mul_output_0                                   0.99999 | 0.1716    1.00000 | 0.0154        
[Add] /core_model/stages/stages.3/blocks/blocks.1/Add_output_0                                    0.99999 | 0.9110    1.00000 | 0.0685        0.99691 | 18.113    1.00000 | 0.0382        
[Conv] /core_model/stages/stages.3/blocks/blocks.2/conv_dw/Conv_output_0                          1.00000 | 1.3217    1.00000 | 0.1748        0.99856 | 36.460    1.00000 | 0.0030        
[exNorm] /core_model/stages/stages.3/blocks/blocks.2/Transpose_output_0_tp_rs_sw                  1.00000 | 0.2768    1.00000 | 0.0810        0.99964 | 7.3444    1.00000 | 0.2561        
[Conv] /core_model/stages/stages.3/blocks/blocks.2/norm/LayerNormalization_output_0_tp_rs_sw      1.00000 | 2.9884    1.00000 | 1.3662        
[exGelu] /core_model/stages/stages.3/blocks/blocks.2/mlp/fc1/MatMul_output_0_sw_sw                0.99999 | 0.0113    1.00000 | 0.0019        0.95814 | 0.9074    0.99286 | 0.3760        
[Conv] /core_model/stages/stages.3/blocks/blocks.2/Mul_output_0                                   0.99999 | 0.1632    1.00000 | 0.0135        
[Add] /core_model/stages/stages.3/blocks/blocks.2/Add_output_0                                    0.99999 | 0.9357    1.00000 | 0.0686        0.99487 | 24.547    1.00000 | 0.0388        
[Conv] /core_model/head/global_pool/pool/GlobalAveragePool_2conv0                                 1.00000 | 0.1376    1.00000 | 0.0183        0.99820 | 3.5928    1.00000 | 0.0002        
[Conv] /core_model/head/global_pool/pool/GlobalAveragePool_output_0                               1.00000 | 0.0141    1.00000 | 0.0034        0.99823 | 0.4880    1.00000 | 0.0           
[exNorm] /core_model/head/norm/LayerNormalization                                                 1.00000 | 0.0275    1.00000 | 0.0048        0.99748 | 0.9598    1.00000 | 0.0067        
[Conv] /core_model/head/fc/Gemm_output_0_mm                                                       1.00000 | 0.6485    1.00000 | 0.2296        
[Sigmoid] output-rs                                                                               1.00000 | 0.0037    1.00000 | 0.0005        0.99068 | 0.3092    0.99649 | 0.1729        
[Reshape] output                                                                                  1.00000 | 0.0037    1.00000 | 0.0004        0.99068 | 0.3092    1.00000 | 0.0           

[happyme531]

问题解决，toolkit2把这个nhwc输入的onnx模型维度变成nwch了😅应该算是bug。

[yuyun2000]

都2.3了还有这个问题啊