欢迎大家关注笔者，你的关注是我持续更博的最大动力

原创文章，转载告知，盗版必究

---

本人环境声明：

• 系统环境：Ubuntu18.04.1
• cuda版本：10.2.89
• cudnn版本：7.6.5
• torch：1.6.0
• torchvision：0.7.0

---

1 错误原因分析

yolov5的3.0版本代码在训练的时候报错：ImportError: cannot import name 'amp' from 'torch.cuda'

(yolov5) shl@zfcv:~/project/yolov5_v3_0820$ ./4_clothes_shoes_hat_v1_train
Traceback (most recent call last):
  File "train.py", line 16, in <module>
    from torch.cuda import amp
ImportError: cannot import name 'amp' from 'torch.cuda' (/home/shl/anaconda3/envs/yolov5/lib/python3.7/site-packages/torch/cuda/__init__.py)

2 错误解决方法一

然后我就开始尝试解决这个问题，在网上看到issues的解决方法，这可能可cuda和pytorch的版本问题，具体解决方法如下：

1、查看你的cuda和pytorch的版本

nvcc --version 或 cat usr/local/cuda-[version]

2、如果你的cuda是10.0，请在这里检查你的pytorch的对应版本

3、重新安装cuda10.0对应的pytorch

pip install torch==1.4+cu100 torchvision==0.5.0+cu100 -f https://download.pytorch.org/whl/torch_stable.html

我的pytorch安装的就是torch==1.4+cu100 torchvision==0.5.0+cu100，但是我的cuda是10.2版本，但是我并不像改变这个版本！

3 错误解决方法二

之前在yolov5的v2版本中，train.py的apex使用代码如下：

(yolov5) shl@zfcv:~/project/yolov5_v3_0820$ cat ~/shl/yolov5/train.py -n
     1  import argparse
     2
     3  import torch.distributed as dist
     4  import torch.nn.functional as F
     5  import torch.optim as optim
     6  import torch.optim.lr_scheduler as lr_scheduler
     7  import yaml
     8  from torch.utils.tensorboard import SummaryWriter
     9
    10  import test  # import test.py to get mAP after each epoch
    11  from models.yolo import Model
    12  from utils.datasets import *
    13  from utils.utils import *
    14
    15  mixed_precision = True
    16  try:  # Mixed precision training https://github.com/NVIDIA/apex
    17      from apex import amp
    18  except:
    19      print('Apex recommended for faster mixed precision training: https://github.com/NVIDIA/apex')
    20      mixed_precision = False  # not installed

于是我把train.py的from torch.cuda import amp修改为：from apex import amp

1、首先安装amp，如果没有安装会提示库包不存在

(yolov5) shl@zfcv:~/project/yolov5_v3_0820$ ./4_clothes_shoes_hat_v1_train
Traceback (most recent call last):
  File "train.py", line 18, in <module>
    from apex import amp
ModuleNotFoundError: No module named 'apex'

2、安装好之后又再执行训练，报错如下

报错：TypeError: Class advice impossible in Python3. Use the @implementer class decorator instead.

(yolov5) shl@zfcv:~/project/yolov5_v3_0820$ ./4_clothes_shoes_hat_v1_train
Traceback (most recent call last):
  File "train.py", line 18, in <module>
    from apex import amp
  File "/home/shl/anaconda3/envs/yolov5/lib/python3.7/site-packages/apex/__init__.py", line 18, in <module>
    from apex.interfaces import (ApexImplementation,
  File "/home/shl/anaconda3/envs/yolov5/lib/python3.7/site-packages/apex/interfaces.py", line 10, in <module>
    class ApexImplementation(object):
  File "/home/shl/anaconda3/envs/yolov5/lib/python3.7/site-packages/apex/interfaces.py", line 14, in ApexImplementation
    implements(IApex)
  File "/home/shl/anaconda3/envs/yolov5/lib/python3.7/site-packages/zope/interface/declarations.py", line 706, in implements
    raise TypeError(_ADVICE_ERROR % 'implementer')
TypeError: Class advice impossible in Python3.  Use the @implementer class decorator instead.
(yolov5) shl@zfcv:~/project/yolov5_v3_0820$

3、解决错误（参考）：TypeError: Class advice impossible in Python3. Use the @implementer class decorator instead.

从apx的github主页的源码进行安装，如下：

git clone https://github.com.cnpmjs.org/NVIDIA/apex.git
cd apex
python setup.py install

然后错误完美解决，但是，紧接着又报错：AttributeError: module 'torch.nn' has no attribute 'Hardswish'

nn.Hardswish这个激活函数应该是torch1.6中刚有的，yolov5官方也说了，在yolov5的3.0版本要保证torch>=1.6，我C C 艹 ！

无奈，我只能重新建一个虚拟环境，然后乖乖安装torch1.6版本了！

#4 最终解决方法

1、新建虚拟环境

conda create -n yolov5-v3 python=3.7

2、激活虚拟环境

conda activate yolov5-v3

3、下载安装pytorch1.6

先去Pytorch官网查看pytorch1.6对应依赖的cuda版本

使用如下的下载命令：

pip install torch===1.6.0 torchvision===0.7.0 -f https://download.pytorch.org/whl/torch_stable.html

4、安装yolov5-v3的依赖库包

pip install -r requirements.txt

torch.nn.modules.module.ModuleAttributeError: ‘BatchNorm2d’ object has no attribute ‘_non_persistent_buffers_set’

查看官网issues，其中说的解决方法是更改torch的版本

https://download.pytorch.org/whl/torch_stable.html

然后我安装了torch1.5.1之后，开始训练时又报错：AttributeError: module 'torch.nn' has no attribute 'Hardswish'这TMD是死循环吗，我决定放弃了！

4 最终错误解决方法

还好没放弃，终于等到你！！！

具体参考这篇博客，我就不在这里在罗列了！

• 博客地址：https://shliang.blog.csdn.net/article/details/108219810

---

---

---

---

♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠