Ubuntu 22.04配置下载apex

我来踩坑啦，最近在下载apex包，遇到的问题步骤记录一下

先尝试按照官网下载配置，不要直接pip install apex, 先从github上下载包，本地安装

git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

接下来，我们可能会报错 No module named 'packaging'

 ModuleNotFoundError: No module named 'packaging'
  error: subprocess-exited-with-error
 
  × Getting requirements to build wheel did not run successfully.
  │ exit code: 1
  ╰─> See above for output.
 
  note: This error originates from a subprocess, and is likely not a problem with pip.
  full command: 'D:\SOFTWARE\Anaconda\Anaconda3\envs\Dunit\python.exe' 'D:\SOFTWARE\Anaconda\Anaconda3\envs\Dunit\lib\site-packages\pip\_vendor\pep517\in_process\_in_process.py' get_requires_for_build_wheel 'C:\Users\admin\AppData\Local\Temp\tmpsb751068'
  cwd: C:\Users\admin\apex
  Getting requirements to build wheel ... error
error: subprocess-exited-with-error
 
× Getting requirements to build wheel did not run successfully.
│ exit code: 1
╰─> See above for output.
 
note: This error originates from a subprocess, and is likely not a problem with pip.

先安装 packaging包，conda install packaging

然后在试运行以下命令行，参考博客

cd apex
pip install -v --no-cache-dir --no-build-isolation --global-option="--cpp_ext" --global-option="--cuda_ext" --global-option="--deprecated_fused_adam" --global-option="--xentropy" --global-option="--fast_multihead_attn" ./

如果可以显示成功就完成，但是笔者还是遇到啦其他问题unsupported operand type(s) for +: 'NoneType' and 'str'

# error
Traceback (most recent call last):
File "setup.py", line 35, in <module>
_, bare_metal_major, _ = get_cuda_bare_metal_version(CUDA_HOME)
File "setup.py", line 14, in get_cuda_bare_metal_version
raw_output = subprocess.check_output([cuda_dir + "/bin/nvcc", "-V"], universal_newlines=True)
TypeError: unsupported operand type(s) for +: 'NoneType' and 'str'

是因为conda安装的cuda是不全的，没有nvcc，解决方案conda install -c nvidia cuda-nvcc，安装成功！！ 参考博客

最后还是运行，就可以啦

pip install -v --no-cache-dir --no-build-isolation --global-option="--cpp_ext" --global-option="--cuda_ext" --global-option="--deprecated_fused_adam" --global-option="--xentropy" --global-option="--fast_multihead_attn" ./

补充小踩坑，之前我试过原本的安装会失败，出现报错

Attr ibuteError: module ' torch.distributed' has no attribute '_ reduce_ scatter_base‘
或者是
AttributeError: module 'torch.distributed' has no attribute '_all_gather_base' 

我后面的解决方案是，下载22.04dev版本，这个参考博客可以提供解决

如果以上问题不能解决，还是出现AttributeError: module ‘torch.distributed‘ has no attribute ‘_all_gather_base‘
请参考以下博客代码，配置vi ~/.bashrc可行