构建python397环境，安装GPU版本 torch

在独立的python环境安装，报错，无法解决

[4885@test ~]$ tar -zxvf Python-3.9.7.tgz
[4885@test ~]$ cd Python-3.9.7/
[4885@test Python-3.9.7]$ ./configure --prefix=/home/4885/python397
[4885@test Python-3.9.7]$ make&&make install
[4885@test Python-3.9.7]$ cd ../python397/bin/

[4885@test torch]$ pwd
/home/4885/python397/bin/torch
[4885@test torch]$ ll
total 2051140
-rw-rw-r--. 1 4885 4885   14886464 Nov  6 11:59 numpy-1.19.5-cp39-cp39-manylinux2010_x86_64.whl
-rw-rw-r--. 1 4885 4885   18186982 Nov  6 12:08 numpy-1.26.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
-rw-rw-r--. 1 4885 4885    3500040 Nov  6 11:59 Pillow-10.1.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
-rw-rw-r--. 1 4885 4885 2041325150 Nov  6 12:09 torch-1.9.1+cu111-cp39-cp39-linux_x86_64.whl
-rw-rw-r--. 1 4885 4885    1899818 Nov  6 12:09 torchaudio-0.9.1-cp39-cp39-manylinux1_x86_64.whl
-rw-rw-r--. 1 4885 4885   20521179 Nov  6 12:09 torchvision-0.10.1+cu111-cp39-cp39-linux_x86_64.whl
-rw-rw-r--. 1 4885 4885      31584 Nov  6 11:59 typing_extensions-4.8.0-py3-none-any.whl

[4885@test torch]$ ../pip install torch*
Processing ./torch-1.9.1+cu111-cp39-cp39-linux_x86_64.whl
Processing ./torchaudio-0.9.1-cp39-cp39-manylinux1_x86_64.whl
Processing ./torchvision-0.10.1+cu111-cp39-cp39-linux_x86_64.whl
Requirement already satisfied: typing-extensions in /home/4885/python397/lib/python3.9/site-packages (from torch==1.9.1+cu111) (4.8.0)
Requirement already satisfied: numpy in /home/4885/python397/lib/python3.9/site-packages (from torchvision==0.10.1+cu111) (1.19.5)
Requirement already satisfied: pillow>=5.3.0 in /home/4885/python397/lib/python3.9/site-packages (from torchvision==0.10.1+cu111) (10.1.0)
Installing collected packages: torch, torchvision, torchaudio
ERROR: Exception:
Traceback (most recent call last):
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/cli/base_command.py", line 180, in exc_logging_wrapper
    status = run_func(*args)
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/cli/req_command.py", line 245, in wrapper
    return func(self, options, args)
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/commands/install.py", line 452, in run
    installed = install_given_reqs(
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/req/__init__.py", line 72, in install_given_reqs
    requirement.install(
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/req/req_install.py", line 847, in install
    install_wheel(
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/operations/install/wheel.py", line 731, in install_wheel
    _install_wheel(
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/operations/install/wheel.py", line 591, in _install_wheel
    file.save()
  File "/home/4885/python397/lib/python3.9/site-packages/pip/_internal/operations/install/wheel.py", line 390, in save
    shutil.copyfileobj(f, dest)
  File "/home/4885/python397/lib/python3.9/shutil.py", line 205, in copyfileobj
    buf = fsrc_read(length)
  File "/home/4885/python397/lib/python3.9/zipfile.py", line 922, in read
    data = self._read1(n)
  File "/home/4885/python397/lib/python3.9/zipfile.py", line 998, in _read1
    data = self._decompressor.decompress(data, n)
zlib.error: Error -3 while decompressing data: invalid block type

解压缩数据时出错-3:无效的块类型

排错链接
https://github.com/PaddlePaddle/Paddle/issues/22749

1、构建与内网conda虚拟环境版本一致的环境

(base) [conda@localhost ~]$ conda create -n python39_gpu python=3.9.7
(base) [conda@localhost ~]$ conda env list
(base) [conda@localhost ~]$ conda activate python39_gpu
(python39_gpu) [conda@localhost ~]$ conda list
(python39_gpu) [conda@localhost ~]$ pip list

安装命令参考官网 https://pytorch.org/get-started/previous-versions/

(python39_gpu) [conda@localhost pkgs]$ pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
Looking in links: https://download.pytorch.org/whl/torch_stable.html
Collecting torch==1.9.1+cu111
  Downloading https://download.pytorch.org/whl/cu111/torch-1.9.1%2Bcu111-cp39-cp39-linux_x86_64.whl (2041.3 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.0/2.0 GB 358.2 kB/s eta 0:00:00
Collecting torchvision==0.10.1+cu111
  Downloading https://download.pytorch.org/whl/cu111/torchvision-0.10.1%2Bcu111-cp39-cp39-linux_x86_64.whl (20.5 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 20.5/20.5 MB 5.1 MB/s eta 0:00:00
Collecting torchaudio==0.9.1
  Downloading torchaudio-0.9.1-cp39-cp39-manylinux1_x86_64.whl (1.9 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.9/1.9 MB 33.4 kB/s eta 0:00:00
Collecting typing-extensions (from torch==1.9.1+cu111)
  Downloading typing_extensions-4.8.0-py3-none-any.whl.metadata (3.0 kB)
Collecting numpy (from torchvision==0.10.1+cu111)
  Downloading numpy-1.26.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (61 kB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 61.2/61.2 kB 29.0 kB/s eta 0:00:00
Collecting pillow>=5.3.0 (from torchvision==0.10.1+cu111)
  Downloading Pillow-10.1.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (9.5 kB)
Downloading Pillow-10.1.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.5 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.5/3.5 MB 28.7 kB/s eta 0:00:00
Downloading numpy-1.26.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (18.2 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 18.2/18.2 MB 41.6 kB/s eta 0:00:00
Downloading typing_extensions-4.8.0-py3-none-any.whl (31 kB)
Installing collected packages: typing-extensions, pillow, numpy, torch, torchvision, torchaudio
Successfully installed numpy-1.26.1 pillow-10.1.0 torch-1.9.1+cu111 torchaudio-0.9.1 torchvision-0.10.1+cu111 typing-extensions-4.8.0

2、安装低版本pandas
pip install numpy==1.19.5 -i https://pypi.tuna.tsinghua.edu.cn/simple

3、因为内网conda虚拟环境，pandas版本是 1.19.5，为了避免升级pandas，安装稍微低一点的版本 torchmetrics==0.11.4
查看历史版本 https://pypi.org/project/torchmetrics/#history
pip install torchmetrics==0.11.4

4、下载torchmetrics
pip download torchmetrics==0.11.4 --no-deps

5、上传到内网机器，进行安装

其他：
pip install torchmetrics==1.0.0

包依赖要求：
numpy>1.20.0
torch>=1.8.1
lightning-utilities>=0.7.0