?一，搭建老版本的环境
首先我的环境是
?
环境，win10，64位，Anaconda4.10.3，Cuda 9.2.148，cudnn应该为v7.4.2.24
python 3.7
搭建基于pytorch的环境。
安装pytorch：
Previous PyTorch Versions | PyTorch
根据我cuda9.2版本找到对应的命令。Anaconda Prompt安装：
(python37) D:\anaconda\envs\python37>pip install torch==1.5.1+cu92 torchvision==0.6.1+cu92 -f https://download.pytorch.org/whl/torch_stable.html
Looking in links: https://download.pytorch.org/whl/torch_stable.html
Collecting torch==1.5.1+cu92
Downloading https://download.pytorch.org/whl/cu92/torch-1.5.1%2Bcu92-cp37-cp37m-win_amd64.whl (694.1 MB)
|████████████████████████████████| 694.1 MB 18 kB/s
Collecting torchvision==0.6.1+cu92
Downloading https://download.pytorch.org/whl/cu92/torchvision-0.6.1%2Bcu92-cp37-cp37m-win_amd64.whl (1.3 MB)
|████████████████████████████████| 1.3 MB 384 kB/s
Requirement already satisfied: future in d:\anaconda\envs\python37\lib\site-packages (from torch==1.5.1+cu92) (0.18.2)
Requirement already satisfied: numpy in d:\anaconda\envs\python37\lib\site-packages (from torch==1.5.1+cu92) (1.21.2)
Requirement already satisfied: pillow>=4.1.1 in d:\anaconda\envs\python37\lib\site-packages (from torchvision==0.6.1+cu92) (8.4.0)
Installing collected packages: torch, torchvision
Successfully installed torch-1.5.1+cu92 torchvision-0.6.1+cu92
(python37) D:\anaconda\envs\python37>conda list torch
# packages in environment at D:\anaconda\envs\python37:
#
# Name Version Build Channel
torch 1.5.1+cu92 pypi_0 pypi
torchvision 0.6.1+cu92 pypi_0 pypi
(python37) D:\anaconda\envs\python37>python
Python 3.7.10 (default, Feb 26 2021, 13:06:18) [MSC v.1916 64 bit (AMD64)] :: Anaconda, Inc. on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
True
>>>测试pytorch的link：以上程序中有测试的函数。
中间因为安装错误的pytorch版本，测试时有报错如下：
>>> # 使用GPU设备
>>> device = torch.device('cuda')
>>> # 将ab搬到GPU
>>> a = a.to(device)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "D:\anaconda\envs\python37\lib\site-packages\torch\cuda\__init__.py", line 166, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled
>>> b = b.to(device)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "D:\anaconda\envs\python37\lib\site-packages\torch\cuda\__init__.py", line 166, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled尝试几次安装都是报错，gpu不可用。于是卸载掉之前安装的pytorch，再次查找安装方法：手残患者如何安装gpu版本的torch - 知乎
先安装驱动，再nvidia版本号，再安装对应版本的pytorch
?甚至报错：
?
最终找到了对应的版本。官网安装指令：Previous PyTorch Versions | PyTorch
划重点：安装的错误版本要完全卸载掉。需要使用pip uninstall torch卸载。之后用conda list torch查看是否有残余。然后-》python-》import torch查看是否失败。失败证明pytorch卸载成功。卸载完后用pip安装（因为多次尝试conda安装，显示失败）。最终使用pip安装1.5.1版本的pytorch成功调用gpu。
尝试使用例子，使gpu可以正常工作。
安装驱动，测试pytorch：手残患者如何安装gpu版本的torch - 知乎
准备好环境后，进行实例运行，看视频：
pytorch 入门学习（目前见过最好的pytorch学习视频）_哔哩哔哩_bilibili
希望我能学习到一些东西。之后有结果再来更贴。