The first is the installation process

Install Anaconda through the mirror source, tsinghua mirror source:

Mirrors.tuna.tsinghua.edu.cn/anaconda/ar…

Once Anaconda is installed, download the Pytorch version from the Pytorch website Here comes the first pothole, CUDA version selection

CUDA is compatible with nvidia’s NVIDIA card. If there is no NVIDIA card, select None. If there is no NVIDIA card, check the VERSION of CUDA in the CONTROL panel of the NVIDIA cardControl Panel — > Help — > System Information — > Components. In my case, 10.1 CUDA

The largest crater

Many tutorials do not specify the official website instructions to download from external sources by default, even if you have added tsinghua source **

The -c pytorch in the command can be removed to download through the domestic image

There are pit

CondaHTTPError:HTTP 000 CONNECTION FAILED**

I was prompted to retry the Anaconda Prompt, but it failed to be tried several times. The Anaconda Prompt kept running too fast and eventually filled up the memory and exploded. After a long time, the problem was not solved, and the Anaconda Prompt was reinstalled angrily

Conda clean-y -all // Delete all installation packages and cache

After the reload, I still encountered this problem and would still get stuck. I changed the source for several times, but the result was similar. Finally, I tried again for several times

It should be the network fluctuation and the network problem of the server on the mirror side, once the network is down, it will not continue to go down

So here’s the mirror image that we’re going to use

channels:
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main

ssl_verify: true
show_channel_urls: true
report_errors: true
Copy the code

The last pit

The import torch is incorrectly reported because the numpy and PIP versions are too old

python -m pip install --upgrade pip

pip install --upgrade numpy

Update it and you can solve it