search

Thinklab-SJTU / ThinkMatch

A research protocol for deep graph matching.
https://thinkmatch.readthedocs.io/
Other
809 stars 121 forks source link

Questions about using docker images #73

Open boss-create opened 2 weeks ago

boss-create commented 2 weeks ago

In WSL, When installing using the recommended docker image: docker pull runzhongwang/thinkmatch:torch1.6.0-cuda10.1-cudnn7-pyg1.6.3-pygmtools0.5.1

torch.cuda is unavailable and there are problems with symbolic links: image

I also tried other different versions, and the above problems still occurred

rogerwwww commented 2 weeks ago

It seems to be an issue when putting WSL + Nvidia driver + docker together. Can you confirm: can you run CUDA with torch outside docker? To my best (but vague) knowledge of WSL, you may need WSL2 to have the Nvidia driver working properly.

boss-create commented 2 weeks ago

It seems to be an issue when putting WSL + Nvidia driver + docker together. Can you confirm: can you run CUDA with torch outside docker? To my best (but vague) knowledge of WSL, you may need WSL2 to have the Nvidia driver working properly.

Thank you for your reply. The WSL I mentioned above is WSL2, and I have also confirmed that CUDA can be used normally in WSL2-Ubuntu22.04 without using docker. I have used other docker environments in WSL2 before and did not encounter similar problems. At the same time, I also tried WSL2-Ubuntu20.04 to re-pull different versions of the docker images and the same problem occurred...

rogerwwww commented 2 weeks ago

Would you mind checking https://github.com/microsoft/WSL/issues/5663

And some StackExchange guys said the error is harmless https://superuser.com/questions/1707681/wsl-libcuda-is-not-a-symbolic-link