[NVIDIA] [CUDA] Warning: NVIDIA GeForce RTX Laptop GPU with CUDA capability sm_86 is not compatible about comboloss HOT 1 CLOSED

You can see the pytorch wheels here and determine a little bit easier the support matrix than from pip.

I want to upgrade to CUDA 11.6 so I press ctrl+F and type cu116. Then I see the available pytorch versions, python versions and if it's for Windows or Linux.

If you want to use the latest pytorch with CUDA 11.6 then it's torch==1.12.0. I am using Python 3.7 so I choose the cp37 one.

Similarly, I need to choose torchaudio==0.12.0 and torchvision==0.13.0. This became quite easy when I found out that it's possible to do it like this.

My install command in my venv:

pip install torch==1.12.0+cu116 torchvision==0.13.0+cu116 -f https://download.pytorch.org/whl/torch_stable.html

I don't need torchaudio but I think you'll be able to find out how to edit the command by reading this thread. I just wanted to share this piece of knowledge so it's searchable on Google and it's helping other people.

from comboloss.