cuda_guide.md

1. Uncomment the "runArgs" section in .devcontainer/devcontainer.json.
2. Uncomment the lines for the nvidia channel and the pytorch-cuda package in the env.yml file.
3. Install Nvidia Container Toolkit and configure Docker by following this guide. You only need to do the steps under "Installing with Apt" and "Configuring Docker" without the "Rootless mode" section.

   ⚠️ Windows users: First verify you can do nvidia-smi in WSL. Then please install and configure Nvidia Container Toolkit in WSL.

   ⚠️ Windows users: For step 2 under "Configuring Docker", instead of doing sudo systemctl restart docker do sudo service docker restart.

You can proceed with the installation of the devcontainer. Once the installation is complete, you can check the access to GPUs by executing in the devcontainer terminal:

nvidia-smi

You should see GPU stats.

And check PyTorch CUDA runtime is installed in Python console:

import torch
torch.cuda.is_available() # should return True