Load apptainer — typically by running module load apptainer.
Run source <(curl -s "https://raw.githubusercontent.com/Ben-Drucker/slurm-gen-internal/refs/heads/main/build_and_nvidia_script.sh?token=GHSAT0AAAAAACTQF4TQOQRUCC5WY5RJVG6IZYRQF6A"). This will do the following automatically:
Download tunneller-support container
Build the container to a .sif file
Install Nvidia drivers (See troubleshooting section if Nvidia drivers not found)
Troubleshooting:
Before continuing, ensure your version of VS Code is up to date.
Problem
Solution
SSH won't connect
• You may need to connect to a VPN and/or HTTPS proxy if your institution requires it • An id_rsa ssh key is required (currently) for authentication. Ensure it is located in ~/.ssh. • Ensure your computer has an ssh client installed.
apptainer not found
• If you get this message (or something like it) you may need to modify the default command string in SLURM++ (I.e., if apptainer is located at a different path on your HPC cluster).
Job seems to quit instantly
• The container could not be run and the job exited. Check output log for the reason why.
CUDA support not working
• During installation, if the Nvidia drivers were not found, they need to be installed manually using a command like cp <path to drivers on host> <container.sif>/<driver paths>.
Other issues
• This extension has not been tested on different HPC clusters with potentially different settings. If stderr differs, for example, when srun is called, the extension will not work.
