Environments availables

scGPT

Only available at FAS computing resources.

Please, reach out to platform to access for the first time to this env:

• Add the environment first:

  • Only the first time, add the env to the notebook kernels:
    • echo "n/holylfs05/LABS/hsph_bioinfo/Lab/shared_resources/scgpt-2" >> ~/.conda/environments.txt

• Start a python notebook on the ondemand web-page.

  • You need to be connected to the VPN to be connected.
  • Use gpu partition and if you need more than one GPU or gpu_test if you need something that won't take a long time
    • add these to the sbatch options --gres=gpu:n in the advance options, they are below in the page.
  • Add gcc/13.2.0-fasrc01 to the list of modules to load
• Once connected, if you get asked to use a password, just close the windows and open again.
  • If you didn't add module gcc/13.2.0-fasrc01, do this now:
    • Initiate a terminal session
    • module load gcc/13.2.0-fasrc01
  • Then, start a notebook choosing as kernel scgpt-2
• Validate everything works with these commands in the python notebook
  • import torch
  • torch.cuda.is_available() -> Should return True
  • torch.cuda.device_count() -> Should return `
  • import scgpt