Basic PyTorch project containerized

installation

here of interest are 4 modes in vscode:

• local/local: normal mode as we know it
• local/container: vscode in a local Docker container, as specified in devcontainer.json/Dockerfile
• ssh/"local": vscode per ssh on Server, ssh-key must be exchanged, so that you can login without pwd. you then can open files and directories on server exactly as on the local machine.
• ssh/container: you can open/create the Docker container on the ssh-server as specified in devcontainer.json/Dockerfile.

local container

• click on >< symbol and select "reopen in container" or "rebuild container"

remote container

• take care that ssh keys are exchanged
• take care that docker runs on server in user mode

  sudo usermod -aG docker $USER

• click >< symbol and select "connect to host..." + enter hostname or ip-address
• git clone this repo on the ssh server (either with ssh or using the shell in vscode which already runs on server)
• open project dir on host
• click >< and select "reopen in container"

with click on >< you can with selecting "reopen in container" and "reopen in local filesystem" switch between container and normal mode

test it

now that the container should run locally/on ssh server, open init.ipynb in order to test if everything is set up correctly, if you have cuda on you host, you should see your GPU listed

Also you can check the GPU with

nvitop