PyTorch Code Server with CUDA

Code Server Docker image for PyTorch with python development on the browser. Contains:

• CUDA 11.1
• Python 3.8.3
• PyTorch 1.81.1
• Code Server 3.10.0 (VSCode 1.56.0)

Requirements

• CUDA device with compute capability 3.5 or higher
• NVIDIA Docker Toolkit

Quickstart

$ docker run --privileged --rm -it --init \
  --gpus=all \
  --ipc=host \
  --user="$(id -u):$(id -g)" \
  --volume="$PWD:/projects" \
  -p 8443:8443 \
  ghcr.io/works-on-my-machine/pytorch-code-server:1.10.0

After running above command open localhost:8443 in your browser. Find your password under ~/.config/code-server/config.yaml

$ docker exec -it <your_container_name> /bin/bash
$ cat ~/.config/code-server/config.yaml

Login with your password. For a working example look at example project folder. Contains recommended extensions and settings. Example project also contains noVNC support. Checkout docker-compose.yml file for more information.

Notes

- Due to a possible bug in ms-python.python extension. After installing it revert it to a previous version by selecting “Install Another Version”. 2020.10.332292344 worked for me.

TODO

• Better documentation
• NO-VNC support for visualization
• Docker compose support