CUDA Code Server

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

• CUDA 12.6.1
• OpenMP Library
• MPICH Library
• Code Server 4.92.2

Requirements

• Docker on Linux or WSL
• CUDA device with compute capability 5.0 or higher
• NVIDIA Docker Toolkit following the installation guide

Quickstart

Create code-server/config.yaml as the configuration as the code-server config file

Optionally modify EXTENSIONS_GALLERY to environment variables in docker-compose.yml. Example usage on VScodium documentation

Then run the script below to build the image locally

$ ./run.sh

After running above command open localhost:8080 in your browser.

Why build the docker image locally?

This is the simplest way I found to install cuda development tools on Windows, specifically WSL. Please regard this as a software installation with docker as the package manager.

docker compose / portainer stack example

services:
  pytorch:
    image: pytorch-code-server:2.4.1 # your image name and tag
    volumes:
      - projects:/projects
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              capabilities: [gpu]
    restart: unless-stopped
    ports:
      - "8080:8080"
volumes:
  projects:
    external: false