PyTorch Docker Image

This repository provides a minimal, customizable Docker image for PyTorch, suitable for various Python, CUDA, and PyTorch versions. The image is built and pushed to Docker Hub automatically using GitHub Actions.

Current Versions

• Python: 3.12
• CUDA: 12.4
• PyTorch: Latest compatible version with the above CUDA version

Customization

Packages

Modify package lists in the requirements folder:

• packages.txt: Standard pip packages installed via pip install -r.
• torch_extensions.txt: PyTorch extensions installed with --no-build-isolation.

Base Image and Versions

Adjust the FROM directive and ARGS in the Dockerfile to change the base OS, CUDA, and PyTorch versions.

Setting Up the Workflow

1. Set GitHub Secrets:

• Navigate to your repository's Settings.
• Go to Secrets and add the following:
  • DOCKERHUB_USERNAME: Your Docker Hub username.
  • DOCKERHUB_TOKEN: Your Docker Hub access token.

2. Create a Release:

• Push your changes to the main branch.
• Go to the "Releases" section in your GitHub repository.
  • Click "Create a new release".
  • Tag your release and publish it.

Upon publishing a release, the workflow is triggered, building and pushing the image to Docker Hub with latest and the release tag.