Deploy instructions

• If necessary: update the minimum PyTorch/Python version, and the minimum linear_operator version:
  • In setup.py
  • In README.md
  • In .github/workflows/*.yml
  • In .conda/meta.yaml
• Create a tag with the release name and push it (e.g. git tag 0.1.0.rc3 && git push origin 0.1.0.rc3)
• On Github, update the new release with some release notes. Include major new features and bug fixes.

That should be it! The deploy actions should take care of the rest.

If the deploy script fails...

Add the new version to ReadTheDocs

• Go to the RTD admin page for GPyTorch and click on the "Versions" tab
• Find the latest tag under "Activate a Version", and click the "Activate" button. Check the "active" check box.
• Under "Admin > Advanced Settings", change the default version to the latest version.

Uploading to PyPI

• Locally create a packaged version (python setup.py sdist)
• Upload the latest package to PyPI (twine upload dist/gpytorch-<version_number>.tar.gz)

Uploading to Anaconda Cloud

• Make sure you have conda-build installed: conda install conda-build
• Let the meta.yaml file know about the pytorch channel conda config --append channels pytorch
• Change in to the .conda directory: cd .conda
• Run conda-build . in the .conda directory
• Upload the latest package to Anaconda Cloud using the command printed out at the end of conda-build.