Publishing

We use Poetry for publishing ModelGauge and its plugins. While we are still in early development, we are publishing to a GCP Python Repository located at https://us-central1-python.pkg.dev/ai-safety-dev/aisafety-base-pypi.

Publishing permissions are connected to GCP users and authentication is handled via the gcloud cli. In other words, you must be successfully logged into a user or service GCP account using these instructions that has the appropriate roles and permissions for this project before you are able to publish these packages.

Configuring Poetry

Configure the MLCommons base repository as a destination

This will add a destination repository named mlcommons to your global Poetry installation.

poetry config repositories.mlcommons https://us-central1-python.pkg.dev/ai-safety-dev/aisafety-base-pypi

Add keychain authentication for Google Cloud plugin

This will add the appropriate keychain authentication plugin to your global Poetry installation.

poetry self add keyrings.google-artifactregistry-auth

Add bumpversion plugin

This will add the poetry-bumpversion plugin to your global Poetry installation.

poetry self add poetry-bumpversion

Publishing

Bump the version of ModelGauge and all plugins by using poetry version <version>, where <version> is one of: "patch", "minor", or "major". Note that this will bump the versions of all plugins referenced in pyproject.toml as well.
Commit those version changes, make a PR and merge it into main.
Check out the version of main corresponding to your PR. Run poetry run pytest --expensive-tests to ensure all tests pass. If they don't, fix the tests and return to the previous step.
Tag the commit with the version number you just created, prefixed by v, e.g. git tag v0.2.6.
git push origin your tag.
In Github create a new release. Select the tag you just created. Write the release notes. For now, also select "Set as a pre-release".
In your local repository use poetry run python publish_all.py to automatically build and publish all packages.