This message was deleted.

Hey Eric! We’ve seen many users integrate BentoML into their retraining pipeline w/ MLFlow/Airflow/Prefect etc or set up CI/CD via something like Github Action. Typically using either the CLI commands or the equivalent Python APIs. We’ve recently put out a few Github Actions to help our users to build such CI/CD pipelines: https://github.com/bentoml/deploy-bento-action https://github.com/bentoml/containerize-push-action https://github.com/bentoml/setup-bentoml-action

Oh these are great. We’re not using Yatai, but this is a great reference for enabling buildx and whatnot in CI. Ah, are you saying that some people will put calls to the GitHub CLI (gh) in their DAG tool that handles training? (To trigger a deployment pipeline) that’s cool, if so. I’m hunting for examples of this. Like, what if you want to have a human approve the deployment before it goes through? Or what if you train lots of models and want to select the best? And could you somehow run tests against both model and bento for every trained model—so that by the time humans are reviewing them, they’ve been validated? Someone could probably write a whole book on this.