The MLflow Model Registry provides a centralized model store, set of APIs, and UI, to collaboratively manage the full lifecycle of an MLflow Model. It provides model lineage, model versioning, model aliasing, model tagging, and annotations. The Model Registry defines four model stages that represent the lifecycle of a model: None, Staging, Production, and Archived. Each model stage has a different meaning and purpose:
None: This is the default stage when a new model version is registered. It indicates that the model version is not ready for any downstream use cases and needs further development or testing.
Staging: This stage indicates that the model version is ready for testing or validation in a pre-production environment. It can be used for quality assurance, user acceptance testing, or performance evaluation before moving to production.
Production: This stage indicates that the model version is ready for deployment in a production environment. It can be used for serving predictions, scoring data, or supporting business decisions. Only one model version per registered model can be in the production stage at a time.
Archived: This stage indicates that the model version is no longer in active use and can be safely deleted or moved to a long-term storage. It can be used for historical or audit purposes, or for restoring a previous model version if needed.
References:
MLflow Model Registry Documentation, p. 1-2
Introducing the MLflow Model Registry - Databricks, p. 1-2
Manage MLflow Models — Databricks Documentation, p. 1-2
