Registering a model to Unity Catalog with MLflow

MLflow supports registering models directly to Unity Catalog, giving them governed, catalog.schema.model_name identity alongside centralized access control, lineage, and discoverability. This is done by setting the MLflow registry URI to 'databricks-uc' and calling mlflow.register_model() (or logging with a registered_model_name) with a three-level name. Unity Catalog model registration is the recommended approach for production GenAI applications on Databricks, replacing the legacy workspace model registry.

1 · Learn the must-know

• 1You must set mlflow.set_registry_uri('databricks-uc') before registering so models are stored in Unity Catalog rather than the legacy workspace registry.
• 2Registered model names must follow the three-level Unity Catalog namespace convention: catalog.schema.model_name.
• 3The user or service principal registering the model needs USE CATALOG, USE SCHEMA, and CREATE MODEL (or MODIFY) privileges on the target catalog/schema.
• 4Unity Catalog model versions use governed aliases (e.g., 'champion', 'challenger') instead of the legacy stage names like 'Production' or 'Staging'.
• 5Models registered in Unity Catalog automatically get lineage tracking to the notebook/job/run and data sources used, visible in the Catalog Explorer UI.
• 6Unity Catalog enforces model signature requirements more strictly, so logging the model with an inferred or explicit input/output signature (and example input) is a common prerequisite for successful registration.

2 · Check your understanding