Model Versioning with MLflow and SageMaker Model Registry
Explore how to enable Model Registry sync between MLflow and SageMaker to unify model versioning, streamline approval gates, and maintain comprehensive audit trails. Understand setting needed IAM permissions, registering models with inference specs and metrics, and locking versions to control deployment staging. This lesson helps you confidently manage model versions across both MLflow and SageMaker environments.
The previous lesson showed how SageMaker Model Registry versions models, gates deployment with ModelApprovalStatus, records lineage, and documents models with Model Cards. Those features see only the models registered in SageMaker. Teams that track runs with MLflow on SageMaker AI, as covered in the Automated Experiments lesson, often register models in MLflow’s own model registry too. If those versions stay only in MLflow, the approval gate, lineage graph, and Model Cards never see them, and auditors face two version histories.
This lesson covers Model Registry sync, which closes that gap. When it is turned on, registering an MLflow model automatically creates a new model package version in SageMaker Model Registry, inside a Model Package Group named after the MLflow registered model (created on first registration). You will learn how to enable the sync, which IAM permissions it needs, how MLflow aliases relate to approval, and how to verify the audit trail.
The diagram below shows how a model registered in MLflow reaches SageMaker Model Registry and the approval gate.
