AWS released an updated model governance framework on July 2023, saying it enables cross-account model registration and governance through MLflow and SageMaker AI Model Registry sync. It is the company's first major governance update since the initial release of the MLflow integration in 2022.
The framework allows organizations to manage models across multiple accounts by using AWS RAM to share MLflow apps and centralize governance. This approach supports both hub-and-spoke and hybrid governance models for regulated environments.
"The hub-and-spoke pattern centralizes governance by sharing one MLflow app across development accounts with AWS RAM," said AWS, highlighting the new topology for larger organizations. The hybrid pattern keeps development accounts fully isolated from the governance hub, meeting strict regulatory requirements.
The announcement follows AWS's earlier release of MLflow integration with SageMaker AI Model Registry, which enabled automatic model registration. The new approach expands on that by supporting cross-account workflows and CI/CD pipelines for model deployment.
AWS did not specify exact performance benchmarks for the new governance framework, and the company raised the question of how to manage lineage across accounts. The framework will be available in the AWS CloudFormation stack, with deployment instructions in the accompanying repository.
Source: awsml
