Hundreds of data scientists store models in SageMaker Model Registry model groups. The team wants to organize existing models into three categories (computer vision, NLP, speech recognition) to improve discoverability without changing model artifacts or existing groupings. What approach satisfies these constraints?

Creating a Model Registry collection for each category and placing existing model groups into these collections is the most suitable approach. SageMaker Model Registry collections allow you to logically group related model packages and model groups, enhancing discoverability without altering the underlying model artifacts or existing model group structures. This meets the requirement of organizing models into categories without changing existing groupings or artifacts. Creating custom tags for model packages would require tagging individual packages, which can be cumbersome for hundreds of models and doesn't provide a hierarchical organization. Creating new model groups and moving models would disrupt existing groupings, violating a key constraint. SageMaker ML Lineage Tracking is for tracking model development steps and dependencies, not for auto-categorizing and organizing models into discoverable categories.