Data Science AI Quick Actions Model Reference

Native Support for Model By Reference: Data Science provides you the option to register model artifacts in your own Object Storage bucket without storing them in Data Science's managed storage. With this release, you can use this feature by way of the native API, too. This improves the maintainability, reduces extra costs incurred during copying the models, especially large models, and also meets compliance requirements. Using this native API support you can register models which are present in your OSS bucket in your respective tenancies.
Model Metadata Storage Enhancement: The AI Quick Actions metadata is now stored in the Model Catalog. Also, a search support for metadata based on user defined keywords is added. When creating the models, you can define the custom keywords to help you search your models.

Service Models Accessibility: You can access the service managed models using the native APIs which reside in the service managed model compartment without any extra permissions. These service modes serve as the foundation models that can be adapted for various tasks such as fine tuning for specific use cases, feature extraction, benchmarking, rapid prototyping and so on. This new feature reduces overly permissive access in AI Quick Actions by restricting the oss-read permission granted to services-managed-models. This enhances security by enforcing least-privilege access. Any unnecessary exposure of model store data is limited. It aligns with compliance and access control best practices.