Model Monitoring & Deployment Security is now GA in CDSW 1.8. Available in this release is native MLOps functionalities for model monitoring, enabling tracking of individual model predictions down to the feature level for calculating model drift and ground truthing to production environments. Data scientists can analyze metrics using their preferred libraries and IDEs in any language, ensuring models are performing optimally and compliantly at scale.
Additionally, this release includes general availability of Resource Quotas and quota management. These features enable administrators to limit users’ aggregate CPU, memory, and GPU consumption to protect against over-usage resulting in critical compute resource shortages. Default quotas can be configured for a Workspace, and overridden on a per-user basis with Custom quotas.
Also in this release:
Ability to use custom command-line arguments for jobs.
Improved security for model deployments allowing user-level access controls to prevent unauthorized access of endpoints.
Read the release notes for the full list of smaller improvements and bug fixes.