1 of 8

Kubeflow 1.10

Model Registry <> Kserve integration

2 of 8

What’s new in Model Registry?

Major Steps Towards Integrating with Kserve

​

Improvements:

  • Model Registry Custom Storage Initializer

​

Next steps: RFE (link)

  • InferenceService controller

3 of 8

Model Registry Custom Storage Initializer

The Model Registry Custom Storage Initializer (CSI) is a custom implementation of the KServe ClusterStorageContainer that allows the users to use Model Registry metadata to download and deploy models

4 of 8

How does it work?

5 of 8

Changelog

  • introduced in kubeflow/manifests (link)
  • refactored to support multiple model registries (link)
  • added a Getting Started and Installation guide to the kubeflow website (link)
  • added an example of a custom storage implementation in the kserve user guide (link)

​

​

6 of 8

InferenceService Controller RFE (link)

The primary role of the InferenceService Controller is to monitor the KServe InferenceService resources associated with the models in the Model Registry. It ensures accurate tracking of each deployment that occurs in the cluster by creating and updating InferenceService records in the Model Registry, thereby enhancing auditing capabilities

​

​

If you like this idea, be sure to vote/comment on github: link as work is always prioritized by community feedback

7 of 8

How does it work?

8 of 8

Changelog

  • introduced the inferenceservice reconcile function (link)
  • improved the reconcile function UX (link)
  • presented a proposal in the model registry community meeting (link)

​

​