Story (Required)

As an application developer trying to build AI infused applications with RHDH,  I want to be able to quickly determine which deployed models my organization has specified as OK to use.

As an platform engineer or AI stack developer of an OpenShift cluster that includes AI services, I want to be able to declare the various modes hosted on my cluster in a central repository which can then somehow accessed by the development tooling used by application developers using my cluster.

Background (Required)

As part of the coupling of Model Registry and RHDH Model Catalog concepts afred@redhat.com started proposing to developer tools AI as far back as https://redhat-internal.slack.com/archives/C06H0E1CT4P/p1720471395677579 , we have been investigating the possible implementations/offerings that might be used for the Model Registry.

One such possible alternative is the KubeFlow Model Registry project ([ https://github.com/kubeflow/model-registry|https://github.com/kubeflow/model-registry] ).  Exec summary:  KubeFlow falls under OpenShift's Open Data Hub umbrella from a portfolio point of view.  KubeFlow's model registry project is a newer offering that is still at an alpha level as of the opening of this Story.

Out of scope

Not targeted at this time for the Nov 2024  Tech Preview

Approach (Required)

Model registry is easily deployable on OpenShift via

kubectl apply -k "https://github.com/kubeflow/model-registry/manifests/kustomize/overlays/db?ref=v0.2.3-alpha"

It provides both a generic REST endpoint (see https://github.com/kubeflow/model-registry/blob/main/docs/logical_model.md ) and a GRPC endpoint with a Golang based client (see https://github.com/kubeflow/model-registry/blob/main/docs/mr_go_library.md

Matteo Motari (won't tag him in our epic here, at least here in the description) is the initial contact given for questions/discussion (see https://redhat-internal.slack.com/archives/C06H0E1CT4P/p1725886054059189 for example)

This spike will aim to

• prototype / validate both the REST and GRPC client flows
• Drill down on their OpenAPI Proxy Server and gather more detail around intended use cases and scenarios that could be applicable to the DEVAI RHDH Model Catalog work.
• prototype golang based clients (including possible kubectl plugins, for use by both devs and admins)
• Work with Matteo on current roadmap around
  • progressing to beta and GA
  • possible accompanyting K8S CRDs that could be leveraged from either K8s controllers or Backstage plugins

Dependencies

Sufficient access to Mateo and the Open Data Hub folks to sort our any roadmap or technical questions.

Acceptance Criteria (Required)

• Any demoable prototypes are socialized as appropriate
• If appropriate fork any prototypes to https://github.com/redhat-ai-dev
• Opening of follow on Refinement Docs, Design Docs, and DEVAI Epics under RHDP-1014 to productize using of KubeFlow Model Registry by Developer Hub based on results of various investigations

Done Checklist

As a spike, see acceptance criteria