-
Story
-
Resolution: Done
-
Normal
-
None
-
None
-
None
-
8
-
False
-
-
False
-
-
-
DEVAI Sprint 3264, DEVAI Sprint 3265, RHDHPAI Sprint 3266
Story (Required)
As a platform or MLOps engineer/administrator, CLI based offerings will facilitate both live and automated population of the RHDH Catalog with AI related artifacts, where there is an increasing array of sources, with varying interfaces and formats, to pull from.
As a developer/tester of an AI infused application, CLI based offerings will facilitate the population of the RHDH catalog in CI/CD Pipelines, testing environment set up, and test framework interaction. Allow for live updates vs. say only using GitOPs based configuration.
Background (Required)
afred@redhat.com started proposing to developer tools AI as far back as https://redhat-internal.slack.com/archives/C06H0E1CT4P/p1720471395677579 integrating various forms of AI Model Registries with RHDH/Backstage.
DEVAI-139 proved out the ability to use the REST APIs of both BackStage and Kubeflow Model Registry to query/import changes.
This spike will, in an environment that includes
- running models via KServe
- running ModelRegistry that has entries for the KServe instances
- running RHDH
The prototype a golang based program (stretch goal, a kubectl plugin) that
- if available, run at the level of Model Registry that stores running kserve inference endpoints in the model registry
- otherwise short term poll the KServe CRDS and correlate on our own the inference endpoint URLs to what is stored
- translate the Model Registry content to the catalog entity API content expected for POSTs to the Catalog REST API's Location URI
- Import to RHDH
Out of scope
Not targeted at this time for the Nov 2024 Tech Preview
Approach (Required)
Work with both upstream and Developer Hub Dev/PMs to gauge the latest on priority, potential roadmaps, with the upstream CLI issues.
Interact with Model Registry as needed to both ask for help or inform on results
Dependencies
Sufficient access to RHDH folks and upstream Backstage SMEs to sort out priorities, roadmaps, and technical details.
Acceptance Criteria (Required)
- Opening of follow on Refinement Docs, Design Docs, and DEVAI Epics under RHDP-1014 to detail where/how CLI based capabilities are determined to be needed. Current thought is Stage 3 of RHDP-1014 is the target timeframe.
- Demo video
- unit test
- create a repo under redhat-devai org
Done Checklist
As a spike, see acceptance criteria
- clones
-
RHIDP-10275 Spike: Investigate Kubeflow Model Registry wrt RHDH AI
-
- Closed
-
- is depended on by
-
RHIDP-10172 RHDH Catalog CLI Fit and Finish
-
- Closed
-
-
RHIDP-10228 RHDH Catalog CLI New Model Metdata sources
-
- Closed
-
-
RHIDP-10828 Catalog CLI and Backstage Upstream Engagement
-
- Closed
-
- links to