Update guidance for layered product seeking conformance#95
Conversation
Revert "Add FAQ about certifying products on existing distributions"
ritazh
force-pushed
the
update-layeredproduct-guidance
branch
from
d765d4d to
1550fa8
Compare
|
/hold @jeefy :) |
|
Question: how many of the AI Conformance requirements need to be met by the layered product on top vs the underlying platform? |
Signed-off-by: Rita Zhang <[email protected]>
puja108
left a comment
puja108
left a comment
There was a problem hiding this comment.
just minor comments for clarity, otherwise LGTM
yuanchen8911
left a comment
yuanchen8911
left a comment
There was a problem hiding this comment.
-
Terminology. Three places say
Kubernetes Conformancewhere the logic needsKubernetes AI Conformance. Please apply @janetkuo's three suggestions. -
The "uniquely provide at least one feature" clause. As written it forces a MUST-level replacement, which pushes toward fragmentation and breaks layered products spanning multiple bases. @puja108's "at least in one place it must be different" is the right intent — please reword so the differentiator can be a SHOULD addition, a variant, or a roadmap feature.
Non-blocking: the MUST-leve typo, the missing "SHOULD already met by base" pattern, examples that do not name what is being replaced, and the NOTE callout being stylistically inconsistent with the rest of faq.md.
Meta: given this is normative guidance with real submission impact (PR #79 is waiting on it) and the PR description calls for community-call discussion.
Signed-off-by: Rita Zhang <[email protected]>
|
Incorporated everyone's feedback. PTAL |
NVIDIA NIM on EKS is a layered AI inference product on top of Amazon EKS. This submission follows the layered-product guidance merged in PR cncf#95: it records per-requirement layer ownership and identifies the layered runtime's unique contributions (KAI Scheduler for gang scheduling, kgateway with inference extension CRDs for inference gateway, NIM Operator for AI-inference robust controller, NIM service metrics). All 9 MUST requirements are evidenced on the layered cluster; 3 SHOULD requirements are implemented via the NVIDIA GPU Operator. Evidence is hosted at: https://github.com/NVIDIA/aicr/tree/main/docs/conformance/cncf/v1.35/nim-eks/evidence Signed-off-by: Yuan Chen <[email protected]>
5 participants
Addendum to #89 since the AI conformance leads and the community have not yet had a chance to review and comment.
Given that this may affect how the community interprets eligibility and certification scope for layered products, this topic would benefit from being discussed in a community call. If you have input, please comment in this PR or join the next AI Conformance project meetings