Repository navigation
feat: add self-service FPM support workflow - #46
Closed
jasonqinzhou wants to merge 1 commit into
Closed
jasonqinzhou wants to merge 1 commit into
jasonqinzhou wants to merge 1 commit into
Conversation
Signed-off-by: Jason Zhou (Engrg-Hardware 1) <jasonzho@nvidia.com>
Contributor
Author
This was referenced Sep 16, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
aisimulate supportworkflow for exact model/GPU support cellspredict/recommendrequests to the self-service workflowValidation
219 passedacross the support workflow and adjacent CLI/runner/Sweeper regressionsQwen/Qwen-AgentWorld-35B-A3Bon an 8xH200 cell:unsupportedwithexact_match=falseforward_model: fpmand the systems overlay through SweeperScope and remaining validation
The MVP onboards arbitrary model/GPU combinations when the GPU already has a packaged AIC system specification. Adding a new GPU architecture is outside this change.
The actual FPM campaign and matched E2E measurements require the requested GPUs. They have not been run because the local Brev session needs reauthentication; no Brev instance was created, stopped, or deleted. This PR remains draft until that evidence is collected or the project owners explicitly split it into follow-up validation issues.
Tracking