docs(vllm-repack): update E2E report §2 with MetaX empty-wheel result - #260
Merged
Conversation
The MetaX E2E is now complete on metax124. Rewrite §2 to match: - Status: end-to-end passed (serve + inference on Qwen3-4B), replacing the earlier "device not visible / incomplete" note. - §2.7: the one real blocker was reshape_and_cache_flash (empty wheel ships no _C_cache_ops kernel). Document why #319 never fired (lazy _OpNamespace makes the hasattr guard always true; C550 disables Triton) and how #333 fixes it by registering the op as flag_gems dispatch (default.flagos) and routing the metax call site through CachedOp. - §2.7b: #325 (activation F.silu/F.gelu) is not needed on the empty wheel — silu/gelu already dispatch to flag_gems, so the vendor _maca impls it rewrites are never called. Verified by reverting #325 on-device: serve + inference still succeed. #325 only matters for the +cpu wheel. - §2.8/2.9/2.11: real launch command, successful inference response, and dispatch log lines; stack table flipped to device-visible / serve / infer OK. - §2.12: PR status (#333 done, #319 and #325 closed). This PR was written in part with the assistance of generative AI.
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Updates §2 (MetaX) of the vllm-repack E2E report to reflect the now-complete end-to-end run on metax124.
What changed
reshape_and_cache_flash(empty wheel ships no_C_cache_opskernel). Documents why #319 never fired (lazy_OpNamespacemakes thehasattrguard always true; C550 disables Triton) and how #333 fixes it via flag_gems dispatch +CachedOp.F.silu/F.gelu) is not needed on the empty wheel: silu/gelu already dispatch to flag_gems, so the vendor_macaimpls it rewrites are never called. Verified on-device by reverting #325 — serve + inference still succeed. #325 only matters for the +cpu wheel.Docs-only change to
vllm-repack/E2E-REPORT.md.This PR was written in part with the assistance of generative AI.