feat(sglang): cut qwen36-27b over to the v0.5.16 engine - #4222
Merged
Conversation
Prod pinned a v0.5.15 digest while main already built v0.5.16, so the next Renovate digest bump would have rolled serving onto an untested engine. The hicache suffix moves with it: page names encode neither engine nor weights.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
🔗 Linked repositories identifiedCodeRabbit considers these linked repositories for cross-repo context during reviews:
📝 WalkthroughWalkthroughUpdates the Qwen3.6-27B SGLang configuration to use the v0.5.16 AWQ hicache storage directory. ChangesSGLang runtime alignment
Estimated code review effort: 1 (Trivial) | ~2 minutes Possibly related PRs
Suggested labels: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
@@ spec.env.SGLANG_HICACHE_FILE_BACKEND_STORAGE_DIR.value @@
# inference.llmkube.dev/v1alpha1/InferenceService/ai/qwen36-27b
! ± value change
- /hicache/sglang-v0.5.15_awq-f541031d
+ /hicache/sglang-v0.5.16_awq-f541031d
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Live-validated on control-1 before this PR: the InferenceService was patched to the v0.5.16
digest, loaded clean, and has been serving production traffic since 21:26 UTC.
Production A/B from VictoriaMetrics, v0.5.15 over the 3h to 21:15 UTC vs v0.5.16 over the
17m to 21:44 UTC:
Decode is flat — TPOT moves under 2% either way, which is what the same kernels on the
same hardware should do. The TTFT halving is not claimed as a win: the windows differ in
concurrency and the v0.5.16 one is only 17 minutes against a cold L3 cache.
The case for merging is the split, not the numbers. #4219 moved the build to v0.5.16 while
serving stayed pinned to a v0.5.15 digest, so the next Renovate digest bump would roll prod
onto an engine nothing had exercised. The hicache
STORAGE_DIRsuffix moves in the samecommit because page names encode neither the engine nor the weights revision, so a bare image
bump would serve v0.5.15 KV pages to a v0.5.16 engine. Verified: the new suffix started a
fresh directory rather than reusing the old pages.
Merging re-applies what is already live, so Flux converges without another reload.
Summary by CodeRabbit