docs(qwen35): add 9B and 27B single-GH200 serving numbers#12
Open
FeathBow wants to merge 1 commit into
Open
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
Qwen3.5-9B Serving LoadandQwen3.5-27B Serving Loadto/models/qwen35-4b/, measured on 1x GH200 120GB (aarch64, sm_90) during the openinfer 509/510 rerun campaign: full c/QPS ladders with TTFT/TPOT p50+p99, long-context and c=120 overload behavior, tool-call round-trips, and greedy-vs-HF plus per-size logits-gate summaries.ffb959c4, Pythonvllm bench serveclient, random 1024/128, greedy, default seed. Saturation ~6.7 req/s / ~857 out tok/s.ffb959c4. Plateau ~2.7 req/s / ~340 out tok/s at qps≥8; the load-time budget line is quoted and explained — 17.2 GB KV, 5.5 GB prefill scratch, 18.8 GB recurrent-state reserve (two ~147 MB states per decode slot across the 64-slot capacity).--max-batchcapacity/KV trade-off, the decodable-vocab selection bound (248077 of the 248320 paddedlm_headrows), and a corrected CUDA-Graph statement — the server always captures and rejects--cuda-graph=false, so the previous "disable for debugging" note no longer holds.Qwen3.5-4B / 9B / 27B → /models/qwen35-4b/.npm run build+ static tests 12/12.