Repository navigation
[NV] MiniMax-M3.1-Flash H100 vLLM AgentX update / [NV] MiniMax-M3.1-Flash H100 vLLM AgentX 更新 #5124
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Changes from 1 commit
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -12993,3 +12993,12 @@ | |
| - "4P5D c660 also sets the sparse-attention indexer logits to BF16 explicitly on prefill and decode (attention-config indexer_logits_dtype bfloat16; BF16 is already the SM10x default, so every point runs BF16 indexer logits), selects the indexer top-k with deep_select on prefill (kernel-config sparse_indexer_topk_backend deep_select), and has the Mooncake master evict 2% of the store per round (--eviction_ratio=0.02 instead of 0.10)." | ||
| - "Every point keeps the model's precision policy: FP8 KV cache, an MXFP4 indexer key cache and FP32 KDA state." | ||
| pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/5108 | ||
|
|
||
| - config-keys: | ||
| - minimaxm31flash-fp4-h100-vllm-agentic-dspark | ||
| scenario-type: | ||
| - agentic-coding | ||
| description: | ||
| - "Update to vllm/vllm-openai:nightly-8ca766326652caf4b8e50d1b9118d8a22f6adc73, which selects FlashInfer's SM90 MSA kernels for the indexer and sparse attention." | ||
| - "Serve flags, DSpark speculation and the search space are unchanged." | ||
| pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/PRNUM | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. 🔴 The changelog merge-prep step will reject this PR, so the image bump cannot land as it is. The new entry's Why this was flaggedThe appended block at inferencex-e2e/perf-changelog.yaml:13004 sets Verification: The diff appends |
||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
🟡 nit (optional): The rewritten comment is a config header comment, which AGENTS.md:56 forbids. It names the image ("vLLM nightly with FlashInfer SM90 MSA") instead of explaining a specific line. It also runs well past the wrap width of the lines around it. Fix: drop the header sentence and keep only the line-level reasons. The FlashInfer MSA reason is already noted next to
kv-cache-dtypein the recipe.Why this was flagged
In nvidia-master.yaml:4133 the diff rewrites the comment above
minimaxm31flash-fp4-h100-vllm-agentic-dsparkto "MiniMax-M3.1-Flash AgentX on H100: vLLM nightly with FlashInfer SM90 MSA. Serve flags follow the". This newly written text is a header comment describing the config, and AGENTS.md:56 bans those ("No header comments describing a file, topology, config, recipe, override, or section"). The image tag already records what the image is. The changelog entry and the new recipe comment at agentic.yaml:38 already explain the FlashInfer MSA reason. Maintainers are left with a stale-prone header that the repo's comment rules say must not exist.Verification: The diff replaces lines 4133-4135 of inferencex-e2e/configs/nvidia-master.yaml with a line that opens
MiniMax-M3.1-Flash AgentX on H100:, a label for the config key below it. AGENTS.md:56 bans this: "No header comments describing a file, topology, config, recipe, override, or section." The rewritten line is about 99 characters, while nearby comment lines wrap near 80.