Skip to content
Open
Show file tree
Hide file tree
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,7 @@ base:
enable-auto-tool-choice: true
# MSA sparse/index cache needs 128-token blocks.
block-size: 128
# fp8 caches enable FlashInfer MSA on SM90.
kv-cache-dtype: fp8
attention-config: '{"indexer_kv_dtype":"fp8"}'
moe-backend: marlin
Expand Down
6 changes: 2 additions & 4 deletions inferencex-e2e/configs/nvidia-master.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -4130,14 +4130,12 @@ qwen3.5-fp8-h100-sglang-agentic-mtp:
search-space:
- { tp: 8, ep: 8, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 4, 8, 12, 16], srt-recipe: agentic.yaml }
- { tp: 8, ep: 8, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: hicache }, conc-list: [4, 8, 12, 16], srt-recipe: agentic.yaml }
# MiniMax-M3.1-Flash AgentX on H100: day-zero vLLM image from
# https://vllm.ai/blog/2026-10-23-minimax-m31-flash, multiarch manifest verified
# 2026-10-23 (amd64 is minimax-m31-flash-x86_64-cu134). Serve flags follow the
# MiniMax-M3.1-Flash AgentX on H100: vLLM nightly with FlashInfer SM90 MSA. Serve flags follow the

Copy link
Copy Markdown
Contributor

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-dtype in the recipe.

Why this was flagged

In nvidia-master.yaml:4133 the diff rewrites the comment above minimaxm31flash-fp4-h100-vllm-agentic-dspark to "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.

# Hopper TP8 recipe: FP8 KV and indexer caches, NVFP4 experts on Marlin MoE.
# Upstream does not enable DSpark on Hopper; this adds the GB200 recipe's
# three-token DSpark, with synthetic acceptance from minimax-m3.1-flash__dspark__native__probabilistic-block.yaml.
minimaxm31flash-fp4-h100-vllm-agentic-dspark:
image: vllm/vllm-openai:minimax-m31-flash@sha256:86f582dae25ea92802353727164ae4eb03b37b18baf9119b805ffaec2a7ef07c
image: vllm/vllm-openai:nightly-8ca766326652caf4b8e50d1b9118d8a22f6adc73@sha256:b6d1cfc60cce1f6405d37ea77645a36a46065230fb643abc2eeb93a5dc8fb79b
model: MiniMaxAI/MiniMax-M3.1-Flash
model-prefix: minimaxm31flash
runner: cluster:h100-dsxe
Expand Down
9 changes: 9 additions & 0 deletions inferencex-e2e/perf-changelog.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -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

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The 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 pr-link ends in pull/PRNUM, and PR_LINK_PLACEHOLDERS in validate_perf_changelog.py:21 accepts only XXX or .../pull/XXX. compare_entries then raises ChangelogValidationError ("new PR entry must use ... or an XXX placeholder"). Fix: set the link to the real PR URL, or to https://github.com/SemiAnalysisAI/InferenceX/pull/XXX as the CONTRIBUTING.md:213 template shows, so the merge step can fill in the canonical link.

Why this was flagged

The appended block at inferencex-e2e/perf-changelog.yaml:13004 sets pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/PRNUM. When the PR is prepared for merge, prepare_perf_changelog_merge.py:88 calls compare_entries(base_entries, head_entries, pr_number). That runs validate_added_pr_link (validate_perf_changelog.py:105-116) on every appended entry. The link is not in PR_LINK_PLACEHOLDERS (only "XXX" and ".../pull/XXX"), and it does not equal .../pull/<pr_number>, so the step raises ChangelogValidationError. Even past that check, prepare_perf_changelog_merge.py:108 raises "appended entry N has unexpected pr-link" for any link outside the placeholder set. On a main-branch run, CANONICAL_PR_LINK (pull/\d+) rejects it as well. The existing 11 entries and every in-repo template use pull/XXX, and PRNUM is the only one of its kind in the file. The result is that the changelog pipeline refuses this PR's entry, and the sweep for minimaxm31flash-fp4-h100-vllm-agentic-dspark cannot be prepared or merged until the link is changed.

Verification: The diff appends pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/PRNUM (perf-changelog.yaml, the final line of the new block). validate_perf_changelog.py:21-24 defines PR_LINK_PLACEHOLDERS = {"XXX", "https://github.com/SemiAnalysisAI/InferenceX/pull/XXX"}. The PRNUM link therefore raises and the merge aborts. The base has no such entry.

Loading