[hshrivastava-droid]

Summary

Update qwen3.5-fp8-h200-sglang-agentic-hicache-mtp (Qwen3.5-397B-A17B-FP8, H200, SGLang AgentX, TP8/EP1 HiCache) to use the FA3 attention backend at concurrency 4, 8, 12, 24, 32 and 48.

Image: lmsysorg/sglang:nightly-dev-20261021-c1c287a4 (pinned by digest).

AI model disclosure

• Model/version: Claude Sonnet 5.5 (claude-sonnet-5-5)
• Role: Prepared the config bundle, changelog entry and PR text.

PR checklist and change type

Type of Change

• Bug fix
• New feature
• Configuration change
• Documentation update
• Other (please describe)

Checklist

• I have completed the AI model disclosure and kept it current
• I have tested my changes locally
• I have updated documentation if necessary
• If python -m infx.matrix.changed --base origin/main lists affected config keys, I have appended one new entry covering them to the physical end of inferencex-e2e/perf-changelog.yaml; if it lists none, I have added none. I have not edited historical entries
• If this PR appends an inferencex-e2e/perf-changelog.yaml entry, it carries exactly one primary sweep label (a maintainer applies it on fork PRs): full-sweep-fail-fast (recommended), full-sweep-enabled, or non-canary-full-sweep-enabled. Optional modifiers all-evals, evals-only, and agentx-fast require a primary label; the last two block reuse while applied.
• Before merging via reuse, an authorized maintainer (OWNER/MEMBER/COLLABORATOR) has commented /use <run_id> (or the legacy /reuse-sweep-run) on this PR. Do this only once there is a final full sweep that is all green with evals passing, since after this comment the primary sweep label will no longer automatically kick off new sweeps. Remove and re-add the primary sweep label to force a new sweep.

中文

更新 qwen3.5-fp8-h200-sglang-agentic-hicache-mtp（Qwen3.5-397B-A17B-FP8，H200，SGLang AgentX，TP8/EP1 HiCache），在并发 4、8、12、24、32、48 使用 FA3 注意力后端。镜像：lmsysorg/sglang:nightly-dev-20261021-c1c287a4（按摘要固定）。

AI 模型使用说明：Claude Sonnet 5.5（claude-sonnet-5-5），负责准备配置、变更日志和 PR 文本。变更类型：配置变更。

🤖 Generated with Claude Code

[github-actions]

Thanks for the contribution!

• Review: If this PR changes files owned by someone other than a repository admin or @SemiAnalysisAI/core, ask one eligible CODEOWNER to complete the latest PR_REVIEW_CHECKLIST.md before contacting a core maintainer on Slack. Follow the template exactly, including As a PR reviewer and CODEOWNER, I have reviewed this and have, so sign-off verification triggers.
• PR verification: Sweeps run only when the PR appends an inferencex-e2e/perf-changelog.yaml entry and carries exactly one primary label: full-sweep-fail-fast (strongly recommended; canary plus per-matrix fail-fast), full-sweep-enabled (canary; matrix jobs continue after a failure), or non-canary-full-sweep-enabled (no canary or fail-fast). The modifiers all-evals, evals-only, and agentx-fast require a primary label. On fork PRs, a maintainer applies the label. See sweep labels and reuse.
• After merging: PR authors must ensure all GitHub Actions jobs pass. Transient failures often pass on rerun; see how to rerun failed jobs.

中文

感谢你的贡献！

• **审阅：**如果 PR 修改的文件归属于仓库管理员及 @SemiAnalysisAI/core 之外的 CODEOWNER，请先联系一位有资格的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，再通过 Slack 联系核心维护者。必须严格遵循模板，并保留 As a PR reviewer and CODEOWNER, I have reviewed this and have，才能触发签核验证。
• **PR 验证：**只有当 PR 在 inferencex-e2e/perf-changelog.yaml 末尾追加了条目，并且恰好带有一个主标签时，才会运行扫描：full-sweep-fail-fast（强烈推荐；canary 加逐矩阵 fail-fast）、full-sweep-enabled（有 canary；矩阵任务在失败后继续运行）或 non-canary-full-sweep-enabled（无 canary，也无 fail-fast）。修饰标签 all-evals、evals-only 和 agentx-fast 必须与主标签一起使用。fork PR 的标签由维护者添加。参见扫描标签与复用。
• **合并后：**PR 作者必须确保所有 GitHub Actions 任务通过。临时性失败通常可以通过重新运行恢复；参见重新运行失败任务的说明。

[claude]

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

[claude]

🟡 nit (optional): The PR adds a header comment that only describes the overrides below it, which AGENTS.md does not allow. The line is "# EP1 HiCache arm on the FA3 attention backend, with the EAGLE depth set per concurrency." and it sits above override_hicache_ep1_fa3_k5_c4. Fix: delete the line. The override names and their args already show the backend and the per-concurrency EAGLE depth.

Why this was flagged

AGENTS.md:64 ("No header comments") says never to add broad comments above a recipe, override or section, or to describe what a config is. AGENTS.md:66 allows a comment only when it explains why a specific line or value was chosen. The new comment at agentic.yaml:119 restates what the six blocks after it set: attention-backend: fa3 plus speculative-num-steps/speculative-num-draft-tokens per CONC. It gives no technical reason for any value. On merge, the recipe carries a comment that breaks the repo's mandatory comment rule. Nothing at runtime depends on it.

Verification: nit: this happens on every merge of this diff. The diff adds # EP1 HiCache arm on the FA3 attention backend, with the EAGLE depth set per concurrency. at agentic.yaml:119, directly above override_hicache_ep1_fa3_k5_c4:. AGENTS.md:64 says "Never add broad, overarching comments ... above a recipe, override, or section". It gives no reason for any value, so the AGENTS.md:66 exception does not cover it.

[github-actions]

View unofficial run (performance): https://inferencex.semianalysis.com/inference?unofficialRun=40040595183

View unofficial run (accuracy): https://inferencex.semianalysis.com/evaluation?unofficialRun=40040595183

[nvpohanh]

[by Claude Code] The SGLang cookbook integration of this PR's AgentX configs (qwen3.5-fp8-h200-sglang-agentic-hicache-mtp, qwen3.5-fp8-h200-sglang-agentic-mtp) will happen in sgl-project/sglang#46506, in the Agentic Long-Context Deployment section of the Qwen3.5 cookbook page.

This PR has no authorized /use <run_id> comment yet, so the cookbook shows the recipe at this PR's head (fad84b252) as "in progress". A later cookbook PR switches it to the measured run once a run is pinned or this PR merges.

[nvpohanh]

[by Claude Code] sgl-project/sglang#46506 has been merged. This PR's AgentX configs (qwen3.5-fp8-h200-sglang-agentic-hicache-mtp, qwen3.5-fp8-h200-sglang-agentic-mtp) are now in the Agentic Long-Context Deployment section of the SGLang Qwen3.5 cookbook: https://docs.sglang.io/cookbook/autoregressive/Qwen/Qwen3.5#agentic-long-context-deployment

This PR has no authorized /use <run_id> comment yet, so the cookbook shows the recipe at this PR's head (fad84b252) as "in progress". A later cookbook PR switches it to the measured run once a run is pinned or this PR merges.

[nvpohanh]

[by Claude Code] The SGLang cookbook update of this PR's AgentX configs (qwen3.5-fp8-h200-sglang-agentic-hicache-mtp, qwen3.5-fp8-h200-sglang-agentic-mtp) will happen in sgl-project/sglang#46547, in the Agentic Long-Context Deployment section of the Qwen3.5 cookbook page. The configs have been live in that section since sgl-project/sglang#46506; sgl-project/sglang#46547 refreshes them with the pin below.

This PR has no authorized /use <run_id> comment yet, so the cookbook shows the recipe at this PR's head (fad84b252) as "in progress". A later cookbook PR switches it to the measured run once a run is pinned or this PR merges.

[nvpohanh]

[by Claude Code] sgl-project/sglang#46547 has been merged. This PR's AgentX configs (qwen3.5-fp8-h200-sglang-agentic-hicache-mtp, qwen3.5-fp8-h200-sglang-agentic-mtp) are now in the Agentic Long-Context Deployment section of the SGLang Qwen3.5 cookbook: https://docs.sglang.io/cookbook/autoregressive/Qwen/Qwen3.5#agentic-long-context-deployment

This PR has no authorized /use <run_id> comment yet, so the cookbook shows the recipe at this PR's head (fad84b252) as "in progress". A later cookbook PR switches it to the measured run once a run is pinned or this PR merges.