[Klaud Cold] Update qwen3.5-fp8-h100-sglang-agentic-mtp SGLang image to nightly-dev-cu13-20260907-30705c00

[functionstackx]

Summary

Update SGLang image for the H100 Qwen3.5-397B-A17B FP8 AgentX MTP recipe from lmsysorg/sglang:v0.5.16-cu130 (v0.5.16 release) to lmsysorg/sglang:nightly-dev-cu13-20260907-30705c00 (2026-09-07 cu13 dev nightly).

• Tag verified on Docker Hub: last pushed 2026-09-07T01:43:42Z, digest sha256:19b8fa1223cc339c1eae7a5b703f1a8c2543b5b119155bf3d7efaef18f77f007; tag commit sgl-project/sglang@30705c00.
• Same tag as the Qwen3.5 SGLang AgentX bumps on B200 ([Klaud Cold] Update qwen3.5-fp8-b200-sglang-agentic-mtp SGLang image to nightly-dev-cu13-20260907-30705c00 #2861, [Klaud Cold] Update qwen3.5-fp4-b200-sglang-agentic-mtp SGLang image to nightly-dev-cu13-20260907-30705c00 #2862) and H200 ([Klaud Cold] Update qwen3.5-fp8-h200-sglang-agentic-mtp (+hicache-mtp) SGLang image to nightly-dev-cu13-20260907-30705c00 #2868), so all Qwen3.5 SGLang AgentX curves share one SGLang commit.
• benchmarks/single_node/agentic/qwen3.5_fp8_h100_mtp.sh is unchanged and pins nothing image-specific: SPEC_V2 EAGLE MTP at 3 steps, golden AL 3.39, flashinfer attention (sm_90) + allreduce fusion, fp8 quantization, fp8_e4m3 KV, HiCache kernel IO / page_first. Concurrency grids unchanged.
• Not bumped: the non-MTP sibling qwen3.5-fp8-h100-sglang-agentic (still on v0.5.12-cu130). Its agentic script qwen3.5_fp8_h100.sh was removed in chore(agentx): sync support with production results #2561, so the key cannot run today, and the arm is slated for retirement under MODELS.md. Flagged for cleanup rather than bumped.

Recipes touched: qwen3.5-fp8-h100-sglang-agentic-mtp

Test plan

• full-sweep-enabled sweep passes on cluster:h100-dgxc.

🤖 Generated with Claude Code

---

Note

Low Risk
Config and changelog-only image pin update for a single benchmark recipe; no application or infra code changes.

Overview
Bumps the H100 Qwen3.5 FP8 AgentX MTP recipe (qwen3.5-fp8-h100-sglang-agentic-mtp) from lmsysorg/sglang:v0.5.16-cu130 to lmsysorg/sglang:nightly-dev-cu13-20260907-30705c00, aligning with the same SGLang tag already used for Qwen3.5 AgentX on B200 and H200.

Adds a matching perf-changelog.yaml entry for the agentic-coding scenario. Benchmark script, search grids, and MTP/HiCache settings are unchanged; the non-MTP H100 agentic key is intentionally not updated (retirement path).

^{Reviewed by Cursor Bugbot for commit 75846b8. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs

---

感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档

[github-actions]

Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs

---

感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档

[claude]

Beyond the inline pr-link finding, I also checked the "model.container must equal image" invariant from AGENTS.md — this recipe has multinode: false and there is no container field anywhere in nvidia-master.yaml, so that invariant doesn't apply here.

Extended reasoning...

A confirmed inline finding already flags that the new perf-changelog.yaml entry's pr-link: PRLINK_PLACEHOLDER will fail utils/validate_perf_changelog.py's placeholder check. Beyond that, I independently verified the diff (a single-line image bump in configs/nvidia-master.yaml plus a tail-only append to perf-changelog.yaml) and specifically checked whether the AGENTS.md invariant "for image bumps, model.container must equal image" applies to this recipe. The recipe block shows multinode: false, and a search of the entire nvidia-master.yaml file found no container field at all, confirming that invariant is scoped to multi-node srt-slurm recipes and is not applicable here — so there is no missed companion field to update.

[claude]

🔴 The new entry uses pr-link: PRLINK_PLACEHOLDER, but utils/validate_perf_changelog.py only accepts the exact PR number (.../pull/<N>) or the literal string XXX / .../pull/XXX as placeholders; this value fails validate_added_pr_link and will fail the changelog validation gate in CI. Fix: use XXX (or https://github.com/SemiAnalysisAI/InferenceX/pull/XXX) as documented in CONTRIBUTING.md, or the real PR link/number.

Extended reasoning...

validate_added_pr_link() in utils/validate_perf_changelog.py checks link not in PR_LINK_PLACEHOLDERS and link != expected, where PR_LINK_PLACEHOLDERS = {"XXX", "https://github.com/SemiAnalysisAI/InferenceX/pull/XXX"} and expected is the canonical URL with the actual PR number. PRLINK_PLACEHOLDER matches neither, so on this PR's CI run the changelog gate raises ChangelogValidationError("new PR entry must use ... or an XXX placeholder; found 'PRLINK_PLACEHOLDER'"), blocking merge until corrected — a concrete, reproducible CI failure introduced by this diff's placeholder text deviating from the documented/enforced convention.

Verification: normal. The new entry's pr-link: PRLINK_PLACEHOLDER (perf-changelog.yaml:6931) is not an accepted value. utils/validate_perf_changelog.py:24-27 defines PR_LINK_PLACEHOLDERS = {"XXX", "https://github.com/SemiAnalysisAI/InferenceX/pull/XXX"}, and validate_added_pr_link (lines 144-160) raises ChangelogValidationError("new PR entry must use {expected!r} or an XXX placeholder; found… | normal…

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=34173727265
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=34173727265

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=34173727265
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=34173727265