[edwingao28]

Summary

Changes

• Add native H3 video sweeps for SGLang on H100/H200/B200 and vLLM-Omni on H200. Configuration support does not qualify an untested row.
• Bind workload, model snapshot, image and engine settings into the service/client recipe. Reject model, port and GPU-count mismatches before submission; disable conditioning-cache reuse across requests.
• Bundle delivered media, client timing windows, serving GPU UUIDs, scheduler allocation and power. Keep executed checkout separate from the Actions head.
• Wait for video completion before statistics. Require every selected cell for full-sweep reuse; retain failures. Limit native jobs to 75 minutes and workflows to 90 minutes, without automatic retry.

flowchart LR
    A["Workload + engine recipe"] --> B["Native server and C1 client"]
    B --> C["Media + request windows"]
    B --> D["GPU UUIDs + allocation"]
    C --> E["Media and power checks"]
    D --> E
    E --> F["Sealed artifacts"]
    F --> G["Statistics"]

Loading

Test setup

• Official SGLang 797af717 image pinned by digest; H3 model 42ed227e, BF16, TP2/Ulysses2, h3-v0.1/release-v1.
• H100, concurrency 1: 20 measured requests + one disjoint client warmup; four participating GPUs, eight allocated GPUs, 192 CPUs.
• Output: 124 frames at 24 fps, 1344×768, 32 kHz stereo. Requested duration is 5 s; decoded duration is 124/24 s.
• C1: submission through downloaded media, 0.5 s polling; excludes subsequent local validation.

Results

Testing: Targeted H100 E2E, attempt 1 passed.

Runtime / original run	GPU	Measured + warmup	C1 p50	Status
797af717 / 41604835531	H100	20 + 1	73.8847 s	Targeted row passed
v0.5.23 / 41280191475	H100	20 + 1	73.8835 s	Historical row passed
v0.5.23 / 41280191475	B200	20 + 1	35.6733 s	Historical row passed

• New H100 audit: six original ZIP digests, 56 file seals, 21 complete audio/video decodes, zero request failures.
• GPU power: four matching UUIDs; 6,604 raw rows pass window and energy checks. Statistics reports 1/1 after video completion.
• One run, no repeated matched A/B trials: the H100 medians do not establish a speedup or equivalence. Historical B200 data does not qualify 797af717 on B200; these rows are not a scaling curve.

Validation scope

• Coverage: the accepted run is the H100 row. Current-runtime H200/B200 and the full matrix are not covered; semantic quality is unjudged.
• GitHub reports a merge conflict; benchmark results have not been published.
• Actual per-rank 49-update/50-sigma traces, complete installed-runtime inventory and DCGM exporter content digest remain unverified.
• Auxiliary host exporter failed with TaskProlog 127; required GPU-power checks passed separately.

Validation details

• Measured benchmark: 739282b. Actions/artifact head: b7e4847, adding only the statistics dependency.
• Current head a6db760e preserves those production files and changes tests; CPU CI passed. No new GPU measurement at this head.
• Review order: workload #4657 → measurement #4658 → GPU evidence #4659 → this PR. Related: admission #4274, App #1564.
• Native delivery guide: recipe, artifact and publication contracts.

PR checklist and change type

Type of Change

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

Checklist

• 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.

[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

[functionstackx]

Sorry, over the weekend, there was 2 major refactors to clean up the technical debt accumalated over the past 11 months of moving at the speed of light. We don't see any major refactors in the forthseeable future besides cleaning up AMD multinode AgentX pile of bash. As much, due to the refactors, u would need to ask your agent to rebase from remote main@latest. Thank you in advance for ur understanding