[edwingao28]

Summary

Move the B200 and H100 H3 video rows from SGLang nightly 797af717 to the lmsysorg/sglang:v0.5.23 release recommended in the MiniMax-H3 recipe. Both rows pin the amd64 digest; VIDEO_RUNTIME_REVISION follows the v0.5.23 tag (297e54fd). Engine flags and the v0.1 workload are unchanged.

Testing: Trusted-mode run 41773473575: B200 36.25 s/clip, H100 74.45 s/clip, 20/20 valid each, no IPC all-to-all fallback, 49/49 steps. Its CI validator job failed on a main-only checkout (fixed in #2916); the validator passes locally on its artifacts.

Dispatched through e2e-tests.yml as for external PRs, since this PR targets #2916.

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.

中文

B200 和 H100 的 H3 视频配置目前使用带日期的 SGLang nightly 镜像（797af717），外部无法复现其结果。本 PR 将这两项配置改为 lmsysorg/sglang:v0.5.23 正式版镜像，即 SGLang Diffusion 团队在 MiniMax-H3 recipe（第 1、2 节）中推荐的 H3 镜像。

• h3-bf16-b200-sglang-video 和 h3-bf16-h100-sglang-video 固定为 lmsysorg/sglang:v0.5.23@sha256:f7876bef696ccf9cf75a2bf7ccb82970f5fac577f082fb8ca95c9a59d2160eb5。该 digest 是 v0.5.23 镜像索引 sha256:f832b5c39559… 中的 linux/amd64 manifest，v0.5.23-cu130 指向同一索引。
• 两个 recipe 的 VIDEO_RUNTIME_REVISION 改为 297e54fd105e974cdf401e5b4d20253a22c16f90，即 sgl-project/sglang 的 v0.5.23 tag 对应的 commit，与镜像标签 org.opencontainers.image.revision 一致。
• 引擎参数不变：B200 使用 Ulysses4，H100 使用 TP2+Ulysses2，均为 --encoder-parallel auto --performance-mode speed --warmup-resolutions 1344x768、eager、BF16，并保留 --disable-conditioning-cache。
• H200 SGLang 和 H200 vLLM-Omni 配置不变。
• perf-changelog.yaml 末尾追加一条记录，包含上述两个 config key。planner 选出两个 C1 部署：cluster:b200-nscale 上的 B200 和 cluster:h100-dsxe 上的 H100。

风险：

• 本 PR 的目标分支是 feat/h3-cross-hardware（[VideoGenX] run native video sweeps from YAML recipes / [VideoGenX] 使用 YAML 配方运行原生视频扫描 #2916），不是 main。run-sweep.yml 和 trusted-external-sweep.yml 只对合入 main 的 PR 触发，因此在 [VideoGenX] run native video sweeps from YAML recipes / [VideoGenX] 使用 YAML 配方运行原生视频扫描 #2916 合入后把本 PR rebase 到 main 之前，或由维护者针对当前 head 手动 dispatch e2e-tests.yml 之前，无法运行 sweep。
• 尚未进行 GPU 运行。H100 在 v0.5.23 上的这一拓扑还没有硬件结果（recipe 中 H100 只做过拓扑对比）。recipe 的审核要求服务端日志中不出现 IPC all-to-all (multi-rank) unavailable，且进度条显示 49/49，请在 sweep 的 logs/service_video-server.out 中核对这两项。

AI 模型使用说明：claude-opus-5-5（Claude Code 运行环境提供的标识），负责阅读贡献文档和 recipe、解析镜像 digest 与 SGLang commit、修改配置、recipe 和 changelog、运行本地检查并撰写本说明，未委派其他 agent。

[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]

Code review found no issues

No high-confidence issues detected in this change.

This review covers commit 6d88916, which is no longer the latest commit on this pull request; later commits are not covered by it.