Skip to content

[VideoGenX] add GB200 H3 SGLang deployment recipes / [VideoGenX] 添加 GB200 H3 SGLang 部署配方 - #4644

Closed
edwingao28 wants to merge 1 commit into
feat/h3-cross-hardwarefrom
feat/videogenx-gb200
Closed

edwingao28 wants to merge 1 commit into
feat/h3-cross-hardwarefrom
feat/videogenx-gb200

Conversation

@edwingao28

@edwingao28 edwingao28 commented Oct 29, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

Add two- and four-GPU GB200 SGLang deployments using a public pinned arm64 image. Preserve the legacy 107-frame, 50-update workload: 20 measured C1 clips plus warmup. These points are separate from v0.1.

Testing:

Check Result
CPU CI Passed at a2be3bcb1
GB200 E2E, 8ce8e3c g2/g4 failed before requests: wrong Python environment; zero clips

The corrected interpreter remains unvalidated on GB200. Depends on #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.

@github-actions

Copy link
Copy Markdown
Contributor

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。

Comment on lines +25 to +26
- bash
- /infmax-workspace/benchmarks/h3/run_server.sh

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i dont understand this? shouldnt the sgl-omni serve + arg command be here?

@edwingao28 edwingao28 changed the title [VideoGenX] run native MiniMax-H3 benchmarks on GB200 / [VideoGenX] 支持 GB200 原生 MiniMax-H3 基准测试 [VideoGenX] configure native H3 serving in YAML and add GB200 / [VideoGenX] 通过 YAML 配置原生 H3 服务并支持 GB200 Oct 29, 2026
@edwingao28
edwingao28 force-pushed the feat/videogenx-gb200 branch from 0e00143 to 657c84c Compare October 29, 2026 20:12
@edwingao28
edwingao28 changed the base branch from main to feat/h3-cross-hardware October 29, 2026 20:12
@edwingao28 edwingao28 changed the title [VideoGenX] configure native H3 serving in YAML and add GB200 / [VideoGenX] 通过 YAML 配置原生 H3 服务并支持 GB200 [VideoGenX] add GB200 H3 SGLang deployment recipes / [VideoGenX] 添加 GB200 H3 SGLang 部署配方 Oct 29, 2026
@edwingao28
edwingao28 added this pull request to stack #4660 October 29, 2026 20:20
@edwingao28
edwingao28 force-pushed the feat/videogenx-gb200 branch from 657c84c to c86a3f5 Compare October 29, 2026 22:03
@edwingao28
edwingao28 force-pushed the feat/h3-cross-hardware branch 2 times, most recently from 6101fe0 to 867ad47 Compare October 29, 2026 22:13
@edwingao28
edwingao28 force-pushed the feat/videogenx-gb200 branch from c86a3f5 to 0512f88 Compare October 29, 2026 22:13
添加 GB200 H3 SGLang 部署配方。
@edwingao28
edwingao28 force-pushed the feat/videogenx-gb200 branch from 0512f88 to a2be3bc Compare October 30, 2026 20:25
@edwingao28
edwingao28 force-pushed the feat/h3-cross-hardware branch from 867ad47 to 5934a67 Compare October 30, 2026 20:25
@functionstackx

Copy link
Copy Markdown
Collaborator

Gb200 queue long, not needed for v0.1

@edwingao28
edwingao28 deleted the feat/videogenx-gb200 branch November 1, 2026 04:23
@edwingao28
edwingao28 removed this pull request from stack #4660 November 3, 2026 22:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

2 participants