Skip to content

chore(amd): refresh GLM-5.2 MI355X AgentX to 20261008 daily / chore(amd): 更新 GLM-5.2 MI355X AgentX 20261008 daily 镜像 - #3851

Open
jiejingzhangamd wants to merge 4 commits into
mainfrom
dev/jiejing/glm52-sglang-daily-1008
Open

jiejingzhangamd wants to merge 4 commits into
mainfrom
dev/jiejing/glm52-sglang-daily-1008

Conversation

@jiejingzhangamd

Copy link
Copy Markdown
Collaborator

Refresh the GLM-5.2 MXFP4 MI355X SGLang AgentX image to lmsysorg/sglang-rocm:v0.5.21-rocm720-mi35x-20261008, keeping the master image and recipe container aligned so the existing curve can be measured against the latest published daily build.

This is an image-only update based on current main, following #3728. Top-K v2, MTP, draft precision settings, TP/EP layouts, HiCache, concurrency, trace workload and evaluation selection are unchanged. The required performance changelog entry records the refresh.

Image provenance
中文

将 GLM-5.2 MXFP4 MI355X SGLang AgentX 镜像更新为 lmsysorg/sglang-rocm:v0.5.21-rocm720-mi35x-20261008,同步 master 配置与 recipe 的 container,使已有曲线可以使用最新发布的 daily 镜像重新测量。

本 PR 基于当前 main,仿照 #3728,仅修改镜像并添加必需的性能变更记录;Top-K v2、MTP、draft 精度设置、TP/EP、HiCache、并发、trace 工作负载和评测选择均保持不变。

AI model disclosure

Prepared by Codex, based on GPT-5, for image verification, the scoped configuration changes and PR preparation. The runtime does not expose the exact model/version identifier, so it could not be verified. No delegated agents contributed.

仅更新 GLM-5.2 MXFP4 MI355X SGLang AgentX 镜像及必要的性能变更记录;其他运行参数保持不变。
@github-actions

github-actions Bot commented Oct 9, 2026

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。

补全 daily 镜像更新的性能变更记录 PR 链接。
@jiejingzhangamd jiejingzhangamd added the full-sweep-fail-fast Full sweep with canary gate; first failure cancels the rest of that matrix (recommended) label Oct 9, 2026
Comment thread inferencex-e2e/perf-changelog.yaml Outdated
description:
- "Update the GLM-5.2 MXFP4 MI355X SGLang AgentX image from lmsysorg/sglang-rocm:v0.5.20-rocm720-mi35x-20260930 to lmsysorg/sglang-rocm:v0.5.21-rocm720-mi35x-20261008."
- "Keep the master image and model.container aligned; preserve Top-K v2, MTP, draft precision settings, TP/EP layouts, HiCache settings, concurrency, trace workload and evaluation selection."
pr-link: TBD

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

🔴 The new changelog entry's pr-link is "TBD", which fails the changelog CI validator and blocks merge. validate_added_pr_link (infx/workflows/validate_perf_changelog.py:139-148) only accepts the exact pull/<pr_number> URL or a placeholder in PR_LINK_PLACEHOLDERS = {"XXX", "https://github.com/SemiAnalysisAI/InferenceX/pull/XXX"} (line 22-25); "TBD" matches neither, so validation raises "new PR entry must use ... or an XXX placeholder; found 'TBD'". Fix: set pr-link to the literal placeholder "XXX" (or the real PR URL) before merging, not the free-text "TBD".

Why this was flagged

The PR run passes pr_number, so validate_added_pr_link at validate_perf_changelog.py:139 is invoked on the newly appended entry at perf-changelog.yaml:9499 with link="TBD". Neither branch of the function accepts it: pr_number is not None so the canonical-link branch is skipped, and "TBD" not in PR_LINK_PLACEHOLDERS (={"XXX", ".../pull/XXX"}) and != the expected pull/ URL, so it raises ChangelogValidationError. On base branch, entries use "XXX" or a real link and pass; this PR's changelog check would fail CI, blocking merge until the author replaces "TBD" with "XXX" or the real link.

Verification: The new entry at inferencex-e2e/perf-changelog.yaml:9499 ends with pr-link: TBD, not an accepted placeholder (PR_LINK_PLACEHOLDERS = {"XXX", ".../pull/XXX"}, validate_perf_changelog.py:22-25). The PR-time CI gate doesn't call validate_added_pr_link and accepts "TBD". Rejection occurs at merge: canonicalize_appended_links raises "unexpected pr-link 'TBD'" (prepare_perf_changelog_merge.py:108-111), run via merge_with_reuse.py:519/677. Base-branch entries use "XXX" and pass.

@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

@jiejingzhangamd jiejingzhangamd added agentx AgentX benchmarks, recipes, and infrastructure AMD full-sweep-fail-fast Full sweep with canary gate; first failure cancels the rest of that matrix (recommended) and removed full-sweep-fail-fast Full sweep with canary gate; first failure cancels the rest of that matrix (recommended) labels Oct 9, 2026
@chunfangamd

Copy link
Copy Markdown
Collaborator

/stage-results 38067578883

@github-actions

github-actions Bot commented Oct 11, 2026 •

Copy link
Copy Markdown
Contributor

@chunfangamd staged run 38067578883: https://inferencemax-app-git-staging-semianalysisai.vercel.app/inference?i_dates=2026-10-10~r38067578883

This run remains available across future /stage-results requests. Staging the same run ID again updates its staged data. Staging workflow

@chunfangamd

Copy link
Copy Markdown
Collaborator

/reuse-sweep-run 38067578883

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

agentx AgentX benchmarks, recipes, and infrastructure AMD full-sweep-fail-fast Full sweep with canary gate; first failure cancels the rest of that matrix (recommended)

Projects

Status: No status

Development

Successfully merging this pull request may close these issues.

2 participants