[Klaud Cold] Update qwen3.5-fp4-b200-sglang-mtp SGLang image to v0.5.20-cu130 / 将 qwen3.5-fp4-b200-sglang-mtp 的 SGLang 镜像更新至 v0.5.20-cu130

[Klaud-Cold]

No description provided.

[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 only run on labeled PRs. Add full-sweep-fail-fast (strongly recommended); use full-sweep-enabled only when matrix jobs should continue after a failure.
• 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：**扫描仅在带有标签的 PR 上运行。强烈建议添加 full-sweep-fail-fast；仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。
• **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。

[Klaud-Cold]

Stopped before GPU dispatch · baseline-point-mismatch
All 9 published 2026-09-01 lmsysorg/sglang:v0.5.14-cu130 points (producer 33451222974 / attempt 2, head 0c32f3c2) were generated before the fixed-sequence SRT-Slurm migration (#3352). Their point identities have no srt-recipe, so the current family can't produce them and final baseline coverage would fail: 8k/1k TP4/EP1 c4; TP2/EP1 c4, c8, c16, c32, c64; TP2/EP2 c16, c32, c64.
Source check: v0.5.20-cu130 (sha256:06e4f2ed…d611f) bundles sglang 0.5.20 at 94602c9c. Its parser accepts every recipe flag once --cuda-graph-max-bs becomes --cuda-graph-max-bs-decode. The alias was removed in v0.5.20; in v0.5.14 it set the same field. Performance compatibility is untested.

中文

在 GPU 调度前停止 · baseline-point-mismatch
2026-09-01 发布的 9 个 lmsysorg/sglang:v0.5.14-cu130 测试点均生成于定长配方迁移至 SRT-Slurm（#3352）之前。其点标识不含 srt-recipe，当前配置族无法生成这些点，最终基线覆盖检查必然失败；受影响的点见上文。镜像源码核对结论见上文：只需将 --cuda-graph-max-bs 改为 --cuda-graph-max-bs-decode，性能兼容性尚未测试。

[Klaud-Cold]

failed · Cleanup pending. Stop and confirm owned runs before closing.

中文

failed · 清理待完成。先停止并确认自有运行结束，再关闭 PR。

[Klaud-Cold]

failed · Reason: baseline-point-mismatch · Repairs: 0 · Runs: —
All owned runs ended. PR closed; branch deleted for retry.

中文

failed · 原因：baseline-point-mismatch · 修复次数：0 · 运行：—
所有自有运行均已结束。PR 已关闭；分支已删除，可重新尝试。