[Klaud Cold] Update qwen3.5-fp8-mi355x-sglang (+mtp) SGLang ROCm image to v0.5.20-rocm720-mi35x-20260924

[functionstackx]

Summary

Update SGLang ROCm image from v0.5.18-rocm720-mi35x-20260828 to v0.5.20-rocm720-mi35x-20260924 (latest nightly) for the Qwen3.5 FP8 MI355X 8k1k recipes.

Recipes touched: qwen3.5-fp8-mi355x-sglang, qwen3.5-fp8-mi355x-sglang-mtp

Changes:

• configs/amd-master.yaml: image bump for both keys.
• srt-slurm-recipes/qwen3.5/sglang/mi355x-fp8{,-mtp}/8k1k.yaml: container bump; cuda-graph-max-bs → cuda-graph-max-bs-decode (SGLang nightlies ≥ 09-15 reject the old flag as ambiguous; matches the b200 siblings).
• perf-changelog.yaml: new entry.

Test plan

• full-sweep-enabled sweep passes.

🤖 Generated with Claude Code

[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。

[claude]

🔴 The new entry's pr-link: PRLINK_PLACEHOLDER will make the merge-prep automation reject this PR instead of auto-filling the real link. infx/workflows/prepare_perf_changelog_merge.py (validate_perf_changelog.py) only accepts "XXX" or "https://github.com/SemiAnalysisAI/InferenceX/pull/XXX" as placeholders (PR_LINK_PLACEHOLDERS); any other value raises ChangelogValidationError(f"appended entry {index + 1} has unexpected pr-link {link!r}"). Fix: use the documented placeholder https://github.com/SemiAnalysisAI/InferenceX/pull/XXX (or XXX) so the merge automation can canonicalize it to the real PR link.

Why this was flagged

perf-changelog.yaml:8894 sets pr-link: PRLINK_PLACEHOLDER for the newly appended entry (config-keys qwen3.5-fp8-mi355x-sglang / -mtp). infx/workflows/prepare_perf_changelog_merge.py:96-109 computes expected_link and, for each appended entry whose link isn't already the expected link, requires it be in PR_LINK_PLACEHOLDERS = {"XXX", "https://github.com/SemiAnalysisAI/InferenceX/pull/XXX"} (validate_perf_changelog.py:21-24); otherwise it raises ChangelogValidationError. PRLINK_PLACEHOLDER matches neither, so the merge-prep step (and validate_perf_changelog.py:138's same check) fails, blocking this PR from merging via the normal automated path where the base branch would have a correctly formatted placeholder or real link.

Verification: normal. The appended entry sets pr-link: PRLINK_PLACEHOLDER (perf-changelog.yaml:8896). The merge-prep automation infx/workflows/prepare_perf_changelog_merge.py command canonicalize calls canonicalize_appended_links -> compare_entries(base, head, pr_number) (line 86). Inside compare_entries, every appended entry passes through validate_added_pr_link(link, pr_number)… | normal. The…

[github-actions]

View unofficial run (performance): https://inferencex.semianalysis.com/inference?unofficialRun=36086631310

View unofficial run (accuracy): https://inferencex.semianalysis.com/evaluation?unofficialRun=36086631310

[functionstackx]

/reuse-sweep-run 36086631310

[functionstackx]

/reuse-sweep-run 36086631310