chore: bump srt-slurm to v2.36.0 and drop the LMCache patch - #3585
Conversation
|
Thanks for the contribution!
中文感谢你的贡献!
|
There was a problem hiding this comment.
Looks good — a mechanical version bump with no logic changes.
- Checked submodule pin update (utils/srt-slurm -> 7b5863a7) matches the new SHA quoted in both RECIPES.md and RECIPES_zh.md, and the bilingual docs stay in sync.
- Checked patch removal against patches/README.md's own rule ("delete the patch and its row" once merged/included) — the removed row matches the deleted patch file exactly.
- Checked precedent: an earlier identical bump+patch-drop commit (e82af63, v2.30.0) also shipped without a perf-changelog.yaml entry, so omitting one here follows established convention rather than being an oversight.
- PR author (cquil11) is a listed CODEOWNER for runners/srt-slurm/patches/.
Extended reasoning...
The change touches only docs (RECIPES.md/RECIPES_zh.md), a submodule pointer bump, and deletion of an obsolete vendor patch plus its README table row — no application code, auth, or security-sensitive surface. Verified the new submodule SHA (7b5863a7) matches the version string quoted in both English and Chinese docs, and confirmed via git history that an earlier equivalent commit (e82af63) made the identical style of change (bump + drop obsolete patch) without a perf-changelog entry, which supports that omission here is consistent with prior practice rather than a gap. The PR author is a CODEOWNER for the patches directory. No bugs were found by the bug hunt, and the two candidate concerns raised were about missing changelog entries, which precedent addresses.
Bumps
utils/srt-slurmfrom v2.35.1 to v2.36.0 (7b5863a7) and deletesrunners/srt-slurm/patches/507-lmcache-server-atom-sglang.patch.The patch carried NVIDIA/srt-slurm#507 and #528. #507 shipped in v2.35.1 and #528 in v2.36.0, so the patch no longer applies to the pinned release (
setup_srt_slurmrunsgit applyon every patch and fails on it at v2.35.1).srtctl(no patch), the MI300X MiniMax-M3 vLLM LMCache point renders onelmcache-serverservice and theLMCacheMPConnectorconfig, and the Kimi-K3 (feat(agentx): bump Kimi-K3 FP4 MI355X ATOM image to 0924 and track recipe #3407) and MiniMax-M3 ATOM (perf(amd): enable MiniMax-M3 ATOM FlyDSL decode / 启用 MiniMax-M3 ATOM FlyDSL 解码 #3388) LMCache points render{"kv_connector":"lmcache_offload","kv_role":"offload"}.infx/tests/srt_slurmshows no new failures.RECIPES.mdandRECIPES_zh.md, and removes the patch row frompatches/README.md.