Files
club-3090/scripts/lib
3291b6099f Bump vllm-stable v0.22.0 → v0.24.0 (overlay-free; marlin-pad native) (#533)
* Bump vllm-stable v0.22.0 → v0.24.0 (engine spec + 10 compose defaults)

Live-validated on 2x3090: TP=2 clean, MTP accept 3.51, KV 622K/2.37x,
NIAH to 240K, soak Cliff2b PASS, toolcall 11/15 / instructfollow 15/15.
marlin-pad confirmed unneeded (native via #45295/#45176).

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01EfF565T9eSLaqGzidyJ1Pm

* arch_patches: add [email protected].0 loads:true rows (keep v0.22.0 as history)

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01EfF565T9eSLaqGzidyJ1Pm

* tests: assert v0.24.0 for vllm-stable pins (gemma stays on v0.22.0)

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01EfF565T9eSLaqGzidyJ1Pm

* docs: record vllm-stable v0.24.0 bump (BENCHMARKS row + UPSTREAM pin status)

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01EfF565T9eSLaqGzidyJ1Pm

---------

Co-authored-by: noonghunna <[email protected]>
Co-authored-by: Claude Opus 4.8 <[email protected]>
2026-07-01 01:23:06 +05:00
..