[None][perf] Custom decode kernels for MinimaxM3 - #17171

Closed

brb-nv wants to merge 10 commits into

NVIDIA:feat/m3_with_msafrom

brb-nv:user/brb/port-vllm-kernels

Commits on Jul 30, 2026

[None][perf] port CuteDSL MiniMax-M3 indexer decode scoring kernel
brb-nv
committed
c50d48f
[None][perf] port Triton MiniMax-M3 sparse block decode attention
brb-nv
committed
a92d786
[None][perf] route MiniMax-M3 dense attention layers through trtllm-gen
brb-nv
committed
34fe9f9

Commits on Jul 31, 2026

minor fixes for tests
brb-nv
committed
4c68348
gate msa-specific part of prepare()
brb-nv
committed
37808fa
[None][chore] pin vLLM upstream commit in vendored file headers
brb-nv
committed
1ba2814

Commits on Aug 1, 2026

Commits on Aug 2, 2026

critical review
brb-nv
committed
ad0ee4a
save minor changes
brb-nv
committed
a949169