[None][perf] Fuse MiniMax-M3 MSA block selection - #18

Open

peihu-nv wants to merge 18 commits into

brb-nv:feat/m3_with_msafrom

peihu-nv:peihengh/m3-msa-selector

Commits on Jul 16, 2026

[TRTLLM-14019][feat] Add MiniMax-M3 MSA sparse attention backend
brb-nv
committed
14d58d2
[TRTLLM-13969][feat] Support MiniMax M3 for Disaggregated Serving
peihu-nv
committed
7deb4f9

Commits on Jul 17, 2026

[None][bench] Add per-layer and intra-layer nvtx markers
brb-nv
committed
b6a85c8
[None][bench] Add per-layer and intra-layer nvtx markers
brb-nv
authored
5bb9930

Commits on Jul 18, 2026

[None][perf] Deduplicate MSA plan builds in eager mode
brb-nv
committed
d9aafa1
perf: avoid Index-K cache materialization for MSA

peihu-nv
authored and
brb-nv
committed
1a2fc60
[None][perf] Avoid paged MSA K/V materialization during prefill

peihu-nv
authored and
brb-nv
committed
ed7375f
[None][perf] Fuse index-q/index-k projections in MinimaxM3
brb-nv
committed
19db591
[None][perf] Fused GEMM + SwiGLU-OAI
brb-nv
committed
552fe2c

Commits on Jul 19, 2026

Merge pull request #13 from brb-nv/user/brb/swiglu-oai
pcicotti
authored
9c29858

Commits on Jul 20, 2026

Commits on Jul 21, 2026

[None][perf] Fuse MiniMax-M3 MSA block selection
peihu-nv
committed
814c387