Repository navigation

[https://nvbugs/6877539][fix] Fold GQA group into SDPA query axis in MiniMax-M3 decode attention - #20029

Merged

chenfeiz0326 merged 2 commits into

NVIDIA:release/1.3from

tensorrt-cicd:repair-bot-bug6877539-full-20261008T015514-f7aba38b-release-1.3-cb2b0e3d6f12

Oct 12, 2026

Merged

[https://nvbugs/6877539][fix] Fold GQA group into SDPA query axis in MiniMax-M3 decode attention#20029
chenfeiz0326 merged 2 commits into
NVIDIA:release/1.3from
tensorrt-cicd:repair-bot-bug6877539-full-20261008T015514-f7aba38b-release-1.3-cb2b0e3d6f12

Commits on Oct 9, 2026
- perf: rc1_m3_dense_decode_sdpa_per_kv_head_loop (+0.00%)
  trtllm-agent
  committed
  021cd19
Commits on Oct 11, 2026
- [https://nvbugs/6877539][test] Add parity test for MiniMax-M3 dense SDPA decode
  
  chenfeiz0326
  and
  claude
  committed
  782c527