decode bf16 smallm: support arbitrary 1<heads_per_group<=8 via direct Q/Y GMEM when TMA unsuitable#37

Open

Religious-J wants to merge 1 commit intoTencent:mainfrom

Religious-J:feat/decode_push

Commits on Apr 2, 2026

decode bf16 smallm: support arbitrary 1<heads_per_group<=8 via direct Q/Y GMEM when TMA unsuitable

Religious-J
authored and
jiangqianchen
committed