Repository navigation

perf(kernel): split triton decode attention by context length and sm count - #729

Merged

jason-fxz merged 2 commits into

mainfrom

perf/triton-attn-kv-splits

Oct 17, 2026

Merged

perf(kernel): split triton decode attention by context length and sm count#729
jason-fxz merged 2 commits into
mainfrom
perf/triton-attn-kv-splits

Commits on Oct 17, 2026
- perf(kernel): split triton decode attention by context length and sm count
  jason-fxz
  committed
  855aedd
- perf(kernel): merge triton decode splits a tile at a time
  jason-fxz
  committed
  2bd6c33