Skip to content

npu: a 17..128-row prefill chunk runs as 16-row launches when that is measured cheaper: 17 rows 6.95 s -> 1.94 s (fixes #1959) - #1963

Merged
bong-water-water-bong merged 8 commits into
mainfrom
cl-npuperf/prefill-narrow-split
Nov 12, 2026
Merged

bong-water-water-bong merged 8 commits into
mainfrom
cl-npuperf/prefill-narrow-split

Commits

  1. Commits on Nov 10, 2026

  2. Commits on Nov 11, 2026

  3. Commits on Nov 12, 2026