Skip to content

Fuse the rotate-half concat into attention and let the K rotary write the KV cache directly - #5379

Open
pfultz2 wants to merge 4 commits into
developfrom
concat-swapped-halves
Open

Fuse the rotate-half concat into attention and let the K rotary write the KV cache directly#5379
pfultz2 wants to merge 4 commits into
developfrom
concat-swapped-halves

Commits

  1. Commits on Oct 12, 2026

  2. Commits on Oct 15, 2026