Skip to content

llama: in-place recurrent (GDN) state update for multi-sequence decode + lazily zeroed RS buffer - #207

Open
alenizy361 wants to merge 4 commits into
PrismML-Eng:prismfrom
alenizy361:gdn-inplace-state
Open

llama: in-place recurrent (GDN) state update for multi-sequence decode + lazily zeroed RS buffer#207
alenizy361 wants to merge 4 commits into
PrismML-Eng:prismfrom
alenizy361:gdn-inplace-state

Commits

  1. Commits on Sep 19, 2026