Skip to content

Commit b08de1b

Browse files
kimishpatelkimishpatel
committed
Update on "Use unfused SDPA for short sequences (q_len <= 128 or kv_len <= 128)"
ATT Differential Revision: [D96044308](https://our.internmc.facebook.com/intern/diff/D96044308/) [ghstack-poisoned]
2 parents 1083d69 + 274df20 commit b08de1b

4 files changed

Lines changed: 615 additions & 317 deletions

File tree

0 commit comments

Comments
 (0)