david/flash-attention

spogulis no https://github.com/Dao-AILab/flash-attention

Autors	SHA1 Ziņojums	Datums
Tri Dao	e45a46a5b7 [Rotary] Implement GPT-J style (interleaved) rotary	1 gadu atpakaļ
Tri Dao	1e712ea8b0 Implement TensorParallel for MHA	2 gadi atpakaļ
Tri Dao	ca81f32e04 Implement rotary embedding in CUDA	2 gadi atpakaļ