david/flash-attention

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

Autors	SHA1 Ziņojums	Datums
Tri Dao	5458c78e6d Remove sink token	4 dienas atpakaļ
Tri Dao	9f313c7073 Move functions getting number of m/n blocks to a separate file	3 nedēļas atpakaļ
Tri Dao	4e8496a781 Support Qv	3 nedēļas atpakaļ
Tri Dao	2a204125ae Implement HeadDim_V != HeadDim_QK, support hdimQK=192, hdimV=128	1 mēnesi atpakaļ
Tri Dao	84f1287e42 Rename bool_constant<true> to true_type, same w bool_constant<false>	1 mēnesi atpakaļ
Tri Dao	7a802796e1 Big refactor and update	1 mēnesi atpakaļ