david/flash-attention

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

Autors	SHA1 Ziņojums	Datums
Tri Dao	3f7d5786ba Pass alibi slopes to flash_attn_with_kvcache during generation	1 gadu atpakaļ
Tri Dao	2c7d7b7396 Implement norm head for Baichuan2	1 gadu atpakaļ
Tri Dao	c3b2196652 Add Alibi to MHA, test with Baichuan-13B	1 gadu atpakaļ
GAOXinyu	a8c35b4f57 FEAT: add codes which supporting for baichuan-inc/Baichuan-7B (#425)	1 gadu atpakaļ