david/flash-attention

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

Autors	SHA1 Ziņojums	Datums
Tri Dao	43ab0b5205 Mention that some CUDA extensions have only been tested on A100s	2 gadi atpakaļ
Tri Dao	2e33fc8e36 Add GPT and ViT models	2 gadi atpakaļ
Tri Dao	fa6d1ce44f Add fused_dense and dropout_add_layernorm CUDA extensions	2 gadi atpakaļ