warp-attention

Warp attention: hardware efficient implementation of scaled dot product attention.


Keywords
transformers, attention, scaled, dot, product, pytorch
Install
pip install warp-attention==0.0.1