SageAttention

Quantized Attention that achieves speedups of 2.1-3.1x and 2.7-5.1x compared to FlashAttention2 and xformers, respectively, without lossing end-to-end metrics across various models.

Pinokio Apps Using This Repo55

unoFeaturedgithub.com/pinokiofactory/unopip install from git in torch.js0 check-ins · updated 5mo agoWan 2.1github.com/remphanstar/For-Geminipip install from git in torch.js17 check-ins · updated 2mo agoFooocus-APIgithub.com/6Morpheus6/Fooocus-APIpip install from git in torch.js3 check-ins · updated 9mo ago · Owner @morpheus+52 more

Community tagsLoading...