AIbase
Product LibraryTool Navigation

SageAttention

Public

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.

Creat2024-10-03T17:33:18
Update2025-03-27T08:33:32
1.2K
Stars
3
Stars Increase