Query regarding attention kernels

#8
by ashwin-r - opened

Is there any information what these optimizations are? I'm curious to know if its using flashattentionv2 for both self-attention in unet and cross-attention for text-conditioning. Thanks!

Sign up or log in to comment