PSA: Sage Attention with 256 header for ideogram avaiable
Sage Attention, a new optimization technique, has been released with a specific header configuration for ideogram. This update claims to offer a 25% speed increase over Flash Attention 2 on a 5090 GPU. The release includes a link to download the necessary files for Windows. AI
IMPACT This optimization could improve inference speeds for image generation models like ideogram, potentially leading to faster workflows for users.