Different attention for each block ๐
retained text
Omg, so apparently in near future we can get SLA's 2x performance boost without loosing too much of quality
document fetched 2026-10-01 00:35 ยท canonical https://reddit.com/r/StableDiffusion/comments/1wtzhwl/different_attention_for_each_block