Reusing Computation in Text-to-Image Diffusion for Efficient Image Generation
41 pointsby gidellav93 days ago2 comments(arxiv.org)
Comments (2)
operator-name•89 days ago
This looks similar to https://github.com/chengzeyi/ParaAttention or https://github.com/ali-vilab/TeaCache.
It’s a shame they don’t compare against or mention them.
transformi•89 days ago
Interesting approach! Remind me the early insights that neurons in DNN that capture similar concepts.