3 ms·
Reusing Computation in Text-to-Image Diffusion for Efficient Image Generation
- transformi 1y agoInteresting approach! Remind me the early insights that neurons in DNN that capture similar concepts.
- operator-name 1y agoThis looks similar to https://github.com/chengzeyi/ParaAttention https://github.com/chengzeyi/ParaAttention or https://github.com/ali-vilab/TeaCache https://github.com/ali-vilab/TeaCache. It’s a shame they don’t compare against or mention them.