What DeepCache Accidentally Tells Us About How Diffusion Models Think
DeepCache is sold as a 2-7× speedup for diffusion models. The more interesting story is what its core finding — that 80% of internal computation is redundant across denoising steps — reveals about what these models are actually doing.
diffusion-modelsstable-diffusionpaper-walkthroughdeep-learninginterpretability
·20 min read·Deepak Reddy Gathpa
