Make Stable Diffusion up to 100% faster with Memory Efficient Attention

That's the explanation behind https://news.ycombinator.com/item?id=32985716, nice work