Training deep neural networks one block at a time cuts memory

Original: Diffusion Blocks

Source: sakana.ai ↗