Fetching the paper…

FasterDiT: Towards Faster Diffusion Transformers Training without Architecture Modification · Around