Tag
ViRDM is a method for few-step causal video generation using representation distribution matching, achieving better results than DMD-based baselines with efficient training on single or multiple GPUs.