Tag
This paper introduces speculative draft trees to accelerate diffusion model sampling by enriching candidate drafts and using greedy rejection sampling, achieving up to 8.3% speedup over baseline methods.