Progressive Refinement Regulation for Accelerating Diffusion Language Model Decoding

arXiv:2603.04514v1 Announce Type: new Abstract: Diffusion language models generate text through iterative denoising under a uniform refinement rule applied to all tokens.