AI RESEARCH

Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models

arXiv CS.AI

ArXi:2506.19037v4 Announce Type: replace-cross Masked diffusion language models (MDLMs) promise fast, non-autoregressive text generation, yet existing samplers, which pick tokens to unmask based on model confidence, ignore interactions when unmasking multiple positions in parallel and effectively reduce to slow, autoregressive behavior.