AI RESEARCH
Diffusion Language Models Are Natively Length-Aware
arXiv CS.LG
•
ArXi:2603.06123v1 Announce Type: cross Unlike autoregressive language models, which terminate variable-length generation upon predicting an End-of-Sequence (EoS) token, Diffusion Language Models (DLMs) operate over a fixed maximum-length context window for a predetermined number of denoising steps. However, this process is independent of the required response length, resulting in computational waste for the majority of short responses common in reasoning and chat tasks.