National University of Singapore Presents "DMax": A New Paradigm For Diffusion Language Models (dLLMs) Enabling Aggressive Parallel Decoding.

TL;DR: DMax cleverly mitigates error accumulation by reforming decoding as a progressive self-refinement process, allowing the model to correct its own erroneous predictions during generation. We present DMax, a new paradigm for efficient diffusion language models (dLLMs). It mitigates error accumulation in parallel decoding, enabling aggressive decoding parallelism while preserving generation quality. Unlike conventional masked dLLMs that decode through a binary mask-to-token transition, DMax reformulates decoding as a progressive self-refinement from mask embeddings to token embeddings.