AI RESEARCH
Dependency-Guided Parallel Decoding in Discrete Diffusion Language Models
arXiv CS.CL
•
ArXi:2604.02560v1 Announce Type: new Discrete diffusion language models (dLLMs) accelerate text generation by unmasking multiple tokens in parallel. However, parallel decoding We propose DEMASK (DEpendency-guided unMASKing), a lightweight dependency predictor that attaches to the final hidden states of a dLLM. In a single forward pass, it estimates pairwise conditional influences between masked positions. Using these predictions, a greedy selection algorithm identifies positions with bounded cumulative dependency for simultaneous unmasking.