AI RESEARCH
Dynamic Infilling Anchors for Format-Constrained Generation in Diffusion Large Language Models
arXiv CS.AI
•
ArXi:2606.04535v1 Announce Type: cross Diffusion large language models (dLLMs) offer bidirectional attention and parallel generation, enabling them to exploit global context and naturally format-constrained tasks like parseable JSON or reasoning templates. While straightforward fixed anchors can enforce such constraints, they often impose rigid spans, leading to truncated reasoning or redundant content. To overcome this, we propose Dynamic Infilling Anchors (DIA), a