PAPO

Back on Track: Aligning Rewards and States for Reasoning in Diffusion Large Language Models

arXiv:2606.08501v1 Announce Type: new Abstract: Reinforcement learning (RL) holds immense promise for enhancing the reasoning capabilities of diffusion large language models (dLLMs). However, progress is fundamentally constrained by a dual misalignment between authentic generation trajectory and the gradient update process: (i) Process-reward misalignment. Sparse, terminal rewards are indiscriminately assigned to all intermediate steps of the generation process, failing to provide...