Break the Block: Dynamic-size Reasoning Blocks for Diffusion Large Language Models via Monotonic Entropy Descent with Reinforcement Learning

Recent diffusion large language models (dLLMs) have demonstrated both effectiveness and efficiency in reasoning via a block-based semi-autoregressive generation paradigm. Despite their progress, the fixed-size block generations remain a critical bottleneck for effective and coherent reasoning. (I) From a global perspective, different reasoning tasks would correspond to different optimal decoding block sizes, which makes a "one-size-fits-all" assumption ineffective. (II) Even within a single reasoning task, the rigid block partitioning would break the logical flow and reduce reasoning coherence. Through empirical observations, we reveal that, for block-wise entropy, incorrect reasoning exhibits a fluctuating and unsteady trend between blocks, while the correctly generated tasks follow a consistent descending paradigm. Therefore, this paper proposes b1, a novel post-training framework that learns dynamic-size reasoning blocks via a Monotonic Entropy Descent objective with reinforcement learning to enhance reasoning coherence. b1 integrates seamlessly as a plug-and-play module with existing dLLM's post-training algorithms. Extensive experiments across various reasoning benchmarks showcase b1's consistent improvement over fixed-size block baselines. Our code has been released at https://github.com/YanJiangJerry/Block-R1.

Recent diffusion large language models (dLLMs) improve reasoning efficiency through block-based semi-autoregressive generation. However, fixed-size blocks remain a bottleneck for effective and coherent reasoning. Different reasoning tasks may require different optimal block sizes, making a “one-size-fits-all” assumption ineffective. Even within one task, rigid block partitioning can break the logical flow and reduce reasoning coherence. This paper proposes b1, a post-training framework that learns dynamic-size reasoning blocks for dLLMs. b1 is guided by Monotonic Entropy Descent, based on the observation that correct reasoning usually follows a consistent descending trend in block-wise entropy, while incorrect reasoning shows fluctuating and unsteady trends. Experiments across various reasoning benchmarks show that b1 consistently improves over fixed-size block baselines and can be integrated as a plug-and-play module with existing dLLM post-training algorithms.