Hierarchical Masked Diffusion Language Models

A new arXiv paper (2610.29255v1) introduces the Hierarchical Masked Diffusion Language Model (HMDLM), which assigns each token position a hierarchy level so higher-priority positions are revealed earlier during parallel decoding. On LM1B, HMDLM reduces perplexity by more than 1.5 points over the Anchored Diffusion Language Model (ADLM) baseline; on WikiHow it cuts perplexity by up to 4.4 points versus ADLM and up to 7 points versus MDLM, and an LLM judge preferred HMDLM generations more than 2x as often as the baseline. On Sudoku puzzles requiring multi-step reasoning, HMDLM improves solution accuracy by up to 8.6 absolute percentage points over existing diffusion baselines.

arXiv:2610.29255v1 Announce Type: new Abstract: Masked diffusion language models have emerged as an alternative to autoregressive models, enabling parallel token generation. However, standard masked diffusion language models use a shared noise schedule across positions without explicitly encoding which positions should be resolved first. Many tasks exhibit a natural priority among positions: document titles and section headings can guide the generation of body text, while Sudoku cells determined directly from the initial clues can help resolve more difficult cells. To incorporate this notion of priority or hierarchy among token positions, we introduce the Hierarchical Masked Diffusion Language Model HMDLM , which augments each token position with a hierarchy level available as part of the training data, and requires zero or minimal additional annotations. It factors generation into two passes: the first predicts a hierarchy distribution over masked positions, the second predicts token distributions conditioned on it. Giving each hierarchy level its own noise schedule induces a soft, controllable preference for revealing higher-priority positions earlier while keeping decoding fully parallel. Our formulation is mathematically grounded, and we re-derive the forward and reverse processes and the NELBO for this factorization. Our evaluation shows that for the task of language generation on LM1B, HMDLM reduces perplexity by more than 1.5 points over Anchored Diffusion Language Model ADLM , a recently proposed strong diffusion baseline; on WikiHow, we see a drop in perplexity of up to 4.4 pts on ADLM and up to 7 pts against MDLM. An LLM judge on WikiHow prefers HMDLM generations the two variants combined more than $2\times$ as often as the baseline. On the task of solving Sudoku puzzles requiring multi-step reasoning, it improves solution accuracy by up to $8.6\%$ absolute points over existing diffusion baselines.