Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Dynamic Chunking for Diffusion Language Models

About

Block discrete diffusion language models factorize a sequence autoregressively over fixed-size positional blocks, decoupling within-block parallel denoising from across-block conditioning. We argue that this rigid partition wastes structure already present in the sequence: blocks defined by position rather than by content separate semantically coherent tokens and group unrelated ones together. We introduce the \textbf{D}ynamic \textbf{C}hunking \textbf{D}iffusion \textbf{M}odel (DCDM), which replaces positional blocks with content-defined semantic chunks. At its core is Chunking Attention, a differentiable layer that routes tokens into $K$ clusters parameterized by learnable subspaces and shaped end-to-end by the diffusion objective. The resulting cluster assignments induce a chunk-causal attention mask under which a discrete diffusion denoiser factorizes the sequence likelihood autoregressively over semantic chunks, strictly generalizing block discrete diffusion. On downstream benchmarks at parameter scales up to 1.5B, DCDM consistently improves over both unstructured and positional-block diffusion baselines, with the advantage stable across scales and visible early in training.

Yichen Zhu, Xiaoming Shi, Peng Zhao, Weiyu Chen, Debing Zhang, James Kwok• 2026

Related benchmarks

TaskDatasetResultRank
Commonsense ReasoningWinoGrande--
1442
Commonsense ReasoningHellaSwag
HellaSwag Accuracy51.28
711
Commonsense ReasoningPIQA
Accuracy66.32
213
Code GenerationHumanEval
pass@13.01
145
Language ModelingPTB (val)
Perplexity87.82
107
Question AnsweringMMLU
Accuracy26.1
74
Language ModelingLM1B (val)
Perplexity66.1
67
Language ModelingWikiText (val)
Perplexity31.94
62
Language ModelingLAMBADA (val)
Perplexity46.43
39
Language ModelingAG News (val)
Perplexity59.85
36
Showing 10 of 16 rows

Other info

Follow for update