扩散语言模型并行解码的实测耦合方法

Measured Coupling for Parallel Decoding in Diffusion Language Models

AI 导读

扩散语言模型并行解码提出用“耦合”替代均值场解码的相似度代理，即衡量提交一个 token 会多大程度改变模型在另一位置的置信度。该方法在主前向传播中用廉价探针测得耦合，无需训练或额外模型，可直接接入原有均值场规则。在三个模型和四个基准上，其所需步数远少于均值场解码，多数设置下速度更快（含 key-value cache 场景），平均精度代价为 1.1 分。

正文 · 原文

View PDF HTML (experimental)

Abstract:Diffusion language models can write many tokens per step, but must choose which tokens are safe to commit together. Mean-field decoding penalizes pairs of positions whose predictions look alike. We replace this proxy with the coupling itself: how much committing one token changes the model's confidence at another. Cheap probes inside the main forward pass measure it, with no training or extra model, and it plugs into the same mean-field rule. On three models and four benchmarks, our method needs far fewer steps than mean-field decoding and is faster in most settings, also with a key-value cache, at an average accuracy cost of 1.1 points. Shuffling the couplings or replacing them with a uniform bonus loses this gain.

Subjects:	Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG); Performance (cs.PF)
Cite as:	arXiv:2611.08307 [cs.CL]
	(or arXiv:2611.08307v1 [cs.CL] for this version)
	https://doi.org/10.48550/arXiv.2611.08307 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Tal Oved [view email]
[v1] Thu, 5 Nov 2026 17:10:03 UTC (257 KB)

来源：arXiv cs.CL · arxiv.org