跳到正文
arXiv cs.AI· Ambroise Aigueperse, Hugo Buurmeijer, Carmen Amo ALonso, Marco Pavone·· 56 分钟前AI 评分42

Token Commitment 作为控制掩码扩散模型生成的控制面

Token Commitment as a Control Surface for Steering Masked Diffusion Generation

AI 导读

一项 arXiv 研究把 masked diffusion 模型的 token commitment 拆为提交哪些 token 和每步提交多少两个决策,提出反事实目标感知 commitment 与基于目标进度和置信度的自适应预算两种机制。在 LLaDA 家族的语言与视觉语言设置中,两种机制在所有配置下均呈现单调的目标质量权衡,并在 Dream-7B 上复现了行为。

正文 · 原文

View PDF HTML (experimental)

Abstract:Masked diffusion models repeatedly predict provisional tokens before a sampler decides which predictions become permanent. Existing inference-time steering methods primarily modify activations, logits, or denoising trajectories, while treating this commitment policy as a fixed decoding detail. We instead study token commitment as a diffusion-specific control surface and separate it into two decisions: which tokens to commit and how many tokens to commit at each step. For the first, we introduce counterfactual target-aware commitment, which ranks unresolved positions by a mixture of model confidence and leave-one-token-out target contribution. For the second, we introduce calibrated adaptive budgets driven by target progress and confidence. Across language and vision-language settings in the LLaDA family, multiple targets, two activation-steering methods, and an unsteered control, both mechanisms yield empirically monotonic target-quality trade-offs in every configuration we evaluate. Adaptive schedules also recover competitive operating points relative to scorer-free stochastic schedules matched in average denoising cost. Because both mechanisms require only a scalar score evaluated on a provisional completion, they apply unchanged when the target is an automatic proxy for task success rather than an output attribute: NLI entailment for dialogue summarization and CLIPScore for captioning, neither of which corresponds to a single reusable steering direction. We reproduce the behavior of both mechanisms on a second masked-diffusion family, Dream-7B, for one control target and one task-success target. These results characterize commitment--which provisional predictions become irreversible, and how rapidly--as an interpretable control surface for steering masked diffusion generation under the conditions we test.
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2611.21669 [cs.CL]
  (or arXiv:2611.21669v1 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2611.21669

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Ambroise Aigueperse [view email]
[v1] Wed, 11 Nov 2026 22:41:39 UTC (9,172 KB)

来源:arXiv cs.AI · arxiv.org