Where and When to Commit: Candidate-Aware Decoding for Diffusion Language Models

📄 arXiv: 2607.28166v1 📥 PDF

作者: Chia-Ming Lee, Ming-Ching Chang, Xin Li, Yu-Lun Liu, Chih-Chung Hsu

分类: cs.CL, cs.AI

发布日期: 2026-07-30

备注: Code is available at https://github.com/ming053l/LATCH-dLLM

💡 一句话要点

提出候选感知解码框架以优化扩散语言模型的生成效率

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 扩散语言模型 提前退出 候选感知 生成效率 自然语言处理 实时应用 速度提升

📋 核心要点

  1. 现有的提前退出机制无法有效处理长链思维输出，导致生成效率低下。
  2. 提出了一种候选感知的提前退出框架LATCH，通过置信度验证和局部规则优化生成过程。
  3. 在11个任务的评估中，LATCH的准确率与全解码相差不超过2.0个百分点，同时实现了9.3-17.8倍的速度提升。

📝 摘要（中文）

扩散语言模型（DLMs）在每个去噪步骤中提供临时预测，为生成时的提前退出创造了机会。现有的提前退出机制依赖于固定区域的置信度统计或调度依赖规则，这些方法在长链思维输出中表现不佳。本文提出了一种训练无关的候选感知提前退出框架，称为LATCH（Localized Acceleration with Tracked-Candidate Halting），通过置信度验证和局部规则来优化生成过程。LATCH在11个任务的零-shot设置下进行评估，结果显示其在保持高准确率的同时，实现了显著的速度提升。

🔬 方法详解

问题定义：本文旨在解决扩散语言模型在生成过程中提前退出的效率问题。现有方法依赖于固定区域的置信度统计，导致在长链思维输出中表现不佳，无法有效判断何时停止解码。

核心思路：提出了一种候选感知的提前退出框架LATCH，分离了置信度验证与解码过程中的承诺决策，使每个决策基于其自身证据进行判断。

技术框架：LATCH框架包括两个主要模块：置信度验证承诺（CVC）和块级提前承诺（BWEC）。CVC负责验证序列停止的时机，而BWEC则应用局部规则加速非最终块的解码。

关键创新：LATCH的创新在于不需要后缀提示构建，且对提示格式敏感，能够在不依赖于训练的情况下实现高效解码。

关键设计：CVC模块使用确定性解析器动态提取候选范围，确保置信度和稳定性得到验证；BWEC模块则通过局部规则加速非最终块的解码过程。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

LATCH在11个任务的零-shot设置下评估，准确率与全解码相差不超过2.0个百分点，且在短答案任务上实现了9.3-17.8倍的速度提升，在长推理任务上实现了2.0-3.3倍的加速，显示出其优越的性能。

🎯 应用场景

该研究的潜在应用领域包括自然语言处理中的对话系统、文本生成和问答系统等。通过提高生成效率，LATCH能够在实时应用中显著提升用户体验，具有广泛的实际价值和未来影响。

📄 摘要（原文）

Diffusion language models (DLMs) expose a provisional prediction at every denoising step, creating an opportunity for generation-time early exit that stops decoding before the schedule is exhausted. Existing early-exit gates decide termination from fixed-region confidence statistics or schedule-dependent rules, evidence too coarse for a decision that freezes every remaining position at once, so they fire prematurely on long chain-of-thought outputs whose answers stabilize only near the end. Adaptive sampling, the other axis of training-free acceleration, paces how quickly positions commit while decoding continues but never verifies that the output itself has stabilized. We introduce a training-free, candidate-aware early-exit framework that keeps the two axes separate and matches each decision to evidence of its own scope. Confidence-Verified Commit (CVC) governs when the sequence may stop by verifying confidence and sustained argmax stability over the dynamically extracted candidate span using a deterministic parser specified from each task's output format. Block-Wise Early Commit (BWEC) governs where to accelerate by applying a cheaper local rule to non-final blocks, while leaving the final block and global termination under CVC. We refer to their combination as LATCH (Localized Acceleration with Tracked-Candidate Halting). Unlike prior methods, LATCH needs no suffix-prompt construction; it is prompt-anchor-free but format-aware. We evaluate LATCH end to end on 11 tasks under zero-shot settings using LLaDA and Dream. LATCH stays within 2.0 percentage points of full-decoding accuracy across all 22 evaluation settings, with one frozen hyperparameter set that transfers cross-backbone untuned, while achieving end-to-end TPS speedups of 9.3-17.8x on short-answer tasks and 2.0-3.3x on long-reasoning tasks.