发现扩散语言模型中间步骤有隐藏正确答案，提出两种方法提升生成质量。
Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Models
- 用多步预测投票选出最一致结果，无需训练
- 通过语义稳定性评分奖励稳定生成，显著提优
- 适合追求高精度文本生成的研究者和开发者
扩散大语言模型通过迭代去噪生成文本，但现有解码策略忽略中间预测。本文发现关键现象：正确答案常出现在中间步骤，却被后续去噪覆盖。为此提出两种互补方法：1）时间自一致性投票，无需训练的测试时解码策略，聚合各去噪步骤预测以选择最一致输出；2）后训练方法时间一致性强化，使用时间语义熵（TSE）作为奖励信号，促进生成过程的语义稳定性。在多个基准上验证有效：仅用负TSE奖励，在Countdown数据集上平均提升24.7%；结合准确率奖励，在GSM8K、MATH500、SVAMP和Countdown上分别实现2.0%、4.3%、6.6%和25.3%的绝对提升。研究揭示了扩散语言模型中时间动态的巨大潜力，并提供两种简单有效的利用工具。
原文摘要 · Abstract (English)
Diffusion large language models (dLLMs) generate text through iterative denoising, yet current decoding strategies discard rich intermediate predictions in favor of the final output. Our work here reveals a critical phenomenon, temporal oscillation, where correct answers often emerge in the middle process, but are overwritten in later denoising steps. To address this issue, we introduce two complementary methods that exploit temporal consistency: 1) Temporal Self-Consistency Voting, a training-free, test-time decoding strategy that aggregates predictions across denoising steps to select the most consistent output; and 2) a post-training method termed Temporal Consistency Reinforcement, which uses Temporal Semantic Entropy (TSE), a measure of semantic stability across intermediate predictions, as a reward signal to encourage stable generations. Empirical results across multiple benchmarks demonstrate the effectiveness of our approach. Using the negative TSE reward alone, we observe a remarkable average improvement of 24.7% on the Countdown dataset over an existing dLLM. Combined with the accuracy reward, we achieve absolute gains of 2.0% on GSM8K, 4.3% on MATH500, 6.6% on SVAMP, and 25.3% on Countdown, respectively. Our findings underscore the untapped potential of temporal dynamics in dLLMs and offer two simple yet effective tools to harness them.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。