动态阈值加速扩散语言模型，提升速度且不牺牲精度。
Beyond Static Cutoffs: One-Shot Dynamic Thresholding for Diffusion Language Models
- 基于单个序列校准阈值，实现快速并行解码。
- 在GSM8K上提速24%，GPQA上提速45%且精度相当。
- 适合追求高效推理的开发者与研究者使用。
掩码扩散语言模型（MDLMs）正逐渐赶上自回归模型，但通常采用固定步数和顺序解码。为加速解码，近期工作如Fast-dLLM通过静态全局置信度阈值实现并行解码，但我们观察到置信度存在显著块级和步骤级波动，且同一数据集内不同输入的置信度轨迹高度相似（余弦相似度高）。受此启发，我们提出一次性动态阈值（OSDT），仅需在单个序列上校准阈值，即可以极低开销应用于后续输入。在GPQA、GSM8K和HumanEval上，OSDT实现了更优的精度-吞吐权衡：在GSM8K上达到最佳精度时提速24%（+24% tokens/s），在GPQA上提速45%且精度相当，在HumanEval上提速50%且精度略有下降。此外，我们的发现表明，可复用的任务级置信度特征为扩散解码的通用算法与系统创新提供了新可能。
原文摘要 · Abstract (English)
Masked diffusion language models (MDLMs) are becoming competitive with their autoregressive counterparts but typically decode with fixed steps and sequential unmasking. To accelerate decoding, recent work such as Fast-dLLM enables parallel decoding via a static global confidence threshold, yet we observe strong block- and step-wise confidence fluctuations and, within a dataset, near-identical confidence trajectories across inputs as measured by cosine similarity. Motivated by these observations, we introduce One-Shot Dynamic Thresholding (OSDT), which calibrates thresholds on a single sequence and applies them to subsequent inputs with negligible overhead. On GPQA, GSM8K, and HumanEval, OSDT attains superior accuracy-throughput trade-offs (+24% tokens/s on GSM8K at the best accuracy, +45% on GPQA with comparable accuracy, and +50% on HumanEval with a modest accuracy gap). Beyond these results, our findings suggest broader opportunities to leverage reusable task-level confidence signatures for more general-purpose algorithmic and systems innovations in diffusion decoding.
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。