动态调整去噪过程中的掩码策略，显著提升扩散语言模型推理速度
STaRR: Spatial-Temporal Token-Dynamics-Aware Responsive Remasking for Diffusion Language Models
- 基于令牌置信度的时空演化动态调整掩码阈值
- 平均提速4.1倍，最高达8.9倍，精度基本不变
- 无需训练，适合部署在资源受限的推理场景
扩散语言模型（DLMs）通过迭代去噪实现并行解码，其中重掩码策略对推理速度与输出质量平衡至关重要。现有方法多依赖静态置信度阈值，忽视了令牌置信度的时空动态变化，导致不必要的重掩码。我们提出无需训练的STaRR框架，根据令牌置信度演化动态调整重掩码决策。STaRR引入时序方差与空间偏离两个指标，实现细粒度、步进式动态阈值设定，并结合响应性优化提升可扩展性与鲁棒性。实验表明，STaRR在保持相近准确率的前提下，平均加速4.1倍，最高可达8.9倍。
原文摘要 · Abstract (English)
Diffusion Language Models (DLMs) enable parallel decoding via iterative denoising, where remasking strategies play a critical role in balancing inference speed and output quality. Existing methods predominantly rely on static confidence thresholds, overlooking the spatial-temporal dynamics of token confidence, causing unnecessary remasking. We propose Spatial-Temporal Token-Dynamics-Aware Responsive Remasking (STaRR), a training-free framework that dynamically adapts remasking decisions based on token confidence evolution. STaRR introduces two metrics, temporal variance and spatial deviance, to guide fine-grained, step-wise dynamic thresholding. We further introduce a step-wise dynamic thresholding strategy, further enhanced with responsiveness optimizations for scalability and robustness. Experiments show that STaRR achieves an average speedup of 4.1 and up to 8.9 while maintaining comparable accuracy.
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。