arXiv cs.CL· Yuzhe Chen, Jiale Cao, Xuyang Liu, Jin Xie, Aiping Yang, Yanwei Pang·· 10 小时前AI 评分32
STDec：面向 dLLM 的时空稳定性引导解码方法
STDec: Spatio-Temporal Stability Guided Decoding for dLLMs
AI 导读
研究者提出免训练的时空稳定性引导解码方法 STDec，通过聚合邻近 token 的解码状态动态生成 token 自适应阈值，并对去噪步骤中预测 ID 保持一致的 token 放宽阈值。在文本推理与多模态理解基准上，STDec 在生成分数相当的情况下提升推理吞吐，在 MBPP 上配合 LLaDA 最高实现 14.17 倍加速，且兼容基于缓存的加速方法。
正文 · 原文
Abstract:Diffusion Large Language Models (dLLMs) have achieved rapid progress, viewed as a promising alternative to the autoregressive paradigm. However, most dLLM decoders still adopt a global confidence threshold, and do not explicitly model local context from neighboring decoded states or temporal consistency of predicted token IDs across steps. To address this issue, we propose a spatio-temporal stability guided decoding approach, named STDec. We observe strong spatio-temporal stability in dLLM decoding: newly decoded tokens tend to lie near decoded neighbors, and their predicted IDs often remain consistent across several denoising steps. Inspired by this stability, our STDec introduces spatial-aware decoding and temporal-aware decoding. The spatial-aware decoding dynamically generates the token-adaptive thresholds by aggregating the decoded states of nearby tokens. The temporal-aware decoding relaxes the decoding thresholds for tokens whose predicted token IDs remain consistent over denoising steps. Our STDec is training-free and remains compatible with cache-based acceleration methods. Across textual reasoning and multimodal understanding benchmarks, STDec achieves improved inference throughput while maintaining comparable generation score. Notably, on MBPP with LLaDA, STDec achieves up to 14.17$\times$ speedup with a comparable score. The homepage is available at this https URL.
| Comments: | Homepage: this https URL |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2604.06330 [cs.CL] |
| (or arXiv:2604.06330v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2604.06330 arXiv-issued DOI via DataCite |
Submission history
From: Yuzhe Chen [view email]
[v1]
Tue, 7 Apr 2026 18:13:31 UTC (760 KB)
[v2]
Tue, 10 Nov 2026 15:39:11 UTC (1,143 KB)
来源：arXiv cs.CL · arxiv.org