arXiv cs.CL· Jui-Chien Tsou, Hsuan-Chi Liu, Wei-Chin Wang, Yu-Syuan Xu, Chun-Yi Lee·· 12 小时前AI 评分38
TeCoP：通过试探性上下文流水线加速扩散大语言模型解码
TeCoP: Accelerating Diffusion Large Language Model Decoding via Tentative Context Pipelining
AI 导读
TeCoP 是一种免训练解码器，通过交错块并行流水线与草稿-验证机制，将稳定但未提交的预测复用为临时跨块上下文。在 LLaDA 和 Dream 等四个 dLLM 上测试 GSM8K、Minerva MATH 和 MBPP，前向评估次数最多减少 87%，无 KV 缓存时吞吐量提升最高 7.50 倍、有缓存时达 11.77 倍。
正文 · 原文
Abstract:Diffusion large language models (dLLMs) predict multiple token positions per forward pass, yet block-wise decoders serialize generation and repeatedly recompute stable predictions. We attribute this inefficiency to coupling a prediction's readiness for irreversible commitment with its utility as context. We introduce Tentative Context Pipelining (TeCoP), a training-free decoder that decouples these roles through staggered block-parallel pipelining and draft-and-verify. TeCoP reuses stable but uncommitted predictions as provisional cross-block context, with commitment governed by confidence and stability checks. Across four LLaDA and Dream dLLMs on GSM8K, Minerva MATH, and MBPP, TeCoP reduces number of forward evaluations by up to 87% and achieves up to 7.50-fold throughput without KV caching and 11.77-fold with caching over the uncached original decoder. It outperforms Fast-dLLM and DSB in throughput across all 12 model-benchmark settings, with and without caching, while keeping average cached accuracy within one percentage point of the original decoder.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.27583 [cs.CL] |
| (or arXiv:2610.27583v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.27583 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Hsuan-Chi Liu [view email]
[v1]
Tue, 20 Oct 2026 03:48:01 UTC (1,030 KB)
来源：arXiv cs.CL · arxiv.org