arXiv cs.CL· Xuning He, Qingyuan Shi, Zinan Sheng, Yongding Tao, Huanyu Liu, Ge Li, Xue Jiang, Yihong Dong·· 2 天前AI 评分24
Archer：为扩散语言模型高效回滚自适应复用缓存隐藏状态
Archer: Adaptive Reuse of Cached Hidden States for Efficient Rollback in Diffusion Language Models
AI 导读
研究者提出免训练缓存方法 Archer，通过在每个更新步重算可变回复、仅在有限回复状态邻域内复用 prompt 的 K/V，实现扩散语言模型的高效回滚。在代码生成基准上，Archer 取得 33.63% 的最佳平均性能与 2.57x 平均加速，单基准最高 2.95x；在非代码基准上相对 LLaDA-Confidence 最高加速 4.33x 并提升平均准确率。代码已公开。
正文 · 原文
Abstract:Diffusion language models (DLMs) can revise earlier predictions as context evolves, but rollback requires costly full-sequence recomputation. Conventional key-value (KV) caching assumes immutable historical states and is difficult to reconcile with such revision. We introduce Adaptive Reuse of Cached Hidden States for Efficient Rollback (Archer), a training-free caching method for rollback-capable DLMs. Archer recomputes the mutable response at every update while reusing prompt K/V within a bounded response-state neighborhood. This asymmetric boundary amortizes prompt computation without restricting rollback. Bounded reuse also delays feedback from tentative tokens while keeping response revision active. Our analysis characterizes the cache boundary, bounds state-dependent approximation error, and gives a decoder-margin condition for preserving full-refresh decisions. On code-generation benchmarks, Archer achieves the best mean performance of 33.63% with a 2.57x mean speedup, reaching 2.95x speedup on a single benchmark. On non-code benchmarks, Archer achieves up to 4.33x speedup over LLaDA-Confidence while improving average accuracy. Controlled analyses support the delayed prompt feedback mechanism. Our code is available at this https URL
| Comments: | 26 pages, 7 figures, 12 tables; revised manuscript, updated author list and affiliations |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2608.08086 [cs.CL] |
| (or arXiv:2608.08086v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2608.08086 arXiv-issued DOI via DataCite |
Submission history
From: Xuning He [view email]
[v1]
Sat, 8 Aug 2026 12:13:10 UTC (1,465 KB)
[v2]
Tue, 11 Aug 2026 12:16:53 UTC (1,465 KB)
[v3]
Thu, 15 Oct 2026 16:56:51 UTC (963 KB)
来源：arXiv cs.CL · arxiv.org