arXiv cs.CL· Minh-Anh Nguyen, Nguyen T. Ha Lan, Xiyan Fu, Dung D. Le, Wei Lu·· 3 天前AI 评分24
能推理却停不下来：剪枝大型推理模型的失效机理与恢复方法
To Reason but Not to Stop: Understanding and Recovering Pruned Large Reasoning Models
AI 导读
研究发现剪枝大型推理模型（LRM）的精度下降主要源于"停不下来"而非推理能力受损：稀疏度升高后，模型常在正确答案已出现后仍陷入重复循环，无法在 token 预算内终止。基于终止与循环续写之间一致的隐状态差异，作者提出免训练方法 TGAS，在推理时将该方向注入残差流。在 6 个 LRM、3 种剪枝流程及数学与代码基准上，TGAS 改善了终止与整体准确率，并减少重复推理和生成长度。
正文 · 原文
Abstract:Weight pruning can degrade the accuracy of large reasoning models (LRMs), yet it remains unclear whether it damages reasoning itself. We find that much of this degradation stems from a failure to stop rather than to reason. As sparsity increases, pruned LRMs increasingly fail to terminate within the token budget, often after the correct answer has appeared in their reasoning, instead falling into repetitive loops, even though they still recognize where the dense model would stop. We further find that the hidden-state difference between terminating and looping continuations is consistent across pruned checkpoints, suggesting a shared direction associated with committing to an answer. Building on this, we propose Termination-Guided Activation Steering (TGAS), a training-free method that adds this direction to the residual stream at inference time. Across six LRMs, three pruning pipelines, and math and code benchmarks, TGAS improves termination and overall accuracy while reducing repetitive reasoning and generation length, showing that this failure is actionable.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.19934 [cs.CL] |
| (or arXiv:2610.19934v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19934 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Minh Nguyen [view email]
[v1]
Wed, 14 Oct 2026 10:19:15 UTC (738 KB)
来源：arXiv cs.CL · arxiv.org