通过微调注意力头权重，解决大模型推理过短问题，提升数学解题准确率。
ThinkEdit: Interpretable Weight Editing to Mitigate Overly Short Thinking in Reasoning Models
- 定位导致推理过短的注意力头，仅修改0.2%参数即可干预
- 在多个数学数据集上，短推理准确率提升6.39%，整体提升3.34%
- 揭示了推理长度由隐藏空间线性方向控制，适合关注模型可解释性的研究者
近期研究表明，带有思维链（CoT）推理的大语言模型展现出强大的问题求解能力。然而，本文发现这些模型时常生成过短的推理过程，导致即使在简单数学题上性能下降。我们分析了推理长度如何嵌入推理模型的隐藏表示中，并发现其受表征空间中一个线性方向的控制，可通过沿该方向引导模型引发过短推理。基于此，我们提出ThinkEdit，一种轻量级权重编辑方法：首先识别约4%主导短推理行为的注意力头，再修改其输出投影权重以消除该方向影响。仅调整模型0.2%参数，便显著减少过短推理现象，在短推理输出上实现+6.39%准确率提升，多数学基准整体提升+3.34%。研究为理解推理长度在大模型中的机制控制提供了新见解，并展示了细粒度干预对提升推理质量的潜力。代码已开源。
原文摘要 · Abstract (English)
Recent studies have shown that Large Language Models (LLMs) augmented with chain-of-thought (CoT) reasoning demonstrate impressive problem-solving abilities. However, in this work, we identify a recurring issue where these models occasionally generate overly short reasoning, leading to degraded performance on even simple mathematical problems. Specifically, we investigate how reasoning length is embedded in the hidden representations of reasoning models and its impact on accuracy. Our analysis reveals that reasoning length is governed by a linear direction in the representation space, allowing us to induce overly short reasoning by steering the model along this direction. Building on this insight, we introduce ThinkEdit, a simple yet effective weight-editing approach to mitigate the issue of overly short reasoning. We first identify a small subset of attention heads (approximately 4%) that predominantly drive short reasoning behavior. We then edit the output projection weights of these heads to remove the short reasoning direction. With changes to only 0.2% of the model's parameters, ThinkEdit effectively reduces overly short reasoning and yields notable accuracy gains for short reasoning outputs (+6.39%), along with an overall improvement across multiple math benchmarks (+3.34%). Our findings provide new mechanistic insights into how reasoning length is controlled within LLMs and highlight the potential of fine-grained model interventions to improve reasoning quality. Our code is available at: https://github.com/Trustworthy-ML-Lab/ThinkEdit
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。