将语言模型控制信号转化为可解释的组件级参数修改
Steer2Edit: From Activation Steering to Component-Level Editing
- 通过激活导向生成诊断信号，实现注意力头与MLP神经元的精准权重编辑
- 在保持性能前提下，安全提升17.2%，真实度提高9.8%，推理长度减少12.2%
- 无需训练，兼容高效推理，适合需要可控且可解释修改的场景
现有控制方法通过推理时的激活干预影响大模型行为，但通常对内部状态进行全局、固定的修改，导致强控制下属性与效用之间的权衡。本文提出Steer2Edit，一种理论严谨、无需训练的框架，将推理时的控制向量转化为组件级秩-1权重编辑的诊断信号。该方法不统一注入控制方向，而是选择性地重新分配行为影响力至单个注意力头和MLP神经元，生成可解释的修改，同时保持标准前向传播并兼容优化后的并行推理。在安全对齐、幻觉缓解和推理效率方面，Steer2Edit在匹配下游性能的前提下，平均实现17.2%的安全提升、9.8%的真实度增益以及12.2%的推理长度缩减。整体上，该方法为表示导向与权重编辑之间提供了原则性桥梁，将控制信号转化为可解释的、无需训练的参数更新。
原文摘要 · Abstract (English)
Steering methods influence Large Language Model behavior by identifying semantic directions in hidden representations, but are typically realized through inference-time activation interventions that apply a fixed, global modification to the model's internal states. While effective, such interventions often induce unfavorable attribute-utility trade-offs under strong control, as they ignore the fact that many behaviors are governed by a small and heterogeneous subset of model components. We propose Steer2Edit, a theoretically grounded, training-free framework that transforms steering vectors from inference-time control signals into diagnostic signals for component-level rank-1 weight editing. Instead of uniformly injecting a steering direction during generation, Steer2Edit selectively redistributes behavioral influence across individual attention heads and MLP neurons, yielding interpretable edits that preserve the standard forward pass and remain compatible with optimized parallel inference. Across safety alignment, hallucination mitigation, and reasoning efficiency, Steer2Edit consistently achieves more favorable attribute-utility trade-offs: at matched downstream performance, it improves safety by up to 17.2%, increases truthfulness by 9.8%, and reduces reasoning length by 12.2% on average. Overall, Steer2Edit provides a principled bridge between representation steering and weight editing by translating steering signals into interpretable, training-free parameter updates. Our code is available at https://github.com/Trustworthy-ML-Lab/Steer2Edit
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。