arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2605.05892cs.CLcs.LG
收藏

超越转向向量:基于流的激活转向用于推理时间干预

Beyond Steering Vector: Flow-based Activation Steering for Inference-Time Intervention

Georgia Institute of Technology(佐治亚理工学院)

Zehao Jin, Ruixuan Deng, Junran Wang, Xinjie Shen, Chao Zhang

AI总结:

本文提出FLAS,一种基于流的激活转向方法,通过学习通用概念条件化的速度场来运输未转向的激活到转向的激活,从而在AxBench上超越了提示方法。

摘要:

激活转向已成为一种有前景的替代方法,用于在推理时间通过修改中间表示而保持模型参数冻结来控制语言模型行为。然而,如AxBench所示的大规模评估表明,现有转向方法往往被简单的上下文提示所超越,并且在未见过的概念上泛化不佳。我们假设这些限制来自于先前方法中未验证的简化假设,这些假设通常限制转向干预为固定、单步、位置不变的转换。我们提出了FLAS(基于流的激活转向),它学习了一个通用、概念条件化的速度场v_t(h,t,c),该速度场在不依赖这些假设的情况下将未转向的激活传输到转向的激活。在AxBench上,FLAS是首个在不进行每种概念调优的情况下一致超越提示方法的学得方法,达到了Gemma-2-2B-IT的保持和谐均值为1.015和Gemma-2-9B-IT的保持和谐均值为1.113。对学习到的流的分析显示出了弯曲、多步、令牌变化的轨迹,这表明以前关于激活空间几何的假设可能不完整。

英文摘要:

Activation steering has emerged as a promising alternative for controlling language-model behavior at inference time by modifying intermediate representations while keeping model parameters frozen. However, large-scale evaluations such as AxBench show that existing steering methods are often outperformed by simple in-context prompting and generalize poorly to unseen concepts. We hypothesize that these limitations arise from unvalidated simplifying assumptions shared across prior methods, which typically restrict steering interventions to fixed, single-step, position-invariant transforms. We propose FLAS (Flow-based Activation Steering), which learns a general, concept-conditioned velocity field $v_t(h,t,c)$ that transports unsteered activations to steered ones without relying on these assumptions. On AxBench, FLAS is the first learned method to consistently outperform prompting, reaching held-out harmonic means of $1.015$ on Gemma-2-2B-IT and $1.113$ on Gemma-2-9B-IT without per-concept tuning. Analysis of the learned flow shows curved, multi-step, token-varying trajectories, which suggests that previous hypotheses on activation space geometry might be incomplete.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑