用文本控制大模型内部表示，实现多种行为调节的统一方法
UniSteer: Text-Guided Flow Matching in Activation Space for Versatile LLM Steering
- 通过文本条件学习激活空间的流动规律，生成通用控制路径
- 在3个主流大模型上实现行为控制、事实性调节、细粒度概念操控等任务
- 适合需要灵活、多场景调节大模型行为的研究者和开发者
基于激活的控制通过推理时干预大语言模型（LLM）的内部表征来调节其行为，如人物设定和风格。然而，现有方法通常依赖固定调节方向或特定任务的干预模块，难以适应细粒度概念和组合约束。我们提出UniSteer，一种文本引导的激活空间流匹配模型，从自然语言条件中学习残差流激活的条件分布。无需为每种目标行为单独设计干预，UniSteer在激活空间中学习一个通用条件速度场。推理时，通过部分逆向传输源激活至潜在状态，并在目标文本条件下重构后注入冻结的LLM。同一条件模型还可用于激活空间分类，通过选择重构能量最低的文本标签实现。在三个目标LLM上的实验表明，UniSteer在行为控制、真实性调节、细粒度概念调节、多约束指令遵循及激活空间分类任务中提供了统一接口。
原文摘要 · Abstract (English)
Activation-based control steers large language models (LLMs) by intervening on their internal representations during inference, and has emerged as an effective paradigm for controlling behaviors such as persona and style. However, existing methods often rely on fixed steering directions or task-specific intervention modules, making them difficult to adapt to fine-grained concepts and compositional constraints. We propose UniSteer, a text-guided activation flow matching model that learns a conditional distribution over residual-stream activations from natural-language conditions. Instead of fitting a separate intervention for each target behavior, UniSteer learns a universal conditional velocity field in activation space. At inference time, UniSteer performs flow inversion by partially transporting a source activation toward a latent state and regenerating it under a target textual condition before injecting it back into the frozen LLM. The same conditional model supports activation-space classification by selecting the textual label with the lowest reconstruction energy. Experiments on three target LLMs show that UniSteer provides a unified interface across behavioral control, truthfulness steering, fine-grained concept steering, multi-constraint instruction following, and activation-space classification.
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。