ASA: Activation Steering for Tool-Calling Domain Adaptation
This addresses the practical challenge of scalable domain adaptation for real-world LLM agents dealing with frequent tool interface changes.
The paper tackles the problem of efficiently adapting LLM agents to rapidly evolving tool ecosystems without costly retraining, proposing Activation Steering Adapter (ASA) as a lightweight, inference-time method that achieves LoRA-comparable adaptation with substantially lower overhead and strong cross-model transferability.
For real-world deployment of general-purpose LLM agents, the core challenge is often not tool use itself, but efficient domain adaptation under rapidly evolving toolsets, APIs, and protocols. Repeated LoRA or SFT across domains incurs exponentially growing training and maintenance costs, while prompt or schema methods are brittle under distribution shift and complex interfaces. We propose \textbf{Activation Steering Adapter (ASA}), a lightweight, inference-time, training-free mechanism that reads routing signals from intermediate activations and uses an ultra-light router to produce adaptive control strengths for precise domain alignment. Across multiple model scales and domains, ASA achieves LoRA-comparable adaptation with substantially lower overhead and strong cross-model transferability, making it ideally practical for robust, scalable, and efficient multi-domain tool ecosystems with frequent interface churn dynamics.