ASA：面向工具调用智能体的免训练表征工程

摘要

在大模型智能体适应领域特定工具调用方面，演化中的接口仍存在显著脆弱性。提示与模式工程虽易于部署，但在分布偏移和严格解析器下往往表现脆弱；而持续的参数高效微调虽能提升可靠性，却需承担训练维护成本及潜在遗忘风险。我们发现了一种关键的"惰性智能体"失效模式：尽管从中层激活状态能近乎完美解码工具使用必要性，模型仍保守地回避进入工具模式，这揭示了表征与行为之间的脱节。我们提出激活引导适配器（ASA），这是一种免训练、推理时控制器，通过单次中层干预，采用路由器调制的引导向量混合方案与探针指导的符号门控机制，在放大真实意图的同时抑制误触发。在Qwen2.5-1.5B模型上的MTU-Bench测试表明，ASA仅需约20KB可移植资源且无需权重更新，就将严格工具使用F1值从0.18提升至0.50，同时将误报率从0.15降至0.05。

English

Adapting LLM agents to domain-specific tool calling remains notably brittle under evolving interfaces. Prompt and schema engineering is easy to deploy but often fragile under distribution shift and strict parsers, while continual parameter-efficient fine-tuning improves reliability at the cost of training, maintenance, and potential forgetting. We identify a critical Lazy Agent failure mode where tool necessity is nearly perfectly decodable from mid-layer activations, yet the model remains conservative in entering tool mode, revealing a representation-behavior gap. We propose Activation Steering Adapter (ASA), a training-free, inference-time controller that performs a single-shot mid-layer intervention and targets tool domains via a router-conditioned mixture of steering vectors with a probe-guided signed gate to amplify true intent while suppressing spurious triggers. On MTU-Bench with Qwen2.5-1.5B, ASA improves strict tool-use F1 from 0.18 to 0.50 while reducing the false positive rate from 0.15 to 0.05, using only about 20KB of portable assets and no weight updates.

ASA：面向工具调用智能体的免训练表征工程

ASA: Training-Free Representation Engineering for Tool-Calling Agents

摘要

Support