发现大模型动态追踪实体的可解释绑定机制
A retrieval conditioned rebinding circuit for dynamic entity tracking in large language models
- 通过因果干预识别出检索条件触发的重绑定电路
- Gemma和Llama模型均存在该机制，但表征方式不同
- 适合关注模型内部状态追踪原理的研究者
为正确理解上下文并检索相关信息，大语言模型必须将实体与其属性绑定，并在状态变化时更新这些绑定。我们分析了大模型在动态状态追踪中实现绑定的过程。通过因果干预，我们识别出一种检索条件触发的重绑定机制——一个紧凑的注意力头电路，能够编码与交换相关的绑定信息，并在读出时重新激活。在Gemma和Llama系列模型中，该机制均支持重绑定行为，但其表征签名在不同模型家族中存在差异：Gemma模型中，绑定特征在相关注意力头的查询/键子空间中清晰体现；而Llama模型中，绑定信息主要由键向量承载。总体而言，我们的研究揭示了大模型中一种可解释的上下文依赖状态追踪机制。
原文摘要 · Abstract (English)
To interpret context correctly and retrieve relevant information, large language models must bind entities to their attributes and update these bindings as state changes. We analyze how LLMs implement this binding process in a dynamic state tracking. Using causal interventions, we identify a retrieval conditioned rebinding mechanism, a compact attention head circuit that encodes swap relevant binding information and reinstates it at readout. Across Gemma and Llama models, this circuit supports rebinding behavior, but the representational signature of the mechanism differs across model families. In Gemma models, the binding signature is clearly expressed in the query/key subspaces of the relevant attention heads, whereas in Llama models, the binding information is carried primarily in key vectors. Overall, our results reveal an interpretable mechanism for context dependent state tracking in LLMs.
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。