The Atlas of In-Context Learning: How Attention Heads Shape In-Context Retrieval Augmentation
This addresses the problem of transparency and safety in language models for AI researchers, but it is incremental as it builds on existing attribution methods.
The paper tackled the unclear inner workings of in-context retrieval augmentation in large language models for question answering, revealing specialized attention heads and their roles, with insights enabling knowledge source tracing for safer models.
Large language models are able to exploit in-context learning to access external knowledge beyond their training data through retrieval-augmentation. While promising, its inner workings remain unclear. In this work, we shed light on the mechanism of in-context retrieval augmentation for question answering by viewing a prompt as a composition of informational components. We propose an attribution-based method to identify specialized attention heads, revealing in-context heads that comprehend instructions and retrieve relevant contextual information, and parametric heads that store entities' relational knowledge. To better understand their roles, we extract function vectors and modify their attention weights to show how they can influence the answer generation process. Finally, we leverage the gained insights to trace the sources of knowledge used during inference, paving the way towards more safe and transparent language models.