Quantifying Retriever-Generator Alignment in RAG with Local Explanations

Korbinian Randl, Guido Rocchietti, Aron Henriksson +3

Retrieval-Augmented Generation (RAG) systems combine dense retrievers and language models to ground outputs in external documents. However, the interaction between these components…