Tag
DynaKRAG introduces a unified framework for multi-hop retrieval-augmented generation that learns a state-conditioned policy to select evidence operations, outperforming baselines on HotpotQA, 2Wiki, and MuSiQue.