R²Adapter routes hybrid RAG, cuts graph calls up to 59%

A new routing adapter for hybrid retrieval-augmented generation claims to cut graph-based retrieval calls by up to 59% while keeping answer accuracy intact.

The paper on arXiv, by Yucan Guo and colleagues, introduces R²Adapter as 'a lightweight plug-in Routing and Rewriting Adapter designed to allocate queries between vanilla and graph-based RAG dynamically.' The pitch is straightforward: vanilla RAG is cheap but 'struggles with relational or multi-hop reasoning,' graph-based RAG handles those but 'incurs higher inference complexity and latency,' and existing hybrid systems 'typically rely on heuristic and LLM-based routing,' which the authors frame as wasteful and tied to the underlying model.

R²Adapter routes only the queries that 'genuinely benefit from graph-based reasoning,' and rewrites uncertain graph-routed ones 'to better expose their multi-hop reasoning requirements, improving retrieval quality without additional supervision.' Experiments on three multi-hop QA benchmarks 'demonstrate that R²Adapter reduces graph-based RAG usage by up to 59% while maintaining comparable answer accuracy.' The abstract does not break that figure down by benchmark, and 'comparable' is not quantified. Two researchers on our Who's Who list circulated the preprint.