Inference Time and Token Count on MuSiQue (Multi-hop Question Answering)
199Inference Time (ms)DR-RAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DR-RAGBaseline=ReAct2026.04 | 199 | 5,687 | |
| DR-RAGBaseline=Search-o12026.04 | 220 | 4,963 | |
| DR-RAGBaseline=Search-R12026.04 | 366 | 5,743 | |
| Step-wiseBaseline=Search-R12026.04 | 734 | 6,259 | |
| RerunBaseline=Search-R12026.04 | 773 | 7,930 | |
| RerunBaseline=Search-o12026.04 | 817 | 8,537 | |
| RerunBaseline=ReAct2026.04 | 927 | 9,820 | |
| Step-wiseBaseline=ReAct2026.04 | 990 | 11,315 | |
| Step-wiseBaseline=Search-o12026.04 | 1,034 | 8,262 | |
| RAG-CriticBaseline=ReAct2026.04 | 7,111 | 17,026 | |
| RAG-CriticBaseline=Search-o12026.04 | 8,790 | 20,711 | |
| RAG-CriticBaseline=Search-R12026.04 | 8,852 | 17,097 | |
| GraphRAG2026.05 | 13,900 | — | |
| Dense RAG2026.05 | 26,900 | — | |
| LLM-Wiki2026.05 | 27,100 | — | |
| None (Closed-book)2026.05 | 30,000 | — | |
| BM25 RAG2026.05 | 30,200 | — | |
| HippoRAG 22026.05 | 38,200 | — | |
| RAPTOR2026.05 | 40,200 | — | |
| LightRAG2026.05 | 51,300 | — |