Multi-hop Question Answering on 2Wiki (MBE, Recall, Searches)
59MBECoRAG
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CoRAG#Train=>100k2025.07 | 59 | 65.4 | 4 | |
| R1-Searcher#Train=8k2025.07 | 52 | 60.4 | 2.36 | |
| FRUGALRAG-7B + CoRAG#Train=10002025.07 | 51.2 | 58.8 | 3.03 | |
| FRUGALRAG-7B + Qwen2.5-32B-Inst#Train=10002025.07 | 50.9 | 58.8 | 3.03 | |
| FRUGALRAG-7B + Qwen2.5-7B-Inst#Train=10002025.07 | 50.4 | 58.8 | 3.03 | |
| SimpleDeepSearcher#Train=8712025.07 | 49.3 | 60.5 | 3.64 | |
| O2 Searcher#Train=14402025.07 | 45.7 | 55.2 | 2.42 | |
| Ret-Robust#Train=10002025.07 | 45.17 | 31.5 | 4.23 | |
| Search-R1#Train=>100k2025.07 | 36.2 | 47.7 | 1.89 | |
| ReAct + DsPy#Train=1002025.07 | 35.65 | 60.03 | 3.27 | |
| ITER-RETGEN#Train=36k2025.07 | 32.1 | 45.3 | 3 | |
| Self RAG 13B#Train=150k2025.07 | 31.5 | 41.3 | 1 | |
| IRCOT#Train=–2025.07 | 30.53 | 57.8 | 3.41 | |
| Self RAG 7B#Train=150k2025.07 | 30.1 | 41.3 | 1 | |
| Zero-Shot CoT#Train=–2025.07 | 29.8 | — | — | |
| Zero-Shot RAG#Train=–2025.07 | 28.8 | 35 | 1 | |
| Zero-Shot#Train=–2025.07 | 28.4 | — | — |