Multi-hop Question Answering on HotPotQA (MBE, Recall, Searches)
28.1MBEZero-Shot
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Zero-Shot#Train=–2025.07 | 28.1 | — | — | |
| Zero-Shot CoT#Train=–2025.07 | 29.1 | — | — | |
| Self RAG 7B#Train=150k2025.07 | 37.6 | 52.6 | 1 | |
| Ret-Robust#Train=10002025.07 | 39.38 | 35.9 | 4.02 | |
| Self RAG 13B#Train=150k2025.07 | 39.7 | 52.6 | 1 | |
| O2 Searcher#Train=14402025.07 | 42.7 | 50.1 | 1.77 | |
| Zero-Shot RAG#Train=–2025.07 | 45.5 | 52.49 | 1 | |
| IRCOT#Train=–2025.07 | 45.6 | 66.9 | 3.31 | |
| ITER-RETGEN#Train=36k2025.07 | 46.1 | 55.4 | 3 | |
| Search-R1#Train=>100k2025.07 | 46.2 | 48.2 | 1.28 | |
| ReAct + DsPy#Train=1002025.07 | 49.3 | 63.4 | 3.23 | |
| SimpleDeepSearcher#Train=8712025.07 | 50.4 | 64.8 | 2.75 | |
| R1-Searcher#Train=8k2025.07 | 57.66 | 69.1 | 2.22 | |
| FRUGALRAG-7B + CoRAG#Train=10002025.07 | 58 | 70.4 | 2.89 | |
| CoRAG#Train=>100k2025.07 | 58.2 | 64.3 | 4 | |
| FRUGALRAG-7B + Qwen2.5-7B-Inst#Train=10002025.07 | 58.5 | 70.4 | 2.89 | |
| FRUGALRAG-7B + Qwen2.5-32B-Inst#Train=10002025.07 | 61.4 | 70.4 | 2.89 |