Multi-hop Question Answering on 2Wiki (EM, LJ)
48.89EMM2R
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| M2RBackbone=Qwen2.5-7B-Instruct2026.04 | 48.89 | 57.01 | |
| ReSearchBackbone=Qwen2.5-7B-Instruct2026.04 | 47.59 | 54.22 | |
| SUREBackbone=Qwen2.5-7B-Instruct2026.04 | 45.65 | 53.93 | |
| COFTBackbone=Qwen2.5-7B-Instruct2026.04 | 41.86 | 48.7 | |
| M2RBackbone=Qwen2.5-3B-Instruct2026.04 | 40.07 | 48.34 | |
| ReSearchBackbone=Qwen2.5-3B-Instruct2026.04 | 38.9 | 47.41 | |
| SUREBackbone=Qwen2.5-3B-Instruct2026.04 | 36.2 | 42.38 | |
| COFTBackbone=Qwen2.5-3B-Instruct2026.04 | 32.82 | 39.76 | |
| Iter-RetGenBackbone=Qwen2.5-7B-Instruct2026.04 | 27.92 | 31.86 | |
| Naive RAGBackbone=Qwen2.5-7B-Instruct2026.04 | 25.78 | 29.52 | |
| Naive GenerationBackbone=Qwen2.5-7B-Instruct2026.04 | 25.76 | 27.87 | |
| IRCoTBackbone=Qwen2.5-7B-Instruct2026.04 | 21.57 | 30.65 | |
| Iter-RetGenBackbone=Qwen2.5-3B-Instruct2026.04 | 14.75 | 28.86 | |
| Naive RAGBackbone=Qwen2.5-3B-Instruct2026.04 | 13.82 | 23.08 | |
| IRCoTBackbone=Qwen2.5-3B-Instruct2026.04 | 12.5 | 23.54 | |
| Naive GenerationBackbone=Qwen2.5-3B-Instruct2026.04 | 10.09 | 19.79 |