Multi-hop Question Answering on MuSiQue (EM, F1, P, R)
35Exact Match (EM)STRIDE-FT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| STRIDE-FTBackbone=Qwen3-8B, Fine-tuning=true2026.04 | 35 | 44.8 | 47.5 | 44.8 | |
| STRIDEBackbone=GPT-4o-mini, Fine-tuning=false2026.04 | 33 | 46.7 | 47.1 | 51.4 | |
| STRIDEBackbone=Qwen3-8B, Fine-tuning=false2026.04 | 28.8 | 40.1 | 40.7 | 46.1 | |
| DualRAGBackbone=GPT-4o-mini, Fine-tuning=false2026.04 | 27.7 | 39.9 | 39.3 | 48.1 | |
| GenGroundBackbone=GPT-4o-mini, Fine-tuning=false2026.04 | 23 | 37.2 | 35.5 | 47.5 | |
| DualRAGBackbone=Qwen3-8B, Fine-tuning=false2026.04 | 22.9 | 31.6 | 32.1 | 34.9 | |
| GenGroundBackbone=Qwen3-8B, Fine-tuning=false2026.04 | 21.8 | 32.5 | 33.1 | 35.3 | |
| RAG w/ CoTBackbone=GPT-4o-mini, Fine-tuning=false2026.04 | 18.6 | 30.8 | 31.3 | 33.5 | |
| SelfAskBackbone=Qwen3-8B, Fine-tuning=false2026.04 | 16.6 | 24 | 23.9 | 26.9 | |
| Iter-RetGenBackbone=GPT-4o-mini, Fine-tuning=false2026.04 | 15.6 | 28.9 | 26.7 | 44.4 | |
| Iter-RetGenBackbone=Qwen3-8B, Fine-tuning=false2026.04 | 14.8 | 23.9 | 25.1 | 26.4 | |
| RAG w/ CoTBackbone=Qwen3-8B, Fine-tuning=false2026.04 | 14.4 | 22.7 | 23.4 | 26.2 | |
| CoTBackbone=GPT-4o-mini, Fine-tuning=false2026.04 | 12 | 22.3 | 23.6 | 23.4 | |
| CoTBackbone=Qwen3-8B, Fine-tuning=false2026.04 | 4.5 | 12.3 | 12.8 | 14.6 |