Multi-Hop Search-Augmented QA on MuSiQue (out-of-domain)
24AccuracyGEPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GEPOType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 24 | |
| StepSearchType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 22.6 | |
| GEPOType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 20.8 | |
| GiGPOType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 18.9 | |
| ZeroSearchType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 18.4 | |
| StepSearchType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 17.4 | |
| Search-R1Type=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 14.6 | |
| GiGPOType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 12.6 | |
| Search-R1Type=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 10.3 | |
| ZeroSearchType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 9.8 | |
| R1-InstructType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 7.2 | |
| R1-InstructType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 6 |