Multi-Hop Question Answering on MuSiQue (Accuracy)
22.6AccuracyStepSearch
Evaluation Results
| Method | Links | |
|---|---|---|
| StepSearchType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 22.6 | |
| GiGPOType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 18.9 | |
| ZeroSearchType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 18.4 | |
| HCAPO (Ours)Type=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 17.7 | |
| StepSearchType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 17.4 | |
| HCAPO (Ours)Type=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 14.8 | |
| Search-R1Type=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 14.6 | |
| GiGPOType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 12.6 | |
| Search-R1Type=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 10.3 | |
| ZeroSearchType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 9.8 | |
| R1-InstructType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 7.2 | |
| R1-InstructType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 6 |