Single-Hop Question Answering on PopQA out-of-domain
52.4AccuracyGEPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GEPOType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 52.4 | |
| ZeroSearchBackbone=Qwen2.5-7B-Instruct2026.02 | 51.5 | |
| ZeroSearchType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 51.5 | |
| GEPOType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 47.2 | |
| GiGPOType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 46.1 | |
| SKILLRLBackbone=Qwen2.5-7B-Instruct2026.02 | 45.9 | |
| ZeroSearchType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 44.8 | |
| ZeroSearch2026.06 | 44.8 | |
| EvolveRBackbone=Qwen2.5-7B-Instruct2026.02 | 44.6 | |
| ECPO2026.06 | 44.1 | |
| GiGPOType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 42.4 | |
| GiGPO2026.06 | 41.5 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2026.02 | 39.7 | |
| Search-R1Type=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 39.7 | |
| Search-R1Type=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 37.8 | |
| Search-R12026.06 | 37.8 | |
| R1-InstructType=RL Training, Model=Qwen2.5-3B-Instruct2025.10 | 19.9 | |
| R1-Instruct2026.06 | 19.9 | |
| RAGBackbone=Qwen2.5-7B-Instruct2026.02 | 17.8 | |
| R1-InstructBackbone=Qwen2.5-7B-Instruct2026.02 | 17.1 | |
| R1-InstructType=RL Training, Model=Qwen2.5-7B-Instruct2025.10 | 17.1 | |
| Search-o1Backbone=Qwen2.5-7B-Instruct2026.02 | 11.4 | |
| CoTBackbone=Qwen2.5-7B-Instruct2026.02 | 3.8 | |
| Qwen2.5Backbone=Qwen2.5-7B-Instruct2026.02 | 1.2 |