General Question Answering on Natural Questions (NQ) (test val)
55.8EMDeepControl
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DeepControlBackbone Model=Qwen2.5-7B2026.02 | 55.8 | 47.9 | |
| DeepControlBackbone Model=Qwen2.5-3B2026.02 | 53.3 | 41.1 | |
| Search-R1-baseBackbone Model=Qwen2.5-7B2026.02 | 48 | 43.1 | |
| Search-R1-baseBackbone Model=Qwen2.5-3B2026.02 | 40.6 | 30.3 | |
| Search-R1-instructBackbone Model=Qwen2.5-7B2026.02 | 39.3 | 38.5 | |
| Rejection SamplingBackbone Model=Qwen2.5-7B2026.02 | 36 | 34.8 | |
| RAGBackbone Model=Qwen2.5-7B2026.02 | 34.9 | 30.4 | |
| RAGBackbone Model=Qwen2.5-3B2026.02 | 34.8 | 27 | |
| Search-R1-instructBackbone Model=Qwen2.5-3B2026.02 | 34.1 | 32.5 | |
| SFTBackbone Model=Qwen2.5-7B2026.02 | 31.8 | 20.7 | |
| R1-baseBackbone Model=Qwen2.5-7B2026.02 | 29.7 | 27.6 | |
| Rejection SamplingBackbone Model=Qwen2.5-3B2026.02 | 29.4 | 26.5 | |
| R1-instructBackbone Model=Qwen2.5-7B2026.02 | 27 | 27.1 | |
| SFTBackbone Model=Qwen2.5-3B2026.02 | 24.9 | 17.6 | |
| Search-o1Backbone Model=Qwen2.5-3B2026.02 | 23.8 | 25.5 | |
| R1-baseBackbone Model=Qwen2.5-3B2026.02 | 22.6 | 22.9 | |
| IRCoTBackbone Model=Qwen2.5-7B2026.02 | 22.4 | 23.9 | |
| R1-instructBackbone Model=Qwen2.5-3B2026.02 | 21 | 22.4 | |
| Search-o1Backbone Model=Qwen2.5-7B2026.02 | 15.1 | 20.6 | |
| Direct InferenceBackbone Model=Qwen2.5-7B2026.02 | 13.4 | 18.1 | |
| IRCoTBackbone Model=Qwen2.5-3B2026.02 | 11.1 | 18.1 | |
| Direct InferenceBackbone Model=Qwen2.5-3B2026.02 | 10.6 | 13.4 | |
| CoTBackbone Model=Qwen2.5-7B2026.02 | 4.8 | 10.6 | |
| CoTBackbone Model=Qwen2.5-3B2026.02 | 2.3 | 1.5 |