General Question Answering on TriviaQA (test) (F1 score)
66.1F1Search-R1 + EKA
Evaluation Results
| Method | Links | |
|---|---|---|
| Search-R1 + EKABackbone=Qwen2.5-7B-Instruct2025.12 | 66.1 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2025.12 | 61 | |
| Rejection SamplingBackbone=Qwen2.5-7B-Instruct2025.12 | 59.2 | |
| Standard RAGBackbone=Qwen2.5-7B-Instruct2025.12 | 58.5 | |
| R1-baseBackbone=Qwen2.5-7B-Instruct2025.12 | 53.9 | |
| R1-instructBackbone=Qwen2.5-7B-Instruct2025.12 | 53.7 | |
| IRCoTBackbone=Qwen2.5-7B-Instruct2025.12 | 47.8 | |
| Search-o1Backbone=Qwen2.5-7B-Instruct2025.12 | 44.3 | |
| Direct InferenceBackbone=Qwen2.5-7B-Instruct2025.12 | 40.8 | |
| SFTBackbone=Qwen2.5-7B-Instruct2025.12 | 35.4 | |
| CoTBackbone=Qwen2.5-7B-Instruct2025.12 | 18.5 |