Question Answering on Average (TriviaQA, HotpotQA, 2Wiki, Musique, Bamboogle)
49.12EM AccuracyICRL
Evaluation results
| Method | Links | |
|---|---|---|
| ICRLModel=Qwen2.5-7B2026.03 | 49.12 | |
| ParallelSearchModel=Qwen2.5-7B2026.03 | 41.78 | |
| ICRLModel=Qwen2.5-3B2026.03 | 40.16 | |
| Search-R1Model=Qwen2.5-7B2026.03 | 38.16 | |
| ZeroSearchModel=Qwen2.5-7B2026.03 | 36.24 | |
| Reject SamplingModel=Qwen2.5-7B2026.03 | 33.94 | |
| Search-R1Model=Qwen2.5-3B2026.03 | 31.1 | |
| R1-baseModel=Qwen2.5-7B2026.03 | 28.66 | |
| R1-instructModel=Qwen2.5-7B2026.03 | 28.62 | |
| RAGModel=Qwen2.5-7B2026.03 | 27.7 | |
| ZeroSearchModel=Qwen2.5-3B2026.03 | 27.14 | |
| Search-o1Model=Qwen2.5-3B2026.03 | 25.7 | |
| Reject SamplingModel=Qwen2.5-3B2026.03 | 24.6 | |
| R1-instructModel=Qwen2.5-3B2026.03 | 23.68 | |
| Search-o1Model=Qwen2.5-7B2026.03 | 23.2 | |
| RAGModel=Qwen2.5-3B2026.03 | 23.04 | |
| IRCoTModel=Qwen2.5-7B2026.03 | 21.12 | |
| SFTModel=Qwen2.5-7B2026.03 | 20.16 | |
| DirectModel=Qwen2.5-7B2026.03 | 19.84 | |
| IRCoTModel=Qwen2.5-3B2026.03 | 19.08 | |
| SFTModel=Qwen2.5-3B2026.03 | 17.64 | |
| DirectModel=Qwen2.5-3B2026.03 | 14.5 | |
| CoTModel=Qwen2.5-7B2026.03 | 12.84 | |
| CoTModel=Qwen2.5-3B2026.03 | 1.52 |