Question Answering on TriviaQA (EM, F1, SD, SE)
65.8EMAutoSearch
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| AutoSearchBackbone=Qwen2.5-7B2026.04 | 65.8 | 73.1 | 1.04 | 63.3 | |
| Search-R1-baseBackbone=Qwen2.5-7B2026.04 | 63.7 | 71.4 | 1.86 | 34.2 | |
| AutoSearchBackbone=Qwen2.5-3B2026.04 | 61.7 | 69 | 1.05 | 58.7 | |
| Search-R1-instructBackbone=Qwen2.5-7B2026.04 | 61 | 68.4 | 1.74 | 35 | |
| HIPRAG-instructBackbone=Qwen2.5-7B2026.04 | 55.8 | 65.2 | 0.97 | 57.8 | |
| StepSearch-baseBackbone=Qwen2.5-7B2026.04 | 54.7 | 63.3 | 1.73 | 31.6 | |
| Search-R1-instructBackbone=Qwen2.5-3B2026.04 | 52.4 | 60.3 | 1.62 | 32.3 | |
| StepSearch-instructBackbone=Qwen2.5-7B2026.04 | 51.6 | 59.9 | 1.99 | 25.9 | |
| Search-R1-baseBackbone=Qwen2.5-3B2026.04 | 51.3 | 59 | 1.29 | 39.9 | |
| StepSearch-instructBackbone=Qwen2.5-3B2026.04 | 51.1 | 59.3 | 1.57 | 32.7 | |
| StepSearch-baseBackbone=Qwen2.5-3B2026.04 | 47.3 | 55 | 2.9 | 16.3 | |
| HIPRAG-baseBackbone=Qwen2.5-3B2026.04 | 23.4 | 34.1 | 1.6 | 14.6 | |
| HIPRAG-instructBackbone=Qwen2.5-3B2026.04 | 14.5 | 24.2 | 1.32 | 11 |