Question Answering on NQ (EM, F1, SD, SE)
47.6EMSearch-R1-base
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Search-R1-baseBackbone=Qwen2.5-7B2026.04 | 47.6 | 56.3 | 1.98 | 24 | |
| AutoSearchBackbone=Qwen2.5-7B2026.04 | 47.6 | 56 | 1.05 | 45.3 | |
| AutoSearchBackbone=Qwen2.5-3B2026.04 | 43.7 | 52.2 | 1.03 | 42.4 | |
| Search-R1-instructBackbone=Qwen2.5-7B2026.04 | 40.2 | 48.5 | 1.95 | 20.7 | |
| StepSearch-baseBackbone=Qwen2.5-7B2026.04 | 37.7 | 47.9 | 1.68 | 22.4 | |
| Search-R1-baseBackbone=Qwen2.5-3B2026.04 | 36 | 44.9 | 1.23 | 29.4 | |
| StepSearch-instructBackbone=Qwen2.5-7B2026.04 | 34.2 | 44.5 | 1.94 | 17.6 | |
| StepSearch-instructBackbone=Qwen2.5-3B2026.04 | 33.9 | 43.5 | 1.46 | 23.3 | |
| Search-R1-instructBackbone=Qwen2.5-3B2026.04 | 32.2 | 42 | 1.76 | 18.3 | |
| HIPRAG-instructBackbone=Qwen2.5-7B2026.04 | 30.6 | 42 | 1.07 | 28.5 | |
| StepSearch-baseBackbone=Qwen2.5-3B2026.04 | 30 | 39.9 | 2.95 | 10.2 | |
| HIPRAG-baseBackbone=Qwen2.5-3B2026.04 | 11 | 20.2 | 1.59 | 6.92 | |
| HIPRAG-instructBackbone=Qwen2.5-3B2026.04 | 3.2 | 9.5 | 1.28 | 2.5 |