Single-Hop Question Answering on NQ (Natural Questions) 2018 Wikipedia dump (dev)
50.2AccuracyMR-Search
Evaluation Results
| Method | Links | |
|---|---|---|
| MR-SearchBackbone=Qwen2.5-7b2026.03 | 50.2 | |
| MR-SearchBackbone=Qwen2.5-3b2026.03 | 47.7 | |
| StepResearchBackbone=Qwen2.5-7b2026.03 | 47.3 | |
| Search-R1Backbone=Qwen2.5-3b2026.03 | 46.2 | |
| Search-R1Backbone=Qwen2.5-7b2026.03 | 45.9 | |
| PPRMBackbone=Qwen2.5-7b2026.03 | 45.8 | |
| StepResearchBackbone=Qwen2.5-3b2026.03 | 44.6 | |
| ReSearchBackbone=Qwen2.5-3b2026.03 | 42.7 | |
| PPRMBackbone=Qwen2.5-3b2026.03 | 42.3 | |
| ReSearchBackbone=Qwen2.5-7b2026.03 | 36.6 | |
| Search-o1Backbone=Qwen2.5-3b2026.03 | 23.8 | |
| Search-o1Backbone=Qwen2.5-7b2026.03 | 15.1 | |
| Direct InferenceBackbone=Qwen2.5-7b2026.03 | 13.4 | |
| Direct InferenceBackbone=Qwen2.5-3b2026.03 | 10.6 |