Multi-Hop Question Answering on Bamboogle 2018 Wikipedia dump (dev)
45.2AccuracyMR-Search
Evaluation Results
| Method | Links | |
|---|---|---|
| MR-SearchBackbone=Qwen2.5-7b2026.03 | 45.2 | |
| StepResearchBackbone=Qwen2.5-7b2026.03 | 43.5 | |
| Search-R1Backbone=Qwen2.5-7b2026.03 | 40 | |
| ReSearchBackbone=Qwen2.5-7b2026.03 | 37.6 | |
| PPRMBackbone=Qwen2.5-7b2026.03 | 35.5 | |
| MR-SearchBackbone=Qwen2.5-3b2026.03 | 34.4 | |
| StepResearchBackbone=Qwen2.5-3b2026.03 | 32.5 | |
| Search-o1Backbone=Qwen2.5-3b2026.03 | 32 | |
| Search-o1Backbone=Qwen2.5-7b2026.03 | 29.6 | |
| PPRMBackbone=Qwen2.5-3b2026.03 | 28 | |
| Search-R1Backbone=Qwen2.5-3b2026.03 | 17.6 | |
| Direct InferenceBackbone=Qwen2.5-7b2026.03 | 12 | |
| ReSearchBackbone=Qwen2.5-3b2026.03 | 11.5 | |
| Direct InferenceBackbone=Qwen2.5-3b2026.03 | 2.4 |