Multi-Hop Question Answering on 2wiki 2018 Wikipedia dump (dev)
43.6Accuracy (%)MR-Search
Evaluation Results
| Method | Links | |
|---|---|---|
| MR-SearchBackbone=Qwen2.5-7b2026.03 | 43.6 | |
| StepResearchBackbone=Qwen2.5-7b2026.03 | 41.8 | |
| MR-SearchBackbone=Qwen2.5-3b2026.03 | 40.1 | |
| Search-R1Backbone=Qwen2.5-7b2026.03 | 38.7 | |
| ReSearchBackbone=Qwen2.5-7b2026.03 | 38.6 | |
| PPRMBackbone=Qwen2.5-7b2026.03 | 35.5 | |
| PPRMBackbone=Qwen2.5-3b2026.03 | 34 | |
| StepResearchBackbone=Qwen2.5-3b2026.03 | 33.8 | |
| Search-R1Backbone=Qwen2.5-3b2026.03 | 31 | |
| ReSearchBackbone=Qwen2.5-3b2026.03 | 27.2 | |
| Direct InferenceBackbone=Qwen2.5-7b2026.03 | 25 | |
| Direct InferenceBackbone=Qwen2.5-3b2026.03 | 24.4 | |
| Search-o1Backbone=Qwen2.5-3b2026.03 | 21.8 | |
| Search-o1Backbone=Qwen2.5-7b2026.03 | 17.6 |