Multi-Hop Question Answering on HotpotQA 2018 Wikipedia dump (dev)
46.8AccuracyMR-Search
Evaluation Results
| Method | Links | |
|---|---|---|
| MR-SearchBackbone=Qwen2.5-7b2026.03 | 46.8 | |
| Search-R1Backbone=Qwen2.5-7b2026.03 | 43.9 | |
| StepResearchBackbone=Qwen2.5-7b2026.03 | 43.9 | |
| MR-SearchBackbone=Qwen2.5-3b2026.03 | 41.9 | |
| PPRMBackbone=Qwen2.5-7b2026.03 | 38.6 | |
| ReSearchBackbone=Qwen2.5-7b2026.03 | 37.8 | |
| StepResearchBackbone=Qwen2.5-3b2026.03 | 37.3 | |
| PPRMBackbone=Qwen2.5-3b2026.03 | 35.3 | |
| Search-R1Backbone=Qwen2.5-3b2026.03 | 32.6 | |
| ReSearchBackbone=Qwen2.5-3b2026.03 | 30.5 | |
| Search-o1Backbone=Qwen2.5-3b2026.03 | 22.1 | |
| Search-o1Backbone=Qwen2.5-7b2026.03 | 18.7 | |
| Direct InferenceBackbone=Qwen2.5-7b2026.03 | 18.3 | |
| Direct InferenceBackbone=Qwen2.5-3b2026.03 | 14.9 |