Single-Hop Question Answering on TriviaQA 2018 Wikipedia dump (dev)
66.6AccuracyMR-Search
Evaluation Results
| Method | Links | |
|---|---|---|
| MR-SearchBackbone=Qwen2.5-7b2026.03 | 66.6 | |
| StepResearchBackbone=Qwen2.5-7b2026.03 | 63.6 | |
| MR-SearchBackbone=Qwen2.5-3b2026.03 | 63.5 | |
| Search-R1Backbone=Qwen2.5-7b2026.03 | 63.2 | |
| Search-R1Backbone=Qwen2.5-3b2026.03 | 62.2 | |
| StepResearchBackbone=Qwen2.5-3b2026.03 | 61.5 | |
| PPRMBackbone=Qwen2.5-7b2026.03 | 61 | |
| ReSearchBackbone=Qwen2.5-7b2026.03 | 60.5 | |
| ReSearchBackbone=Qwen2.5-3b2026.03 | 59.7 | |
| PPRMBackbone=Qwen2.5-3b2026.03 | 56.5 | |
| Search-o1Backbone=Qwen2.5-3b2026.03 | 47.2 | |
| Search-o1Backbone=Qwen2.5-7b2026.03 | 44.3 | |
| Direct InferenceBackbone=Qwen2.5-7b2026.03 | 40.8 | |
| Direct InferenceBackbone=Qwen2.5-3b2026.03 | 28.8 |