Multi-Hop Question Answering on Bamboogle out-of-domain (val test)
56.4Exact Match (EM)Search-R2
Evaluation Results
| Method | Links | |
|---|---|---|
| Search-R2Backbone=Qwen2.5-32B2026.02 | 56.4 | |
| Search-R2Backbone=Qwen3-8B2026.02 | 51.2 | |
| Search-R2Backbone=Qwen2.5-7B2026.02 | 46.2 | |
| Search-R1Backbone=Qwen2.5-32B2026.02 | 45 | |
| Search-R1Backbone=Qwen3-8B2026.02 | 43 | |
| Search-R1Backbone=Qwen2.5-7B2026.02 | 36 | |
| Rejection SamplingBackbone=Qwen2.5-7B2026.02 | 35.5 | |
| Search-o1Backbone=Qwen2.5-7B2026.02 | 29.6 | |
| R1-baseBackbone=Qwen2.5-7B2026.02 | 29.6 | |
| R1-instructBackbone=Qwen2.5-7B2026.02 | 29.3 | |
| CoTBackbone=Qwen2.5-7B2026.02 | 23.2 | |
| IRCoTBackbone=Qwen2.5-7B2026.02 | 22.4 | |
| RAGBackbone=Qwen2.5-7B2026.02 | 20.8 | |
| Direct InferenceBackbone=Qwen2.5-7B2026.02 | 12 | |
| SFTBackbone=Qwen2.5-7B2026.02 | 11.2 |