Multi-Hop Question Answering on Bamboogle (CEM %)
58.7CEM (%)R1-Searcher++
Evaluation results
| Method | Links | |
|---|---|---|
| R1-Searcher++2025.10 | 58.7 | |
| R1-Searcher2025.10 | 55.8 | |
| β-GRPO-step*Reward=β-GRPO's reward, Output Format=HiPRAG's format2025.10 | 54.4 | |
| HiPRAG-7BModel Size=7B2025.10 | 52.8 | |
| β-GRPO2025.10 | 52 | |
| Search-R1-step*Search Strategy=Search-R1 v0.3, Output Format=HiPRAG's format2025.10 | 49.6 | |
| Search-R12025.10 | 48.4 | |
| HiPRAG-3BModel Size=3B2025.10 | 41.6 | |
| IRCoT2025.10 | 20.6 | |
| Search-o12025.10 | 19.3 | |
| Standard RAG2025.10 | 18.6 | |
| Direct Inference2025.10 | 15.9 |