Multi-hop Question Answering on HotpotQA (Answer F1, Citation F1, Joint F1)
76.18Answer F1RAG-RL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RAG-RLEval. Setting=Ideal Retrieval, Curriculum=Min-Max2025.03 | 76.18 | 93.13 | 83.81 | |
| RAG-RLEval. Setting=Ideal Retrieval, Curriculum=Linear2025.03 | 75.67 | 89.34 | 81.94 | |
| RAG-RLEval. Setting=Ideal Retrieval, Curriculum=Base-Unanswerable2025.03 | 75.23 | 82.99 | 78.92 | |
| RAG-RLEval. Setting=Ideal Retrieval, Curriculum=Base-Answerable2025.03 | 74.76 | 83.69 | 78.98 | |
| RAG-RLEval. Setting=Ideal Retrieval, Curriculum=Max2025.03 | 74.25 | 86.26 | 79.81 | |
| RAG-RLEval. Setting=Distractor, Curriculum=Min-Max2025.03 | 68.87 | 81.64 | 74.72 | |
| RAG-RLEval. Setting=Distractor, Curriculum=Linear2025.03 | 68.71 | 78.54 | 73.3 | |
| Qwen2.5-7B-InstructEval. Setting=Ideal Retrieval, Curriculum=Baseline2025.03 | 67.9 | 63.26 | 65.5 | |
| RAG-RLEval. Setting=Distractor, Curriculum=Base-Answerable2025.03 | 66.19 | 72.81 | 69.34 | |
| RAG-RLEval. Setting=Distractor, Curriculum=Max2025.03 | 66.04 | 73.93 | 69.76 | |
| RAG-RLEval. Setting=Distractor, Curriculum=Base-Unanswerable2025.03 | 65.25 | 71.13 | 68.06 | |
| Qwen2.5-7B-InstructEval. Setting=Distractor, Curriculum=Baseline2025.03 | 60.65 | 36.47 | 45.55 |