Multi-hop Question Answering on HotpotQA (EM, F1, Acc)
49.6Exact Match (EM)CompactRAG GPT-4-Reading
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CompactRAG GPT-4-ReadingToken / Sample=1.9K2026.02 | 49.6 | 69.54 | 77.2 | |
| Iter-RetGenToken / Sample=4.7K2026.02 | 46.8 | 52.24 | 72.4 | |
| CompactRAG LLaMA-3.1-8b-ReadingToken / Sample=1.9K2026.02 | 45.2 | 66.21 | 70.4 | |
| IRCoTToken / Sample=10.2K2026.02 | 42.8 | 48.95 | 65.2 | |
| Vanilla-RAGToken / Sample=2.7K2026.02 | 27.6 | 30.32 | 50.8 | |
| Self-AskToken / Sample=6.9K2026.02 | 23.6 | 26.3 | 40.8 |