Multi-hop Question Answering on 2WikiMultihopQA (Exact Match/RAG Improvement)
44.4Exact Match (EM)PyRAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PyRAGBackbone=Qwen2.5-72B-Instruct2026.05 | 44.4 | 14 | |
| ITER-RETGENBackbone=Qwen2.5-72B-Instruct2026.05 | 40.2 | — | |
| IRCoTBackbone=Qwen2.5-72B-Instruct2026.05 | 35.8 | — | |
| PyRAGBackbone=Qwen2.5-7B-Instruct2026.05 | 33.4 | 14.5 | |
| Self-AskBackbone=Qwen2.5-72B-Instruct2026.05 | 32.5 | — | |
| Vanilla RAGBackbone=Qwen2.5-72B-Instruct2026.05 | 30.4 | — | |
| ITER-RETGENBackbone=Qwen2.5-7B-Instruct2026.05 | 28.9 | — | |
| CoTBackbone=Qwen2.5-72B-Instruct2026.05 | 25.4 | — | |
| IRCoTBackbone=Qwen2.5-7B-Instruct2026.05 | 24.8 | — | |
| Self-AskBackbone=Qwen2.5-7B-Instruct2026.05 | 21.5 | — | |
| Direct InferenceBackbone=Qwen2.5-72B-Instruct2026.05 | 20.6 | — | |
| Vanilla RAGBackbone=Qwen2.5-7B-Instruct2026.05 | 18.9 | — | |
| Direct InferenceBackbone=Qwen2.5-7B-Instruct2026.05 | 12.6 | — | |
| CoTBackbone=Qwen2.5-7B-Instruct2026.05 | 10.8 | — |