Multi-hop Question Answering on MuSiQue (Exact Match, RAG Improvement)
16.9Exact MatchPyRAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PyRAGBackbone=Qwen2.5-72B-Instruct2026.05 | 16.9 | 6.3 | |
| IRCoTBackbone=Qwen2.5-72B-Instruct2026.05 | 14.2 | — | |
| ITER-RETGENBackbone=Qwen2.5-72B-Instruct2026.05 | 13.8 | — | |
| PyRAGBackbone=Qwen2.5-7B-Instruct2026.05 | 11.8 | 7.1 | |
| Self-AskBackbone=Qwen2.5-72B-Instruct2026.05 | 11.8 | — | |
| Vanilla RAGBackbone=Qwen2.5-72B-Instruct2026.05 | 10.6 | — | |
| CoTBackbone=Qwen2.5-72B-Instruct2026.05 | 9.9 | — | |
| IRCoTBackbone=Qwen2.5-7B-Instruct2026.05 | 9.1 | — | |
| ITER-RETGENBackbone=Qwen2.5-7B-Instruct2026.05 | 8.7 | — | |
| Self-AskBackbone=Qwen2.5-7B-Instruct2026.05 | 6.8 | — | |
| Direct InferenceBackbone=Qwen2.5-72B-Instruct2026.05 | 5.5 | — | |
| Vanilla RAGBackbone=Qwen2.5-7B-Instruct2026.05 | 4.7 | — | |
| Direct InferenceBackbone=Qwen2.5-7B-Instruct2026.05 | 3.1 | — | |
| CoTBackbone=Qwen2.5-7B-Instruct2026.05 | 2.2 | — |