Multi-hop Question Answering on MuSiQue (Hop-Specific F1 Breakdown)
40.62-hop F1SAKE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SAKEBackbone=Qwen3-4B-Thinking2026.02 | 40.6 | 24.1 | 12 | |
| SAKEBackbone=QwQ-32B2026.02 | 37.6 | 23 | 20.4 | |
| Search-o1Backbone=QwQ-32B2026.02 | 35.8 | 15 | 12.4 | |
| SAKEBackbone=Qwen3-30B-A3B-Thinking2026.02 | 35.2 | 18 | 14.6 | |
| Search-o1Backbone=Qwen3-4B-Thinking2026.02 | 34.9 | 18.8 | 9.58 | |
| Standard RAGBackbone=Qwen3-30B-A3B-Thinking2026.02 | 32 | 22 | 15.8 | |
| Standard RAGBackbone=Qwen3-4B-Thinking2026.02 | 31.5 | 21.6 | 10.9 | |
| Standard RAGBackbone=QwQ-32B2026.02 | 29.6 | 18.4 | 15.3 | |
| Search-o1Backbone=Qwen3-30B-A3B-Thinking2026.02 | 25.3 | 14.8 | 5.94 |