Multi-hop Question Answering on MuSiQue (F1, EM)
37.68F1 ScoreGraphAnchor
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GraphAnchorBackbone=Qwen3-32B2026.01 | 37.68 | 26.4 | |
| GraphAnchorBackbone=Qwen2.5-14B-Instruct2026.01 | 33.73 | 23.49 | |
| GraphAnchorBackbone=Qwen2.5-7B-Instruct2026.01 | 32.6 | 21.2 | |
| GraphAnchorBackbone=Llama3.1-8B-Instruct2026.01 | 28.91 | 19.96 | |
| DeepNoteBackbone=Llama3.1-8B-Instruct2026.01 | 26.81 | 16.6 | |
| DeepNoteBackbone=Qwen2.5-7B-Instruct2026.01 | 26.67 | 15.8 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2026.01 | 24.95 | 17.6 | |
| Iter-RetGenBackbone=Qwen2.5-7B-Instruct2026.01 | 21.36 | 13 | |
| IRCoTBackbone=Llama3.1-8B-Instruct2026.01 | 19.91 | 12.4 | |
| Vanilla RAGBackbone=Qwen3-32B2026.01 | 19.56 | 9.8 | |
| Vanilla RAGBackbone=Qwen2.5-14B-Instruct2026.01 | 18.6 | 9.6 | |
| Iter-RetGenBackbone=Llama3.1-8B-Instruct2026.01 | 18.15 | 11.2 | |
| Vanilla RAGBackbone=Qwen2.5-7B-Instruct2026.01 | 18.05 | 9.8 | |
| IRCoTBackbone=Qwen2.5-7B-Instruct2026.01 | 17.91 | 8.2 | |
| Vanilla RAGBackbone=Llama3.1-8B-Instruct2026.01 | 17.43 | 9.4 |