Multi-hop Question Answering on Average (MuSiQue, HotpotQA, 2WikiMQA, Bamboogle)
55.34F1 ScoreGraphAnchor
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GraphAnchorBackbone=Qwen3-32B2026.01 | 55.34 | 43.75 | |
| GraphAnchorBackbone=Qwen2.5-14B-Instruct2026.01 | 50.86 | 40.86 | |
| GraphAnchorBackbone=Qwen2.5-7B-Instruct2026.01 | 48.63 | 38 | |
| GraphAnchorBackbone=Llama3.1-8B-Instruct2026.01 | 48.19 | 38.47 | |
| DeepNoteBackbone=Llama3.1-8B-Instruct2026.01 | 45.11 | 34.35 | |
| DeepNoteBackbone=Qwen2.5-7B-Instruct2026.01 | 43.26 | 32.95 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2026.01 | 38.6 | 30.4 | |
| IRCoTBackbone=Llama3.1-8B-Instruct2026.01 | 38.11 | 28.6 | |
| Iter-RetGenBackbone=Qwen2.5-7B-Instruct2026.01 | 37.04 | 28.7 | |
| Vanilla RAGBackbone=Qwen3-32B2026.01 | 36.12 | 26.6 | |
| Vanilla RAGBackbone=Qwen2.5-14B-Instruct2026.01 | 35.95 | 26.7 | |
| Iter-RetGenBackbone=Llama3.1-8B-Instruct2026.01 | 33.63 | 25.4 | |
| Vanilla RAGBackbone=Llama3.1-8B-Instruct2026.01 | 33.18 | 25.05 | |
| IRCoTBackbone=Qwen2.5-7B-Instruct2026.01 | 33.02 | 22.6 | |
| Vanilla RAGBackbone=Qwen2.5-7B-Instruct2026.01 | 32.29 | 24.35 |