Multi-hop Question Answering on WebQ 2013 (test)
48.3F1 ScoreCIRAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CIRAGBackbone=Qwen2.5-7B-Instruct2026.01 | 48.3 | 34 | |
| MetaRAGBackbone=Qwen2.5-7B-Instruct2026.01 | 48.2 | 33.8 | |
| Dualrag-FTBackbone=Qwen2.5-7B-Instruct, Fine-tuned=true2026.01 | 47.4 | 31.2 | |
| DualragBackbone=Qwen2.5-7B-Instruct2026.01 | 45.1 | 29.5 | |
| KiragBackbone=Qwen2.5-7B-Instruct2026.01 | 44.6 | 27.5 | |
| FLAREBackbone=Qwen2.5-7B-Instruct2026.01 | 42.3 | 26 | |
| NativeBackbone=Qwen2.5-7B-Instruct2026.01 | 42.1 | 27.8 | |
| IRCOTBackbone=Qwen2.5-7B-Instruct2026.01 | 41.6 | 25 |