Question Answering on HotpotQA (EM, ROUGE, F1)
32EMGRIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GRIPBackbone=Qwen2.5-7B-Instruct2026.04 | 32 | 37.6 | 44.6 | |
| GRIPBackbone=Qwen-3-4B-Instruct2026.04 | 31.9 | 39.7 | 46.2 | |
| Single-RAGBackbone=Qwen2.5-7B-Instruct2026.04 | 30.6 | 34.9 | 40.9 | |
| Robust-RAGBackbone=Qwen-3-4B-Instruct2026.04 | 27.3 | 30.8 | 36.7 | |
| Robust-RAGBackbone=Qwen2.5-7B-Instruct2026.04 | 26.1 | 30.8 | 35.5 | |
| R1-SearcherBackbone=Qwen2.5-7B-Instruct2026.04 | 20.2 | 24.3 | 28.8 | |
| Single-RAGBackbone=Qwen-3-4B-Instruct2026.04 | 18.4 | 24.6 | 28.8 | |
| InstructBackbone=Qwen2.5-7B-Instruct2026.04 | 18.3 | 21.5 | 26 | |
| R1-SearcherBackbone=Qwen-3-4B-Instruct2026.04 | 18.2 | 21.2 | 25.4 | |
| InstructBackbone=Qwen-3-4B-Instruct2026.04 | 8 | 13.6 | 16.5 |