Question Answering on PopQA (EM, ROUGE, F1)
28.7EMGRIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GRIPBackbone=Qwen2.5-7B-Instruct2026.04 | 28.7 | 30.9 | 36.3 | |
| GRIPBackbone=Qwen-3-4B-Instruct2026.04 | 27.8 | 30.9 | 36.6 | |
| Robust-RAGBackbone=Qwen2.5-7B-Instruct2026.04 | 27.5 | 28.6 | 33.5 | |
| Robust-RAGBackbone=Qwen-3-4B-Instruct2026.04 | 27.3 | 28.2 | 33.2 | |
| Single-RAGBackbone=Qwen2.5-7B-Instruct2026.04 | 23.6 | 26.6 | 31.1 | |
| Single-RAGBackbone=Qwen-3-4B-Instruct2026.04 | 18.3 | 21.7 | 25.5 | |
| R1-SearcherBackbone=Qwen-3-4B-Instruct2026.04 | 12.5 | 14.4 | 18 | |
| R1-SearcherBackbone=Qwen2.5-7B-Instruct2026.04 | 11.6 | 13.8 | 17 | |
| InstructBackbone=Qwen2.5-7B-Instruct2026.04 | 11.5 | 13 | 16.3 | |
| InstructBackbone=Qwen-3-4B-Instruct2026.04 | 5.3 | 8 | 10.3 |