Question Answering on PopQA Normal (test)
25.95F1 ScoreOpenDecoder
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OpenDecoderBackbone=Qwen-2.5-3B-Instruct2026.01 | 25.95 | 22.8 | |
| RbFTBackbone=Qwen-2.5-3B-Instruct2026.01 | 24.73 | 21.42 | |
| AstuteRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 23.92 | 19.9 | |
| Vanilla SFTBackbone=Qwen-2.5-3B-Instruct2026.01 | 20.46 | 17.37 | |
| InstructRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 16.19 | 21.7 | |
| No RAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 11.07 | — | |
| RobustRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 10.91 | 15.9 | |
| Vanilla RAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 7.35 | 21.87 |