Question Answering on NQ Extreme
22.5F1 ScoreOpenDecoder
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OpenDecoderBackbone=Qwen-2.5-3B-Instruct2026.01 | 22.5 | 18.06 | |
| RbFTBackbone=Qwen-2.5-3B-Instruct2026.01 | 21.49 | 17.1 | |
| Vanilla SFTBackbone=Qwen-2.5-3B-Instruct2026.01 | 19.78 | 16.73 | |
| AstuteRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 16.06 | 9.5 | |
| InstructRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 5.52 | 7.4 | |
| RobustRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 3.84 | 3.93 | |
| Vanilla RAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 3.33 | 10.14 |