Question Answering on NQ Noisy (test)
37.71F1 ScoreOpenDecoder
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OpenDecoderBackbone=Qwen-2.5-3B-Instruct2026.01 | 37.71 | 33.82 | |
| RbFTBackbone=Qwen-2.5-3B-Instruct2026.01 | 35.5 | 30.7 | |
| Vanilla SFTBackbone=Qwen-2.5-3B-Instruct2026.01 | 34.98 | 32.83 | |
| AstuteRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 32.36 | 29 | |
| InstructRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 28.09 | 29.33 | |
| RobustRAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 25.21 | 30.2 | |
| Vanilla RAGBackbone=Qwen-2.5-3B-Instruct2026.01 | 15.22 | 32.7 |