Short-answer Question Answering on Natural Questions (NQ)
40.8EMAPCD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| APCDBackbone=Llama-2-13b-chat2026.05 | 40.8 | 39.2 | |
| APCDBackbone=Llama-3-8b-it2026.05 | 40.5 | 37.8 | |
| DoLaBackbone=Llama-2-13b-chat2026.05 | 39.6 | 36.8 | |
| ADBackbone=Llama-2-13b-chat2026.05 | 39.2 | 37 | |
| DoLaBackbone=Llama-3-8b-it2026.05 | 39 | 34.5 | |
| DoLa+ADBackbone=Llama-2-13b-chat2026.05 | 38.8 | 36.8 | |
| DoLa+ADBackbone=Llama-3-8b-it2026.05 | 38.5 | 34.1 | |
| ADBackbone=Llama-3-8b-it2026.05 | 38.4 | 34 | |
| Llama-3-8b-itDecoding Strategy=Greedy2026.05 | 37.9 | 33.5 | |
| Llama-2-13b-chatDecoding Strategy=Greedy2026.05 | 36.9 | 33.8 | |
| CADBackbone=Llama-3-8b-it2026.05 | 36.5 | 33 | |
| APCDBackbone=Llama-2-7b-chat2026.05 | 36 | 34 | |
| DoLaBackbone=Llama-2-7b-chat2026.05 | 35.2 | 32.2 | |
| DoLa+ADBackbone=Llama-2-7b-chat2026.05 | 34.9 | 31.7 | |
| ADBackbone=Llama-2-7b-chat2026.05 | 34.8 | 31.6 | |
| Llama-2-7b-chatDecoding Strategy=Greedy2026.05 | 34.2 | 31.2 | |
| ITIBackbone=Llama-2-7b-chat2026.05 | 33.5 | 28.1 | |
| ITIBackbone=Llama-2-13b-chat2026.05 | 31.3 | 26.9 | |
| CADBackbone=Llama-2-7b-chat2026.05 | 30 | 25.5 | |
| CADBackbone=Llama-2-13b-chat2026.05 | 30 | 25 | |
| ITIBackbone=Llama-3-8b-it2026.05 | 26.5 | 21.6 |