Short-answer Question Answering on TriviaQA
68.5Exact Match (EM)APCD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| APCDBackbone=Llama-3-8b-it2026.05 | 68.5 | 66.9 | |
| APCDBackbone=Llama-2-13b-chat2026.05 | 68.3 | 66.7 | |
| DoLaBackbone=Llama-2-13b-chat2026.05 | 67.4 | 64.1 | |
| DoLa+ADBackbone=Llama-2-13b-chat2026.05 | 67 | 64 | |
| ADBackbone=Llama-2-13b-chat2026.05 | 66.8 | 64 | |
| DoLaBackbone=Llama-3-8b-it2026.05 | 66.8 | 63.1 | |
| DoLa+ADBackbone=Llama-3-8b-it2026.05 | 66.5 | 62.5 | |
| ADBackbone=Llama-3-8b-it2026.05 | 66.4 | 62.5 | |
| Llama-2-13b-chatDecoding Strategy=Greedy2026.05 | 66.2 | 62.7 | |
| Llama-3-8b-itDecoding Strategy=Greedy2026.05 | 65.4 | 62.1 | |
| CADBackbone=Llama-3-8b-it2026.05 | 63.5 | 59.4 | |
| CADBackbone=Llama-2-13b-chat2026.05 | 62 | 57.8 | |
| APCDBackbone=Llama-2-7b-chat2026.05 | 61 | 59.4 | |
| ITIBackbone=Llama-2-13b-chat2026.05 | 60.9 | 58.8 | |
| DoLaBackbone=Llama-2-7b-chat2026.05 | 60.2 | 56.8 | |
| ADBackbone=Llama-2-7b-chat2026.05 | 58.6 | 55.5 | |
| DoLa+ADBackbone=Llama-2-7b-chat2026.05 | 58.5 | 55.5 | |
| Llama-2-7b-chatDecoding Strategy=Greedy2026.05 | 58.2 | 54.8 | |
| ITIBackbone=Llama-2-7b-chat2026.05 | 57.6 | 52.2 | |
| CADBackbone=Llama-2-7b-chat2026.05 | 54 | 48 | |
| ITIBackbone=Llama-3-8b-it2026.05 | 53.1 | 50.9 |