Question Answering on TAQA 2021
29.2F2021LLaMA2-70B + target-year finetuning
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LLaMA2-70B + target-year finetuningAlignment Target Year=20212024.02 | 29.2 | 42.5 | 56.7 | |
| LLaMA2-70B + time-aware promptingAlignment Target Year=20212024.02 | 25.9 | 42.4 | 56.7 | |
| LLaMA2-70B + time-aware promptingAlignment Target Year=20222024.02 | 23.9 | 40.5 | 54.1 | |
| ChatGPTTraining Cutoff=Sep. 20212024.02 | 23.8 | 35.5 | 46 | |
| LLaMA2-70B + target-year finetuningAlignment Target Year=20222024.02 | 23.7 | 40.8 | 55.4 | |
| LLaMA2-70B + temporal-adaptive finetuningAlignment Strategy=adaptive2024.02 | 23.5 | 40.9 | 57.9 | |
| LLaMA2-70BTraining Cutoff=Sep. 20222024.02 | 19.8 | 37 | 55.9 | |
| Tülu2-70BTraining Cutoff=-2024.02 | 19.1 | 35.1 | 50.1 | |
| LLaMA1-65BTraining Cutoff=Aug. 20222024.02 | 13.2 | 29.5 | 54.5 | |
| GPT-3Training Cutoff=Sep. 20212024.02 | 11.4 | 25.1 | 46.5 |