Machine Translation on WMT En-De 2016
41.2BLEUSupervised Model
Evaluation Results
| Method | Links | |
|---|---|---|
| Supervised Modelmode=supervised2021.09 | 41.2 | |
| FLAN 137Bprotocol=few-shot, template_selection=best dev, k=11, number_of_templates=32021.09 | 40.7 | |
| GPT-3 175Bprotocol=few-shot, k=642021.09 | 40.6 | |
| FLAN 137Bprotocol=few-shot, template_selection=average, k=11, number_of_templates=32021.09 | 40.6 | |
| FLAN 137Bprotocol=zero-shot, template_selection=average, number_of_templates=32021.09 | 38.9 | |
| FLAN 137Bprotocol=zero-shot, template_selection=best dev, number_of_templates=32021.09 | 38.9 | |
| Supervised Modelmode=supervised2021.09 | 38.6 | |
| REP(SIM)+WDROPPositions=both2021.04 | 37.96 | |
| REP(UNI)Positions=both2021.04 | 37.73 | |
| REP(UNI)+WDROPPositions=both2021.04 | 37.68 | |
| WDROPPositions=both2021.04 | 37.65 | |
| REP(SIM)Positions=both2021.04 | 37.4 | |
| ADVPositions=both2021.04 | 37.2 | |
| LaMDA-PTprotocol=few-shot, k=52021.09 | 36.8 | |
| REP(SS)Positions=dec2021.04 | 36.59 | |
| w/o perturbation2021.04 | 36.47 | |
| GPT-3 175Bprotocol=few-shot, k=642021.09 | 29.7 | |
| GPT-3 175Bprotocol=zero-shot2021.09 | 27.2 | |
| FLAN 137Bprotocol=zero-shot, template_selection=best dev, number_of_templates=52021.09 | 27 | |
| FLAN 137Bprotocol=few-shot, template_selection=average, k=11, number_of_templates=52021.09 | 26.8 | |
| LaMDA-PTprotocol=few-shot, k=52021.09 | 26.7 | |
| FLAN 137Bprotocol=few-shot, template_selection=best dev, k=11, number_of_templates=52021.09 | 26.1 | |
| FLAN 137Bprotocol=zero-shot, template_selection=average, number_of_templates=52021.09 | 25.4 | |
| GPT-3 175Bprotocol=zero-shot2021.09 | 24.6 | |
| LaMDA-PTprotocol=zero-shot2021.09 | 20.8 | |
| LaMDA-PTprotocol=zero-shot2021.09 | 7.7 |