Machine Translation on WMT Czech-to-English 2016 (test)
0.611BLEU-1GPT-3.5-turbo
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-3.5-turborole=Victim Model2024.09 | 0.611 | 0.313 | 0.604 | 0.957 | |
| LoRDMEA method=Locality Reinforced Distillation, query samples=16, initial local model=Llama3-8B2024.09 | 0.545 | 0.249 | 0.538 | 0.906 | |
| MLEMEA method=Maximum Likelihood Estimation, query samples=16, initial local model=Llama3-8B2024.09 | 0.535 | 0.245 | 0.526 | 0.899 | |
| Llama3-8Brole=Local Model2024.09 | 0.255 | 0.105 | 0.348 | 0.868 |