Machine Translation on WMT Finnish-to-English 2016 (test)
55.8BLEU-1GPT-3.5-turbo
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-3.5-turborole=Victim Model2024.09 | 55.8 | 25.2 | 55.7 | 95.3 | |
| LoRDMEA method=Locality Reinforced Distillation, query samples=16, initial local model=Llama3-8B2024.09 | 49.8 | 19.6 | 48.5 | 90.5 | |
| MLEMEA method=Maximum Likelihood Estimation, query samples=16, initial local model=Llama3-8B2024.09 | 44.4 | 17.3 | 44.9 | 90.5 | |
| Llama3-8Brole=Local Model2024.09 | 24.2 | 8.5 | 32 | 86.6 |