Machine Translation on WMT EN-ZH 22 (COMET/COMETkiwi)
88.2COMETGoogleMT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GoogleMT2023.11 | 88.2 | 82.7 | — | |
| OracleSelection=Upper bound2024.02 | 87.7 | — | — | |
| TIM-13Bbackbone=tigerbot-13b-base2023.11 | 87.6 | 82.3 | — | |
| WMT-Best2023.11 | 86.7 | 82 | — | |
| MicroMT2023.11 | 86 | 81.3 | — | |
| BayLing-7Bbackbone=7B2023.11 | 84.4 | 79.6 | — | |
| MBRCandidates=10242024.02 | 84.2 | — | — | |
| PruneMBRPruning=Confidence-based2024.02 | 84.1 | — | — | |
| CBMBRInitialization=kmeans++, Centroids (k)=642024.02 | 84.1 | — | — | |
| CBMBRInitialization=random, Centroids (k)=642024.02 | 84 | — | — | |
| QEModel=COMETKIWI2024.02 | 82.9 | — | — | |
| MAPDecoding=Beam Search, Beam Size=2562024.02 | 77.4 | — | — | |
| C-DPOSize=8B, Evaluation Protocol=5-shot2024.09 | — | — | 24.89 | |
| C-DPOSize=70B, Evaluation Protocol=5-shot2024.09 | — | — | 29.49 | |
| C-SFTSize=8B, Evaluation Protocol=5-shot2024.09 | — | — | 24.95 | |
| C-SFTSize=70B, Evaluation Protocol=5-shot2024.09 | — | — | 29.17 | |
| Llama-3 InstructSize=8B, Evaluation Protocol=5-shot2024.09 | — | — | 22.07 | |
| Llama-3 InstructSize=70B, Evaluation Protocol=5-shot2024.09 | — | — | 24.05 |