Machine Translation on WMT En-Ja 2023 (test)
88.5COMETDUAL-REFLECT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DUAL-REFLECTRefinement=Dual-Reflection2024.06 | 88.5 | 63.6 | — | |
| ChatGPT + MADRefinement=Multi-Agent Debate2024.06 | 88 | 63.1 | — | |
| ChatGPT + 5shotPrompting=5-shot2024.06 | 87.8 | 61.5 | — | |
| ChatGPTPrompting=Zero-shot2024.06 | 87.3 | 60.2 | — | |
| X-ALMA-13B (Only SFT, MoE)Parameters=13B, Training Strategy=Only SFT, Architecture=MoE2025.02 | 86.8 | — | 22.3 | |
| Aya-23-8BParameters=8B2025.02 | 86.5 | — | 19.3 | |
| Aya-23-35BParameters=35B2025.02 | 86.4 | — | 20.6 | |
| X-ALMA-13B-DAT (MoE)Parameters=13B, Training Strategy=DAT, Architecture=MoE2025.02 | 86 | — | 19 | |
| X-ALMA-13B-DATM (MoE)Parameters=13B, Training Strategy=DATM, Architecture=MoE2025.02 | 86 | — | 19 | |
| Aya-1012025.02 | 84.6 | — | 14.1 | |
| LLaMAX3-Alpaca-8BParameters=8B2025.02 | 81.8 | — | 11.9 | |
| NLLB-3.3BParameters=3.3B2025.02 | 81.6 | — | 13.8 |