Machine Translation on Flores-200 Romance group (en->xx, test)
39.83BLEUX-ALMA (ARPO)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| X-ALMA (ARPO)Backbone=Llama22026.01 | 39.83 | 89.38 | |
| ARPO-z'-V1 + CLewRBackbone=Gemma2-9B2026.01 | 37.45 | 88.4 | |
| ARPO-z'-V2 + CLewRBackbone=Gemma2-9B2026.01 | 37.45 | 88.37 | |
| ARPO-z'-V2 + CLewR-zBackbone=Gemma2-9B2026.01 | 37.08 | 88.38 | |
| ARPO-z'-V1 + CLewR-zBackbone=Gemma2-9B2026.01 | 36.98 | 88.38 | |
| ARPO + CLewRBackbone=Gemma2-9B2026.01 | 36.63 | 88.12 | |
| ARPO + CLewR-zBackbone=Gemma2-9B2026.01 | 36.5 | 88.12 | |
| ARPO-z'-V2Backbone=Gemma2-9B2026.01 | 36.46 | 88.43 | |
| ARPO-z'-V1Backbone=Gemma2-9B2026.01 | 36.35 | 88.4 | |
| CPO + CLewRBackbone=Gemma2-9B2026.01 | 36.24 | 87.85 | |
| ARPO-z'-V2 + CLewRBackbone=LLaMA3.1-8B2026.01 | 35.53 | 87.77 | |
| ARPO-z'-V1 + CLewRBackbone=LLaMA3.1-8B2026.01 | 35.49 | 87.77 | |
| ARPO-z'-V2 + CLewR-zBackbone=LLaMA3.1-8B2026.01 | 35.43 | 87.8 | |
| ARPOBackbone=Gemma2-9B2026.01 | 35.37 | 88.12 | |
| ARPO-z'-V1 + CLewR-zBackbone=LLaMA3.1-8B2026.01 | 35.34 | 87.77 | |
| ARPO + CLewRBackbone=LLaMA3.1-8B2026.01 | 34.49 | 87.44 | |
| ARPO + CLewR-zBackbone=LLaMA3.1-8B2026.01 | 34.33 | 87.57 | |
| ARPO-z'-V2Backbone=LLaMA3.1-8B2026.01 | 33.96 | 87.49 | |
| ARPO-z'-V1Backbone=LLaMA3.1-8B2026.01 | 33.68 | 87.44 | |
| CPOBackbone=Gemma2-9B2026.01 | 33.53 | 87.43 | |
| ARPOBackbone=LLaMA3.1-8B2026.01 | 33.5 | 87.48 | |
| ARPO-z'-V2 + CLewRBackbone=Qwen2.5-7B2026.01 | 32.7 | 85.78 | |
| ARPO-z'-V1 + CLewRBackbone=Qwen2.5-7B2026.01 | 32.68 | 85.81 | |
| ARPO-z'-V1 + CLewR-zBackbone=Qwen2.5-7B2026.01 | 32.68 | 85.93 | |
| ARPO-z'-V2 + CLewR-zBackbone=Qwen2.5-7B2026.01 | 32.67 | 85.93 | |
| CurriDPOBackbone=LLaMA3.1-8B2026.01 | 32.46 | 86.94 | |
| ARPO + CLewRBackbone=Qwen2.5-7B2026.01 | 31.56 | 85.38 | |
| DPOPBackbone=LLaMA3.1-8B2026.01 | 31.53 | 87.05 | |
| ARPO + CLewR-zBackbone=Qwen2.5-7B2026.01 | 31.14 | 85.47 | |
| ARPO-z'-V1Backbone=Qwen2.5-7B2026.01 | 31.08 | 85.67 | |
| ARPO-z'-V2Backbone=Qwen2.5-7B2026.01 | 30.95 | 85.67 | |
| DPOP + CLewRBackbone=LLaMA3.1-8B2026.01 | 30.87 | 86.85 | |
| ARPOBackbone=Qwen2.5-7B2026.01 | 30.41 | 85.64 | |
| CPO + CLewRBackbone=Qwen2.5-7B2026.01 | 30.05 | 84.15 | |
| CPOBackbone=Qwen2.5-7B2026.01 | 27.68 | 83.93 | |
| DPOPBackbone=Qwen2.5-7B2026.01 | 24.43 | 82.9 | |
| CurriDPOBackbone=Qwen2.5-7B2026.01 | 23.8 | 81.9 | |
| DPOP + CLewRBackbone=Qwen2.5-7B2026.01 | 23.59 | 82.74 | |
| DPOPBackbone=Gemma2-9B2026.01 | 23.26 | 85.94 | |
| DPOP + CLewRBackbone=Gemma2-9B, Base Algorithm=DPOP2026.01 | 22.35 | 86.11 | |
| CurriDPOBackbone=Gemma2-9B, Base Algorithm=DPOP2026.01 | 21.81 | 85.64 | |
| Gemma2-9BMode=Base Model2026.01 | 19.85 | 61.92 | |
| Qwen2.5-7BMode=Base Model2026.01 | 4.1 | 45.94 | |
| CPO + CLewRBackbone=LLaMA3.1-8B2026.01 | 3.97 | 59.66 | |
| LLaMA3.1-8BMode=Base Model2026.01 | 1.7 | 46 | |
| CPOBackbone=LLaMA3.1-8B2026.01 | 1.47 | 54.97 |