Machine Translation on Flores-200 Romance group (xx->en, test)
42.88BLEUX-ALMA (ARPO)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| X-ALMA (ARPO)Backbone=Llama22026.01 | 42.88 | 89.28 | |
| ARPO-z'-V2 + CLewR-zBackbone=Gemma2-9B2026.01 | 41.45 | 88.87 | |
| ARPO-z'-V1 + CLewRBackbone=Gemma2-9B2026.01 | 41.28 | 88.78 | |
| ARPO-z'-V1 + CLewR-zBackbone=Gemma2-9B2026.01 | 41.22 | 88.84 | |
| ARPO-z'-V2 + CLewRBackbone=Gemma2-9B2026.01 | 41.09 | 88.75 | |
| ARPO + CLewRBackbone=Gemma2-9B2026.01 | 40.44 | 88.56 | |
| CPO + CLewRBackbone=Gemma2-9B2026.01 | 40.33 | 88.49 | |
| ARPO-z'-V1Backbone=Gemma2-9B2026.01 | 40.32 | 88.62 | |
| ARPO-z'-V2Backbone=Gemma2-9B2026.01 | 40.1 | 88.58 | |
| ARPO + CLewR-zBackbone=Gemma2-9B2026.01 | 39.87 | 88.62 | |
| ARPO-z'-V1 + CLewRBackbone=LLaMA3.1-8B2026.01 | 39.05 | 88.22 | |
| ARPO-z'-V2 + CLewRBackbone=LLaMA3.1-8B2026.01 | 38.89 | 88.17 | |
| ARPO-z'-V1 + CLewR-zBackbone=LLaMA3.1-8B2026.01 | 38.02 | 88.15 | |
| ARPO + CLewRBackbone=LLaMA3.1-8B2026.01 | 37.99 | 87.94 | |
| ARPO-z'-V2 + CLewR-zBackbone=LLaMA3.1-8B2026.01 | 37.87 | 88.13 | |
| ARPO-z'-V1Backbone=LLaMA3.1-8B2026.01 | 37.82 | 87.91 | |
| ARPOBackbone=Gemma2-9B2026.01 | 37.68 | 88.3 | |
| ARPO-z'-V2Backbone=LLaMA3.1-8B2026.01 | 37.47 | 87.84 | |
| ARPO-z'-V1 + CLewRBackbone=Qwen2.5-7B2026.01 | 37.18 | 87.6 | |
| ARPO-z'-V2 + CLewRBackbone=Qwen2.5-7B2026.01 | 37.16 | 87.64 | |
| ARPO + CLewR-zBackbone=LLaMA3.1-8B2026.01 | 36.75 | 87.93 | |
| ARPO + CLewRBackbone=Qwen2.5-7B2026.01 | 36.69 | 87.62 | |
| ARPO-z'-V1 + CLewR-zBackbone=Qwen2.5-7B2026.01 | 36.55 | 87.56 | |
| CPOBackbone=Gemma2-9B2026.01 | 36.44 | 87.76 | |
| ARPO-z'-V2Backbone=Qwen2.5-7B2026.01 | 36.4 | 87.75 | |
| CPO + CLewRBackbone=Qwen2.5-7B2026.01 | 36.28 | 87.33 | |
| ARPO-z'-V2 + CLewR-zBackbone=Qwen2.5-7B2026.01 | 36.15 | 87.51 | |
| CurriDPOBackbone=LLaMA3.1-8B2026.01 | 36.01 | 87.75 | |
| ARPOBackbone=LLaMA3.1-8B2026.01 | 35.92 | 87.75 | |
| DPOPBackbone=LLaMA3.1-8B2026.01 | 35.78 | 87.82 | |
| DPOP + CLewRBackbone=LLaMA3.1-8B2026.01 | 35.46 | 87.83 | |
| ARPO-z'-V1Backbone=Qwen2.5-7B2026.01 | 35.12 | 87.37 | |
| Gemma2-9BMode=Base Model2026.01 | 34.98 | 85.57 | |
| ARPO + CLewR-zBackbone=Qwen2.5-7B2026.01 | 34.88 | 87.45 | |
| DPOPBackbone=Gemma2-9B2026.01 | 33.17 | 87.77 | |
| DPOPBackbone=Qwen2.5-7B2026.01 | 33.06 | 87.42 | |
| ARPOBackbone=Qwen2.5-7B2026.01 | 32.12 | 86.71 | |
| DPOP + CLewRBackbone=Gemma2-9B2026.01 | 31.86 | 87.65 | |
| CPOBackbone=Qwen2.5-7B2026.01 | 31.66 | 86.36 | |
| DPOP + CLewRBackbone=Qwen2.5-7B2026.01 | 30.63 | 87.1 | |
| CurriDPOBackbone=Qwen2.5-7B2026.01 | 29.12 | 86.77 | |
| CurriDPOBackbone=Gemma2-9B2026.01 | 28.33 | 87.27 | |
| CPO + CLewRBackbone=LLaMA3.1-8B2026.01 | 13.08 | 76.76 | |
| CPOBackbone=LLaMA3.1-8B2026.01 | 10.36 | 74.32 | |
| Qwen2.5-7BMode=Base Model2026.01 | 9.59 | 62.83 | |
| LLaMA3.1-8BMode=Base Model2026.01 | 1.84 | 46.3 |