Machine Translation on FLORES-200 (devtest) (Delta Evaluation)
4.88Delta chrF++GRPO for Seq2Seq MT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GRPO for Seq2Seq MTModel Size=1.3B, Language=zho_Hant, Language Group=Isol.2026.05 | 4.88 | 0.084 | 0.106 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=zho_Hant, Language Group=Isol.2026.05 | 3.45 | 0.086 | 0.109 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=swh, Language Group=Aggl.2026.05 | 3 | 0.012 | 0.011 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=eus, Language Group=Aggl.2026.05 | 2.81 | 0.014 | 0.006 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=eus, Language Group=Aggl.2026.05 | 2.57 | 0.011 | 0.006 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=bod, Language Group=Aggl.2026.05 | 2.47 | 0.007 | 0.006 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=swh, Language Group=Aggl.2026.05 | 2.22 | 0.009 | 0.009 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=ces, Language Group=Fus.2026.05 | 1.8 | 0.006 | 0.015 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=tur, Language Group=Aggl.2026.05 | 1.79 | 0.005 | 0.006 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=tur, Language Group=Aggl.2026.05 | 1.69 | 0.002 | 0.005 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=ben, Language Group=Fus.2026.05 | 1.68 | 0.004 | 0.006 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=zho_Hans, Language Group=Isol.2026.05 | 1.41 | 0.03 | 0.037 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=zho_Hans, Language Group=Isol.2026.05 | 1.15 | 0.029 | 0.036 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=ben, Language Group=Fus.2026.05 | 1.03 | 0.002 | 0.003 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=ces, Language Group=Fus.2026.05 | 0.96 | 0.001 | 0.009 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=pol, Language Group=Fus.2026.05 | 0.91 | 0.003 | 0.002 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=bod, Language Group=Aggl.2026.05 | 0.68 | 0.001 | 0.002 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=jpn, Language Group=Aggl.2026.05 | 0.68 | 0.014 | 0.011 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=jpn, Language Group=Aggl.2026.05 | 0.48 | 0.015 | 0.011 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=bel, Language Group=Fus.2026.05 | 0.47 | 0.001 | 0.001 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=arb, Language Group=Root-pat.2026.05 | 0.43 | 0.002 | 0.004 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=pol, Language Group=Fus.2026.05 | 0.42 | 0.002 | 0.001 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=bel, Language Group=Fus.2026.05 | 0.37 | 0.004 | 0.001 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=yor, Language Group=Isol.2026.05 | 0.36 | 0.001 | 0.002 | |
| GRPO for Seq2Seq MTModel Size=600M, Language=arb, Language Group=Root-pat.2026.05 | 0.13 | 0.002 | 0.001 | |
| GRPO for Seq2Seq MTModel Size=1.3B, Language=yor, Language Group=Isol.2026.05 | 0.12 | 0.007 | 0.001 |