Fluency Evaluation on mimir-instruct Norwegian Bokmål (test)
91.8Win Rateon-policy RL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| on-policy RLopponent=Mistral Nemo2025.12 | 91.8 | — | |
| translated SFTopponent=Mistral Nemo2025.12 | 87.5 | — | |
| on-policy RLopponent=translated SFT2025.12 | 67.5 | — | |
| translated SFTopponent=on-policy RL2025.12 | 32.5 | — | |
| mistral Nemoopponent=translated SFT2025.12 | 12.5 | — | |
| mistral Nemoopponent=on-policy RL2025.12 | 8.2 | — | |
| mistral Nemoaggregation=Average2025.12 | — | 10.3 | |
| on-policy RLaggregation=Average2025.12 | — | 79.7 | |
| translated SFTaggregation=Average2025.12 | — | 60 |