Machine Translation on Romansh-to-German Seen Varieties
60.17chrF ScoreOur SFT
Evaluation results
| Method | Links | |
|---|---|---|
| Our SFTBase Model=Qwen3-4B-Base, Training Method=SFT, Context=full2026.06 | 60.17 | |
| Our SFTBase Model=Llama-3.2-3B-Instruct, Training Method=SFT, Context=full2026.06 | 55.77 | |
| Our RLBase Model=Qwen3-4B-Base, Training Method=RL, Context=full2026.06 | 51.6 | |
| Our RLBase Model=Llama-3.2-3B-Instruct, Training Method=RL, Context=full2026.06 | 47.65 | |
| Our SFTBase Model=Qwen3-4B-Base, Training Method=SFT, Context=none2026.06 | 46.44 | |
| Our SFTBase Model=Llama-3.2-3B-Instruct, Training Method=SFT, Context=none2026.06 | 44.11 | |
| Our RLBase Model=Llama-3.2-3B-Instruct, Training Method=RL, Context=none2026.06 | 34.7 | |
| Qwen3-4B-BaseBase Model=Qwen3-4B-Base, Training Method=Base (untuned), Context=full2026.06 | 34.13 | |
| Our RLBase Model=Qwen3-4B-Base, Training Method=RL, Context=none2026.06 | 30.48 | |
| Llama-3.2-3B-InstBase Model=Llama-3.2-3B-Instruct, Training Method=Instruct-tuned, Context=full2026.06 | 30.12 | |
| Llama-3.2-3B-InstBase Model=Llama-3.2-3B-Instruct, Training Method=Instruct-tuned, Context=none2026.06 | 22.22 | |
| Qwen3-4B-BaseBase Model=Qwen3-4B-Base, Training Method=Base (untuned), Context=none2026.06 | 21.5 |