from transformers import pipeline

translator = pipeline("translation", model="raulilma/nllb-200-osing-indonesia")

# Osing → Indonesian
result = translator("Aku tresno kowe", src_lang="jav_Latn", tgt_lang="ind_Latn")
print(result[0]["translation_text"])

# Indonesian → Osing
result = translator("Aku cinta kamu", src_lang="ind_Latn", tgt_lang="jav_Latn")
print(result[0]["translation_text"])

Training

Base model: facebook/nllb-200-distilled-600M
Dataset: 2,925 Osing-Indonesian pairs from weling.id kamus
Method: LoRA (r=16, alpha=32, target: q_proj, v_proj)
Epochs: 5, Final loss: 1.23
Language codes: Osing uses jav_Latn (Javanese Latin script as proxy), Indonesian uses ind_Latn