VanModers114/Gronings_NLLB

Gronings NLLB

Bidirectional Dutch–Gronings translation, fine-tuned jointly on human and language-filtered synthetic parallel data. Four variety tokens are supported:

Variety	Token
Hogelandsters	<gos_hogelandsters>
Oldambtsters	<gos_oldambtsters>
Veenkoloniaals	<gos_veenkoloniaals>
Westerkwartiers	<gos_westerkwartiers>

Use NllbTokenizer with legacy_behaviour=False. Set src_lang and tgt_lang to nld_Latn and the appropriate variety token (reverse for Gronings → Dutch). During generation, force the target language token with forced_bos_token_id. Load the saved tokenizer together with the model; do not add new tokens.

Model repository: VanModers114/Gronings_NLLB