IbRo-nllb / README.md
agaliano's picture
Update README.md
f73ce2b verified
metadata
license: apache-2.0

Overview

This model was presented at the WMT24 Shared Task on Translation into Low-Resource Languages of Spain as a submission by the Transducens team from the Universitat d'Alacant. It is a many-to-many model capable of translating between several languages of the Iberian Peninsula.

The model is based on NLLB-1.3B, fine-tuned for the following languages:

  • Spanish ↔ Asturian
  • Spanish ↔ Aragonese
  • Spanish ↔ Aranese
  • Spanish ↔ Galician
  • Spanish ↔ Catalan
  • Spanish ↔ Valencian
  • Catalan ↔ Aranese

The new language tokens are:

  • Aragonese: arg_Latn
  • Aranese: arn_Latn
  • Valencian: val_Latn

Usage

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

model = AutoModelForSeq2SeqLM.from_pretrained("Transducens/IbRo-nllb")
tokenizer = AutoTokenizer.from_pretrained("Transducens/IbRo-nllb")

tokenizer.src_lang = "spa_Latn"

sentence = "«Actualmente, tenemos ratones de cuatro meses de edad que antes solían ser diabéticos y que ya no lo son», agregó."
inputs = tokenizer(sentence, return_tensors="pt")

translated_tokens = model.generate(**inputs, forced_bos_token_id=tokenizer.lang_code_to_id["arg_Latn"])

print(tokenizer.batch_decode(translated_tokens, skip_special_tokens=True))

Citation

If you use this model, please cite it as follows:

@inproceedings{wmt2024-galiano-jimenez,
    title = "Universitat d'{A}lacant's Submission to the {WMT} 2024 {S}hared {T}ask on {T}ranslating into {L}ow-{R}esource {L}anguages of {S}pain",
    author = "Galiano-Jim{\'e}nez, Aar{\'o}n and S{\'a}nchez-Cartagena, V{\'i}ctor M and P{\'e}rez-Ortiz, Juan Antonio and S{\'a}nchez-Mart{\'i}nez, Felipe",
    editor = "Koehn, Philipp  and Haddow, Barry  and Kocmi, Tom  and  Monz, Christof",
    booktitle = "Proceedings of the Ninth Conference on Machine Translation",
    month = nov,
    year = "2024",
    address = "Miami",
    publisher = "Association for Computational Linguistics",
}

Acknowledgements

This model has been produced as part of the research project Lightweight neural translation technologies for low-resource languages (LiLowLa) (PID2021-127999NB-I00) funded by the Spanish Ministry of Science and Innovation (MCIN), the Spanish Research Agency (AEI/10.13039/501100011033) and the European Regional Development Fund A way to make Europe.