AfriNLLB-train-distilled Dataset

AfriNLLB is a series of efficient multilingual open-source models for African languages. AfriNLLB-train-distilled is one of two datasets we curated and used for training AfriNLLB models. We created AfriNLLB-train-distilled through knowledge distillation, translating the authentic dataset AfriNLLB-train with NLLB-200 3.3B. More details about data sources and processing can be found in the paper.

Supported Languages

AfriNLLB supports 15 language pairs (30 translation directions), including Swahili, Hausa, Yoruba, Amharic, Somali, Zulu, Lingala, Afrikaans, Wolof, and Egyptian Arabic, as well as other African Union official languages such as Arabic (MSA), French, Portuguese, and Spanish. Our training data covers bidirectional translation between English and 13 languages, and between French and two languages (Lingala and Wolof).

Citation

If you use any of AfriNLLB models, datasets, or approaches, please cite the following paper:

@inproceedings{moslem-etal-2026-afrinllb,
    title = "{A}fri{NLLB}: Efficient Translation Models for African Languages",
    author = "Moslem, Yasmin  and
      Wassie, Aman Kassahun  and
      Gizachew, Amanuel",
    booktitle = "Proceedings of the Seventh Workshop on African Natural Language Processing (AfricaNLP)",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://openreview.net/forum?id=hVJZNUZBur"
}