Masakhane Hugging Face models Benchmarks¶
Overview¶
Masakhane Hugging Face models are a collection of models for natural language processing (NLP) tasks in African languages, created by the Masakhane NLP community¹.
Masakhane is a grassroots organization whose mission is to strengthen and spur NLP research in African languages, for Africans, by Africans¹.
Masakhane Hugging Face models¶
- masakhaner²: A named entity recognition (NER) dataset consisting of PER, ORG, LOC, and DATE entities annotated by Masakhane for ten African languages: Amharic, Hausa, Igbo, Kinyarwanda, Luganda, Luo, Nigerian-Pidgin, Swahili, Wolof, and Yoruba.
- masakhaner2³: An updated version of masakhaner, with more languages, more data, and more annotations. It covers 21 African languages and 6 entity types: PER, ORG, LOC, DATE, TIME, and MISC.
- masakhanews⁴: A machine translation dataset for low-resource African languages, consisting of news articles from the BBC and Africanews, covering topics such as politics, sports, health, and entertainment. It includes 13 languages: Amharic, Akan, Hausa, Igbo, Kinyarwanda, Lingala, Luganda, Luo, Ndebele, Nigerian-Pidgin, Swahili, Wolof, and Yoruba.
- masakhane/afroxlmr-large-ner-masakhaner-1.0_2.0⁵: A NER model for 21 African languages, fine-tuned on an aggregation of masakhaner and masakhaner2 datasets. It is based on the Davlan/afro-xlmr-large model, which is a multilingual model trained on 100 languages, including several African languages.
- masakhane/afriqa⁶: A question answering dataset for African languages, consisting of questions and answers extracted from Wikipedia articles. It includes 8 languages: Amharic, Hausa, Igbo, Swahili, Wolof, Yoruba, Somali, and Oromo.
- masakhane/afri-mbart50: A machine translation model for 50 African languages, fine-tuned on the JW300 dataset. It is based on the facebook/mbart-large-50 model, which is a multilingual model trained on 50 languages, including several African languages.
- masakhane/afri-mt5-base: A text-to-text generation model for African languages, fine-tuned on the JW300 dataset. It is based on the google/mt5-base model, which is a multilingual model trained on 101 languages, including several African languages.