Inkuba-Instruct dataset

Dataset Sources

• We collected downstream datasets for 5 African languages from the following public repositories:

Task	Datasets	Sources	Combined Size (# of samples)
Machine Translation	WMT-22-African, Mafand-MT, Menyo-20k	allenai/wmt22_african, masakhane/mafand, menyo20k_mt	359 M
NER	MasakhaNER2, Hausa VoA NER, isiXhosa NER Corpus	masakhane/masakhaner2, hausa_voa_ner, https://repo.sadilar.org/handle/20.500.12185/312	~64k
POS	MasakhaPOS	masakhane/masakhapos	6.5k
Question-Answering	afriqa	masakhane/afriqa	4.45k
Topic Classification	SIB-200, MasakhaNEWS, Hausa News Classification	Davlan/sib200,masakhane/masakhanews, hausa_voa_topics	22.8k
Sentiment Analysis	AfriSenti, NaijaSenti, Swhaili-Tweet-Sentiment	shmuhammad/AfriSenti-twitter-sentiment,HausaNLP/NaijaSenti-Twitter,Davis/Swahili-tweet-sentiment	46.62k

• Paper: InkubaLM

Dataset statistics

Language	Number of samples
Hausa	5.8 M
Yoruba	6.4 M
Swahili	62.41 M
isiZulu	16.20 M
isiXhosa	25.35 M
English **	95.42 M

** Only for Machine Translation eng-xxx & xxx-eng

Uses

• This dataset can be utilized to instruct finetune language models. Only the train and dev set are currently open sourced. The test set will be made open-source at a later date.

How to Use

from datasets import load_dataset
data = load_dataset("lelapa/lelapa_instruct_datasets_no_test_set", "swahili_train,)

Structure

{
    "instruction": "Ainisha mada ya ...",
    "input": "Je chanjo ya corona ...",
    "output": "afya",
    "data_source": "masakhanews",
    "task": "news_classification",
}

Datasets task keys are as follows:

{
    "mmt": machine translate,
    "ner": maned entity recognition,
    "pos": part-of-speech tagging,
    "sentiment": sentiment analysis,
    "news_classification": topic classification
}

The keyword news_classification is used for the topic classification task because we pulled together the data from both topic classification and news classification datasets. We also considered/casted the latter as a topic classification problem.

Citation

@article{tonja2024inkubalm,
  title={InkubaLM: A small language model for low-resource African languages},
  author={Tonja, Atnafu Lambebo and Dossou, Bonaventure FP and Ojo, Jessica and Rajab, Jenalea and Thior, Fadel and Wairagala, Eric Peter and Anuoluwapo, Aremu and Moiloa, Pelonomi and Abbott, Jade and Marivate, Vukosi and others},
  journal={arXiv preprint arXiv:2408.17024},
  year={2024}
}

Dataset Card Authors

Lelapa AI - Fundamental Research Team

Dataset Card Contact

Lelapa AI