Atlaset: A Curated Dataset of Moroccan Darija

This dataset is a comprehensive, carefully curated collection of text data specifically for Moroccan Darija, the Arabic dialect spoken in Morocco. It combines various sources to provide a diverse and accurate representation of the language.

This dataset was curated by Abdelaziz Bounhar and is particularly suited for tasks such as:

• Learning word embeddings for Moroccan Darija
• Training NLP models for tasks like language modeling, text generation (Pretraining large language models (LLMs) for instance), and more

As of today, the total number of tokens (using Jais-13B)is:

• For the training set: 155,501,098 tokens.
• For the test set: 19,187 tokens.

Features

• Textual Data: The dataset contains a rich collection of text, including social media posts, news articles, and more, providing a diverse range of language use.
• Language Variety: The dataset spans various registers, capturing both formal and informal expressions across different domains.

Data Sources

Training

Training Dataset Source	Count
atlasia/moroccan_darija_domain_classifier_dataset	188,777
goud-sum	158,277
NQ-Open-AYA	136,780
sharegpt	125,488
cot	94,430
MTCD	58,629
atlasia/Social_Media_Darija_DS/facebook_comments	55,647
flan_v2	37,746
wizardlm	37,577
sawalni-ai/fw-darija	37,352
gpt4_alpaca	36,211
aboutaleb/moroccantourism	30,433
MArSum	24,229
common-crawl	19,128
oasst1	17,657
atlasia/DODa-audio-dataset-V3	11,517
tachicart/mo_darija_merged	11,496
medmac01/moroccan_history_qa	11,156
atlasia/Social_Media_Darija_DS/default	10,678
Lyte/Darija-QA	10,439
atlasia/Moroccan-Darija-Wiki-Dataset	9,460
DarijaBridge	6,587
Darija-Stories-Dataset	6,142
dataset_dyal_darija	5,306
moroccan_darija_wikipedia_dataset	4,862
abdeljalilELmajjodi/darija_qa_ds	4,538
MAC Corpus	3,632
atlasia/Social_Media_Darija_DS/facebook_uncleand_posts	3,360
atlasia/darija_english/stories	2,961
JasperV13/Darija_instruct	1,683
learnmoroccan.com	1,673
lima	1,559
ArabicDarija_xP3x	997
atlasia/darija_english	867
SoftAge-AI/sft-conversational_dataset	792
atlasia/Moroccan-Darija-Wiki-Audio-Dataset	551
loecsen.com	428
auto-math	415
abdeljalilELmajjodi/darija_topic_ds	384
stanford	352
bourbouh/moroccan-darija-youtube-subtitles	335
darija_speech_to_text	322
darija_youtube_subtitles	247
translation-rows	246
wikihow	140
atlasia/facebook_darija_dataset	18
ML101	7
DarijaEnglish-xP3x	3
abdeljalilELmajjodi/abdeljalil_darija_classification_ds	1

Test

Test Dataset Source	Count
tachicart/mo_darija_merged	1,799
atlasia/TerjamaBench	830
MAC Corpus	54

Citation

@misc{bounhar2025atlaset,
  title = {Atlaset: A Moroccan Darija Text Corpus},
  author = {Abdelaziz Bounhar},
  year = {2025},
  howpublished = {\url{https://huggingface.co/datasets/atlasia/Atlaset/}}
}

Acknowledgments

Special thanks to Abdeljalil El Majjodi for his contribution. We would like to extend our sincere thanks and appreciation to all the contributors and members of the Moroccan AI community who provided these valuable datasets. Their efforts in gathering, curating, and sharing this data have been crucial in making this resource possible. Special thanks to the individuals and organizations behind each dataset for their commitment to advancing research and development in Moroccan Darija. Your contributions are deeply appreciated.

Contribution

We welcome contributions to improve and expand the dataset. If you have additional text resources in Moroccan Darija or suggestions for improvements, feel free to submit a pull request.