Logo Lanfrica
  • Accueil
  • Atlas
  • Analyses
  • Documentation
  • Sign in

© 2026 Lanfrica. Tous droits réservés. Tous les droits d'auteur des ressources affichées sur le site Web Lanfrica appartiennent aux détenteurs de droits d'auteur d'origine, sauf indication contraire explicite.

atlasia-ma/Al-Atlas-Moroccan-Darija-Pretraining

Domaine:

natural language processing

Type de record:

datasetmodel
Créateur:
atl
Hôte:
Training and data analysis code for Al-Atlas dataset and models # Al-Atlas: Data, Models, Evals ## TL;DR We present a comprehensive dataset for Moroccan darija, addressing the lack of resources for this widely spoken dialect. We detail our collection methodology, provide thorough data analysis, and demonstrate performance improvements in both masked and causal language models after training on this dataset. 1. Try our **demo** showcasing our **AlAtlas LLM** for Moroccan darija. 2. Try our **demo** showcasing our **Masked Language Model** for Moroccan darija. 3. Our **Atlaset** dataset on Hugging Face. 4. **Code** is available on Github. ## 1. Challenges in Moroccan Darija Moroccan darija stands at a fascinating linguistic crossroads, embodying a rich tapestry of influences from Arabic foundations interwoven with Berber, French, and Spanish elements. This vibrant dialect, though widely spoken throughout Morocco, presents unique challenges in the computational linguistics domain. Unlike Modern Standard Arabic (MSA) and other well-resourced languages, Moroccan darija exists in a digital resource vacuum, with **no standardized writing system** to guide its representation. This **orthographic fluidity** where texts may appear in Arabic script, Latin characters, or creative hybridizations, creates significant hurdles for traditional Natural Language Processing (NLP) approaches. The dialect's natural tendency toward **code-switching** and **regional variations** further compounds these challenges, placing Moroccan darija squarely in the category of **under-resourced languages** despite its **cultural significance and widespread use** in daily Moroccan life. These challenges create a significant barrier for developing AI applications for the millions of Moroccan darija speakers, highlighting the **need for dedicated resources** such as datasets, foundational models and NLP tools. ## 2. Data Collection: Atlaset, A Mother Dataset We've assembled what we believe is the most complete collection of publicly available Moroccan darija …

Visit

github.com

Tasks

language modeling

Languages

Arabic, Algerian SpokenArabic, Moroccan SpokenBerber

Similaires

AhmedH005/Al-Atlas-Moroccan-Darija-Pretrainingatlasia/Moroccan-Darija-LLM-Battle-Al-Atlasatlasia-ma/darija-translatoroelbourki/atlasia-Al-Atlas-0.5B_classification_mental_darijaoelbourki/atlasia-Al-Atlas-0.5B_64_3_classification_mental_darijaoelbourki/atlasia-Al-Atlas-0.5B_128_classification_mental_darija

AhmedH005/Al-Atlas-Moroccan-Darija-Pretraining

Pretraining code and dataset analysis for the Al-Atlas family of open-source LLMs trained on authent

atlasia/Moroccan-Darija-LLM-Battle-Al-Atlas

atlasia-ma/darija-translator

# Darija Translator Fine-tuning LFM2.5-230M for English↔Darija translation with LoRA (Unsloth + TRL

oelbourki/atlasia-Al-Atlas-0.5B_classification_mental_darija

oelbourki/atlasia-Al-Atlas-0.5B_64_3_classification_mental_darija

oelbourki/atlasia-Al-Atlas-0.5B_128_classification_mental_darija