UBC-NLP/nilechat-fw-edu-mor

fineweb-edu-Morocco: A Resource for Advancing Moroccan Arabic Language Models

fineweb-edu-Morocco is a substantial dataset specifically developed to foster the creation and improvement of language models for the Moroccan Arabic dialect (Darija). This resource consists of a machine-translated version of the renowned FineWeb-Edu dataset.

Dataset Snapshot:

• Language: Moroccan Arabic Dialect (Darija)
• Corpus Size: 5.25 million rows
• License: Creative Commons Attribution-NonCommercial 4.0 International (CC-BY-NC-4.0)
• Origin: The Hugging Face FineWeb-Edu dataset.
• Translation Method: Machine-translated using the CohereLabs/c4ai-command-r-plus-08-2024 model.

Applications and Utility:

The fineweb-edu-Morocco dataset is primarily designed for pre-training large language models (LLMs). It has notably been utilized in the development of NileChat-3B, a significant LLM focused on Moroccan and Egyptian Arabic dialects. Researchers and developers can leverage this dataset to significantly enhance the dialectal capabilities and coverage of their Arabic language models, particularly for the Moroccan context.

Python Implementation:

Accessing and utilizing the fineweb-edu-Morocco dataset is straightforward using the Hugging Face datasets library in Python. The following snippet demonstrates how to load the dataset in a streaming fashion:

from datasets import load_dataset

# Load the dataset with streaming
dataset = load_dataset("UBC-NLP/nilechat-fw-edu-mor", streaming=True)

# Print the first example from the training split
print(next(iter(dataset["train"])))

Citation:

If you use this dataset in your research, please cite the following paper:

@misc{mekki2025nilechatlinguisticallydiverseculturally,
      title={NileChat: Towards Linguistically Diverse and Culturally Aware LLMs for Local Communities}, 
      author={Abdellah El Mekki and Houdaifa Atou and Omer Nacar and Shady Shehata and Muhammad Abdul-Mageed},
      year={2025},
      eprint={2505.18383},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.18383}, 
}