awras/lahja-it-dataset-v1

lahja-it-dataset-v1 is an open-source instruction-tuning dataset for Algerian Darija. It contains around 89.9k conversation examples (80.9k train, 9k eval) designed to teach a language model how to understand and respond naturally in Darija, including the typical code-switching between Arabic, French, and local expressions that Algerians use every day.