awras/lahja-dataset-v1

An extensively cleaned and deduplicated text corpus specifically curated for the Algerian dialect (Darja), combining code-switching patterns between Arabic and French. Noise, duplicate entries, and excessively short sentences have been systematically filtered out to optimize the corpus for training and evaluating language models on low-resource Algerian African dialects. Total Rows: 1,641,779 Train Split: 1,477,601 rows