Transformer-based architecture for a translation model from english to moroccan darija, written in arabizi (latin alphabet)
Transformer-based model to translate english to darija (moroccan arabic) in latin alphabet. A pdf report of the proejct is available in english and french.
To test the model (with the according tokenizer) : HF Space
To see the dataset generated : HF Dataset
Training took 4.5 hours on one A100 GPU. Further details in the project report.