lost-in-darija-marian
Fine-tuned version of Helsinki-NLP/opus-mt-en-ar for English โ Moroccan Darija translation.
๐ Try it live on HuggingFace Spaces
Fine-tuned version of Helsinki-NLP/opus-mt-en-ar for English โ Moroccan Darija translation.
Model Details
- Base model: Helsinki-NLP/opus-mt-en-ar
- Language pair: English โ Moroccan Darija (Arabic script)
- License: Apache 2.0
Training
- Dataset: DODa (~39,000 pairs, eng โ darija_ar)
- Epochs: 8 (best checkpoint saved at epoch 7 via validation loss tracking)
- Learning rate: 2e-5
- Optimizer: AdamW
- Batch size: 16
- Hardware: Google Colab T4
Results
| Metric | Score |
|---|---|
| BLEU (max_ngram_order=3) | 18.30 |
Usage
from transformers import MarianTokenizer, MarianMTModel
tokenizer = MarianTokenizer.from_pretrained("ILyass-42/lost-in-darija-marian")
model = MarianMTModel.from_pretrained("ILyass-42/lost-in-darija-marian")
phrase = ">>ary<< Where is the market?"
inputs = tokenizer(phrase, return_tensors="pt")
translated = model.generate(**inputs)
print(tokenizer.decode(translated[0], skip_special_tokens=True))
# ููู ูุงูู ุณููุ
Project
Part of lost-in-darija โ a speech-to-speech pipeline for English โ Moroccan Darija translation targeting English-speaking tourists.
- Downloads last month
- 7
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support
Model tree for ILyass-42/lost-in-darija-marian
Base model
Helsinki-NLP/opus-mt-en-ar