Logo Lanfrica

mwkhettab/darija-translator

Domain:

natural language processing

Record type:

softwaremodel
Creator:
mwk
Host:
A full-stack Moroccan Arabic translator # Darija Translator ๐Ÿ‡ฒ๐Ÿ‡ฆ A free, open-source Darija (Moroccan Arabic) translation app supporting 20+ languages. The system combines a fine-tuned Darija model with high-quality multilingual models to deliver accurate translations for real, informal Darija. Try it here. --- ## Features - ๐ŸŒ **Multiple Languages** โ€“ Translate to/from Darija and major world languages **๐Ÿ” Model-Aware Pipeline** โ€“ Uses a Darija-specific model for Moroccan Arabic and multilingual models for other languages - ๐Ÿง  **Custom Darija Model** โ€“ Fine-tuned specifically on real Darija data - ๐Ÿ–ฅ **Multilingual UI** โ€“ English, French, Arabic - ๐Ÿ’ธ **Free & Open Source** --- ## Architecture `client/` - SvelteKit frontend `server/` - FastAPI backend `model/` - Training & inference scripts --- ## Custom Model - **Base**: facebook/nllb-200-distilled-600M - **Training Data**: - Darija Open Dataset - Curated real-world Darija sentence pairs - **Goal**: Accurate informal Darija, including code-switching ๐Ÿ‘‰ Models on Hugging Face: huggingface.co --- ## Examples ### Darija โ†’ English **Input** Kidayr ta9s barra daba? **Output** Howโ€™s the weather outside right now? --- ### English โ†’ Darija **Input** Where can I find a restaurant? I'm really hungry. **Output** ููŠู† ู†ู‚ุฏุฑ ู†ู„ู‚ู‰ ุฑูŠุณุทูˆุฑุฉุŸ ุฃู†ุง ุฌูˆุน ุจุฒุงู **Latin** fin n9dr nl9a ristoura? ana jou3an bzaf --- ## Tech Stack - Frontend: SvelteKit + Paraglide - Backend: FastAPI - ML: PyTorch, Hugging Face - Models: NLLB + OpenAI --- If you find this useful, please leave a โญ