williampepple1/ibani-nllb-model

# Ibani-English NLLB Translator A state-of-the-art translation model for Ibani ↔ English using Meta's No Language Left Behind (NLLB-200) architecture. ## 🌟 Features - **Bidirectional Translation**: Ibani → English and English → Ibani - **NLLB-200 Based**: Leverages Meta's multilingual model fine-tuned for Ibani - **Tonal Mark Support**: Properly handles Ibani special characters (á, ḅ, etc.) - **FastAPI Backend**: Production-ready REST API - **Google Colab Training**: Train on free GPU resources - **Local Inference**: Run the model on your machine ## 📋 Requirements - Python 3.10+ (3.11 recommended for training) - 8GB+ RAM for inference - GPU recommended for training (Colab/Kaggle provides free GPUs) ## 🚀 Quick Start ### 1. Installation ```bash # Clone the repository git clone github.com cd ibani-nllb-model # Create virtual environment python -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate # Install dependencies pip install -r requirements.txt ``` ### 2. Prepare Training Data Your `ibani_eng_training_data.json` file should already be in the root directory (Bible translation data): ```json [ { "translation": { "en": "This is the genealogy of Jesus the Messiah...", "ibani": "Mịị anịị diri bie anị fịnị ḅara Jizọs tádọ́apụ..." } } ] ``` The scripts automatically handle multiple formats: - `{"translation": {"en": "...", "ibani": "..."}}` (your current format) ✅ - `{"ibani_text": "...", "english_text": "..."}` (Bible format with metadata) - `{"ibani": "...", "english": "..."}` (simple format) ### 3. Train the Model **Option A: Google Colab (Recommended)** 1. Open `notebooks/train_ibani_nllb.ipynb` in Google Colab 2. Upload your training data 3. Run all cells 4. Download the trained model **Option B: Local Training** ```bash python scripts/train.py --data ibani_eng_training_data.json --output models/ibani-nllb ``` ### 4. Run the API ```bash python app.py ``` The API will be available a …