Wilsonafolabi/Rule-based-...Roman-numerals-to-Yoruba-

# Yoruba Roman Numerals Translation System This project is a fine-tuned sequence-to-sequence model that translates **Roman numerals** into their **Yoruba language representations**, trained on over 6000 samples. It includes both the model training code and a deployed web demo using Flask and Hugging Face. --- ## 🔍 Project Overview - **Model**: Davlan/byt5-base-eng-yor-mt fine-tuned on Roman-to-Yoruba numeral pairs. - **Dataset**: 6,000+ entries mapping Roman numerals to Yoruba equivalents. - **Goal**: Build an expert system that automatically translates Roman numerals (including extended Unicode variants like `V̅`) into Yoruba language. - **Approach**: - Fine-tune a pretrained multilingual model (ByT5). - Deploy the trained model via Flask as a local web app and Hugging Face Space. --- ## 🚀 Online Demo - 🧠 Model: Hugging Face Model Page - 🌍 Live Web App: Hugging Face Spaces Demo --- ## 🧠 Model Training - Framework: PyTorch + Hugging Face Transformers - Tokenizer: `AutoTokenizer` from `Davlan/byt5-base-eng-yor-mt` - Loss: Cross-entropy - Optimizer: AdamW - Evaluation: BLEU, ROUGE, and exact match accuracy - Features: - Gradient clipping - Early stopping - Train-validation split - Performance tracking with loss plots ### 📈 Sample Metrics: - **Validation Accuracy**: ~`XX%` *(Replace with actual value)* - **BLEU Score**: `X.XXX` - **ROUGE-L**: `X.XXXX` --- ## 🛠 Local Setup & Usage ### 🔧 Requirements - Python 3.8+ - PyTorch - Transformers - Flask - Evaluate (for metrics) ### 🔨 Run Locally 1. Clone this repo: ```bash git clone github.com cd yoruba-roman-numerals ``` 2. Install dependencies: ```bash pip install -r requirements.txt ``` 3. Start the Flask app: ```bash python app.py ``` 4. Visit `localhost` in your browser. --- ## 📡 API Usage **Endpoint:** `/convert` **Method:** `POST` **Request (JSON):** ```json { "input": "XIV" }