0xanwar/egypt-culture-prompt-tuning

# Egyptian Cultural Understanding with Prompt Tuning This repository implements a **parameter-efficient approach** to Egyptian cultural understanding using **prompt tuning** on BERT. The system classifies social intents in Egyptian cultural contexts into 5 core themes, addressing bias in vision-language models through culturally-aware text modeling. ## 🌟 Features - **Prompt Tuning Architecture**: Adapts BERT with only 0.01% trainable parameters - **Egyptian Cultural Themes**: Recognizes 5 core cultural dimensions: - Religious Celebration (Eid, Mawlid, Ramadan) - Family and Respect (elder care, familial bonds) - National Pride (Revolution Day, patriotism) - Cultural Heritage (Sham El Nessim, traditional foods) - Community Generosity (neighborhood gift-giving) - **Parameter Efficiency**: Updates only 11,520 parameters out of 110M - **Bias Mitigation**: Counters Western defaults in pretrained language models - **Modular Design**: Clean OOP structure for easy extension ## 📁 Project Structure ``` egypt-culture-prompt-tuning/ ├── src/ # Core source code │ ├── config.py # Configuration classes │ ├── data/ # Dataset handling │ ├── models/ # Model architectures │ ├── training/ # Training and evaluation logic │ └── utils/ # Utility functions ├── scripts/ # Executable scripts │ ├── train.py # Training script │ └── evaluate.py # Evaluation script ├── notebooks/ # Experiment notebooks ├── models/ # Trained models ├── data/ # Raw dataset files └── configs/ # Configuration files ``` ## 🚀 Quick Start ### Installation ```bash # Clone the repository git clone github.com cd egyptian_cultural_ai # Create virtual environment (recommended) python -m venv env source env/bin/activate # Linux/MacOS # env\Scripts\activate # Windows # Install dependencies pip install -r r …