LLM Amharic

The best Amharic Large language model! Our goal is to help African businesses by using new technology in AI. By using advanced AI, this project aims to provide smooth, Amharic support across different platforms.

This repository contains scripts and instructions to fine-tune the LLaMA-2-7b-chat model for Amharic customer support using data stored in a PostgreSQL database.

Project Structure

llm-amharic/
├── data/
│ ├── tokenized_dataset/
│ └── load_data_to_db.py
├── docker/
│ ├── Dockerfile
│ └── docker-compose.yml
├── scripts/
│ ├── evaluate_modle.py
│ ├── inference_script.py
│ ├── tokenize_data.py
│ ├── train_model.py
│ └── train_tokenizer.py
├── utils/
│ ├── data_preprocessing.py
│ └── fetch_data_from_db.py
├── .gitignore
├── amharic.model
├── amharic.vocab
├── README.md
└── README.md

Getting Started

Prerequisites

• Python 3.8+
• PostgreSQL
• CUDA-enabled GPU (optional but recommended for training)

Installation

1. Clone the repository:

   git clone https://github.com/10-academy-w5-group-2/llm-amharic.git
   cd llm-amharic

2. Set up a virtual environment:

   python3 -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install Requirements:

   pip install -r requirements.txt

4. Train Tokenizer

   python scripts/train_tokenizer.py

5. Fine-Tune the Model

   python scripts/train_model.py

6. Evaluate the model

   python scripts/evaluate_model.py

Database Setup

Ensure your PostgreSQL database is set up with the required data. The table should have a column containing the Amharic text data for training.

Dockerfile

Use docker/Dockerfile to containerize and run the entire project

Contributing

1. Fork the repository.
2. Create your feature branch (git checkout -b feature/your-feature).
3. Commit your changes (git commit -m 'Add your feature').
4. Push to the branch (git push origin feature/your-feature).
5. Open a pull request.

License

This project is licensed under the MIT License.

Contributors

• @abyt101 - Abraham Teka
• Melaku Alehegn
• Grace Nyutu
• Henock Kinfegebriel

Challenge by