A Python chess application that utilizes Llama 2 models as LLM players. The goal of this application is to demonstrate the chess skills of Llama 2 models. This repository is a great example for RAG (Retrieval Augmented Generation) implementations and prompt engineering.
To play the game, run start_game.py inside the directory:
python start_game.pyFor your information, the chess skills of Llama 2 models are very limited. Their skills are equivalent to 120 ELO player. In the future, I plan to finetune a Llama model to see whether I can turn them into Bobby Fischer. Since I am too lazy, I did not integrate a chess GUI for this application, rather I used Python's chess library. Sometimes the game crashes, as the models do not respond with a proper format. Feel free to send pull requests.
The application uses Replicate's models. To run this application, a Replicate API key is required. Other package requirements are included in requirements.txt. Install them by the following command:
pip install -r requirements.txt