Hugging Face Overview
Model Overview
Bilmokhtar23/chess-qwen2.5-0.5b-v2 is a specialized language model built upon the Qwen2.5-0.5B-Instruct architecture. It has been fine-tuned specifically for the task of predicting optimal chess moves.
Key Capabilities
- Chess Move Prediction: Given a chess board state in FEN (Forsyth-Edwards Notation), the model predicts the most appropriate next move.
- UCI Output: Predicted moves are provided in standard UCI (Universal Chess Interface) format, such as
e2e4. - Compact Size: With 0.5 billion parameters, this model offers a relatively lightweight solution for chess AI integration.
Training Details
The model was fine-tuned using SFT (Supervised Fine-Tuning) with LoRA (Low-Rank Adaptation) for efficiency. The training dataset comprised 500,000 FEN-move pairs derived from Stockfish-vs-Stockfish games. Loss calculation during training was specifically masked to focus only on the move tokens, ignoring the prompt input.
Good For
- Integrating chess move prediction into applications or games.
- Developing chess analysis tools.
- Educational platforms demonstrating chess AI.