A fine-tuned Qwen2.5-0.5B-Instruct model for chess move prediction. Given a
board position in FEN notation, the model predicts the best move in UCI
format (e.g., e2e4).
Model Details
Base model: Qwen/Qwen2.5-0.5B-Instruct
Fine-tuning method: SFT with LoRA (r=64, alpha=128)
Task: Chess move prediction
Training
Data: 500,000 FEN-move pairs from Stockfish-vs-Stockfish games
Loss masking: Loss computed only on move tokens, not the prompt
