EvelienUU/chess-qwen-finetuned-v2
EvelienUU/chess-qwen-finetuned-v2 is a causal language model developed by Evelien van Driel, fine-tuned from Qwen2.5-0.5B-Instruct. This model is specifically optimized for chess move prediction, taking FEN notation and legal moves as input. It outputs the best move in UCI format, making it suitable for integration into chess-playing AI systems.
Loading preview...
Model Overview
EvelienUU/chess-qwen-finetuned-v2 is a specialized causal language model, developed by Evelien van Driel, designed for chess move prediction. It is fine-tuned from the Qwen/Qwen2.5-0.5B-Instruct base model, focusing on generating optimal chess moves given a board state.
Key Capabilities
- Chess Move Prediction: The model accepts a board position in standard FEN (Forsyth-Edwards Notation) along with a list of legal moves.
- UCI Output: It outputs the predicted best move in UCI (Universal Chess Interface) format.
- Fine-tuned Performance: This version (v2) was further fine-tuned from an earlier iteration (v1) on an expanded dataset, building on 100,000 chess positions from the
aicrowd/ChessExplaineddataset. All training moves are validated by Stockfish.
Training Details
The model was trained using the LoRA (Low-Rank Adaptation) method over 3 epochs with a batch size of 16 and a learning rate of 2e-4, utilizing a Google Colab T4 GPU. It specifically targets the INFOMTALC 2026 project at Utrecht University for integration into a TransformerPlayer class.
Ideal Use Cases
- Chess AI Development: Integrating into chess engines or AI players that require a model to suggest or predict moves.
- Research in Game AI: Exploring the application of fine-tuned language models for strategic game decision-making.