Show HN: Tiny chess model, 392 KiB, ~1523 elo (github.com)

A new project called chesslm has been introduced, featuring a compact neural chess evaluator that operates with a mere 392 KiB footprint. Utilizing a small NumPy neural network along with a fixed material evaluator, chesslm effectively chooses chess moves and allows users to play in a Tkinter desktop environment or via command line. In a 40-game test series against a strength-limited version of Stockfish, the model achieved a nominal Elo rating of approximately 1523, showcasing its competitive potential despite being a simplified implementation. Significantly, chesslm highlights the drive toward lightweight AI models that can function efficiently on consumer-grade hardware without reliance on external engines like Stockfish. It uses a straightforward architecture with 100,353 parameters and employs techniques like alpha-beta pruning for move selection. This accessibility makes chessml a valuable tool for hobbyists and developers interested in AI and machine learning, particularly in the realm of game AI development. With the project being open-source, it encourages community contributions and experimentation, promising further advancements in compact AI solutions for gaming.