Mate-in-Two (Qwen + RL Chess Puzzle Solver)

Instructions

Setup Stockfish
- Download link
- On Mac, allow app in Systems Settings -> Privacy & Security
- Set STOCKFISH_PATH environment variable to the Stockfish executable

From TinyZero
- Twitter thread
- Model size, base vs. instruct, RL algorithm
Effect of curriculum training
- Increasing elo vs. random ordering
Board representation
- FEN vs. piece-list vs. ascii
- Qwen VL/2.5-VL + image of board (?)
Reward shaping
- Binary 0/1 vs. Stockfish score delta (centipawns)

Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
data		data
train		train
.gitattributes		.gitattributes
README.md		README.md
requirements.txt		requirements.txt