Ted Wong publishes machine-learning projects through wtedw, including nanoAlphaZero, an implementation for learning game play through self-play and search.[1][3]
Start with nanoAlphaZero – Train a grandmaster-level chess model in 24h with TPUs →
GitHub profile 
X · _wtedw_ Contributions
nanoAlphaZero combines a model’s move preferences and value estimates with tree search, then trains from games it plays against itself. Its JAX implementation fuses self-play, search and training into one compiled call, reducing coordination overhead between stages. The repository supplies game environments and example commands; its claims about chess strength and training time refer to the reported hardware and evaluation setup.[3]
Tools, writing & talks
Writing, research & talks
Start here
A JAX implementation combining self-play, tree search and model training for board games.
Begin with a small supplied game and trace the self-play/search/training loop before trying chess. Python, reinforcement learning and access to suitable accelerator hardware help; compare the exact configuration behind reported strength and training-time claims.
Creator and repository owner, linked from personal site[1][3]Original source