archeryue/AI-ZeroToOne
learning AI, from zero to one.
57Stars
11Forks
14Claude Commits
Jupyter NotebookLanguage
First Claude commit: Apr 8, 2026Last Claude commit: 21d agoDiscovered: Apr 9, 2026
Recent Claude Commits
Add pinned memory for fast CPU→GPU transfers in parallel self-play
38bb6e021d agoauthor_emailAdd multi-threaded parallel self-play with batched GPU inference
e17423721d agoauthor_emailAdd AlphaZero neural network and self-play training pipeline
61c999321d agoauthor_emailTrim PLAN.md: remove verbose cost justification, keep tech plan clean
99521b722d agoauthor_emailRemove stale pre-benchmark cost estimates, clean up PLAN.md
c9c6a5522d agoauthor_emailFix game counts: 13x13 gets more games than 9x9
0d6265e22d agoauthor_emailRevise training strategy: pure self-play for 9x9/13x13, pretrained 19x19
a60831522d agoauthor_emailSwitch from A100 to RTX 4090, add revised cost estimates from benchmarks
84f434522d agoauthor_emailOptimize MCTS: deque game pool, fixed-size paths, bulk expand
a7aa3b722d agoauthor_emailAdd test_mcts binary to .gitignore
3fdd2d822d agoauthor_emailImplement C++ MCTS with PUCT selection, virtual loss, and tree reuse (Step 6)
2f39ce822d agoauthor_emailAdd test-results to .gitignore
b10567c22d agoauthor_emailAdd Playwright e2e tests and frontend AI resign handling
e0e2b6722d agoauthor_emailAdd AI resign feature with MC Score Estimation for Go game
7ef441c22d agoauthor_email