Aleph is a chess engine combining a cross-attention transformer evaluator with a search-contempt MCTS algorithm. It is designed to be strong and efficient, built on modern research in computer chess.
The name "Aleph" refers first to Aleph-Null, the smallest infinity from set theory in mathematics, and is second a nod to AlphaZero and Leela Chess Zero, whose empirical data informed the decisions for the neural network architecture.
Aleph is currently under extremely early development. The code within this repo is not a functional engine yet. More is written than is shown here, however, I am taking time to ensure a high standard before adding code from my local copy into staging, and then an even higher quality before adding it into the main branch. Often the staging branch is broken. However, main is always functional.
Aleph requires no special runtime dependencies beyond the standard C++20-compatible runtime. A GPU is optional, however CPU-only usage of the neural network is considerably slower and worse-performing than the GPU-accelerated performance.
See ./docs/src/developer/building.md.
See ./SECURITY.md.
See ./CONTRIBUTING.md.
See ./CODE_OF_CONDUCT.md.
The initial inspiration for this project came from reading through Lc0's update about transformers here.
See ./docs/appendix/acknowledgements.md.
Aleph is licensed under the GPLv3 license. See ./LICENSE.