An open-source Go engine that replicates AlphaGo Zero's architecture, learning solely through self-play without human knowledge.
Leela Zero is an open-source Go engine that replicates DeepMind's AlphaGo Zero algorithm. It learns to play Go at a superhuman level through self-play reinforcement learning, without any human game data or domain knowledge. The project enables community-driven distributed training to recreate the capabilities of the original AlphaGo Zero system.
Go enthusiasts, AI researchers, and developers interested in reinforcement learning, game AI, or distributed computing projects. It's particularly valuable for those studying the AlphaGo Zero architecture.
Leela Zero provides a fully open-source, transparent implementation of the groundbreaking AlphaGo Zero algorithm. Unlike proprietary systems, it allows community participation in training and offers educational insights into state-of-the-art game AI techniques.
Go engine with no human-provided knowledge, modeled after the AlphaGo Zero paper.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Implements the exact Monte Carlo Tree Search and deep residual neural network from the paper, providing a transparent educational tool for AI research.
Leverages volunteer GPU computing worldwide to collaboratively train networks, enabling open participation in recreating superhuman AI.
Supports OpenCL for efficient neural network inference on consumer GPUs, making high-performance Go analysis accessible.
Compatible with GTP interfaces like Lizzie and Sabaki, allowing seamless integration with existing Go software for gameplay and analysis.
Weights must be obtained separately, and generating them from scratch requires immense computational resources—estimated at 1700 years on commodity hardware—relying entirely on community effort.
Optimal performance demands a recent GPU; the README warns that CPU-only use on older hardware results in 'outright bad' performance, limiting accessibility.
On macOS and Linux, installation requires compiling from source with multiple dependencies like Boost and OpenCL, which is less user-friendly compared to Windows releases.
leela-zero is an open-source alternative to the following products: