yaneurao/YaneuraOu

What it solves

YaneuraOu is a high-performance Shogi (Japanese chess) AI engine designed to compete at the highest levels of computer Shogi. It provides a professional-grade search and evaluation framework that has won numerous world championships, including WCSC29, and serves as the foundation for many of the world's strongest Shogi AI players.

How it works

The project consists of two primary components: the main YaneuraOu engine and FukauraOu. The engine is USI (Universal Shogi Interface) compliant, allowing it to work with various GUIs. It supports advanced search techniques like Pondering (thinking during the opponent's turn) and MultiPV (outputting multiple candidate moves). For evaluation, it supports multiple formats including NNUE (Efficiently Updatable Neural Networks) and SFNN, as well as traditional KPPT and KPP_KKPT functions. It is designed for extreme parallelism, supporting up to 256 threads, and utilizes a highly efficient on-the-fly book (opening book) system to avoid loading massive files into memory.

Who it’s for

• Competitive Shogi AI Developers: Those looking to build or refine world-class Shogi engines.
• Shogi Enthusiasts: Players who want to use the strongest available AI for analysis or practice.
• Researchers: Individuals interested in high-performance game AI and search algorithms.

Highlights

• World-Class Performance: Multiple-time winner of the World Computer Shogi Championship (WCSC) and Denryu tournaments.
• Broad Compatibility: Supports Windows, Ubuntu, macOS, and ARM platforms.
• Hignly Scalable: Capable of handling up to 256 threads and transposition tables up to 33TB.
• Flexible Evaluation: Supports various neural network-based evaluation functions (NNUE, SFNN) and traditional methods.
• USI Compliant: Fully compatible with the standard Universal Shogi Interface protocol.