Pgx: Hardware-Accelerated Parallel Game Simulators for Reinforcement Learning

TL;DR

Pgx is a GPU/TPU-optimized suite of parallel board game simulators in JAX, enabling 10-100x faster reinforcement learning environment simulations for research and development.

Contribution

The paper introduces Pgx, a novel high-performance, hardware-accelerated RL environment suite for board games, leveraging JAX's auto-vectorization for scalable simulation.

Findings

Simulates RL environments 10-100x faster than existing Python implementations

Supports popular benchmark games like chess and Go

Enables rapid training of RL algorithms like Gumbel AlphaZero

Abstract

We propose Pgx, a suite of board game reinforcement learning (RL) environments written in JAX and optimized for GPU/TPU accelerators. By leveraging JAX's auto-vectorization and parallelization over accelerators, Pgx can efficiently scale to thousands of simultaneous simulations over accelerators. In our experiments on a DGX-A100 workstation, we discovered that Pgx can simulate RL environments 10-100x faster than existing implementations available in Python. Pgx includes RL environments commonly used as benchmarks in RL research, such as backgammon, chess, shogi, and Go. Additionally, Pgx offers miniature game sets and baseline models to facilitate rapid research cycles. We demonstrate the efficient training of the Gumbel AlphaZero algorithm with Pgx environments. Overall, Pgx provides high-performance environment simulators for researchers to accelerate their RL experiments. Pgx is available at http://github.com/sotetsuk/pgx.