This paper addresses the use of games as benchmarks and test environments for artificial intelligence research. Existing game description languages have enabled researchers to generalize their algorithms and approaches across multiple games by compiling game-specific code into executable and simulable environments. In this paper, we present a domain-specific language for board games, taking into account advances in hardware acceleration that have contributed to the advancement of reinforcement learning (RL). This framework, called Ludax, combines the generality of game description languages with the speed of modern parallel processing hardware, and is designed to fit into existing deep learning pipelines. Ludax aims to accelerate game research across the spectrum from RL to cognitive science, providing fast simulations and flexible representations. The Ludax framework and existing board game implementations are open-sourced, along with a detailed description of the Ludax description language, compilation process, speed benchmarking, and RL agent training demos.
