LLM-based methods for the creation of unit tests in game development
Ciprian Păduraru, Adelina Nicoleta Staicu, Alin Ştefănescu
Abstract
Problems related to the quality of games, whether on the initial release or after updates, can lead to player dissatisfaction, media attention, and potential financial setbacks. These issues can stem from software bugs, performance bottlenecks, or security vulnerabilities. Despite these challenges, game developers often rely on manual playtesting, highlighting the need for more robust and automated processes in game development. This research explores the application of Large Language Models (LLMs) to automate the creation of unit tests in game development, focusing on strongly typed programming languages such as C++ and C#, which are widely used in the industry. The study focuses on fine-tuning Code Llama, an advanced code generation model, to address common scenarios in game development, including game engines and specific APIs or backends. Although the prototyping and evaluations primarily took place within the Unity game engine, the proposed methods can be adapted to other internal or publicly available solutions. The evaluation results demonstrate these methods’ effectiveness in improving existing unit test suites or automatically generating new tests based on natural language descriptions of class contexts and targeted methods.