Optimizing LLMs for Code Generation: Which Hyperparameter Settings Yield the Best Results?

• Published in: Proceedings / Asia Pacific Software Engineering Conference pp. 281 - 290
• Main Authors: Arora, Chetan, Sayeed, Ahnaf Ibn, Licorish, Sherlock, Wang, Fanyu, Treude, Christoph
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.12.2024
• Subjects: Code Generation; Codes; Empirical Study; Generative AI; Hyperparameter Analysis; Large language models; LLM Temperature; LLMs; Python; Regression tree analysis; Requirements management; Software development management; Software engineering; Systematics; Temperature distribution; Tuning

Large Language Models (LLMs), such as GPT models, are increasingly used in software engineering for various tasks, such as code generation, requirements management, and debugging. While automating these tasks has garnered significant attention, a systematic study on the impact of varying hyperparame...