Leveraging Large Language Models for the Automatic Implementation of Problems in Optimization Frameworks

Abstract

With the growing complexity of optimization tasks across domains, practitioners increasingly rely on established frameworks that offer state-of-the-art algorithms. However, a gap remains between domain expertise and the programming skills needed to implement problems within such frameworks. This paper presents a novel approach that leverages large language models (LLMs) to automate the implementation of continuous multi-objective optimization problems in the jMetal framework. The methodology accepts a textual description of a problem and uses a fine-tuned version of the Mistral LLM to generate the corresponding code. Its effectiveness is validated on a set of real-world engineering optimization problems. To enhance usability, the model is integrated into a graphical tool that allows domain experts to seamlessly translate their problems into jMetal-compatible code. Both the fine-tuned model and the tool are released as open-source software, facilitating broader adoption and enabling more accessible use of advanced optimization techniques by non-programmers.