METHODS FOR IMPROVING LARGE LANGUAGE MODELS TO IMPROVE THE QUALITY OF CODE REFACTORING
DOI:
https://doi.org/10.34185/1991-7848.itmm.2025.01.110Keywords:
refactoring, development, code, LLM, artificial intelligence, machine learning, large language model.Abstract
This study discusses methods for improving large language models (LLMs) for software refactoring. Using the methods of fine-tuning and indexing source code files, the paper addresses the issue of improving the results of using LLM for the task of refactoring code bases and improving the use of the context of language models. The proposed approaches are aimed at improving the quality of the source code after refactoring, as well as improving various large language models by changing the model itself or integrating it with additional software. The results of the methods implementation will be evaluated using Code Health and F1 metrics. This allows us to determine the effectiveness of the proposed solutions. The research results open up new perspectives for academic research and effective implementation in projects of various sizes.
References
Codescene. Refactoring-vs-Refuctoring-Advancing-the-state-of-AI-automated-code-improvements. URL: https://codescene.com/hubfs/whitepapers/Refactoring-vs-Refuctoring-Advancing-the-state-of-AI-automated-code-improvements.pdf
Rishab Ramesh, Akarsh Thejasvi Raju M, Harsha Vardhan Reddy, Sandeep Varma N. Fine-Tuning Large Language Models for Task Specific Data. URL: https://ieeexplore.ieee.org/document/10730913 DOI: 10.1109/ICNEWS60873.2024.10730913
Fu Q., Cho M., Merth T., Mehta S., Rastegari M., Najibi M. LazyLLM: DYNAMIC TOKEN PRUNING FOR EFFICIENT LONG CONTEXT LLM INFERENCE. arXiv preprint arXiv:2407.14057, 2024. URL: https://arxiv.org/pdf/2407.14057
Codescene. Code Health – How easy is your code to maintain and evolve? URL: https://codescene.io/docs/guides/technical/code-health.html
ENCORD. F1 Score in Machine Learning. URL: https://encord.com/blog/f1-score-in-machine-learning/
