Better Python Programming for all: With the focus on Maintainability and Code Style

The use of Large Language Models tailored for coding tasks has brought the quality of the code they generate into sharp focus, with concern for its maintainability and adherence to coding standards. Current studies have primarily centred on the functionality of Code LLMs, measured by their ability to pass predefined tests. This research seeks to shift this focus, improving Code LLMs to produce Python code that is not only functional but also maintainable and stylistically consistent.

We propose a refined experimental approach to fine-tune Code LLMs with the custom extended dataset we curated to aid in training and a more nuanced evaluation of these models. This method specifically targets the often-overlooked dimensions of maintainability and Code Style. While our findings improve code quality in Python, they also offer insights that could apply to other programming languages and various qualities of code generation.