Improving LLM-Generated Code Quality with GRPO
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866909634709684224 |
|---|---|
| author | Robeyns, Maxime Aitchison, Laurence |
| author_facet | Robeyns, Maxime Aitchison, Laurence |
| contents | Large Language Models (LLMs) are gaining widespread use for code generation. Recent training procedures use execution feedback as a reward signal, typically focusing on the functional correctness of the code, using unit test pass rate as a reward signal. However, this reward signal fails to capture notions of maintainability, quality and safety of the code produced. We address this under-explored area and develop a comprehensive library to quantify various aspects of code quality, and use it as a reward in GRPO. We find GRPO increases code quality according to this measure, which is confirmed by expert, blinded human annotators. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_02211 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Improving LLM-Generated Code Quality with GRPO Robeyns, Maxime Aitchison, Laurence Artificial Intelligence Large Language Models (LLMs) are gaining widespread use for code generation. Recent training procedures use execution feedback as a reward signal, typically focusing on the functional correctness of the code, using unit test pass rate as a reward signal. However, this reward signal fails to capture notions of maintainability, quality and safety of the code produced. We address this under-explored area and develop a comprehensive library to quantify various aspects of code quality, and use it as a reward in GRPO. We find GRPO increases code quality according to this measure, which is confirmed by expert, blinded human annotators. |
| title | Improving LLM-Generated Code Quality with GRPO |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2506.02211 |