一种基于代码语言模型的程序缺陷自动修复方法及系统
By constructing an automatic program defect repair method based on a code language model, and using open-source project corpora and defect-repair code datasets for multi-round iterative fine-tuning training, the method solves the problems of long repair time, high cost and poor applicability in existing technologies, and achieves efficient and universally applicable program defect repair.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN INST OF TECH
- Filing Date
- 2023-06-19
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for fixing program defects are time-consuming, costly, and lack universality. Traditional manual repair relies on the professional skills of developers, heuristic rule-based methods have limited applicability, and patches generated by neural networks are of poor quality and have a low success rate.
We construct an automatic program defect repair method based on a code language model. We build the model using an open-source project corpus, construct a dataset using program defects and correct code samples, establish a defect repair difficulty discrimination strategy and a sample scheduling function, conduct multiple rounds of iterative fine-tuning training, and generate and verify candidate patches.
It improves the success rate of program defect repair and the quality of generated patches, enables unified repair of multiple types of defects, reduces the time cost of manual repair, and provides auxiliary tools for developers.
Smart Images
Figure CN116755753B_ABST