Distilrr: Transferring Code Repair For Low-resource Programming Languages

Wong Kyle, Amayuelas Alfonso, Pan Liangming, Wang William Yang. Arxiv 2024

[Paper]    
Applications Tools

Large language models (LLMs) have shown remarkable performance on code generation tasks. A recent application of LLMs for code generation is iterative code repair, where a model fixes an incorrect program by rationalizing about errors and generating a new program. However, code repair is primarily studied on high-resource languages like Python, and the framework’s efficacy is under-explored on low-resource languages. To apply code repair for low-resource languages, we propose Distilling Low-Resource Repairs (DistiLRR), an approach that transfers the reasoning and code generation ability from a teacher model to a student model. Our results show that DistiLRR consistently outperforms baselines on low-resource languages, but has similar performance on high-resource languages. To investigate this behavior, we perform a further analysis and find that the correlation between rationale quality and code correctness is weaker than previously perceived. We hypothesize this weakness is magnified in low-resource settings where base models lack deep knowledge of a programming language, leading to wavering benefits of code repair between high-resource and low-resource languages.

Similar Work