基于大模型知识蒸馏的代码摘要自动生成

打开文本图片集

引用格式：,,,等.基于大模型知识蒸馏的代码摘要自动生成[J].指挥控制与仿真,2025,47（4):27-33.YOU G,LIUWJ,etalodemmaaedrgeodelkolegedistilionJmandorol&iulatio,47 (4) :27-33.

中图分类号：TP311;TP18 文献标志码：A DOI:10.3969/j.issn.1673-3819.2025.04.005

Code summarization based on large model knowledge distillation YOU Gang，LIU Wenjie，LI Meipeng，SUN Liqun，WANG Lian， TIAN Tieku （Unit 96941 of PLA，Beijing ,China）

Abstract：Code summarization is a short natural language description ofsource code.Summariesareusuallyonlyone sentencelong，buttheyaretheprimarywayfordevelopers tounderstandcode.Recently，productsbasedonlargelanguagemodels（such as ChatGPT）havedemonstratedastrong abilitytogenerate these descriptions.However,touse these tools，programmersmustsend theircodetoanuntrustedthirdpartyforprocessing（forexample,throughAPIcalls），butthismethod is unacceptable to many organizations.This paper presents an alternative:weuse the exampleoutput generated by GPT-3.5 totrainanopensourcemodel throughaprocessrelatedtoknowledgedistilation.Enablingsmall models（with 350millon parameters） to also be comparable to GPT-3.5 in code summarization tasks.

Keywords: code summarization；large model；knowledge distillation

代码摘要是源代码的简短自然语言描述。（剩余15134字）