Paper · arXiv 2502.16645
CODESYNC: Synchronizing Large Language Models with Dynamic Code Evolution at Scale
Chenlong Wang, Zhaoyang Chu, Zhengxiang Cheng, Xuyi Yang, Kaiyue Qiu, Yao Wan, Zhou Zhao, Xuanhua Shi, Dongping Chen
CODESYNCBENCH is introduced to evaluate LLMs' adaptation to evolving third-party library APIs, revealing their challenges with dynamic code changes.
Abstract
Large Language Models (LLMs) have exhibited exceptional performance in software engineering yet face challenges in adapting to continually evolving code knowledge, particularly regarding the frequent updates of third-party library APIs. This limitation, stemming from static pre-training datasets, often results in non-executable code or implementations with suboptimal safety and efficiency. To this end, this paper introduces CODESYNC, a data engine for identifying outdated code patterns and collecting real-time code knowledge updates from Python third-party libraries. Building upon CODESYNC, we develop CODESYNCBENCH, a comprehensive benchmark for assessing LLMs' ability to stay synchronized with code evolution, which covers real-world updates for 220 APIs from six Python libraries. Our benchmark offers 3,300 test cases across three evaluation tasks and an update-aware instruction tuning dataset consisting of 2,200 training samples. Extensive experiments on 14 state-of-the-art LLMs reveal that they struggle with dynamic code evolution, even with the support of advanced knowledge updating methods (e.g., DPO, ORPO, and SimPO). We believe that our benchmark can offer a strong foundation for the development of more effective methods for real-time code knowledge updating in the future. The experimental code and dataset are publicly available at: https://github.com/Lucky-voyage/Code-Sync.