Reproduce “Private-Library-Oriented Code Generation with Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Daoguang Zan, Bei Chen, Yongshun Gong, Junzhi Cao, Fengji Zhang, Bingchao Wu, Bei Guan, Yilong Yin and 1 more
Researchers introduced a human-inspired framework enabling Large Language Models (LLMs) to generate code utilizing private or proprietary libraries, a domain previously unaddressed. The system, comprising an API retrieval module and a code generation module, achieves enhanced performance on newly established benchmarks through targeted continuous pre-training and leverages relevant API documentation.
Key insights
- LLMs, despite no prior exposure, inherently possess the potential to invoke private APIs if provided with correct and relevant documentation.
- API Basic (name, signature, description) and API Examples are the most beneficial components of documentation for LLMs, outperforming parameters or related APIs.
- Continuous pre-training specifically targeting API invocation significantly improves LLMs' ability to integrate and utilize new API knowledge from prompts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.