CoLT-132K
收藏资源简介:
CoLT-132K是一个大规模的代码补全数据集,由北京大学的研究团队创建。该数据集包含120,000个训练样本和12,000个测试样本,是从2,000个高质量的开源代码仓库中构建的。每个样本包含一个长达128K tokens的代码上下文作为输入,覆盖了Python、Java、C++和Go四种流行的编程语言,以及三种代码补全场景:API调用、单行代码补全和结构化跨度补全。数据集的设计旨在为仓库级别的代码补全训练提供足够的数据支持,并包含更多现实世界的代码补全场景。
CoLT-132K is a large-scale code completion dataset created by a research team from Peking University. It consists of 120,000 training samples and 12,000 test samples, which are constructed from 2,000 high-quality open-source code repositories. Each sample contains a code context of up to 128K tokens as input, covering four widely used programming languages: Python, Java, C++, and Go, as well as three code completion scenarios: API invocation, single-line code completion, and structured span completion. The dataset is designed to provide sufficient data support for repository-level code completion training and includes more real-world code completion scenarios.




