CodeXHug
收藏资源简介:
CodeXHug是由拉奎拉大学研究团队构建的一个精心策划的数据集,旨在系统化地关联Hugging Face平台上的预训练模型与其在GitHub生态系统中的实际代码使用模式。该数据集涵盖了7,325个不同的预训练模型,并收集了来自71,748个代码仓库的372,063个Python文件,数据源整合了Hugging Face模型元数据(如模型卡、管道标签)和GitHub仓库的源代码及项目元数据。其创建过程首先对Hugging Face数据进行了清洗和基于下载量的流行度采样,随后通过GitHub API映射并筛选出实际使用这些模型的Python代码文件。该数据集主要应用于软件工程领域,旨在解决预训练模型在实际项目集成中因缺乏具体代码示例而导致的采用障碍,支持模型卡增强、代码助手开发和模型使用模式分析等任务。
CodeXHug is a carefully curated dataset constructed by the research team from the University of L'Aquila, aiming to systematically correlate pre-trained models on the Hugging Face platform with their actual code usage patterns in the GitHub ecosystem. This dataset covers 7,325 distinct pre-trained models, and collects 372,063 Python files from 71,748 code repositories. The data sources integrate Hugging Face model metadata (such as model cards and pipeline tags) as well as source code and project metadata from GitHub repositories. Its creation process first cleans the Hugging Face data and performs popularity sampling based on download counts, then maps and filters Python code files that actually use these models via the GitHub API. This dataset is primarily applied in the field of software engineering, aiming to address the adoption barriers of pre-trained models in actual project integration caused by the lack of specific code examples, and supports tasks such as model card enhancement, code assistant development, and model usage pattern analysis.




