depthapi_technical_corpus
收藏资源简介:
DepthAPI Technical Corpus 是一个为检索增强生成(RAG)优化的综合技术语料库数据集。它包含约24万个高质量、可信技术文档和书籍的语义块,旨在支持技术编码任务和知识检索应用。数据内容来源于多个权威技术资源,包括MDN Web Docs、Kubernetes文档、CPython文档、Node.js API文档、React.dev内容,以及各种编程语言(如Java、Python、SQL、JavaScript、TypeScript等)的专业笔记和算法与系统设计入门材料。数据集规模中等,样本数量在10万到100万之间,具体提供248,982个示例。每个数据样本包含11个字段:chunk_id(块唯一标识符)、doc_id(源文档标识符)、collection_id(集合标识符)、collection_name(集合名称)、namespace(类别/命名空间,如“trusted”)、source_name(源的人类可读名称,如“CPython Docs”)、source_url(URL或源定位器)、raw_text(块的原始文本内容)、cleaned_text(清理后可用于嵌入的解析markdown/文本内容)、tags(逗号分隔的标签列表,如“python, stdlib, P0”)和chunk_order(块在文档中的顺序整数)。该数据集适用于训练和评估大型语言模型(LLMs)执行技术编码任务,并可作为混合搜索或RAG管道的高质量知识库,特别针对编程和计算机科学领域。
DepthAPI Technical Corpus is a comprehensive technical corpus dataset optimized for Retrieval-Augmented Generation (RAG). It contains approximately 240,000 high-quality, credible semantic chunks extracted from technical documents and books, aiming to support technical coding tasks and knowledge retrieval applications. The data content is sourced from multiple authoritative technical resources, including MDN Web Docs, Kubernetes documentation, CPython documentation, Node.js API documentation, React.dev content, as well as professional notes for various programming languages (such as Java, Python, SQL, JavaScript, TypeScript, etc.) and introductory materials for algorithms and system design. The dataset is of medium scale, with a sample size ranging from 100,000 to 1,000,000, and specifically contains 248,982 valid examples. Each data sample includes 11 fields: chunk_id (unique identifier of the text chunk), doc_id (source document identifier), collection_id (collection identifier), collection_name (collection name), namespace (category/namespace, e.g., "trusted"), source_name (human-readable name of the source, e.g., "CPython Docs"), source_url (URL or source locator), raw_text (original text content of the chunk), cleaned_text (parsed markdown or plain text content cleaned for embedding), tags (comma-separated tag list, e.g., "python, stdlib, P0"), and chunk_order (an integer representing the order of the chunk in the source document). This dataset is suitable for training and evaluating Large Language Models (LLMs) to perform technical coding tasks, and can serve as a high-quality knowledge base for hybrid search or RAG pipelines, particularly targeting the programming and computer science domains.




