abouelgoud/NextecCode

Agent-API-Frontend-SFT-Dataset是一个由Saad Abouelgoud策划的数据集，包含100万行数据，以JSONL文件格式存储，采用标准OpenAI聊天模板风格。数据集主要用于微调大型语言模型（如Llama 3 Coder、Qwen 2.5 Coder和DeepSeek Coder），使其在“代理模式”下充当专家软件工程师。它涵盖了多种编程语言和技术栈，包括英语、C#、SQL、Dart、TypeScript、JavaScript、HTML和CSS。具体应用包括教授.NET 9 WebAPI中的Clean Architecture原则（如存储库、MediatR CQRS处理程序、EF Core配置和OpenAPI Swagger文档）、微调现代前端框架（如Angular（使用Signals和HttpClient）、React（使用Hooks、Redux切片和Tailwind CSS）和Flutter小部件）的UI生成模型，以及提升复杂关系数据库布局上的文本到SQL能力。数据集结构基于对话式交互，每个条目包含系统、用户和助理角色的消息。技术关键词分布显示，SQL和数据库相关占42.32%，C#/.NET 9/EF Core占19.02%，其他技术如React、Angular、Flutter等占较小比例。数据集在Apache 2.0许可证下发布。Agent-API-Frontend-SFT-Dataset is a dataset curated by Saad Abouelgoud, containing 1,000,000 rows in a single JSONL file formatted in the standard OpenAI chat template style. It is designed for fine-tuning Large Language Models (LLMs) such as Llama 3 Coder, Qwen 2.5 Coder, and DeepSeek Coder to act as expert software engineers in Agent Mode. The dataset covers multiple programming languages and technologies, including English, C#, SQL, Dart, TypeScript, JavaScript, HTML, and CSS. Key uses include teaching models the structural principles of Clean Architecture in .NET 9 WebAPIs (e.g., repositories, MediatR CQRS handlers, EF Core configuration, and OpenAPI Swagger documentation), fine-tuning UI generation models for modern frontend frameworks like Angular (with Signals and HttpClient), React (with Hooks, Redux slices, and Tailwind CSS), and Flutter widgets, and improving text-to-SQL capabilities on complex relational database layouts. The structure is conversational, with each entry containing messages in roles of system, user, and assistant. Technology keyword distribution shows SQL & Databases at 42.32%, C# / .NET 9 / EF Core at 19.02%, and smaller percentages for other technologies like React, Angular, and Flutter. The dataset is licensed under Apache 2.0.