summykai/tailwind-v4-sft-mix-001

Tailwind-SFT是一个包含5,610个示例的指令调优数据集，用于将自然语言UI意图映射到经过验证的Tailwind CSS v4实用类和React/JSX片段。每个示例都经过编译器验证，确保所有实用类在Tailwind v4编译器中的语法正确性，避免了幻觉现象。数据集填补了现有CSS数据集缺乏真实验证的空白，适用于训练与现代CSS框架交互的领域特定模型。数据集支持多种任务类型，包括选择优化类字符串、生成生产就绪的JSX片段以及解释类选择理由。技术规格包括硬件要求、模型优化和验证方法等。数据集结构包含消息和元数据字段，支持按组件原型和类别进行筛选。

Tailwind-SFT is a 5,610-example instruction-tuning dataset mapping natural language UI intents to validated Tailwind CSS v4 utility classes and React/JSX snippets. Each example is compiler-verified to ensure syntactic correctness against the Tailwind v4 compiler, eliminating hallucination. The dataset addresses the gap in existing CSS datasets by providing ground-truth validation, making it suitable for training domain-specific models interfacing with modern CSS frameworks. It supports multiple task types including choosing optimized class strings, generating production-ready JSX snippets, and explaining class selection rationale. Technical specifications cover hardware requirements, model optimizations, and validation methods. The dataset structure includes messages and metadata fields, supporting filtering by component archetype and category.