jalasoft/typst-instruct

Typst Instruct是一个合成的指令遵循数据集，旨在为大型语言模型进行微调，以生成Typst标记代码。该数据集包含307个指令-完成对，每个对由一个详细的自然语言指令和相应的Typst文档实现组成。数据集适用于各种用途，包括为Typst代码生成微调大型语言模型、训练技术写作和文档格式化的助手模型、基准测试以及教育应用。它还讨论了数据集的局限性和偏见。

Typst Instruct is a synthetic instruction-following dataset designed for fine-tuning large language models to generate Typst markup code. The dataset contains 307 instruction-completion pairs, each consisting of a detailed natural language instruction and the corresponding Typst document implementation. It is intended for various uses, including fine-tuning LLMs for Typst code generation, training assistant models for technical writing and document formatting, benchmarking, and educational applications. The dataset also discusses its limitations and biases.