keisuke-miyako/text-commands-2026-0425

该数据集是对4D语言参考的清理总结，旨在通过持续预训练（CPT）和微调技术，帮助大型语言模型（LLM）更好地理解和识别4D代码中的特定模式。数据集基于Grok 4.20 Reasoning生成，使用了来自4D公共文档、syntaxEN.json文件和特定命令的原始输入。处理过程中特别关注了版本指定、移除Markdown表格以及以中性格式分解和解释语法等规则。数据集的目标是微调模型，使其能够执行诸如“使用版本Y命令集重写版本X代码”等任务。

This dataset is a clean summary of the 4D language reference, designed to help large language models (LLMs) better understand and recognize distinct patterns in 4D code through continued pre-training (CPT) and fine-tuning techniques. The dataset was generated using Grok 4.20 Reasoning, with raw inputs from 4D public documentation, the syntaxEN.json file, and specific commands. Special attention was paid to rules such as specifying versions, removing Markdown tables, and decomposing and explaining syntax in a neutral format. The goal of the dataset is to fine-tune models to perform tasks such as rewrite this version X code using version Y command set.