bombman/thwiki-2026-super-clean-4k

该数据集是从泰语维基百科（2026年版本）中提取的“超级清洁”数据集，专注于“专业清洁”级别，专为蒸馏和微调大型语言模型（LLM）而设计。关键特征包括：高信息密度，仅选择内容长度超过1000字符且泰语比例大于50%的文章；专业清洁，移除图像文件名（如.jpg、.png）、维基符号垃圾（如“==”、“”）和所有空括号；实体保留，保留引号“ ”以维持专有名词和地点的边界；随机化，从整个文件中随机抽取数据以确保内容多样性。

This dataset is a Super-Clean extraction from the Thai Wikipedia (Dump 2026), focusing on a Pro-Clean level specifically designed for distillation and fine-tuning of Large Language Models (LLMs). Key features include: high information density, selecting only articles with content length exceeding 1,000 characters and Thai language proportion > 50%; pro-cleaned, removing image file names (e.g., .jpg, .png), Wiki symbol clutter (e.g., ==, "), and all empty brackets; entity preserved, retaining quotation marks " " to maintain boundaries of proper nouns and locations; randomized, randomly sampling data from the entire file to ensure content diversity.