bombman/thwiki-2026-super-clean-5k

该数据集是从2026年版泰语维基百科中提取的“超级清洁”版本，专门用于蒸馏和微调大语言模型。它具有高信息密度，仅选取长度超过1000字符且泰语内容比例超过50%的文章；经过专业清洁处理，移除了图像文件名（如.jpg、.png）、维基符号（如“==”、“”）和空括号等噪声；同时保留引号以保护专有名词和地点的边界；并通过随机抽样确保内容的多样性。This dataset is extracted from the Thai Wikipedia (Dump 2026) with a focus on Pro-Clean level cleanliness, specifically designed for distillation and fine-tuning LLMs. It features high information density by selecting only articles longer than 1,000 characters and with Thai language proportion > 50%; pro-cleaned by removing image filenames (.jpg, .png), wiki symbols (`==`, ``), and empty parentheses; entity preserved by keeping quotation marks `" "` to maintain boundaries of proper nouns and locations; and randomized by sampling from the entire file for content diversity.