NotoriousH2/gsm8k-ko

GSM8K Korean是一个基于GSM8K生成的韩语数学问题解答数据集，采用GSM8K风格，排除了源生成输出中的失败行。数据集包含三个配置：full（完整翻译/生成元数据，仅包含成功行）、sft（训练分割，包含问题、答案、解析答案及原始GSM8K字段）、eval（测试分割，包含问题、简短答案及原始GSM8K字段）。行数统计：full训练集7,309行，测试集1,296行；sft训练集7,309行；eval测试集1,296行；排除的失败行：训练集1,184行，测试集187行。数据集适用于问答和文本生成任务，专注于数学领域，语言为韩语。GSM8K Korean is a Korean GSM8K-style dataset generated from GSM8K, with failed rows from the source generation outputs excluded. It includes three configs: full (full translation/generation metadata for ok rows only), sft (train split with question, answer, parsed_answer, and original GSM8K fields), and eval (test split with question, short answer, and original GSM8K fields). Row counts: full train 7,309, full test 1,296; sft train 7,309; eval test 1,296. Excluded failed rows: 1,184 train and 187 test. The dataset is designed for question-answering and text-generation tasks, focusing on math, and is in the Korean language.