agentlans/ibndias-DeepSeek-Distilled-40M-prompt-sample

该数据集是从一个更大的数学、编码和指令遵循数据集中提取的子集，包含了前10,000行数据。保留了原始数据集中的关键列：问题、问题来源、类别（数学、编码、科学、指令遵循或其他）以及三个不同模型的通过率。通过对通过率进行主成分分析（PCA），将整体通过率缩放到[0,1]范围。数据集被划分为80%的训练集和20%的测试集。

This dataset was drawn from a larger math, coding, and instruction-following dataset, containing the first 10,000 rows. Key columns from the original dataset were retained: question, question source, category (math, code, science, instruction follow, or other), and pass rates of three different models. Principal components analysis (PCA) of the pass rates was performed, scaling the overall pass rate to the range [0,1]. The dataset is split 80% training, 20% testing.