mlfoundations-dev/instruction_filtering_scale_up_math_base_embedding_filter_mean_per_domain_16K

该数据集包含多个字段，如指令种子(instruction_seed)、来源(source)、嵌入(embeddings)、正面分数均值(mean_positive_score)、负面分数均值(mean_negative_score)、分数差(difference_score)、推理(reasoning)、DeepSeek解决方案(deepseek_solution)、原始行索引(__original_row_idx)、最终推理轨迹(final_reasoning_trace)和对话(conversations)。对话字段包含两个子字段：from和value。数据集分为训练集，共有约16000个示例，总数据大小约为1.17GB。但README文件中并未提供数据集的具体描述。

The dataset includes multiple fields such as instruction_seed, source, embeddings, mean_positive_score, mean_negative_score, difference_score, reasoning, deepseek_solution, __original_row_idx, final_reasoning_trace, and conversations. The conversations field contains two sub-fields: from and value. The dataset is split into a training set with approximately 16,000 examples and a total dataset size of about 1.17GB. However, the README file does not provide a specific description of the dataset.