mlfoundations-dev/instruction_filtering_scale_up_code_base_embedding_filter_mean_per_domain_16K

该数据集包含了多个字段，如指令种子、来源、嵌入向量、正面评分平均值、负面评分平均值、评分差值、推理过程、DeepSeek解决方案、原始行索引和最终推理轨迹等。这些字段表明数据集可能与某种推理或评分任务相关。数据集被划分为训练集，包含16000个样本。然而，具体的应用场景和详细描述在README中并未提供。

The dataset includes multiple fields such as instruction seed, source, embedding vectors, mean positive score, mean negative score, difference score, reasoning process, DeepSeek solution, original row index, and final reasoning trace. These fields suggest that the dataset may be related to some kind of reasoning or rating task. The dataset is split into a training set with 16,000 samples. However, the specific application scenario and detailed description are not provided in the README.