mlfoundations-dev/instruction_filtering_scale_up_code_base_embedding_filter_mean_4K

该数据集是一个包含多种特征的数据集，其中包括指令种子(instruction_seed)、来源(source)、嵌入向量(embeddings)、正面评分均值(mean_positive_score)、负面评分均值(mean_negative_score)、评分差(difference_score)、推理过程(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)。对话包含发送者(from)和消息内容(value)。数据集分为训练集(train)，包含4000个示例。

The dataset is a collection of various features including instruction_seed, source, embeddings, mean_positive_score, mean_negative_score, difference_score, reasoning, deepseek_solution, __original_row_idx, final_reasoning_trace, and conversations. Conversations consist of sender (from) and message content (value). The dataset is split into a training set (train) containing 4000 examples.