mlfoundations-dev/instruction_filtering_scale_up_code_base_gemini_length_1K

该数据集包含了一系列的指令种子(instruction_seed)、来源(source)、Gemini模型的响应(gemini_response)等信息。数据集中的每个样本还包括了原始行索引、长度、推理过程、DeepSeek解决方案和最终的推理轨迹等字段。此外，样本之间可能存在对话，对话包括发送者(from)和消息内容(value)。数据集分为训练集(train)，共有1000个样本。

The dataset includes a series of instruction seeds (instruction_seed), sources (source), responses from the Gemini model (gemini_response), and other information. Each sample in the dataset also includes fields such as the original row index, length, reasoning process, DeepSeek solution, and the final reasoning trace. Moreover, there may be conversations between samples, which include the sender (from) and the message content (value). The dataset is split into a training set (train) with a total of 1000 samples.