mlfoundations-dev/a1_math_hendrycks

该数据集包含了指令种子(instruction_seed)、解决方案(solution)、是否正确(is_correct)、解决方案步骤(solution_steps)、奖励回报率(rtg)、未折扣奖励回报率(undiscounted_rtg)、目标答案(target_answer)、完成度(completion)、轨迹(trajectory)、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)等字段。数据集分为训练集(train)，共有39985个示例，大小为1,145,387,806字节。该数据集可用于训练机器学习模型，处理与问题解决和推理相关的任务。

The dataset includes fields such as instruction_seed, solution, is_correct, solution_steps, reward-to-go (rtg), undiscounted reward-to-go (undiscounted_rtg), target_answer, completion, trajectory, reasoning, DeepSeek solution (deepseek_solution), original row index (__original_row_idx), final reasoning trace (final_reasoning_trace), and conversations. The dataset is split into a training set (train) with a total of 39,985 examples and a size of 1,145,387,806 bytes. This dataset can be used to train machine learning models for tasks related to problem-solving and reasoning.