TAUR-dev/solution-trees__short-and-wide_p3_batch33

该数据集包含了多个字符串类型的特征字段，包括解决方案（solution）、问题（question）、类型（cot_type）、来源类型（source_type）等。此外，还包括了元数据（metadata）、两种方法的思考轨迹（gemini_thinking_trajectory和deepseek_thinking_trajectory）、尝试次数（gemini_attempt和deepseek_attempt）、两种方法的评分（gemini_grade和deepseek_grade）及评分原因（gemini_grade_reason和deepseek_grade_reason）。数据集分为训练集，包含1000个示例，文件大小为521,328,446字节。提供了默认配置文件，指定了训练数据的文件路径。

The dataset includes several string-typed feature fields such as solution, question, type (cot_type), source type (source_type), etc. It also includes metadata, thinking trajectories for two methods (gemini_thinking_trajectory and deepseek_thinking_trajectory), attempt counts (gemini_attempt and deepseek_attempt), ratings for two methods (gemini_grade and deepseek_grade), and rating reasons (gemini_grade_reason and deepseek_grade_reason). The dataset is split into a training set with 1000 examples, totaling 521,328,446 bytes in size. A default configuration file is provided, specifying the file path for the training data.