Andrew613/PICABench

该数据集是一个针对物理学科知识的应用数据集，它包含了物理类别、物理定律等文本信息，以及与问题解答相关的编辑操作和提示语句。每个样本中还包括了注释的问答对，这些问答对包含了问题的分割索引、置信度、错误信息、是否与编辑无关、预测耗时、原始预测结果和推理过程等详细信息。此外，数据集中的每个样本还与一个图像相关联，图像信息包括路径、尺寸和坐标等。数据集分为picabench等不同的划分，适合进行相关物理知识处理的模型训练和评估。

This dataset is an application-oriented dataset for physics knowledge, which includes text information such as physics categories and physics laws, as well as edit operations and prompts related to question answering. Each sample also contains annotated QA pairs with details such as split index, confidence, error information, whether it is edit-agnostic, latency, prediction on the original, and reasoning process. In addition, each sample in the dataset is associated with an image, with image information including path, size, and coordinates. The dataset is divided into different splits like picabench, suitable for model training and evaluation for related physics knowledge processing.