This repository contains the PHYSICS dataset we introduced, which covers five physics disciplines and includes physics problems ranging from high school to graduate-level physics courses, with rigorous quality control.

The dataset is divided into training and testing parts, At this stage, we currently provide only the test portion. The training portion will be open-sourced in the future. The test portion corresponds to PHYSICS_test.jsonl in the repository.

The field names in the files are explained as follows:

• id: A unique identifier for each original data entry. Both translated and original data share the same id.
• question: The physics problem.
• solution: The step-by-step solution process extracted from the data source.
• answer: The correct answer to the question. Each sub-question's answer is stored in a list.
• answer_type: The type of each answer, which can be one of the following: Interval, Expression, Equation, True/False, Multiple Choice, Numerical, Open-End.
• language: The language of the question, either Chinese (zh) or English (en).
• domain: The physics domain the question belongs to, including Modern Physics, Mechanics, Electromagnetism, Thermodynamics, and Optics.
• difficulty: The difficulty level of the question, categorized as High School and Below, High School Olympiad, Undergraduate (Non-Physics Major), or Undergraduate/Postgraduate (Physics Major).
• translate: Whether the question was obtained via translation. true means the question was translated; false indicates it is original data.
• reason_path (only in the training set): The detailed reasoning path generated by QwQ-32B for questions, provided to facilitate model training.