bermaneh/pde-mc-logprob-results-v2

该数据集名为pde-mc-logprob-results-v2，是一个完整的多项选择（MC）对数概率结果集，专门用于偏微分方程（PDE）模拟代码的评估。数据集包含7776行数据，由9个不同模型、96个基础行和9个问题类型组合而成。核心内容是针对PDE模拟代码的物理有效性进行重新评估，使用更新后的提示问题：这段代码是否运行并为PDE产生正确的物理解？，替代了原来的这个模拟在物理上有效吗？。数据集包含19列，涉及模型名称、标题、PDE类别、修改类型、问题类型、候选答案、字母选项、正确字母、各选项的对数概率（logprob_A到logprob_D）、预测字母、正确性标志、正确选项的对数概率、边际值、熵、完成原因和评分方法等。这些数据用于分析模型在PDE相关任务中的表现，特别是通过对数概率来评估模型输出的物理有效性和准确性。The dataset named pde-mc-logprob-results-v2 is a full multiple-choice (MC) log probability results set, specifically designed for evaluating partial differential equation (PDE) simulation code. It contains 7776 rows, derived from 9 models × 96 base rows × 9 question types. The core focus is on re-evaluating the physical validity of PDE simulation code using an updated prompt: Does this code run and produce a correct physical solution for the PDE?, which replaces the original Is this simulation physically valid?. The dataset comprises 19 columns, including model name, title, PDE class, modification type, question type, candidate answers, letter options, correct letter, log probabilities for each option (logprob_A to logprob_D), predicted letter, correctness flag, log probability of the correct option, margin, entropy, finish reason, and scoring method. These data are used to analyze model performance on PDE-related tasks, particularly assessing the physical validity and accuracy of model outputs through log probabilities.