JessicaE/OpenSeeSimE-Fluid-Small

OpenSeeSimE-Fluid-Small是一个从父数据集cmudrc/OpenSeeSimE-Fluid中提取的10%分层子集，用于在减少计算资源的同时评估视觉语言模型。数据集包含9881个样本，覆盖了五种源文件类型（Bent Pipe, Converging Nozzle, Heat Exchanger, Heat Sink, Mixing Pipe）和两种媒体类型（image, video）。数据集的采样方法保证了源文件、问题类型、媒体类型和问题ID的联合分布得以保留。数据集的特征包括文件名、源文件、问题文本、问题类型、问题ID、答案、答案选项、正确选项索引、图像或视频数据以及媒体类型。数据集的主要用途包括视觉语言模型的基准评估、评估管道的烟雾测试以及在存储或带宽受限情况下的比较研究。

OpenSeeSimE-Fluid-Small is a stratified 10% subset of the parent dataset cmudrc/OpenSeeSimE-Fluid, designed to evaluate vision-language models at a reduced compute footprint while preserving the joint distribution of simulation type, question type, media type, and question id. The dataset contains 9,881 samples covering five source file types (Bent Pipe, Converging Nozzle, Heat Exchanger, Heat Sink, Mixing Pipe) and two media types (image, video). The sampling method ensures the joint distribution of source file, question type, media type, and question id is preserved. Features include file name, source file, question text, question type, question id, answer, answer choices, correct choice index, image or video data, and media type. The primary uses of the dataset include benchmark evaluation of vision-language models, smoke-testing of evaluation pipelines, and comparative studies under storage or bandwidth constraints.