JessicaE/OpenSeeSimE-Structural-Small

OpenSeeSimE-Structural-Small是cmudrc/OpenSeeSimE-Structural数据集的一个分层10%子集，用于以较低的计算成本评估视觉语言模型，同时保持模拟类型、问题类型、媒体类型和问题ID的联合分布。数据集包含10,343行数据，分为4个Parquet分片，存储大小约为15.60 GB。数据集的组成包括不同的source_file（如Beams、Dog Bone等）、media_type（image和video）以及question_type（Binary、Multiple Choice、Spatial）。数据集的特征包括file_name、source_file、question、question_type、question_id、answer、answer_choices、correct_choice_idx、image、video和media_type。数据集的主要用途包括评估视觉语言模型在工程模拟问题回答上的性能、在运行完整基准测试前进行烟雾测试，以及在存储或带宽受限的情况下进行比较研究。数据集采用MIT许可证，允许学术和商业使用，但需注明出处。

OpenSeeSimE-Structural-Small is a stratified 10% subset of the cmudrc/OpenSeeSimE-Structural dataset for evaluating vision-language models at a reduced compute footprint while preserving the joint distribution of simulation type, question type, media type, and question id. The subset contains 10,343 rows, divided into 4 Parquet shards with a storage size of approximately 15.60 GB. The dataset composition includes different source_files (e.g., Beams, Dog Bone, etc.), media_types (image and video), and question_types (Binary, Multiple Choice, Spatial). The dataset features include file_name, source_file, question, question_type, question_id, answer, answer_choices, correct_choice_idx, image, video, and media_type. The primary uses of the dataset are to benchmark vision-language models on engineering simulation question answering, smoke-test evaluation pipelines before running the full benchmark, and comparative studies where storage or bandwidth constraints matter. The dataset is licensed under MIT, free for academic and commercial use with attribution.