DinoDS/dino-data-vision-tooling-preview

该数据集是一个专注于视觉工具预览的数据集，由两个Dino Data能力切片构建：图像上下文理解和图像工具使用。目的是训练或检查助手在图像相关任务中的行为，这些任务涉及视觉上下文、多模态解释或工具感知的图像处理。数据集包含40行数据，分为训练集（36行）、验证集（2行）和测试集（2行），每行数据包含多个重要列，如sample_id、user_message、assistant_response等。数据集可用于多模态助手微调实验、图像上下文指令处理、视觉任务响应行为评估以及工具感知图像工作流研究。

This dataset is a focused vision-tooling preview built from two Dino Data capability slices: image context understanding and image tooling. The goal is to train or inspect assistant behavior for image-related tasks where visual context, multimodal interpretation, or tool-aware image handling is relevant. The dataset contains 40 rows of data, divided into training (36 rows), validation (2 rows), and test (2 rows) sets. Each row includes important columns such as sample_id, user_message, assistant_response, etc. The dataset can be used for multimodal assistant fine-tuning experiments, image-context instruction handling, evaluation of visual-task response behavior, and tool-aware image workflow studies.