SaltySander/HISTAI-Instruct

HISTAI-Instruct是一个多语言、多模态的指令调优数据集，专为计算病理学设计，基于开放的HISTAI数据集构建。它支持视觉语言模型（VLMs）在病理学任务中的应用，包括详细描述、鉴别诊断和多轮对话。数据集包含24,259个病例和全切片图像（WSIs），涵盖9个器官，生成1,175,524个对话属性和2,153,699个问答对，支持7种语言。数据集结构包括主数据集、原始和中间数据、审计跟踪以及数据分割。数据集创建使用了Polysome框架，相关代码和模型可在指定仓库中找到。

HISTAI-Instruct is a multilingual, multimodal instruction-tuning dataset for computational pathology built on top of the open HISTAI dataset. It is designed to support Vision-Language Models (VLMs) in histopathology tasks, including detailed description, differential diagnosis, and multi-turn conversation. The dataset contains 24,259 cases and WSIs across 9 organs, generating 1,175,524 conversational attributes and 2,153,699 question answering pairs in 7 languages. The dataset structure includes the main dataset, raw and intermediate data, audit trails, and data splits. The dataset was created using the Polysome framework, with related code and models available in specified repositories.