danielfdias98/derm-reasoning-cot

这是一个皮肤病学推理数据集的链式思考（Chain-of-Thought）格式变体。数据集包含图像集、每行的临床内容和诊断标签，与完整推理版本相同，仅在助手回合的格式上有所不同。每个样本的助手回合包含自由形式的临床推理叙述和结构化的答案，以鼓励在标签承诺之前进行链式思考生成。数据集用于视觉语言模型的微调，特别是在医学推理能力方面。数据集包含25,637个训练样本和2,849个验证样本。

This is the chain-of-thought formatted variant of the Dermatology Reasoning Dataset. The dataset includes the same image set, per-row clinical content, and diagnostic labels as the full-reasoning version, differing only in the assistant-turn formatting. Each samples assistant turn contains a free-form clinical reasoning narrative and a structured answer, encouraging chain-of-thought generation prior to label commitment. The dataset is used for fine-tuning vision-language models, particularly in medical reasoning capabilities. It consists of 25,637 training samples and 2,849 validation samples.