drkolesnikov/russian-nmo-medical-mcq

这是一个俄语医学NMO（可能指непрерывное медицинское образование，即持续医学教育）测试多选题数据集。数据集包含两种类型的问题：单选题（只有一个正确答案）和多选题（有多个正确答案）。数据集专门设计用于实际应用，包括对大语言模型（LLM）进行监督微调、评估答案质量以及进行医学问答（QA）实验。核心任务很简单：给定一个主题、一个问题以及编号的答案选项，模型需要返回正确选项的编号（对于单选题）或所有正确选项的编号（对于多选题）。数据集提供了多种配置视图：sft_all（包含所有问题，主要用于微调）、sft_single（仅单选题）、sft_multi（仅多选题）和mcq（结构化规范数据，用于分析、过滤和评估）。数据集中的答案选项从0开始编号，模型的输出补全也采用相同的0起始编号格式。数据集明确声明其用途仅限于教育和实验目的，不适用于临床决策支持。This is a Russian-language medical NMO (likely referring to непрерывное медицинское образование, meaning continuous medical education) test dataset with multiple-choice answer options. It contains both single-correct questions (only one correct answer) and multiple-correct questions (more than one correct answer). The dataset is packaged for practical applications, including supervised fine-tuning of large language models (LLMs), evaluation of answer quality, and medical QA experiments. The core task is simple: given a topic, a question, and numbered answer choices, the model should return the number of the correct choice (for single-correct questions) or the numbers of all correct choices (for multiple-correct questions). The dataset offers multiple configuration views: sft_all (includes all questions, mainly for fine-tuning), sft_single (only single-correct questions), sft_multi (only multiple-correct questions), and mcq (structured canonical data for analysis, filtering, and evaluation). Answer choices are numbered from 0, and model completions use the same zero-based numbering format. The dataset explicitly states that it is intended for educational and experimental purposes only and is not suitable for clinical decision support.