摘要
研究主要探讨了研究生课程思政大模型的设计与数据集生成方法。为将个人创新与社会发展需求相结合,引导学生将科研成果转化为对社会有益的实践,作者设计并开发了研究生课程思政大模型,收集了包括《习近平新时代中国特色社会主义思想学习纲要》在内的相关思政教材,通过对这些文本进行格式处理和数据生成,构建了一个研究生课程思政的专用数据集。采用智谱AI推出的GLM-4-Flash作为数据集生成模型,并利用精心设计的prompt优化数据集输出,生成了用于微调的对话类数据集。通过LoRA微调技术优化了模型在课程思政领域的性能,增强了模型的问答能力。
This study mainly explores the design of a large model for ideological and political education in graduate courses and the methods for generating datasets.To combine personal innovation with societal development needs,and to guide students in transforming research outcomes into socially beneficial practices,a large model for ideological and political education in graduate courses was designed and developed.This research collected relevant ideological and political textbooks,including the Outline for Learning Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era.By processing these texts and generating data,a dedicated dataset for ideological and political education in graduate courses was constructed.The GLM-4-Flash model launched by Zhipu AI was used for dataset generation,and carefully designed prompts were employed to optimize the dataset output,resulting in a dialogue dataset for fine-tuning.The LoRA fine-tuning technique was utilized to enhance the model’s performance in the field of ideological and political education,improving its question-and-answer capa-bilities.
作者
蒋磊
孙俊忍
周翔
王子
JIANG Lei;SUN Junren;ZHOU Xiang;WANG Zi
出处
《科教文汇》
2025年第16期51-55,共5页
Journal of Science and Education
基金
2024年中国矿业大学(北京)研究生课程思政建设项目(YKCSZ2024017)。
关键词
研究生课程思政
大模型微调
问答大模型
ideological and political education in graduate courses
fine-tuning of large models
large models for question-and-answer
