针对高校大语言模型教学中存在理解不深入、实践环节缺失的问题，提出一种面向系统能力培养的“三段式”实验教学设计框架，以ChatGLM-6B本地微调为例，具体阐述如何通过构建认知理解、工程实践、任务应用3个阶段，逐步提升学生对大语言模型的理解与应用能力，通过教学结果说明该设计具有良好的可操作性与教学成效，可为人工智能类课程教学提供参考。

To address the insufficient understanding and lack of practical training in large language model(LLM) teaching in universities, this study proposes a three-stage experimental teaching framework aimed at developing system-level competence. Taking the local fine-tuning of ChatGLM-6B as an example, the study specifically illustrates how to enhance students' understanding and application abilities by constructing three progressive phases: cognitive understanding, engineering practice, and task-oriented application. Based on the teaching outcomes, the proposed design demonstrates strong feasibility and teaching effectiveness, providing a useful reference for instructional practices in artificial intelligence-related courses.