《动手学大模型》系列教程为大模型初学者提供免费的编程实践指导，旨在帮助学生和研究人员快速入门大模型技术，支持课程设计和学术研究。该项目源自上海交通大学的《人工智能安全技术》课程，并已扩展至包含国产化大模型开发全流程的公益教程。

How It Works

该教程通过一系列编程实践，涵盖了预训练模型微调与部署、提示学习与思维链、知识编辑、模型水印、越狱攻击、多模态模型以及大模型智能体与安全等核心大模型技术。其方法论侧重于通过实际操作，让学习者能够动手实践，理解并掌握大模型在不同场景下的应用和安全挑战。

Quick Start & Requirements

• Installation: No explicit installation instructions are provided, suggesting a focus on understanding concepts and potentially running code snippets within existing Python environments.
• Prerequisites: Python environment, familiarity with machine learning concepts. Specific hardware requirements (e.g., GPU) are not detailed but are implied for practical model training and deployment.
• Resources: Links to slides and tutorials are provided for each topic. A separate "大模型开发全流程" series, developed with Huawei Ascend, offers PPT, lab manuals, and videos, targeting different skill levels.
• Links:
  • 《大模型开发全流程》系列课程: https://www.hiascend.com/community/topic/detail/1210

Highlighted Details

• Comprehensive coverage of LLM lifecycle: from fine-tuning and deployment to advanced topics like jailbreaking and AI agents.
• Includes a new series on the full LLM development process, co-developed with Huawei Ascend, offering a localized perspective.
• Content is derived from university coursework, emphasizing academic rigor and practical application.
• Focus on AI safety, including adversarial attacks and model alignment.

Maintenance & Community

The project is actively maintained and welcomes contributions via Issues and Pull Requests. Key contributors are listed from Shanghai Jiao Tong University and the National University of Singapore, with significant contributions from Huawei Ascend for the new series.

Licensing & Compatibility

The project is offered as a free, public good. Specific licensing details are not explicitly stated in the README, but the "公益性质、完全免费" (public welfare nature, completely free) suggests a permissive stance for educational and research use. Commercial use compatibility is not detailed.

Limitations & Caveats

The content is based on contributors' personal experience and internet data, with a disclaimer that accuracy is not guaranteed. The project is described as ongoing, implying potential for omissions or changes.