大模型开发教程：从基础到昇腾国产化

💡 最新动态

💡 Updates

2025年6月6日，我们衷心感谢各位朋友们的关注与积极反馈！基于此，我们从以下两个方面对本教程进行了重要更新：

On June 6, 2025, we sincerely thank all friends for your attention and positive feedback! Based on this, we have made significant updates to this tutorial in the following two aspects:

上线国产化《大模型开发全流程》公益教程（含PPT、实验手册和视频），此处特别感谢华为昇腾社区的大力支持！
- Launched the localized public welfare tutorial "Full Process of Large Model Development" (including PPTs, lab manuals, and videos). Special thanks to the Huawei Ascend Community for their strong support!
在原系列编程实践教程的基础上进行内容更新，并增加了新的主题（数学推理、GUI Agent、大模型对齐、隐写术等）！
- Updated content based on the original series of programming practice tutorials and added new topics (Mathematical Reasoning, GUI Agent, Large Model Alignment, Steganography, etc.)!

🎯 项目动机

《动手学大模型》系列编程实践教程，由上海交通大学《自然语言处理前沿技术》（NIS8021）、《人工智能安全技术》课程（NIS3353）讲义拓展而来（主讲教师：张倬胜），旨在为大模型初学者提供一套实用的编程入门参考。本教程完全免费，属公益性质。我们希望通过简洁明了的实践案例，帮助学习者快速入门大模型技术，从而更高效地开展课程设计或学术研究。

The "Hands-on Large Models" series of programming practice tutorials is expanded from the lecture notes of Shanghai Jiao Tong University's courses "Frontier Technologies in Natural Language Processing" (NIS8021) and "Artificial Intelligence Security Technology" (NIS3353) (Instructor: Zhuosheng Zhang). It aims to provide a practical programming reference for beginners in large models. This tutorial is completely free and public welfare in nature. We hope that through concise and clear practical cases, learners can quickly get started with large model technology, thereby conducting course design or academic research more efficiently.

📚 教程目录

📚 Tutorial Catalog

教程内容概览

Overview of Tutorial Content

教程内容	简介	资源链接
微调与部署	预训练模型微调与部署指南：想提升预训练模型在指定任务上的性能？让我们选择合适的预训练模型，在特定任务上进行微调，并将微调后的模型部署成方便使用的Demo！	[课件] [教程] [脚本]
提示学习与思维链	大模型的API调用与推理指南：“AI在线求鼓励？大模型对一些问题的回答令人大跌眼镜，但它可能只是想要一句「鼓励」”	[课件] [教程] [脚本]
知识编辑	语言模型的编辑方法和工具：想操控语言模型对指定知识的记忆？让我们选择合适的编辑方法，对特定知识进行编辑，并验证编辑后的模型！	[课件] [教程] [脚本]
数学推理	如何让大模型学会数学推理？让我们快速蒸馏一个迷你R1！	[课件] [教程] [脚本]
模型水印	语言模型的文本水印：在语言模型生成的内容中嵌入人类不可见的水印	[课件] [教程] [脚本]
越狱攻击	想要得到更好的安全，要先从学会攻击开始。让我们了解越狱攻击如何撬开大模型的嘴！	[课件] [教程] [脚本]
大模型隐写	“看不见的墨水”！想让大模型在流畅回答的同时，悄悄携带只有“自己人”能识别的信息吗？大模型隐写告诉你！	[课件] [教程] [脚本]
多模态模型	作为能够更充分模拟真实世界的多模态大语言模型，其如何实现更强大的多模态理解和生成能力？多模态大语言模型是否能够帮助实现AGI？	[课件] [教程] [脚本]
GUI智能体	想要饭来张口、解放双手？那么让我们一起来让AI Agent替你点外卖、回消息、购物比价吧！	[课件] [教程] [脚本]
智能体安全	大模型智能体迈向了未来操作系统之旅。然而，大模型在开放智能体场景中能意识到风险威胁吗？	[课件] [教程] [脚本]
RLHF安全对齐	基于PPO的RLHF实验指南：本教程”十分危险“，阅读后请检查你的大模型是否在冷笑。	[课件] [教程] [脚本]

Tutorial Content Introduction Resource Links

Fine-tuning & Deployment A Guide to Fine-tuning and Deploying Pre-trained Models: Want to improve the performance of a pre-trained model on a specific task? Let's select a suitable pre-trained model, fine-tune it on a particular task, and deploy the fine-tuned model into a user-friendly Demo! [Slides] [Tutorial] [Code]

Prompt Learning & Chain-of-Thought A Guide to API Calling and Inference for Large Models: "AI seeking encouragement online? The answers from large models to some questions can be shocking, but maybe it just wants a word of 'encouragement'." [Slides] [Tutorial] [Code]

Knowledge Editing Methods and Tools for Editing Language Models: Want to manipulate a language model's memory of specific knowledge? Let's choose an appropriate editing method, edit the specific knowledge, and verify the edited model! [Slides] [Tutorial] [Code]

Mathematical Reasoning How to make large models learn mathematical reasoning? Let's quickly distill a mini R1! [Slides] [Tutorial] [Code]

Model Watermarking Text Watermarking for Language Models: Embedding human-invisible watermarks in content generated by language models. [Slides] [Tutorial] [Code]

Jailbreak Attacks To achieve better security, one must first learn to attack. Let's understand how jailbreak attacks pry open the mouth of large models! [Slides] [Tutorial] [Code]

Large Model Steganography "Invisible ink"! Want a large model to carry information, recognizable only by "insiders," while answering fluently? Large model steganography tells you how! [Slides] [Tutorial] [Code]

Multimodal Models As multimodal large language models that can more fully simulate the real world, how do they achieve stronger multimodal understanding and generation capabilities? Can multimodal LLMs help achieve AGI? [Slides] [Tutorial] [Code]

GUI Agents Want to be served and free your hands? Then let's get AI Agents to order takeout, reply to messages, and compare prices for you! [Slides] [Tutorial] [Code]

Agent Security Large model agents are embarking on a journey toward the future operating system. However, can large models perceive risk threats in open agent scenarios? [Slides] [Tutorial] [Code]

RLHF Safety Alignment An Experimental Guide to RLHF based on PPO: This tutorial is "very dangerous." After reading, please check if your large model is sneering. [Slides] [Tutorial] [Code]

Tutorial Content	Introduction	Resource Links
Fine-tuning & Deployment	A Guide to Fine-tuning and Deploying Pre-trained Models: Want to improve the performance of a pre-trained model on a specific task? Let's select a suitable pre-trained model, fine-tune it on a particular task, and deploy the fine-tuned model into a user-friendly Demo!	[Slides] [Tutorial] [Code]
Prompt Learning & Chain-of-Thought	A Guide to API Calling and Inference for Large Models: "AI seeking encouragement online? The answers from large models to some questions can be shocking, but maybe it just wants a word of 'encouragement'."	[Slides] [Tutorial] [Code]
Knowledge Editing	Methods and Tools for Editing Language Models: Want to manipulate a language model's memory of specific knowledge? Let's choose an appropriate editing method, edit the specific knowledge, and verify the edited model!	[Slides] [Tutorial] [Code]
Mathematical Reasoning	How to make large models learn mathematical reasoning? Let's quickly distill a mini R1!	[Slides] [Tutorial] [Code]
Model Watermarking	Text Watermarking for Language Models: Embedding human-invisible watermarks in content generated by language models.	[Slides] [Tutorial] [Code]
Jailbreak Attacks	To achieve better security, one must first learn to attack. Let's understand how jailbreak attacks pry open the mouth of large models!	[Slides] [Tutorial] [Code]
Large Model Steganography	"Invisible ink"! Want a large model to carry information, recognizable only by "insiders," while answering fluently? Large model steganography tells you how!	[Slides] [Tutorial] [Code]
Multimodal Models	As multimodal large language models that can more fully simulate the real world, how do they achieve stronger multimodal understanding and generation capabilities? Can multimodal LLMs help achieve AGI?	[Slides] [Tutorial] [Code]
GUI Agents	Want to be served and free your hands? Then let's get AI Agents to order takeout, reply to messages, and compare prices for you!	[Slides] [Tutorial] [Code]
Agent Security	Large model agents are embarking on a journey toward the future operating system. However, can large models perceive risk threats in open agent scenarios?	[Slides] [Tutorial] [Code]
RLHF Safety Alignment	An Experimental Guide to RLHF based on PPO: This tutorial is "very dangerous." After reading, please check if your large model is sneering.	[Slides] [Tutorial] [Code]

🔥 新上线：国产化《大模型开发全流程》公益教程

🔥 Newly Launched: Localized Public Welfare Tutorial "Full Process of Large Model Development"

我们联合华为昇腾推出的《大模型开发全流程》公益教程现已正式上线！本教程融合前沿技术与代码实践，手把手带你玩转AI大模型。

The public welfare tutorial "Full Process of Large Model Development," jointly launched by us and Huawei Ascend, is now officially online! This tutorial integrates cutting-edge technology with code practice, guiding you step-by-step to master AI large models.

在《动手学大模型》原系列教程的基础上，我们与华为合作开发了《大模型开发全流程》系列课程。该系列教程基于昇腾基础软硬件平台开发，覆盖PPT课件、实验手册、视频讲解等多种形式。课程分为初级、中级、高级三个系列，面向不同层次的大模型实践需求，旨在通过代码实践的方式，为相关研究者与开发者提供由浅入深的快速上手指南，涵盖应用昇腾已支持模型和进行全新模型迁移调优的全流程。

Building upon the original "Hands-on Large Models" series, we collaborated with Huawei to develop the "Full Process of Large Model Development" series. This series is developed based on the Ascend basic software and hardware platform, covering various formats such as PPT slides, lab manuals, and video explanations. The course is divided into three series: Beginner, Intermediate, and Advanced, catering to different levels of large model practice needs. It aims to provide researchers and developers with a step-by-step quick-start guide through code practice, covering the entire process from applying Ascend-supported models to migrating and optimizing new models.

🚀 前往昇腾社区探索《大模型开发全流程》系列课程：
👉 《大模型开发学习专区》@ 昇腾社区 👈

🚀 Explore the "Full Process of Large Model Development" series on the Ascend Community:
👉 "Large Model Development Learning Zone" @ Ascend Community 👈

🙏 免责声明

本教程所有内容均来源于贡献者的个人经验、互联网公开数据及日常科研工作的相关积累。所有提供的方法与技巧仅供参考，不保证其绝对正确性或适用于所有场景。若发现任何问题，欢迎通过提交 Issue 或 Pull Request 进行反馈。此外，本项目所使用的徽章等资源来自互联网，如涉及图片版权问题，请联系我们删除，谢谢。

All content in this tutorial is derived from the contributors' personal experience, publicly available internet data, and accumulated knowledge from daily research work. All provided methods and techniques are for reference only, and we do not guarantee their absolute correctness or applicability to all scenarios. If you find any issues, please feel free to provide feedback by submitting an Issue or Pull Request. Furthermore, resources such as badges used in this project are sourced from the internet. If there are any image copyright concerns, please contact us for removal. Thank you.

🤝 欢迎贡献

本教程是一个持续进行的开源项目，疏漏之处在所难免。我们热烈欢迎任何形式的贡献，包括但不限于提交 Pull Request 修复错误、补充内容，或创建 Issue 进行讨论。

This tutorial is an ongoing open-source project, and omissions are inevitable. We warmly welcome contributions in any form, including but not limited to submitting Pull Requests to fix errors or add content, or creating Issues for discussion.

❤️ 贡献者列表

衷心感谢以下老师和同学对本项目的鼎力支持与宝贵贡献：

We sincerely thank the following teachers and students for their tremendous support and valuable contributions to this project:

《动手学大模型》系列教程开发团队：

"Hands-on Large Models" Series Tutorial Development Team:

上海交通大学：张倬胜、袁童鑫、马欣贝、何志威、杜巍、赵皓东、吴宗儒、吴铮、董凌众、张玉龙
新加坡国立大学：费豪

Shanghai Jiao Tong University: Zhuosheng Zhang, Tongxin Yuan, Xinbei Ma, Zhiwei He, Wei Du, Haodong Zhao, Zongru Wu, Zheng Wu, Lingzhong Dong, Yulong Zhang

National University of Singapore: Hao Fei

《大模型开发全流程》系列教程开发团队：

"Full Process of Large Model Development" Series Tutorial Development Team:

上海交通大学：张倬胜、刘功申、陈星宇、程彭洲、董凌众、何志威、鞠天杰、马欣贝、吴铮、吴宗儒、闫子赫、姚杳、袁童鑫、赵皓东;
华为昇腾社区：ZOMI、谢乾、程黎明、楼梨华、焦泽昱

Shanghai Jiao Tong University: Zhuosheng Zhang, Gongshen Liu, Xingyu Chen, Pengzhou Cheng, Lingzhong Dong, Zhiwei He, Tianjie Ju, Xinbei Ma, Zheng Wu, Zongru Wu, Zihe Yan, Yao Yao, Tongxin Yuan, Haodong Zhao;

Huawei Ascend Community: ZOMI, Qian Xie, Liming Cheng, Lihua Lou, Zeyu Jiao

🌟 项目星标历史

🌟 Star History

（此处通常放置一个显示项目GitHub Star增长历史的图片链接，格式一般为 ![Star History Chart](https://api.star-history.com/svg?...)）

(A link to an image showing the project's GitHub Star growth history is typically placed here, usually in the format ![Star History Chart](https://api.star-history.com/svg?...))