Foundations-of-LLMs is an open-source book for learners of large language models, created and maintained by a team from Zhejiang University. The book aims to systematically explain the relevant foundational knowledge for readers interested in large language models and to introduce cutting-edge technologies. The author team states that it will continue monthly updates, listen to suggestions from the open-source community and experts, and strive to create an easy-to-read, rigorous, and in-depth textbook on large models.
The first edition of this book includes six chapters: traditional language models, the architectural evolution of large language models, Prompt engineering, parameter-efficient fine-tuning, model editing, and retrieval-augmented generation. To increase readability, each chapter uses a different animal as a background to illustrate specific technologies with examples, which is why the book uses six animals as its cover. The content of the book comes from the author team's exploration and understanding of the relevant directions.
In terms of usage, the project repository provides the current complete PDF version of the book at the path "大模型基础.pdf". In addition, it also provides two folders: "大语言模型分章节内容" contains the PDF version of each chapter; "大语言模型相关论文" contains related papers for each chapter and is continuously updated. Each chapter is also accompanied by a related Paper List to track the latest progress of related technologies.
Applicable scenarios include: readers learning the foundational knowledge of large language models, teaching references for related courses, introductions to scientific research, and more. The author team also states that it will subsequently explore directions such as large model inference acceleration and large model agents, and the related content will be gradually added to future versions.
In addition, the team has also open-sourced a multi-agent development framework, Agent-Kernel. According to the description, one hundred agents can run on one's own laptop, making it suitable for scenarios such as scientific research, graduation projects, college student innovation and entrepreneurship projects, and SRTP. The GitHub address of Agent-Kernel ishttps://github.com/ZJU-LLMs/Agent-Kernel。
Restricted or not yet provided information: the book's pricing, purchase methods, specific platform support, usage effects, etc. are not explained in the materials. If you have other questions, you can send an email to xuwenyi@zju.edu.cn.
