ai-infra-book
《深入理解 AI Infra:量化分析与系统设计》(李博杰 著)开源书稿:从硬件约束和模型架构出发,量化推导 LLM 推理与训练系统设计。含全书正文、PDF、配套计算工具与实验
What is ai-infra-book?
《深入理解 AI Infra:量化分析与系统设计》(李博杰 著)开源书稿:从硬件约束和模型架构出发,量化推导 LLM 推理与训练系统设计。含全书正文、PDF、配套计算工具与实验
Overview of ai-infra-book
What it is
This is the open-source manuscript of the book Deep Understanding of AI Infra: Quantitative Analysis and System Design by Bojie Li. It provides a quantitative approach to designing LLM inference and training systems, starting from hardware constraints and model architectures. The repository includes the full text, PDF, EPUB, companion calculation tools, and experiments.
Key features
- Quantitative derivation of system design from hardware constraints and model architecture
- Covers 12 chapters on topics from model architecture to edge-cloud collaboration
- Includes a calculation tool for estimating resource requirements without a GPU
- Experiments organized by chapter with run instructions and result descriptions
- Community translations available in English, Traditional Chinese, and Russian
Use cases
This book is suitable for developers who have called model APIs or run models locally and want to understand why models are slow or how to reduce costs. It is also useful for engineers working on systems, networking, and chips, as well as researchers and students in related fields. Readers can focus on specific chapters based on their interests: model application and inference service, system or networking, chip and architecture, or systematic learning.
Getting started
bashgit clone https://github.com/bojieli/ai-infra-book.gitcd ai-infra-bookpython3 calculations/calc.py modelspython3 calculations/calc.py forward --model qwen3-8b --tokens 8192 --format md
License and pricing
Licensed under Apache-2.0. The project is open source and free.
Based on the project's public README and repository metadata.