中文 | English

---

Machine Learning Systems: Design and Implementation

An open-source book explaining the design principles and implementation experience of modern machine learning systems, covering the complete technology stack from programming interfaces and computational graphs to compilers and distributed training.

Read Online: openmlsys.github.io

Table of Contents

• Target Audience
• Content Overview
• Build Guide
• Contributing
• Community
• License

Target Audience

• Students: Those who have mastered machine learning fundamentals and want to deeply understand the design and implementation of modern ML systems.
• Researchers: Those who need to develop custom operators or leverage distributed execution for large model development.
• Engineers: Those responsible for building ML infrastructure and need to tune system performance or customize ML systems for business needs.

Content Overview

The book is organized into three parts: Fundamentals, Advanced Topics, and Extensions.

Part I: Fundamentals

Chapter	Content
Programming Interface	Framework API design, ML workflows, deep learning model definition, C/C++ framework development
Computational Graph	Graph components, generation methods, scheduling strategies, automatic differentiation

Part II: Advanced Topics

Chapter	Content
Compiler Frontend & IR	Type inference, intermediate representation (IR), automatic differentiation, common optimization passes
Compiler Backend & Runtime	Graph optimization, operator selection, memory allocation, compute scheduling and execution
Hardware Accelerators	GPU/Ascend architecture, high-performance programming interfaces (CUDA/CANN)
Data Processing	Usability, efficiency, order preservation, distributed data processing
Model Deployment	Model conversion, compression, inference, and security
Distributed Training	Data parallelism, model parallelism, pipeline parallelism, collective communication, parameter servers

Part III: Extensions

Chapter	Content
Recommender Systems	Recommendation principles, large-scale industrial architecture
Federated Learning	Federated learning methods, privacy protection, system implementation
Reinforcement Learning Systems	Single-agent and multi-agent RL systems
Explainable AI Systems	XAI methods and production practices
Robot Learning Systems	Robot perception, planning, control, and system safety

Build Guide

Prerequisites

• Python >= 3.10
• pandoc >= 2.19

Installation

# Clone the repository
git clone https://github.com/openmlsys/openmlsys-zh.git
cd openmlsys-zh

# Install d2lbook
git clone https://github.com/openmlsys/d2l-book.git
cd d2l-book && pip install . && cd ..

# Install Python dependencies
pip install -r requirements.txt

Build HTML

sh build_html.sh
# Output is in _build/html/

For more details, see the Build Guide.

Contributing

We welcome all forms of contributions, including:

• Errata: If you find text or figure errors, please open an Issue and @ the chapter editors, or submit a PR directly.
• Content updates: Submit PRs to update or add Markdown files.
• New chapters: We welcome community contributions on topics such as meta-learning systems, automatic parallelism, cluster scheduling, green AI, and graph learning.

Before contributing, please read:

• Writing Style Guide
• Terminology Guide

Community

Join our WeChat group by scanning the QR code in info/mlsys_group.png.

License

This project is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.