Learning Ray

Foreword

For the past decade, the computation demands of machine learning and data applications have vastly outgrown the capabilities of a single server or a single processor, including hardware accelerators such as GPUs and TPUs. This trend leaves us no choice but to distribute these applications. Unfortunately, building such distributed applications is notoriously difficult.

Over the past few years, Ray has emerged as the framework of choice to simplify the development of such applications. Ray includes a flexible core and a set of powerful libraries that enable the developers to easily scale a variety of workloads, including training, hyperparameter tuning, reinforcement learning, model serving, and batch processing of unstructured data. Ray is one of the most popular open source projects and has been used by thousands of companies to implement everything from machine learning platforms to recommendation systems, fraud detection, and training some of the largest models, including Open AI’s ChatGPT.

In this book, Max Pumperla, Edward Oakes, and Richard Liaw have done an outstanding job in providing a gentle and comprehensive introduction to Ray and its libraries using easy to follow examples. At the end of this book, you will master the key concepts and abstractions in Ray and be able to develop and quickly scale end-to-end machine learning applications from your laptop to large on-premise clusters or to the cloud.

Ion Stoica

Cofounder of Anyscale and Databricks, and Professor ...