Accelerate Deep Learning Workloads with Amazon SageMaker: Train, deploy, and scale deep learning models effectively using Amazon SageMaker

Plan and design model serving infrastructure to run and troubleshoot distributed deep learning training jobs for improved model performance.

Key Features

• Explore key Amazon SageMaker capabilities in the context of deep learning
• Train and deploy deep learning models using SageMaker managed capabilities and optimize your deep learning workloads
• Cover in detail the theoretical and practical aspects of training and hosting your deep learning models on Amazon SageMaker

Book Description

Over the past 10 years, deep learning has grown from being an academic research field to seeing wide-scale adoption across multiple industries. Deep learning models demonstrate excellent results on a wide range of practical tasks, underpinning emerging fields such as virtual assistants, autonomous driving, and robotics. In this book, you will learn about the practical aspects of designing, building, and optimizing deep learning workloads on Amazon SageMaker. The book also provides end-to-end implementation examples for popular deep-learning tasks, such as computer vision and natural language processing. You will begin by exploring key Amazon SageMaker capabilities in the context of deep learning. Then, you will explore in detail the theoretical and practical aspects of training and hosting your deep learning models on Amazon SageMaker. You will learn how to train and serve deep learning models using popular open-source frameworks and understand the hardware and software options available for you on Amazon SageMaker. The book also covers various optimizations technique to improve the performance and cost characteristics of your deep learning workloads.

By the end of this book, you will be fluent in the software and hardware aspects of running deep learning workloads using Amazon SageMaker.

What you will learn

• Cover key capabilities of Amazon SageMaker relevant to deep learning workloads
• Organize SageMaker development environment
• Prepare and manage datasets for deep learning training
• Design, debug, and implement the efficient training of deep learning models
• Deploy, monitor, and optimize the serving of DL models

Who this book is for

This book is relevant for ML engineers who work on deep learning model development and training, and for Solutions Architects who design and optimize end-to-end deep learning workloads. It assumes familiarity with the Python ecosystem, principles of Machine Learning and Deep Learning, and basic knowledge of the AWS cloud.

Table of Contents

1. Introducing Deep Learning with Amazon SageMaker
2. Deep Learning Frameworks and Containers on SageMaker
3. Managing SageMaker Development Environment
4. Managing Deep Learning Datasets
5. Considering Hardware for Deep Learning Training
6. Engineering Distributed Training
7. Operationalizing Deep Learning Training
8. Considering Hardware For Inference
9. Implementing Model Servers
10. Operationalizing Inference Workloads

Vadim Dabravolski is a Solutions Architect and Machine Learning Engineer. He has over 15 years of career in software engineering, specifically data engineering and machine learning. During his tenure in AWS, Vadim helped many organizations to migrate their existing ML workloads or engineer new workloads for the Amazon SageMaker platform. Vadim was involved in the development of Amazon SageMaker capabilities and adoption of them in practical scenarios. Currently, Vadim works as an ML engineer, focusing on training and deploying large NLP models. The areas of interest include engineering distributed model training and evaluation, complex model deployments use cases, and optimizing inference characteristics of DL models.