Search icon CANCEL
Subscription
0
Cart icon
Your Cart (0 item)
Close icon
You have no products in your basket yet
Save more on your purchases! discount-offer-chevron-icon
Savings automatically calculated. No voucher code required.
Arrow left icon
Explore Products
Best Sellers
New Releases
Books
Videos
Audiobooks
Learning Hub
Newsletter Hub
Free Learning
Arrow right icon
timer SALE ENDS IN
0 Days
:
00 Hours
:
00 Minutes
:
00 Seconds
Distributed Machine Learning with Python
Distributed Machine Learning with Python

Distributed Machine Learning with Python: Accelerating model training and serving with distributed systems

eBook
€22.99 €25.99
Paperback
€31.99
Subscription
Free Trial
Renews at €18.99p/m

What do you get with a Packt Subscription?

Free for first 7 days. $19.99 p/m after that. Cancel any time!
Product feature icon Unlimited ad-free access to the largest independent learning library in tech. Access this title and thousands more!
Product feature icon 50+ new titles added per month, including many first-to-market concepts and exclusive early access to books as they are being written.
Product feature icon Innovative learning tools, including AI book assistants, code context explainers, and text-to-speech.
Product feature icon Thousands of reference materials covering every tech concept you need to stay up to date.
Subscribe now
View plans & pricing
Table of content icon View table of contents Preview book icon Preview Book

Distributed Machine Learning with Python

Chapter 1: Splitting Input Data

Over the recent years, data has grown drastically in size. For instance, if you take the computer vision domain as an example, datasets such as MNIST and CIFAR-10/100 consist of only 50k training images each, whereas recent datasets such as ImageNet-1k contain over 1 million training images. However, having a larger input data size leads to a much longer model training time on a single GPU/node. In the example mentioned previously, the total training time of a useable state-of-the-art single GPU training model on a CIFAR-10/100 dataset only takes a couple of hours. However, when it comes to the ImageNet-1K dataset, the training time for a GPU model will take days or even weeks.

The standard practice for speeding up the model training process is parallel execution, which is the main focus of this book. The most popular in-parallel model training is called data parallelism. In data parallel training, each GPU/node holds the full copy of a model. Then, it partitions the input data into disjoint subsets, where each GPU/node is only responsible for model training on one of the input partitions. Since each GPU only trains its local model on a subset (not the whole set) of the input data, we need to conduct a procedure called model synchronization periodically. Model synchronization is done to ensure that, after each training iteration, all the GPUs involved in this training job are on the same page. This guarantees that the model copies that are held on different GPUs have the same parameter values.

Data parallelism can also be applied at the model serving stage. Given that the fully-trained model may need to serve a large number of inference tasks, splitting the inference input data can reduce the end-to-end model serving time as well. One major difference compared to data parallel training is that in data parallel inference, all the GPUs/nodes involved in a single job do not need to communicate anymore, which means that the model synchronization phase during data parallel training is completely removed.

This chapter will discuss the bottleneck of model training with large datasets and how data parallelism mitigates this.

The following topics will be covered in this chapter:

  • Single-node training is too slow
  • Data parallelism – the high-level bits
  • Hyperparameter tuning
Left arrow icon Right arrow icon
Download code icon Download Code

Key benefits

  • Accelerate model training and interference with order-of-magnitude time reduction
  • Learn state-of-the-art parallel schemes for both model training and serving
  • A detailed study of bottlenecks at distributed model training and serving stages

Description

Reducing time cost in machine learning leads to a shorter waiting time for model training and a faster model updating cycle. Distributed machine learning enables machine learning practitioners to shorten model training and inference time by orders of magnitude. With the help of this practical guide, you'll be able to put your Python development knowledge to work to get up and running with the implementation of distributed machine learning, including multi-node machine learning systems, in no time. You'll begin by exploring how distributed systems work in the machine learning area and how distributed machine learning is applied to state-of-the-art deep learning models. As you advance, you'll see how to use distributed systems to enhance machine learning model training and serving speed. You'll also get to grips with applying data parallel and model parallel approaches before optimizing the in-parallel model training and serving pipeline in local clusters or cloud environments. By the end of this book, you'll have gained the knowledge and skills needed to build and deploy an efficient data processing pipeline for machine learning model training and inference in a distributed manner.

Who is this book for?

This book is for data scientists, machine learning engineers, and ML practitioners in both academia and industry. A fundamental understanding of machine learning concepts and working knowledge of Python programming is assumed. Prior experience implementing ML/DL models with TensorFlow or PyTorch will be beneficial. You'll find this book useful if you are interested in using distributed systems to boost machine learning model training and serving speed.

What you will learn

  • Deploy distributed model training and serving pipelines
  • Get to grips with the advanced features in TensorFlow and PyTorch
  • Mitigate system bottlenecks during in-parallel model training and serving
  • Discover the latest techniques on top of classical parallelism paradigm
  • Explore advanced features in Megatron-LM and Mesh-TensorFlow
  • Use state-of-the-art hardware such as NVLink, NVSwitch, and GPUs

Product Details

Country selected
Publication date, Length, Edition, Language, ISBN-13
Publication date : Apr 29, 2022
Length: 284 pages
Edition : 1st
Language : English
ISBN-13 : 9781801815697
Category :
Languages :
Tools :

What do you get with a Packt Subscription?

Free for first 7 days. $19.99 p/m after that. Cancel any time!
Product feature icon Unlimited ad-free access to the largest independent learning library in tech. Access this title and thousands more!
Product feature icon 50+ new titles added per month, including many first-to-market concepts and exclusive early access to books as they are being written.
Product feature icon Innovative learning tools, including AI book assistants, code context explainers, and text-to-speech.
Product feature icon Thousands of reference materials covering every tech concept you need to stay up to date.
Subscribe now
View plans & pricing

Product Details

Publication date : Apr 29, 2022
Length: 284 pages
Edition : 1st
Language : English
ISBN-13 : 9781801815697
Category :
Languages :
Tools :

Packt Subscriptions

See our plans and pricing
Modal Close icon
€18.99 billed monthly
Feature tick icon Unlimited access to Packt's library of 7,000+ practical books and videos
Feature tick icon Constantly refreshed with 50+ new titles a month
Feature tick icon Exclusive Early access to books as they're written
Feature tick icon Solve problems while you work with advanced search and reference features
Feature tick icon Offline reading on the mobile app
Feature tick icon Simple pricing, no contract
€189.99 billed annually
Feature tick icon Unlimited access to Packt's library of 7,000+ practical books and videos
Feature tick icon Constantly refreshed with 50+ new titles a month
Feature tick icon Exclusive Early access to books as they're written
Feature tick icon Solve problems while you work with advanced search and reference features
Feature tick icon Offline reading on the mobile app
Feature tick icon Choose a DRM-free eBook or Video every month to keep
Feature tick icon PLUS own as many other DRM-free eBooks or Videos as you like for just €5 each
Feature tick icon Exclusive print discounts
€264.99 billed in 18 months
Feature tick icon Unlimited access to Packt's library of 7,000+ practical books and videos
Feature tick icon Constantly refreshed with 50+ new titles a month
Feature tick icon Exclusive Early access to books as they're written
Feature tick icon Solve problems while you work with advanced search and reference features
Feature tick icon Offline reading on the mobile app
Feature tick icon Choose a DRM-free eBook or Video every month to keep
Feature tick icon PLUS own as many other DRM-free eBooks or Videos as you like for just €5 each
Feature tick icon Exclusive print discounts

Frequently bought together


Stars icon
Total 104.97
Machine Learning Techniques for Text
€35.99
Deep Learning with PyTorch Lightning
€36.99
Distributed Machine Learning with Python
€31.99
Total 104.97 Stars icon

Table of Contents

16 Chapters
Section 1 – Data Parallelism Chevron down icon Chevron up icon
Chapter 1: Splitting Input Data Chevron down icon Chevron up icon
Chapter 2: Parameter Server and All-Reduce Chevron down icon Chevron up icon
Chapter 3: Building a Data Parallel Training and Serving Pipeline Chevron down icon Chevron up icon
Chapter 4: Bottlenecks and Solutions Chevron down icon Chevron up icon
Section 2 – Model Parallelism Chevron down icon Chevron up icon
Chapter 5: Splitting the Model Chevron down icon Chevron up icon
Chapter 6: Pipeline Input and Layer Split Chevron down icon Chevron up icon
Chapter 7: Implementing Model Parallel Training and Serving Workflows Chevron down icon Chevron up icon
Chapter 8: Achieving Higher Throughput and Lower Latency Chevron down icon Chevron up icon
Section 3 – Advanced Parallelism Paradigms Chevron down icon Chevron up icon
Chapter 9: A Hybrid of Data and Model Parallelism Chevron down icon Chevron up icon
Chapter 10: Federated Learning and Edge Devices Chevron down icon Chevron up icon
Chapter 11: Elastic Model Training and Serving Chevron down icon Chevron up icon
Chapter 12: Advanced Techniques for Further Speed-Ups Chevron down icon Chevron up icon
Other Books You May Enjoy Chevron down icon Chevron up icon

Customer reviews

Top Reviews
Rating distribution
Full star icon Full star icon Full star icon Full star icon Half star icon 4.3
(14 Ratings)
5 star 78.6%
4 star 0%
3 star 7.1%
2 star 0%
1 star 14.3%
Filter icon Filter
Top Reviews

Filter reviews by




Baron C. Jun 03, 2022
Full star icon Full star icon Full star icon Full star icon Full star icon 5
This book covers an area that isn't taught much, and especially not in academia. Distributed ML is going to be how you get the performance you need. Python is naturally synchronous and this book teaches how to scale up ML to be asynchronous (a necessary addition to anyone's toolset). It also does a great job in covering the pros and cons of each approach. Understanding why you do something is paramount in tech as explaining tradeoffs is a critical part of the job.At a high level, this book covers data parallelism, model synchronization, parallel training, bottlenecks and solutions, pipeline parallelism, parallel serving, elastic model training, and various other ways to speed up the process. You get the picture from the 30k foot view and in great detail.
Amazon Verified review Amazon
Haoran YU May 26, 2022
Full star icon Full star icon Full star icon Full star icon Full star icon 5
An awesome book for ML engineers in cloud computing for practices of the machine learning algorithms on modern distributed computing platforms. I find this book a great source of information regarding for algorithm and system design! Highly Recommend!!
Amazon Verified review Amazon
@maxgoff May 24, 2022
Full star icon Full star icon Full star icon Full star icon Full star icon 5
Although distributed computing has become de rigueur in most modern web applications, the fact remains that most training and reference materials for ML/AI programming still focus on single node architectures. One undeniable trend is the growing girth of data required to train some of the most interesting models emerging today. In order to rapidly innovate and compete, distributed ML will become table stakes in the near future as we move forward.If you write ML/AI code, implement smart data pipelines, architect systems in order to scale or simply want to learn techniques beyond the common core ML/AI training available, this book is a must-have for your shelf. Wang covers a lot of territory and does so clearly with excellent examples. He also provides the technical foundation for the WHY.As more data and processing capabilities accumulate at the edge, the exponentially expanding universe of data processing demands distributed computing. Machine Learning must follow a distributed pattern if it is to continue to provide value. Wang's text provides a solid foundation and reference point.Distributed computing is awesome. We use distributed computing applications every day. Wang's text provides the lessons you will need to ensure that modern ML innovations will utilize resources with much greater productivity. Time is our most precious resource. Distributed Machine Learning with Python will save you LOTS of it!
Amazon Verified review Amazon
Kenan May 14, 2022
Full star icon Full star icon Full star icon Full star icon Full star icon 5
As someone who works with large NLP models everyday, I found this book extremely helpful in industry settings. Not only it provides detailed explanation on different parallel training techniques with clear and simple design-flow pictures, the book also contains code snippets and error messages. One thing I love most about this book is that it takes a very practical perspective. The discussion on outputs and errors with screenshots just makes the process of re-implementing those techniques so much easier for me!I would recommend this book to all researchers and young ML engineers 100%!
Amazon Verified review Amazon
Hitesh Hinduja Aug 11, 2022
Full star icon Full star icon Full star icon Full star icon Full star icon 5
Interesting book with a need of hour in today's age of data. Must read for all the distributed systems enthusiasts.
Amazon Verified review Amazon
Get free access to Packt library with over 7500+ books and video courses for 7 days!
Start Free Trial

FAQs

What is included in a Packt subscription? Chevron down icon Chevron up icon

A subscription provides you with full access to view all Packt and licnesed content online, this includes exclusive access to Early Access titles. Depending on the tier chosen you can also earn credits and discounts to use for owning content

How can I cancel my subscription? Chevron down icon Chevron up icon

To cancel your subscription with us simply go to the account page - found in the top right of the page or at https://subscription.packtpub.com/my-account/subscription - From here you will see the ‘cancel subscription’ button in the grey box with your subscription information in.

What are credits? Chevron down icon Chevron up icon

Credits can be earned from reading 40 section of any title within the payment cycle - a month starting from the day of subscription payment. You also earn a Credit every month if you subscribe to our annual or 18 month plans. Credits can be used to buy books DRM free, the same way that you would pay for a book. Your credits can be found in the subscription homepage - subscription.packtpub.com - clicking on ‘the my’ library dropdown and selecting ‘credits’.

What happens if an Early Access Course is cancelled? Chevron down icon Chevron up icon

Projects are rarely cancelled, but sometimes it's unavoidable. If an Early Access course is cancelled or excessively delayed, you can exchange your purchase for another course. For further details, please contact us here.

Where can I send feedback about an Early Access title? Chevron down icon Chevron up icon

If you have any feedback about the product you're reading, or Early Access in general, then please fill out a contact form here and we'll make sure the feedback gets to the right team. 

Can I download the code files for Early Access titles? Chevron down icon Chevron up icon

We try to ensure that all books in Early Access have code available to use, download, and fork on GitHub. This helps us be more agile in the development of the book, and helps keep the often changing code base of new versions and new technologies as up to date as possible. Unfortunately, however, there will be rare cases when it is not possible for us to have downloadable code samples available until publication.

When we publish the book, the code files will also be available to download from the Packt website.

How accurate is the publication date? Chevron down icon Chevron up icon

The publication date is as accurate as we can be at any point in the project. Unfortunately, delays can happen. Often those delays are out of our control, such as changes to the technology code base or delays in the tech release. We do our best to give you an accurate estimate of the publication date at any given time, and as more chapters are delivered, the more accurate the delivery date will become.

How will I know when new chapters are ready? Chevron down icon Chevron up icon

We'll let you know every time there has been an update to a course that you've bought in Early Access. You'll get an email to let you know there has been a new chapter, or a change to a previous chapter. The new chapters are automatically added to your account, so you can also check back there any time you're ready and download or read them online.

I am a Packt subscriber, do I get Early Access? Chevron down icon Chevron up icon

Yes, all Early Access content is fully available through your subscription. You will need to have a paid for or active trial subscription in order to access all titles.

How is Early Access delivered? Chevron down icon Chevron up icon

Early Access is currently only available as a PDF or through our online reader. As we make changes or add new chapters, the files in your Packt account will be updated so you can download them again or view them online immediately.

How do I buy Early Access content? Chevron down icon Chevron up icon

Early Access is a way of us getting our content to you quicker, but the method of buying the Early Access course is still the same. Just find the course you want to buy, go through the check-out steps, and you’ll get a confirmation email from us with information and a link to the relevant Early Access courses.

What is Early Access? Chevron down icon Chevron up icon

Keeping up to date with the latest technology is difficult; new versions, new frameworks, new techniques. This feature gives you a head-start to our content, as it's being created. With Early Access you'll receive each chapter as it's written, and get regular updates throughout the product's development, as well as the final course as soon as it's ready.We created Early Access as a means of giving you the information you need, as soon as it's available. As we go through the process of developing a course, 99% of it can be ready but we can't publish until that last 1% falls in to place. Early Access helps to unlock the potential of our content early, to help you start your learning when you need it most. You not only get access to every chapter as it's delivered, edited, and updated, but you'll also get the finalized, DRM-free product to download in any format you want when it's published. As a member of Packt, you'll also be eligible for our exclusive offers, including a free course every day, and discounts on new and popular titles.