Wang Distributed Machine Learning with Python

Accelerating model training and serving with distributed systems
1. Auflage 2024
ISBN: 978-1-80181-721-9
Verlag: De Gruyter
Format: EPUB
Kopierschutz: 0 - No protection

E-Book, Englisch, 284 Seiten

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.

Autoren/Hrsg.

Wang, Guanhua

Fachgebiete

• Mathematik | Informatik EDV | Informatik Informatik Künstliche Intelligenz Spracherkennung, Sprachverarbeitung
• Mathematik | Informatik EDV | Informatik Daten / Datenbanken Data Warehouse
• Mathematik | Informatik EDV | Informatik Informatik Mensch-Maschine-Interaktion Informationsarchitektur
• Mathematik | Informatik EDV | Informatik Daten / Datenbanken Datenbankdesign & Datenbanktheorie

Inhaltsverzeichnis

Table of Contents - Splitting Input Data

- Parameter Server and All-Reduce
- Building a Data Parallel Training and Serving Pipeline
- Bottlenecks and Solutions
- Splitting the Model
- Pipeline Input and Layer Split
- Implementing Model Parallel Training and Serving Workflows
- Achieving Higher Throughput and Lower Latency
- A Hybrid of Data and Model Parallelism
- Federated Learning and Edge Devices
- Elastic Model Training and Serving
- Advanced Techniques for Further Speed-Ups

Leseproben

: 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 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

Single-node training is too slow

The vanilla model training process is to load both the training data and ML model into the same accelerator (for example, a GPU), which is called single-node training. There are mainly three steps that occur in a single node training model:

1. Input pre-processing
2. Training
3. Validation

The following diagram shows what this looks like in a typical model training workflow:

Figure 1.1 – Model training workflow on a single node

As you can see, after input pre-processing, the augmented input data is loaded into the memory of the accelerators (such as GPUs). Following that, the model is trained on the loaded input data batch and validates our trained model iteratively. The goal of this section is to discuss why single-node training is way too slow. First, we will show the real bottleneck in single-node training and then describe how data parallelism mitigates this bottleneck.

The mismatch between data loading bandwidth and model training bandwidth

Now, let's focus on the two kinds of bandwidth (BW) in this data pipeline, namely data loading bandwidth and model training bandwidth, as shown in the preceding diagram. Nowadays, we have more and more input data. Hence, we would ideally want the to be as large as possible (the wide gray arrow in the preceding diagram). However, due to the limited on-device memory of the GPUs or other accelerators, the real is also limited (the narrow gray arrow in the preceding diagram).

Although it is generally believed that the larger input data size leads to a longer training time in single-node training, this is not true from the data flow perspective. From a system perspective, the mismatch between data loading bandwidth and model training bandwidth is the real issue. .

Real Bottleneck

A large input data size is not the fundamental cause of long training times in terms of single nodes. The mismatch between and is the key issue.

Now that we know the reason behind the delay in single-node training when faced with large input data, let's move on to the next subtopic. Next, we will quantitively show the training times of some classic deep learning models by using standard datasets. This should help you understand why data parallel training is a must-have to deal with the mismatch between data loading bandwidth and model training bandwidth.

Single-node training time on popular datasets

Let's directly jump into training time analysis using a single GPU. We will use an NVIDIA Tesla M60 GPU as the accelerator. First, we will train both VGG-19 and ResNet-164 on the CIFAR-10 and CIFAR-100 datasets. The following diagram shows the corresponding total training time for reaching a model test accuracy over 91%:

Figure 1.2 – Model training time of a single node on the CIFAR-10/100 datasets

As we can see, the total training time of VGG-19 is around 2 hours for both the CIFAR-10 and CIFAR-100 datasets, while for ResNet-164, the total training time for both the CIFAR-10 and CIFAR-100 datasets is around 10 hours.

It seems that the standard model training time, when using a single GPU on the CIFAR-10/100 dataset, is neither short nor long, which is acceptable. This is mainly because of low image resolution. For the CIFAR-10/100 datasets, the resolution of each image is very low at 32x32. Thus, the intermediate results that are generated during the model training stage are relatively small, since the activation matrices in the intermediate results are always less than 32x32. Since we generate smaller activations during training in a given fixed hardware memory size, we can train more input images at once. Consequently, we can achieve a higher , which mitigates the mismatch between and .

Now, let's look at a modern ML model training dataset, such as ImageNet-1K. We have maintained a similar training environment setup to what we had for our CIFAR-10/100 training jobs. The difference is that we are training the VGG-19 and ResNet-50 models. The following diagram shows the corresponding total training time with a single GPU setting:

Figure 1.3 – Model training time for a single node on the ImageNet-1K dataset

As we can see, the training time on a single GPU is unacceptable. It takes around to train a single model, such as VGG-19 or ResNet-50. The main reason for this much slower training speed on the ImageNet-1K dataset is the higher image resolution, which is now around 256x256. Having a higher image resolution means that each training image will have a bigger memory footprint for storing its activations, which means that we can only train a smaller amount of images at once. Thus, the gap between model training bandwidth and data loading bandwidth is larger. Furthermore, the training time can be even longer for wider and deeper model training.

For our machine learning practitioners, the whole model updating cycle is way too long if we only limit ourselves to using a single GPU. This long training time is amplified since we need to try multiple sets of hyperparameters and find the best training recipes.

Therefore, we need to adopt the data parallel training paradigm to mitigate this mismatch between data loading bandwidth and model training bandwidth.

Accelerating the training process with data parallelism

So far, we have discussed why data parallel training is a must-have due to the mismatch between and . Before we dive into the details of how data parallel training works, let's look at the speed-ups that data parallelism can achieve over single node training.

Let's take ResNet-50 training on the ImageNet-1K dataset as an example. By using a proper hyperparameter setup, the following diagram shows the normalized speedups over different GPU training baselines:

Figure 1.4 – Normalized speedups over a single GPU baseline

As we can see, we have tested the system throughput for the data parallel training process over a single GPU training baseline. By incorporating multiple GPUs into the same training job, we expanded our model training bandwidth significantly with parallelism. Ideally, the extended model training bandwidth should be linearly increased by the number of GPUs involved. Due to system control overheads and network communications introduced in data parallel training,...

Über Autor(innen)

Wang Guanhua :

Guanhua Wang is a final-year Computer Science PhD student in the RISELab at UC Berkeley, advised by Professor Ion Stoica. His research lies primarily in the Machine Learning Systems area including fast collective communication, efficient in-parallel model training and real-time model serving. His research gained lots of attention from both academia and industry. He was invited to give talks to top-tier universities (MIT, Stanford, CMU, Princeton) and big tech companies (Facebook/Meta, Microsoft). He received his master's degree from HKUST and bachelor's degree from Southeast University in China. He also did some cool research on wireless networks. He likes playing soccer and runs half-marathon multiple times in the Bay Area of California.