Other Useful Deep Learning Libraries

TensorFlow from Google is not the only framework available for deep learning tasks. There is a good range of libraries and frameworks available, each with its special features, capabilities, and use cases. In this chapter, we will explore some of the popular deep learning libraries and compare their features.

The chapter will include:

• Hugging Face
• H2O
• PyTorch
• ONNX
• Open AI

All the code files for this chapter can be found at https://packt.link/dltfchp16.

Let’s begin!

Hugging Face

Hugging Face is not new for us; Chapter 6, Transformers, introduced us to the library. Hugging Face is an NLP-centered startup, founded by Delangue and Chaumond in 2016. It has, in a short time, established itself as one of the best tools for all NLP-related tasks. The AutoNLP and accelerated inference API are available for a price. However, its core NLP libraries datasets, tokenizers, Accelerate, and transformers (Figure 16.1) are available for free. It has built a cool community-driven open-source platform.

Figure 16.1: NLP libraries from Hugging Face

The core of the Hugging Face ecosystem is its transformers library. The Tokenizers and Datasets libraries support the Transformers library. To use these libraries, we need to install them first. Transformers can be installed using a simple pip install command:

pip install transformers

Some of the out-of-the-box models available with Hugging Face are text summarization, question answering, text classification...

OpenAI

OpenAI is another well-known name for people working in the field of reinforcement learning. Their Gym module is a standard toolkit used by developers across the globe for developing and comparing reinforcement learning algorithms. In Chapter 11, Reinforcement Learning, we have already covered the Gym module in detail. In this chapter, we will explore two more offerings by OpenAI.

OpenAI GPT-3 API

“OpenAI GPT3 is a machine learning platform that allows developers to build custom algorithms for deep learning. This platform was released in December of 2017 and has been widely used by businesses and individuals in the field of artificial intelligence. One of the primary reasons that GPT3 has been so successful is because it is easy to use and has a wide range of features. This platform is able to learn from data and can be used for a variety of tasks, including deep learning, natural language processing, and image recognition. GPT3 is also popular because it is...

PyTorch

Like TensorFlow, PyTorch is a full-fledged deep learning framework. In AI-based social groups, you will often find die-hard fans of PyTorch and TensorFlow arguing that theirs is best. PyTorch, developed by Facebook (Meta now), is an open-source deep learning framework. Many researchers prefer it for its flexible and modular approach. PyTorch also has stable support for production deployment. Like TF, the core of PyTorch is its tensor processing library and its automatic differentiation engine. In a C++ runtime environment, it leverages TorchScript for an easy transition between graph and eager mode. The major feature that makes PyTorch popular is its ability to use dynamic computation, i.e., its ability to dynamically build the computational graph – this gives the programmer flexibility to modify and inspect the computational graphs anytime.

The PyTorch library consists of many modules, which are used as building blocks to make complex models. Additionally, PyTorch...

ONNX

Open Neural Network Exchange (ONNX) provides an open-source format for AI models. It supports both deep learning models and traditional machine learning models. It is a format designed to represent any type of model, and it achieves this by using an intermediate representation of the computational graph created by different frameworks. It supports PyTorch, TensorFlow, MATLAB, and many more deep learning frameworks. Thus, using ONNX, we can easily convert models from one framework to another. This helps in reducing the time from research to deployment. For example, you can use ONNX to convert a PyTorch model to ONNX.js form, which can then be directly deployed on the web.

H2O.ai

H2O is a fast, scalable machine learning and deep learning framework developed by H2O.ai, released under the open-source Apache license. According to the company website, as of the time of writing this book, more than 20,000 organizations use H2O for their ML/deep learning needs. The company offers many products like H2O AI cloud, H2O Driverless AI, H2O wave, and Sparkling Water. In this section, we will explore its open-source product, H2O.

It works on big data infrastructure on Hadoop, Spark, or Kubernetes clusters and it can also work in standalone mode. It makes use of distributed systems and in-memory computing, which allows it to handle a large amount of data in memory, even with a small cluster of machines. It has an interface for R, Python, Java, Scala, and JavaScript, and even has a built-in web interface.

H2O includes a large number of statistical-based ML algorithms such as generalized linear modeling, Naive Bayes, random forest, gradient boosting, and...

Summary

In this chapter, we briefly covered the features and capabilities of some other popular deep learning frameworks, libraries, and platforms. We started with Hugging Face, a popular framework for NLP. Then we explored OpenAI’s GPT-3 and DALL-E 2, both very powerful frameworks. The GPT-3 API can be used for a variety of NLP-related tasks, and DALL-E 2 uses GPT-3 to generate images from textual descriptions. Next, we touched on the PyTorch framework. According to many people, PyTorch and TensorFlow are equal competitors, and PyTorch indeed has many features comparable to TensorFlow. In this chapter, we briefly talked about some important features like the NN module, Optim module, and Autograd module of PyTorch. We also discussed ONNX, the open-source format for deep learning models, and how we can use it to convert the model from one framework to another. Lastly, the chapter introduced H2O and its AutoML and explain modules.

In the next chapter, we will learn about...