At a recent PyTorch developer conference in San Francisco, Facebook released a developer preview version of PyTorch 1.0. PyTorch is an open source deep learning framework used to reduce friction when taking research projects to production. In this release, many investments have been made by public cloud and hardware companies to better support the PyTorch ecosystem as the popularity of PyTorch is on the rise.
There have been many recent investments made by Facebook and third parties in the PyTorch ecosystem. Joseph Spisak, a product manager at Facebook, shared the following update at the conference:
PyTorch 1.0 accelerates the workflow involved in taking breakthrough research in artificial intelligence to production deployment. With deeper cloud service support from Amazon, Google, and Microsoft, and tighter integration with technology providers ARM, Intel, IBM, NVIDIA, and Qualcomm, developers can more easily take advantage of PyTorch’s ecosystem of compatible software, hardware, and developer tools. The more software and hardware that is compatible with PyTorch 1.0, the easier it will be for AI developers to quickly build, train, and deploy state-of-the-art deep learning models.
Amazon has made investments in AWS SageMaker, a fully managed platform for training and deploying machine learning models at scale. The service now provides preconfigured environments for PyTorch 1.0 which include automatic model tooling.
The investments that Google has made are in the areas of virtual machine (VM) images in the Google Cloud Platform, which include NVIDIA drivers, and pre-installed tutorials. In addition, Google is working with Facebook’s PyTorch team to enable support for PyTorch on custom application-specific integrated circuits (ASIC) hardware.
Microsoft was an early partner with Facebook on another AI initiative called ONNX, which allows developers to represent and transport deep learning models across different tooling. In addition to the work on ONNX, Microsoft is providing pre-configured Data Science Virtual Machines (DSVM) that come preinstalled with PyTorch. Microsoft is also providing developers with an on-ramp to learning PyTorch without having to install software. This is accomplished through free cloud-hosted Jupyter Notebook solutions that include PyTorch tutorials. Lastly, Microsoft has included an AI extension for Visual Studio Code that includes the integration of Azure ML and PyTorch APIs for streamlined PyTorch code development and training.
Beyond the advancements made by cloud providers, hardware companies have also made investments. These investements have been made by ARM, IBM, Intel, NVIDIA and Qualcomm who are adding support for PyTorch 1.0 through direct optimizations, kernel library integration and support for compilers and inference runtimes. Spisak emphasized the value of these investments:
This extra support ensures that PyTorch developers can run models across a broad array of hardware, optimized for training and inference, for both data center and edge devices.
PyTorch has also seen an increase in students enrolling in online learning programs and university programs and Facebook is working with Udacity to publish a free online course. Spisak attributes this interest to:
The framework’s approachability and deep integration into Python have made it easier for students to understand and experiment with various deep learning concepts. With the evolution of PyTorch 1.0, we’re thrilled that more partners will be further focusing their curricula around it.
Developers can get started by downloading the developer preview of PyTorch 1.0, or through one of the cloud providers mentioned earlier in this article.