PyTorch, Facebook’s Python-based neural network library, is getting deeper integration with Google Cloud as it approaches its 1.0 release.
PyTorch is a Tensor library, developed by Facebook and built on Python for building (and easily rebuilding) neural networks that can be computed by the CPU or GPU. The library is actually a direct alternative to Google’s own TensorFlow library.
Google Cloud has had some support for PyTorch up to this point, but to coincide with PyTorch’s upcoming 1.0 release, Google Cloud is announcing a variety of upcoming features to improve the developer experience using the two together.
To get developers up and running quickly, Google Cloud’s Deep Learning VM service now offers a virtual machine preloaded with PyTorch 1.0, all the necessary drivers and software, and some examples to help getting started.
In recent years, Google has developed dedicated Tensor Processing Units (or TPUs) to accelerate Tensor calculations in the Google Cloud infrastructure. Going forward, Google will work directly with the PyTorch team to get their software running well on the Cloud TPUs.
Today, we’re pleased to announce that engineers on Google’s TPU team are actively collaborating with core PyTorch developers to connect PyTorch to Cloud TPUs. The long-term goal is to enable everyone to enjoy the simplicity and flexibility of PyTorch while benefiting from the performance, scalability, and cost-efficiency of Cloud TPUs.
Google’s initial tests of the library on Cloud TPUs have been successful, but more needs to be done before open sourcing the work for the community to use and improve on it.
The PyTorch developers are also collaborating with Google Cloud’s TensorBoard team to create useful visualization tools to watch PyTorch instances as they train. This helps the results of machine learning be less of a surprise, and let developers catch issues early on, rather than wasting precious time (and money) on flawed designs.
This seems mundane on first read, but keep in mind that PyTorch is primarily developed by Facebook. The announcement is basically saying that Google is collaborating directly with Facebook. This partnership could be for nothing more than the benefit of machine learning developers around the world, but perhaps this could be the start of an interesting relationship between the tech giants.