Facebook’s PyTorch 1.0 Will Offer End-to-End A.I. Platform, Tools

If you’re an established A.I. researcher (and you know who you are; tech companies are trying to back dump-trucks full of cash into your driveway as we speak, if you believe current reports), it’s no doubt an exciting time: tech companies, anxious to jump aboard the A.I. and machine-learning bandwagon, are releasing all sorts of tools designed to make research and development a little bit easier.

Facebook is the latest company to jump into this fray. Its upcoming PyTorch 1.0 is an open-source A.I. framework that’s geared toward production and research. “With PyTorch 1.0, A.I. developers can both experiment rapidly and optimize performance through a hybrid front end that seamlessly transitions between imperative and declarative execution modes,” reads Facebook’s latest posting on the matter. “The technology in PyTorch 1.0 has already powered many Facebook products and services at scale, including performing 6 billion text translations per day.”

The beta, scheduled to roll out later this year, will supposedly come with tools, libraries, datasets and pre-trained models, which will hopefully spare researchers a lot of time and effort related to setup.

The earlier version of PyTorch enjoyed more than a million downloads; researchers used it for deep-learning tasks such as machine vision (i.e., training a platform to recognize, and even edit, images). For example, researchers at UC Berkeley used it to build a tool capable of transforming objects within images into other objects, such as zebras to horses and summer landscapes into winter wonderlands:

PyTorch version 1.0 will feature a more refined front-end designed to speed up productivity, and supposedly process researcher requests at “production scale.”

Facebook plans on open-sourcing other A.I. tools, including Translate (a PyTorch programming-language library), ELF (a game platform for A.I. reasoning applications), and Glow (a machine-learning compiler for speeding up framework performance). Also in the works: open-sourced libraries such as Detectron, which is for object-detection research.

For Facebook, providing A.I. researchers with an end-to-end platform for A.I. and deep learning has obvious benefits: the more tech pros using its datasets and tools, the more it can improve the underlying technology—ultimately giving it an advantage in the marketplace. That will allow Facebook to more effectively push back against Google, Microsoft, and other tech firms that want to become the “go to” hubs for everything A.I.-related.