FACEBOOK IS PUSHING its top artificial intelligence (AI) code and research to the open source community, which will give developers the scope to make smarter software.
The code in question is PyTorch 1.0, the social networking giant's most powerful AI framework in its arsenal.
The framework mixes the Python-based PyTorch open source machine learning library and mixes it with Caffe2 deep learning framework, which Facebook has used to train large machine learning models. It also includes the Open Neural Network Exchange (ONXX), which Facebook and Microsoft worked on to make moving Al development from the research and training stage to deployment easier.
And that's the crux of PyTorch 1.0. Normally developers would need a framework for AI research and training, and another for putting the clever code into production for use in smart systems and software.
This would involve a complicated and time-consuming export process, but by fusing the training, research and production frameworks together, PyTorch 1.0 effectively provides the tools for developers to speed up the move from researching an AI system to bringing to the market.
"With PyTorch 1.0, AI developers can both experiment rapidly and optimise performance through a hybrid front end that seamlessly transitions between imperative and declarative execution modes. The technology in PyTorch 1.0 has already powered many Facebook products and services at scale, including performing six billion text translations per day," explained Bill Jia, vice president at Facebook's infrastructure division.
PyTorch 1.0 will be available in beta form in the next couple of months and will give developers a suite of code libraries, tools, datasets, and pre-trained machine learning models to help get their AI ambition on the road.
It also means devs will get the chance to tap into some of the tech that underpins Facebook's own AI work, which it is increasingly looking at pushing into its main social network, apps, and other platforms.
While Facebook may be under fire for how it handles user and data privacy, the firm does get the nod for being pretty generous with open sourcing its tech and software, as seen with its connect-everyone-to-the-web OpenCellular open source kit. µ
But we probably won't see it until next year
Why stick a finger in a dyke when you can ram the entire boy in the hole, eh?
Reminds us that we're supposed to be able to trust them
'Exclusive' model starts shipping on 29 June