- Facebook has announced it’ll be releasing PyTorch 1.0 and making it open source, along with other AI tools
- PyTorch 1.0 is already in use at Facebook; it is used for 6 billion language translations everyday
- Microsoft Azure ML and Amazon Web Services have already indicated that they will be using this in their flagship offerings
Moving your machine learning project from research to production has historically been a challenge – both budget wise as well as framework wise. It can also be time consuming along with being complicated to test when new approaches need to be deployed. Doesn’t sound promising, does it?
A research team at Facebook is aiming to smoothen this process for the community.
PyTorch 1.0, announced by Facebook at the F8 event yesterday, aims to provide developers a seamless path to take their project from research to production in a single framework! With this tool, data scientists and AI developers can experiment and optimize performance through a hybrid front end. It might not surprise you to know that this technology is already in use at Facebook where it is performing close to 6 billion language translations per day (this translation tool is called, appropriately, Translate).
So how is PyTorch 1.0 different from their latest release – v0.4.0? v1.0 takes the existing super flexible framework and combines it with the production-oriented capabilities from Caffe2 and ONNX. The releases so far have had one challenge – their performance at production. Developers often need to first translate their code to a graph mode representation in Caffe2 in order to make it production ready. This issue has been eliminated in PyTorch 1.0.
Microsoft and Amazon Web Services have already indicated that they will be using PyTorch 1.0 in their flagship offerings.
PyTorch 1.0 will be available in beta version in the next few months. It is expected to include tools, libraries, datasets and pre-trained models for each stage of development. You can read more about this latest release by Facebook on it’s blog post here.
Our take on this
Facebook, from time to time, pioneers new tools and techniques that are state-of-the-art. I love that they oen source their research so that the entire ML community can benefit from it. Deploying your code to production from research has commonly been a major challenge so this release should get rid of that headache soon.
Time will tell how many organizations decide to integrate PyTorch into their existing services but all things considered, this is a major announcement in the wider perspective. Soon, you won’t be able to say that TensorFlow is better for production than PyTorch!
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!
You can also read this article on our Mobile APP