PyTorch 1.0: Facebook’s Very Own AI Framework Built On Python

At last! Good news about Facebook.

via GIPHY

Mark Zuckerberg’s can take a breath of relief.

Today the company has announced PyTorch 1.0, their open source AI framework which is the advanced version of its predecessor. Made with Python language, Latest version of PyTorch has been integrated with caffe 2 and ONNX, renowned deep learning framework.

phython

The most astounding factor of this integration is that it will remove all the migration related issues of the previous version and might equip the tool to process more data.

The news was announced at F8 developer conference at San Jose, California on Wednesday.

 

PyTorch.org the official website of the framework after the announcement put up a brief note regarding the latest version and explained the features of the new version like this.

One of its biggest downsides has been production-support. What we mean by production-support is the countless things one has to do to models to run them efficiently at massive scale:

  • Exporting to C++-only runtimes for use in larger projects

 

  • Optimizing mobile systems on iPhone, Android, Qualcomm and other systems

 

  • Using more efficient data layouts and performing kernel fusion to do faster inference (saving 10% of speed or memory at scale is a big win)

 

  • Quantized inference (such as 8-bit inference)

 

The framework PyTorch 1.0 is already put to use and is responsible for over 6 billion language translation per day and that too from 48 languages.

via GIPHY

The latest version will equip developers to optimize efficiency with the help of hybrid entrance finish that can transition between modes.

The first version of the platform was introduced a year back. So far, more than a million downloads were made.

The biggest drawback of the first version was that the platform was not equipped to be implemented on a production scale which is expected not to prevail in this version of the framework.

via GIPHY

It is expected that soon Facebook will make a lot of their AI related developing platform open source so that people can make use of it.

AI (Artificial Intelligence) team of Facebook has been indulging themselves in a lot of research for quite some time. They mainly have 2 research team with the name Facebook AI Research group (FAIR) and the company’s Applied Machine Learning team (AML). FAIR usually deals with hardcore AI related computation and on the other hand, AML deals with lightweight machine learning models.

During the f8 developer conference, Mark said that Facebook uses the AI tools developed by the team for many purposes. One of the most important is content moderation.

probytes

Talking about moderation, apart from the huge manpower assistance, there is software with the name “Computer Vision” developed by Facebook that helps in flagging inappropriate content so that it can be either informed or removed.

It is expected that more tools will be introduced in the days to come to offer a better user experience in Facebook. However, the latest allegation against Facebook has indeed tarnished their image. So this is desperate times for Facebook and for desperate measures.