Meta Launches PyTorch Live To build AI-Powered Mobile Apps
“PyTorch's mission is to accelerate the transition from research prototypes to production deployments. A VentureBeat spokesperson said via email that this is more important than ever as the mobile machine learning ecosystem grows. "To help mobile developers reduce friction when building new machine learning solutions, we are introducing PyTorch Live, a tool for building, testing and (future) demonstrating AI on PyTorch-powered devices."
PyTorch, released publicly by Meta in January 2017, is an open source machine learning library based on Torch, a scientific computing framework and scripting language based on the Lua programming language. TensorFlow has been around for a bit longer since November 2015, but PyTorch continues to quickly gain popularity with the developer and data science community. Octoverse's 2018 GitHub report ranks it as one of the fastest growing open source projects, and Meta recently reported that the platform grew over 50% per annum in 2019, reaching nearly 1,200.
4. PyTorch Live is based on PyTorch Mobile, a React Native library for building runtimes and visual user interfaces that allow developers to move from model training to deployment while staying in the PyTorch ecosystem. PyTorch Mobile provides on-device output for PyTorch Live.
PyTorch Mobile was released in October 2019 following the previous release of Caffe2go, a mobile CPU and GPU optimized version of Meta's Caffe2 framework for machine learning. PyTorch Mobile can run with its own runtime and is built on the assumption that whatever developers want to do on a mobile or edge device, they can also do on the server.
“For example, if we wanted to demonstrate a model of a mobile application that runs on Android and iOS, it would take several days to set up the project and create the user interface. PyTorch Live cuts costs in half and requires no Android and iOS development experience.
PyTorch Live comes with a command line interface (CLI) and data processing API. The CLI allows developers to customize their mobile development environment and launch mobile application projects. In terms of data processing APIs, you can prepare and integrate custom models for use with the PyTorch Live API and then embed them into AI-powered mobile apps for Android and iOS.
Going forward, Meta plans to provide the community with the ability to open and share PyTorch models and demos via PyTorch Live, as well as provide a more customizable data processing API and support the machine learning domain to process audio and video data. .