Amazon Web Services and Facebook have recently announced plans to join hands for a few new projects for PyTorch. PyTorch is a well-known, open-source ML (machine learning) framework that has been adopted to train AI (artificial intelligence) models.
The framework has been created by the AI research team of Facebook as an ML library of various functions for Python, the programming language. It has been primarily designed for a branch of ML, deep learning, that tries to emulate the way of functioning for the human brain and has led to various major breakthroughs in areas including voice recognition, image, and language translation. The framework has been built to increase the development process of various kinds of artificial intelligence capabilities. It has been used previously to create realistic avatars for the Oculus virtual reality headset of Facebook. For instance, researchers at the University of California, Berkeley have also adopted the framework to increase the work speed on the image-to-image transformation.
The recently announced PyTorch projects include TorchServe, which enables developers to shift new models to production, and TorchElastic, which creates fault-tolerant training jobs on the Kubernetes clusters such as EC2 spot instances of Amazon or Amazon Web Services Elastic Kubernetes Service. TorchServe allows developers to run various model versions at a time, or even roll back to the model’s previous version. On the other hand, TorchElastic is meant to be adopted in large distributed ML projects including computer vision and natural language processing.
The release of PyTorch 1.5 includes a stable C++ front-end API (app programming interface) that allows the translation of the models in the framework from Python API to C++ API. The company has reportedly stated that PyTorch 1.5 also comes with an upgraded version of torchaudio, torchtext, and torchvision libraries. The torch_xla package that allows PyTorch to link with the Google Cloud TPU (Tensor Processing Unit) chips has also received an update. Google Cloud TPU support was first added by Facebook at an annual developer conference for PyTorch held in San Francisco in 2019.
Source credit:
https://siliconangle.com/2020/04/21/aws-facebook-announce-new-open-source-projects-pytorch/