PyTorch

PyTorch  is library  of machine learning  the library  for  the Python language  open source created based on  Torch. It develops under a wing of Facebook and is used for  natural languag processing. PyTorch provides two main high-level models:

• Tensor calculations (by analogy with NumPy) with the developed support of acceleration on GPU;
• Deep neural networks, based on the autodiff system.

2018: An exit of PyTorch 1.0 for developers

At the beginning of October, 2018 Facebook released the final version of the open machine learning PyTorch platform for Facebook developers. It contains the mass of tools and integration tools which will facilitate compatibility with cloud services Google Cloud, Amazon Web Services (AWS) and Microsoft Azure Machine Learning.

Besides, the project was supported by leading manufacturers of chips ARM Nvidia, Qualcomm and Intel which use a framework for integration with library of a core and tracking of runtime of a logical output.

Facebook provided to developers the open platform of machine learning. It supports clouds of Microsoft, Google, Amazon

Simplification of interaction between different stages of machine learning is promoted by a combination in  the version of PyTorch 1.0 of the modular and  focused on  development opportunities of a framework  of Caffe2  and   the ONNX standard  to  the flexible, aimed at  researches structure of library. Thanks to existence of these functions in  one framework need to switch between libraries vanishes, note in Facebook.

Google not only implemented support of PyTorch in the several cloud services, but also started cooperation with Facebook for development    of Tensor Processing Unit (TPU) accelerators for users of PyTorch.

The managed service of end-to-end machine learning  Sagemaker from Amazon will provide to users of PyTorch previously configured environments for automatic adjustment of models of machine learning and other purposes.^[1]

Notes