Open Neural Network Exchange (ONNX)

 Open Neural Network Exchange (ONNX) is the open standard for representation of models of machine learning. By means of it it is possible to tell training models from frameworks of Caffe2, PyTorch and CNTK  (Cognitive Toolkit). For example, researchers can construct a pilot model through PyTorch having big flexibility and ample opportunities or by means of a framework of ConvNet allowing in interactive the mode quickly to create prototypes, and during creation of an end product to transfer to Caffe2, differing in the high performance and an opportunity for work on mobile devices. 

Initially this OpenSource-project developed Microsoft and Facebook, and then several other companies, including Amazon were connected to ONNX.

2018: Microsoft opened the source code of runtime environment

At the beginning of December, 2018 Microsoft opened source codes of runtime environment of ONNX which represents the highly effective mechanism of logical processing for models of machine learning in the ONNX format.

This Wednesday includes API Python and provides hardware acceleration of the central and graphic processor.  It supports the ONNX 1.2 models and 16.04 Linux are executed in the Ubuntu system.  Packets  of the CPU  and  GPU  are available on  the website PyPi.org. Source codes of the runtime environment are submitted on the GitHub portal at the expense of what as writes ZDNet, developers can configure and integrate it in the systems and compile/create an environment on different operating systems.^[1]

Microsoft opened source codes of the mechanism of logical processing in Windows platform "heart" for machine learning

The ONNX models can be unrolled  in a cloud using machine learning of Azure and runtime environment of ONNX.  Using   their Windows ML it is also possible to unroll on Windows 10 devices.  In community ONNX there are also converters for deployment on other platforms.

Microsoft notes that thanks to ONNX and  other  services Azure AI, infrastructure and  tools, such as  Azure Machine Learning  and  Visual Studio for  AI,  developers and  scientists will be able quicker to create new and  interesting things  in the field of  artificial intelligence.^[2]

Notes