Microsoft remains committed to making AI more accessible for everyone. Microsoft offers many tools like their Cognitive Toolkit (an open source framework for building deep neural networks) for developing various AI related applications. In the same approach, Microsoft has announced Open Neural Network Exchange format in conjunction with Facebook. Open Neural Network Exchange (ONNX) will provide a shared model representation for interoperability and innovation in the AI framework. Various toolkit from Microsoft like Cognitive Toolkit, PyTorch and Caffe2 will be supporting ONNX.
There are various frameworks that provide interfaces which make it easier for developers to construct and run computation graphs to represent the neural network. But, one of the limitations with so many frameworks is that they have their own format for representing these graphs. This is where ONNX id bit different. The two-unique feature provided by ONNX are –
Developers can easily shift between frameworks and use the best tool available. Each framework is optimised for some particular functions like speed, flexibility. Many times, restricting yourself to only one framework prevent you to take advantage of all these available benefits. But, ONNX will simply t5his process by allowing developers to use the best technology available and become more agile in their approach.
While using different framework frequently optimizations need to be integrated separately into each framework which can be a time-consuming process. But, ONNX will make it easier for optimisation to reach more developers.
ONNX is beneficial for all developer, framework vendor, and hardware vendor as Developers can choose the right framework for their task, framework authors can focus on innovative enhancements, and hardware vendors can streamline optimizations. The first version of ONNX is available on GitHub so that developers can get started right away. Microsoft has assured that they will work on this in conjunction with Facebook to contribute reference implementations, examples, tools.
GitHub Link of ONNX : Click Here
GitHub link for ONNX -caffe 2 : Click Here