Khronos Announces NNEF 1.0 Standard For Neural Networks

Last year The Khronos Group announced NNEF as a open-source, royalty-free neural network format to combat the proprietary formats used today. In their last standards update of 2017, NNEF 1.0 is now available.

The Neural Network Exchange Format 1.0 (NNEF 1.0) is designed to facilitate the porting of trained neural networks across different inference engines and frameworks. The goal of NNEF is to reduce the fragmentation in the machine learning space by having this standard, interoperable format for more easily transferring networks across platforms.

NNEF 1.0 was designed with import/export in mind for Torch, TensorFlow, Caffe, Caffe2, Thano, Chainer, PyTorch, and MXNet.

From Khronos' announcement, "The NNEF 1.0 Provisional specification covers a wide range of use-cases and network types with a rich set of functions and a scalable design that borrows syntactical elements from Python but adds formal elements to aid in correctness. NNEF includes the definition of custom compound operations that offers opportunities for sophisticated network optimizations. Future work will build on this architecture in a predictable way so that NNEF tracks the rapidly moving field of machine learning while providing a stable platform for deployment."

For those into machine learning, you can learn more about NNEF 1.0 at Khronos.org.