Skip to content or navigation


Khronos Group Releases new NNEF Converters, Extensions and Model Zoo

This month Khronos Group Releases New Open Standards for Neural Networks.

Advertisement

The Khronos Group is advancing the interest of neural networks with the new expansion of the ecosystem for the NNEF™ (Neural Network Exchange Format), the open and royalty-free standard that enables hardware manufacturers to reliably exchange trained neural networks between training frameworks and inference engines.

Neural Networks and Standards

New and improved NNEF open source converters, including for TensorFlow Lite and ONNX, enables NNEF to be used to carry trained frameworks from a wider range of training frameworks. A set of extensions to the NNEF 1.0 specification enable NNEF files to contain a  richer network of operations and topologies. Finally, an openly available NNEF Model Zoo enables inferencing engines to test their reliable import of NNEF models. More information on NNEF can be found at the NNEF Home Page.

“Since the release of NNEF 1.0 last year, the NNEF ecosystem has been steadily growing, and the release of these new tools and the Model Zoo is another pivotal step in continuing its mission to reduce machine learning deployment fragmentation,” says Viktor Gyenes, the Khronos NNEF working group specification editor. “With a quorum of supporting hardware companies and its proven multi-company governance model, Khronos is enabling NNEF to effectively evolve and respond to the needs of the diverse and fast-growing field of hardware inferencing companies.”

NNEF Tools Ecosystem diagram as part of Khronos Group standards initiatives.

The NNEF Working Group has unified the GitHub library of open-source converter tools to use a common interface and a common underlying code-base to facilitate easy maintenance and further updates to support additional formats. The set of available converters now enable bi-directional conversion between NNEF and TensorFlow, Caffe/Caffe2, and ONNX. There are also exporters to TensorFlow Lite and an open-source syntax parser/validator. Data scientists and engineers can now easily work with custom networks and use these new tools to seamlessly move from training to deployment using NNEF. These new open source tools for NNEF provide a firm foundation for new converters in the future to meet growing market needs.

NNEF’s list of operations has also been updated through extensions included in version 1.0.2 of the NNEF specification to now include sine/cosine and further reduce and pad/tile operations. NNEF’s capabilities will continue to grow as the machine learning field evolves and increasingly complex networks need to be supported.

The NNEF Model Zoo can be accessed via GitHub and consists of a significant set of pre-converted networks to enable inferencing engine vendors to test their reliable import of NNEF models and inferencing engine operations. The NNEF community is encouraged to visit the NNEF GitHub repository to provide feedback on the new convertors, request new functionality, and help evolve the Model Zoo to be most useful for their use cases and applications.

To provide deeper feedback on NNEF and its new tools under the Khronos IP Framework, Khronos encourages interested companies or individuals to share feedback via GitHub or inquire about joining Khronos membership at the Khronos membership page.

Reader Comments

There are no comments yet on this story

Leave a Reply

architosh

INSIDER Xpresso 

Monthly newsletter with a focus on emTech (AI, machine learning, AAD, AR, robotics, 3d printing, and smart cities) its impact on CAD professionals.)

Subscribe now — it’s free!.

INSIDER Xpresso keeps CAD industry professionals up-to-date on next-gen emerging technologies (emTech) that will revolutionize the worlds of AEC and manufacturing and design. As an Xpresso reader, you will hear from some of the most important voices inventing and using the very latest tech in areas such as AI, machine learning, algorithmic-aided design (AAD), AR, VR, MR, 3D printing, 3D computer vision, robotics, and SmartCities technologies.

Each issue arrives in your inbox on the first Sunday of the month. Issue #1 in your inbox March 3, 2019.

Sign-up for our monthly newsletter
architosh INSIDER Xpresso.

  • Architosh will never pass any of your information onto third parties.
  • For more information read our privacy policy.
  • It is easy to unsubscribe at any time. Follow the links in the newletter footer.

(Recommended. These infrequent sponsored emails help us to provide our Xpresso newsletter for free.)

 
INSIDER Membership

Read 3 free Feature or Analysis articles per month.

Or, subscribe now for unlimited full access to Architosh.