Skip to content

[Announcement] ONNX working groups established

Dmytro Dzhulgakov edited this page Mar 28, 2018 · 5 revisions

ONNX Working Groups

We are excited to announce the formation of community working groups. Working groups will bring together ONNX partners and members of the community to help steer the direction of ONNX. We have created 4 new working groups to provide guidance and feedback on the topics of Quantization, RNNs and Control Flow, Test and Compliance, and Training. If you are interested in participating in one of these working groups, please contact the group leader directly. As the project evolves we plan to announce more working groups, so if there is a topic you'd like to explore in a working group, let us know. We look forward to innovating with you in the working groups!

Quantization

Leader: Ke Zhang kezhan@microsoft.com

Discussion: Quantization Gitter Room

The goal of the group is to enhance ONNX to support quantized data types and operators on a variety of runtimes and hardware devices.

RNNs and Control Flow

Leader: James Reed jamesreed@fb.com

Discussion: Control Flow Gitter Room

The goal of the working group is to enable dynamic control structures including support for RNNs—beyond optimized GRU and LSTM implementations—that are used in applications such as NLP and video analysis. Through standardizing the representation of control flow operations, the group will create a common representation and push forward the flexibility and performance of models that use control flow.

Test and Compliance

Leader: Lu Fang lufang@fb.com

Discussion: Test and Compliance Gitter Room

The Test and Compliance working group will create tools, tests and apis to provide interoperability across the ONNX ecosystem. The goal of the group is to ensure models using the .onnx format, meet and comply with the ONNX specification and that testing is in place for supported backend frameworks.

Training

Leader: Sukwon Kim sukwokim@amazon.com

Discussion: Training Gitter Room

The goal of the group is to explore the possibility of expanding ONNX to support training as well as inference.

You can’t perform that action at this time.