Hello,
Here is our proposed abstract & title:

Extending MXNet for new features and performance
Sam Skalicky, Serge Panev

We will present MXNet Extensions: custom operators, custom partitioners,
custom graph passes, and APIs for using those extensions. We’ll give a
concrete example of how the TensorRT backend leverages the built-in
partitioning and accelerates parts of a model.

Thanks
Sam

Reply via email to