Deep gradient compression

Hi All,

I would like to know if anyone has successfully implemented Deep Gradient Compression on MXNet in this community.
I have made a post in Reddit regarding this. Check it out.

It would be great if someone can help me on this.


Have you come across the work on gradient compression in MXNet already?

Yes. I believe the compression techniques implemented in MXNet are quantization techniques. What I’m looking for specifically the Deep Gradient Compression technique published in this paper.