Port to CMix-NN

[mrgabich]

Hy dear,

I did a little port to CMix-NN with 8-8-8 bit precision.
However, it is hard to understand how did you define the quantization layers to improve the accuracy in the CMix-NN paper.
Could you provide any detail about this configuration?
Also, could you provide the scripts to generate the quantization data from tensorflow?

Best Regards,
Geancarlo Abich