Weights quantization on the fly

[francisco muñoz martinez]

Hi all,

I want to read the quantized weights with python but I am not able to do it since the weights are copied and quantized on the fly. Why is the quantization done on the fly? Why not quantizing the layer just while loading ir and storing it on blobs_[0]? I have thought about doing that, but I am not sure about whether it can affect another part of the code. Could please anybody help me? Can I just modify the loading in order to fill the weights with the quantized values?

Thank you very much