caffe
caffe copied to clipboard
intel
weight quantization
Hello! I am new to Intel Caffe! As i read Intel document "LOWER NUMERICAL PRECISION DEEP LEARNING INFERENCE AND TRAINING". It said that "quantizing the weights is done before inference starts. Quantizing the activations efficiently requires precomputing the quantization factors". However, when i use Calibrator tool, i just get the quantized prototxt. I don't know where the weights is quantized.
Could you show me where the weights is quantized? Thanks you so much!
You can find "scale_params" in the quantized prototxt
yes, i saw it. do we have to quantize the weight firstly before running inference? or the weight is quantized through reorder primitive? thanks you!