>>106363201
quantization is a mapping of the models weights down to a smaller size.
weights are basically floating point numbers.
basically it is like images, the lower the amount of bits you can store for the image, the less accurate the picture will be to the original.