Abstract: By quantizing weights with different precision for different parts of a network, mixed-precision quantization promises to reduce the hardware cost and improve the speed of deep neural ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback