Text this: Smoothed per-tensor weight quantization: a robust solution for neural network deployment