Quantized model would be around 4x smaller and hardware support computations is typically 2-4 times faster with INT8. Quantization techniques