Post-Training Quantization with Low-precision Minifloats and Integers on FPGAs