Is (Selective) Round-to-Nearest Quantization All You Need?