NeurIPS2022AlgoZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers