Towards Next-Level Post-Training Quantization of Hyper-Scale Transformers