Adaptive Layer-Wise Transformations for Post-Training Quantization of Large Language Models