Interactions Across Blocks in Post-Training Quantization of Large Language Models