Improving Quantization with Post-Training Model Expansion