Benchmarking the Robustness of Quantized Models