How to user quantize to imporve inference performance on tensorflow-serving?

Prior to filing: check that this should be a bug instead of a feature request. Everything supported, including the compatible versions of TensorFlow, is listed in the overview page of each technique. For example, the overview page of quantization-aware training is here. An issue for anything not supported should be a feature request.

Describe the bug
I use the "Quantization aware training in Keras example"
and save 4 models: model(model without quantization aware training), q_aware_model(model with quantization aware training), quantized_tflite_model(TFLite model with quantization aware training ), float_tflite_model(TFLite model without quantization), and depoly the models with tensorflow-serving,but I got the same performance. I dont konw the reason. How can I use the quantized model to imporve inference performance on tensorflow-serving?
System information

TensorFlow version (installed from source or binary): nightly

TensorFlow Model Optimization version (installed from source or binary): binary

Python version: python3.8