Input data range after quantization

I have a trained float32 model with input data range from 0-1. I want to do a quantization aware training. My question is for the quantized model, the input should be still 0-1 or need to scale to 0-255? should I retrain the float32 model with input 0-255 before the qat? Thanks.

you don’t need to change the input, it should still be 0-1