Evaluate model with half precision while training in mixed precision

If I train the model using mixed precision Trainer(precision=16) and I want to evaluate the model using half precision (model.half()), do I have to set to model back to model.float() or just leave it in model.half() after evaluation process?

def on_validation_epoch_start(self):
    if self.precision == 16:

def on_validation_epoch_end(self):
    if self.precision == 16: