when i use the cswin as backbone, but the input resolution is different with 224, can i change the width from 1, 2, 7, 7 to 1, 2, 8, 8?

the pretrained model's width is 1, 2, 7, 7, if i change to 1, 2, 8 , 8 in downstream task, does it reduce the performance?
thanks!~~@LightDXY