I am now trying to change the cross-scan mode of the author's pytorch version, instead of scanning the previous 4 kinds of scans only once, and then executing the code, I find that the model parameter number in the log only changed from 30.24M to 29.08M, and the GFLPs changed from 4.906 to 4.522, is this normal? Or do I need to rewrite the script to calculate the number of parameters and Flops?