paras of resnet101

[xxyh1993]

The paras of resnet50 is 25.56M, but I get 23.53M here. Am I using it the wrong way?

[sovrasov]

@xxyh1993 number of params depends on number of classes you've specified when creating your model and maybe other minor differences in torch implementation. Which implementation of ResNet are you using (timm, torchvision, other?) and what is the number of classes?
The readme was created using standard imagenet classes (1k), as far as I remember.