How to count model parameters?

Uncategorized
1

how to count model parameters in pytorch, torchstat packages didn’t update for a long time, meet some error when i use it.

2

Hi,

you can count them as follows:

num_params = sum(param.numel() for param in model.parameters())

or:

num_params = sum(param.numel() for param in model.parameters() if param.requires_grad)

to only consider trainable parameters.

3

thank you!! if i want to see parameters in each layer, how to do it?

4

Maybe listing all modules in a model can be helpful if you want to see parameters in each layer:

for name, module in model.named_modules():
     print(name, sum(param.numel() for param in module.parameters()))

It will print all modules and modules’ number of parameters including activation functions or dropout.

5

You can use torchsummary library for that.

from torchsummary import summary
print(summary(model_name,  (3,224,224) , 'cpu'))