Model Size of Pruned model about torch-pruning HOT 3 OPEN

So you can either use torch_summary(package) if you want all the parameter details and model size which outputs something like this

if you only want size you can do something like this

def print_size_of_model(model):
    torch.save(model.state_dict(), "temp.p")
    print('Size (MB):', os.path.getsize("temp.p")/1e6)
    os.remove('temp.p')
print_size_of_model(model)

from torch-pruning.

By using the print_size_of_model method, the pruned model size shown did'nt reduce because I think this is cause by even the value is prune （turn to 0），but is still store with fp32. So I wonder how did you get your pruned model size?

from torch-pruning.

Yeah so you're using Pytorch's pruning which is unstructured . what unstructured pruning does is that it will put zeros in your weight matrix but it won't remove them. In essence you have a sparse matrix but you're still saving the zeros(those floating point fp32's). If you want model size reduction use this library(torch_pruning), which uses structured pruning or you might want to look into quantization

from torch-pruning.