model size after finetuning

[kaishijeng]

@foolwood

I noticed model size after finetuning is the same as the original model size.
Shouldn't the size be the same as the pruned model?

Thanks,

[angelamin]

I meet the same problem as yours, and I solved it by clearing the pycache, and I hope it will help you~

[MrLinNing]

I can get this, the pruned model size is 9.3M which is not 2.2M? @angelamin @kaishijeng is it same as your model?

[RichardMrLu]

emmm, after I run 'python main.py -sr --s 0.0001', my checkpoint.pth.tar is 153M...
Can you help me? Thanks.

[RichardMrLu]

And it didn't save model_best.pth.tar...

[RichardMrLu]

I check it, because of pytorch0.4, after test,type(prec1) is Tensor, so change it to type int.
But models are still 153M....

[RichardMrLu]

@MrLinNing Hello, I only change dataset's path to own Cifar10, how do you save so small model? Thanks for your help.

[RichardMrLu]

Parameters are 20M, so model size should be (20x1000000x32)/(1024x1024x8)=76.29M
soga

[Coderx7]

@RichardMrLu : how did you run the script with version 0.4? Can you share your changes?

[RichardMrLu]

@Coderx7
def test():
model.eval()
test_loss = 0
correct = 0
for data, target in test_loader:
if args.cuda:
data, target = data.cuda(), target.cuda()
data, target = Variable(data, volatile=True), Variable(target)
output = model(data)
test_loss += F.cross_entropy(output, target, size_average=False).data[0] # sum up batch loss
pred = output.data.max(1, keepdim=True)[1] # get the index of the max log-probability
correct += pred.eq(target.data.view_as(pred)).cpu().sum()

test_loss /= len(test_loader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.1f}%)\n'.format(
    test_loss, correct, len(test_loader.dataset),
    100. * correct / len(test_loader.dataset)))
return 100. * correct / len(test_loader.dataset)

def save_checkpoint(state, is_best, filename='checkpoint.pth.tar'):
torch.save(state, filename)
if is_best:
shutil.copyfile(filename, 'model_best.pth.tar')

best_prec1 = 0.
print(model)
for epoch in range(args.start_epoch, args.epochs):
if epoch in [args.epochs0.5, args.epochs0.75]:
for param_group in optimizer.param_groups:
param_group['lr'] *= 0.1
train(epoch)
prec1 = test()
is_best = int(prec1) > best_prec1
best_prec1 = max(int(prec1), int(best_prec1))
save_checkpoint({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'best_prec1': best_prec1,
'optimizer': optimizer.state_dict(),
}, is_best)

[helloelliot]

@RichardMrLu My original model is 160M . Why our models such big with only 20M parameters?