Loading a model, getting different results

I saved and loaded a model using

torch.save({'state_dict': model.state_dict()}, "classifier.pth")

state_dict = torch.load('classifier.pth')

I was getting 93% accuracy in a classification on the validation set which got reduced to 76% after I restarted the kernel… Am I making some mistake…Please help

That shouldn’t be. Are you sure that you are feeding the same dataset?

Yes I didnt change anything , just restarted the kernel…
I am adding some screenshots of new results…
Here I loaded the model

Here I restarted the kernel and loaded the model

I see it’s an old discussion but I’d recommend setting seed value

I saved it this way, so was working fine

torch.save(model, ‘filename.pt’)

1 Like