Cifar10 tutorial accuracy is different at each run

I am following official tutorial https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html “Training a classifier”.
I used google colab to run the code. Each time I click run all cells, I get very different accuracy.
Like 1 time I got:
Accuracy of plane : 53 %
Accuracy of car : 68 %
Accuracy of bird : 37 %
Accuracy of cat : 56 %
Accuracy of deer : 37 %
Accuracy of dog : 41 %
Accuracy of frog : 63 %
Accuracy of horse : 53 %
Accuracy of ship : 74 %
Accuracy of truck : 68 %

and another time:
Accuracy of plane : 54 %
Accuracy of car : 62 %
Accuracy of bird : 49 %
Accuracy of cat : 30 %
Accuracy of deer : 34 %
Accuracy of dog : 36 %
Accuracy of frog : 71 %
Accuracy of horse : 63 %
Accuracy of ship : 82 %
Accuracy of truck : 70 %

I understand that seed is not used, but why we are getting so much different accuracy?. I mean what part of code is responsible for such a large variation and why?
Thanks a lot :slight_smile:

help please :slightly_smiling_face: