How to fix the training inputs same when running the 'mnist' example on gpu?

1

I want to fixed all input the same when I run the code ‘example/mnist/main.py’ next time.
I fix np.random seed, torch cpu cuda random seed, and use shuffle=False for DataLoader,but the test result is different when I run the code next time with cuda,but normal on cpu,is there something to be set?

1 Like