I am training and testing an autoencoder. I run the exact same code in Jupyter notebooks and also in Googlecolab. I have these settings:
np.random.seed(0)
random.seed(0)
torch.manual_seed(0)
torch.cuda.manual_seed(0)
torch.cuda.manual_seed_all(0)
torch.backends.cudnn.enabled = False
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
and
train_loader = torch.utils.data.DataLoader(
dataset=train_set,
batch_size=batch_size,
shuffle=False,
num_workers=2,
worker_init_fn=np.random.seed(0))
test_loader = torch.utils.data.DataLoader(
dataset=test_set,
batch_size=batch_size,
shuffle=False,
num_workers=2,
worker_init_fn=np.random.seed(0))
Also, I am using GPU on both environments.
The performance in each environment is deterministic but they don’t produce the exact same accuracy: Colab=0.7068, JupyterNotebook=0.6689
I wonder why they behave like this? shouldn’t they produce the exact same values?
I appreciate any help and guidance.