Zero_grad() gives None gradient

1

I’m trying to implement this tutorial http://pytorch.org/tutorials/beginner/blitz/neural_networks_tutorial.html

when I arrived to this point and print the output it gave me None, I couldn’t find a similar problem on the internet that explains why.
net.zero_grad() # zeroes the gradient buffers of all parameters

print('conv1.bias.grad before backward')
print(net.conv1.bias.grad)

but after applying backward the results are good
loss.backward()

print('conv1.bias.grad after backward')
print(net.conv1.bias.grad)
2

The .grad field is None by default and corresponds to being full of 0.
So until you actually compute any gradient for this Variable, this field will remain None.

2 Likes
3

thanks for clearing this up. it is just the tutorial said I should get zero when I zero the gradient.