Hi, I am training a NN with PyTorch and I want to check whether the gradients of the loss total_loss
(on which I call total_loss.backward()
), are constant with respect to the weights contained in an nn.Sequential
block, call it block
.
How can I easily achieve this? Thanks!
You can iterate over the model parameters:
for name,param in model.named_parameters()
and then inspect param.grad
Which should be either a torch.Tensor or None (if your network is breaking backprop due to any bug).