TypeError: type Tensor doesn't define __round__ method

[Deep1994]

Hi, I found an error when run your code, the error log is:

TRAINING : Epoch 1 Learning rate : 0.1 train_nli.py:245: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number all_costs.append(loss.data[0]) Traceback (most recent call last): File "train_nli.py", line 351, in train_acc = trainepoch(epoch) File "train_nli.py", line 276, in trainepoch round(100.*correct/(stidx+k), 2))) TypeError: type Tensor doesn't define round method

I change round(100.correct/(stidx+k), 2))) to round(100.correct.item()/(stidx+k), 2))), it can run, but the loss cannot update, and always be a constant, do you know why this happen? Thanks!

[fooSynaptic]

i encountered sample exception, how you solved it?

[Punchwes]

same here, changed to item() but the loss not updated anymore.

[jakeandersonbell]

I was receiving this issue when using the built-in round() function rather than the torch.round() function. The built-in function will not work with tensors.