LabelSmoothing running error about annotated-transformer

[liquor233]

I encountered the same error, have u fixed it yet?

[Unilmalas]

apparently the problem lies in the dimension check of the mask in class LabelSmoothing(nn.Module):
mask.dim() returns 1 for empty tensors; for a quick fix change if mask.dim() > 0: to if mask.dim() > 1:

Edit: while I am at it: to fix the deprecation warning change
self.criterion = nn.KLDivLoss(size_average=False)
to
self.criterion = nn.KLDivLoss(reduction = 'none')

the plot then throws an exception, change
plt.plot(np.arange(1, 100), [loss(x) for x in range(1, 100)])
to
plt.plot(np.arange(1, 100), [torch.mean(loss(x)) for x in range(1, 100)])
to fix that

[Unilmalas]

fixed some more deprecated errors etc, working up to copy task,
The Annotated Transformer.zip

also fixed the GPU issues:
The Annotated Transformer.zip

also attention vis fixed:
The Annotated TransformerV02.zip