problem:RuntimeError: 1only batches of spatial targets supported (non-empty 3D tensors) but got targets of size: : [20]

(1)when i run the architecture squeezenet1_0 and squeezenet1_1 ,I met the problem above.
(2)then I print the shape of output and target as below:
output.shape torch.Size([20, 1000, 1, 1])
target.shape torch.Size([20])
obviously, it' s the reason that the shape is not matched.
(3)so i add the code:
output = model(input)
output =output.reshape(-1,1000)
and it worked
(4)but there maybe remains some better solution,such as modify the code of squeezenet1_0 and squeezenet1_1(ps: because other architetures don't have the problems above)

Start by reproducing the reported error with squeezenet1_0 and squeezenet1_1, then inspect their output shapes alongside the target shape shown in the issue. Compare these architectures with the others that do not fail; done means the reported training path accepts the model output without requiring an ad hoc reshape.

Written by the indexing model from the issue text.