RuntimeError: dim() called on undefined Tensor

[LukeB42]

The most recent PyTorch build requires the dim argument to be specified when using torch.nn.functional.softmax. I've specified the 0th dimension in both calls to torch.nn.functional.softmax within pytorch_model.py and get this error:

bash-4.3# python pytorch_run.py --nogpu --start
Converting data to one-hot representation
Data Loaded
Dim Training Data (11258, 1995)
Dim Test Data (7487, 1995)
Traceback (most recent call last):
  File "pytorch_run.py", line 153, in <module>
    train()
  File "pytorch_run.py", line 93, in train
    loss.backward()             # backprop
  File "/usr/lib/python2.7/site-packages/torch/autograd/variable.py", line 128, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph, retain_variables)
  File "/usr/lib/python2.7/site-packages/torch/autograd/__init__.py", line 98, in backward
    variables, grad_variables, retain_graph)
RuntimeError: dim() called on undefined Tensor

Any help would be appreciated. You may also want to update those two calls to softmax.

Regards, Luke

[bhomass]

is there any resolution to this? facing the same problem. it would really help if there is some documentation on which version of pytorch the code does work with.

[hyqneuron]

It was written before pytorch 1.0. Since then pytorch upgraded its tensor interface.

[bhomass]

I am using torch 0.4.0, and its failing. Maybe 0.3.0? if you have working code, please provide that version.

[hyqneuron]

I think it was for 0.1.10 or around that.

Yunqing

On 14 Jul 2018, at 10:08 AM, Bruce Ho [email protected] wrote:

I am using torch 0.4.0, and its failing. Maybe 0.3.0? if you have working code, please provide that version.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[bhomass]

finally! 0.1.12 worked. Thank you very much.

[erlebach]

I thought the softmax was done on dim=1. The 0th dimension is the batch dimension.