Torch.cat() returning None. (Solved by pytorch reinstallation)

vision
1

I’ve been using torch.cat() for my unet implementation, but in pytorch 0.4.0 it has begun to return None and is defined as such:

def cat(param, dim):
return None

in torch’s init.py for me. How would I go about fixing this? I tried to use torch.stack() but it expects tensors to have the same shape. I’m getting this error:

RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1. Got 512 and 256 in dimension 2 at c:\programdata\miniconda3\conda-bld\pytorch_1524549877902\work\aten\src\th\generic/THTensorMath.c:3586

from this line:

x = torch.stack((x, prev_feature_map), dim=1)

Hope to hear from you soon! And thank you in advance!

2

Could you tell me, where you have found the init.py defining cat?
Also, could you try to run the following code:

a = torch.randn(10, 2)
b = torch.randn(2, 2)

c = torch.cat((a, b), dim=0)

and post the result?

1 Like
3

I’m getting this back:
image

Thank you for reply, ptrblck!
Oh, and sorry about the delay. I was out helping my father cut down some trees for the garage he’s building.
I’m gonna try reinstalling pytorch. Will be back with results.

4

Reinstalling pytorch seems to have fixed the issue. Sorry about creating the thread! I was just really confused.

5

No worries. It was a weird bug.
Did you build PyTorch from source or installed the binaries?

6

Sure was… Anyway, thank you for your quick reply! Thought I might’ve updated the library and that they had renamed it or something first. I use conda to install it. Bit easier since I’m on windows 10 at the moment.