While using torch.norm
my code is giving nans. So, I tried changing the dtype
argument as mentioned here torch.norm — PyTorch 1.7.0 documentation to dtype=torch.float64
.
But it is giving the following error:
ValueError: dtype argument is not supported in frobenius norm
.
As mentioned as WARNING on this page as torch.norm — PyTorch 1.7.0 documentation torch.norm is deprecated and maybe removed in future release.
But the following snippet is also giving as ImportError : cannot import name 'linalg'
from torch import linalg
Please look into the issue.
Thanks.
Which torch version are you using?
linalg
is introduced in pytorch 1.7.0 I think.
Also, can you give a short snippet of code to reproduce the issue?
I’m using torch 1.3.1 version
Even I am facing the same issue.
from torch import linalg
gives me the error
from torch import linalg
---------------------------------------------------------------------------
ImportError Traceback (most recent call last)
<ipython-input-11-fe7ec55944c3> in <module>
----> 1 from torch import linalg
ImportError: cannot import name 'linalg' from 'torch' (/home/nithin/anaconda3/envs/Pytorch_cuda_env/lib/python3.7/site-packages/torch/__init__.py)
Which PyTorch version are you using?
As given by @InnovArul it seems that this namespace was introduced in PyTorch>=1.7
.
Oh. Okay.
I’m using 1.3.1.
Thanks