CUDA Device Assert Trigger when moving tensor from CPU to GPU

nlp
1

Hi,

I tried to move a tensor from cpu to gpu by doing:

>>> t = torch.tensor(1)
>>> t.cuda()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
RuntimeError: CUDA error: device-side assert triggered

All cpu tensors can’t be moved to GPU. My pytorch cuda is 10.2, pytorch version 1.5.1. Nvidia-SMI info as follows:

NVIDIA-SMI 450.102.04   Driver Version: 450.102.04   CUDA Version: 11.0

Could someone help me out?

2

This error points to an installation (or unlikely a hardware) issue.
Did you restart your workstation after installing or upgrading some drivers?
If so, make sure you are able to run some CUDA samples.

3

Thanks for your reply. Reinstalling pytorch and cudatoolkit solved everything.