[PyTorch1.13] [CUDA 11.7] Could not run 'aten::empty_strided' with arguments from the 'CUDA' backend

1

Hi All,

I have a issu to execute an inférence on torch under Cuda 11.7.
Previously I use PyTorch with cuda 10.2 on GPU on Turing and Pascal Nvidia Card.
To use PyTorch endvironnment in Ampere génération, I must install Cuda 11.7

To Load the model :
torch:jit:Load('yolo.pt")

with the error : Could not run ‘aten::empty_strided’ with arguments from the ‘CUDA’ backend.

Could you please give a piece of advise ?

3080 TI
Visual Studio 2017
Cuda 11.7
Pytorch 1.13

2

Could you post a minimal and executable code snippet to reproduce the issue, please?

3

Hi All,

I have the following error with the current code

It’s located on load of my classifier…

Thanks in adavance.

Ben

code
code1154×1081 54.8 KB