(Resolved)Libtorch1.5_cu92 win7: torch.cat RuntimeError: error in LoadLibraryA

1

Win7 64bit
CUDA9.2
Libtorch1.5+cu92

when I use “module.forward(inputs).toTensor()”, crashed.
I print out the exception, something like it below:
"
input = torch.cat([argument_2, x1], 1)
~~~~~~~~~ <— HERE
return (_17).forward(input, )
…
…
…
RuntimeError: error in LoadLibraryA
"

But I can run cpu traced model well.

2

This error should be fixed in PyTorch 1.5.0, if I’m not mistaken.
If you are using an older version, you could update or use this workaround.

3

Someone else found out that it could also be caused by the antivirus. See https://blog.csdn.net/weixin_38314865/article/details/103840899.

4

I double checked that, I use libtorch1.5, and I use pytorch 1.5.0 to trace and saved my model.
Today I downloaded the latest nightly build libtorch version, has the same issue.

5

I tried it, not work for me. Pytorch is OK, no issue, Libtorch has this situation.

6

For libtorch, just copy all the DLLs to the location of your executable.

7

I will try it. thanks

8

Fixed, thank you very much.