error: ‘class at::Context’ has no member named ‘getCurrentCUDAStream…

[jiapei100]

No description provided.

[YexingWan]

your commits work for me, thank you

[clarle]

For folks trying to get vid2vid running on PyTorch 1.0.0, you need to make the changes in this pull request.

Thanks @jiapei100!

[MoonBlvd]

Thank you for your pull request! I'm using pytorch 1.0.0 and CUDA 10. After merging this pull request, I was able to install the three customized layers (correlation, resample2d and channelnorm). However I had to use sudo /path/to/virtualenv/python setup.py install since I was facing permission denied issue in virtualenv. After those, when I try python main.py in my virtual env, it gives errors: resample2d_cuda.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN2at18SparseCUDATensorIdEv. It seems that the compiled binary is till not usable. Could you please help?

[fabiopk]

I am having the exact same problem as @MoonBlvd , with the resample2d_cuda.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN2at18SparseCUDATensorIdEv.

Could anyone run this with CUDA 10?

[awaelchli]

Why do I get a different message?:
undefined symbol _ZN2at19UndefinedTensorImpl10_singletonE

[Bingmang]

No!!!
It does not work actually!
The Error was caused by PyTorch version, the code should never be running in PyTorch 1.0.0.
And these codes will cause the undefined symbol problem in master branch using PyTorch 0.4.1.
So please revert the commit...

[huangbiubiu]

Same problem as @MoonBlvd. Is there any workaround?

undefined symbol: _ZN3c105ErrorC1ENS_14SourceLocationERKSs