Closed SkyeFromKindergarten closed 3 years ago
The same issue, what I currently know is I have 'fatal error: ATen/ATen.h: No such file or directory' when run the chamferdist.cu independently. It means I don't have ATen though my PyTorch is 1.4.0.
I'm trying to reinstall PyTorch from source now.
Could you check torch.cuda.is_available()
first? If you're able to run pytorch code on CUDA, then a fix of this sort could help (See https://github.com/NVIDIAGameWorks/kaolin/issues/31#issuecomment-557776519)
Also, curious: what GPU archs are you using? Newer RTX 2080Ti etc? (I'm looking to update the package to suit those archs)
Same problem here. Sure my cuda works fine, and torch.cuda.is_available() returns True. I installed it with pip, but even though it's in my conda environment (conda list). I cannot import it.
I tried reinstalling this today (PyTorch 1.6) and it works fine. Closing for now, but feel free to reopen this if anything.
Exactly the same issue installing on Windows10, python 3.8, pytorch 1.6
Hi~ I find this work really will solve my problem.But I failed when installing.
When I use pip to install, it comes up like below.![image](https://user-images.githubusercontent.com/37540879/71151844-badd4900-226f-11ea-8db4-64a02b57191c.png)
When I install with source code, it comes nvcc fatal
nvcc fatal : Unsupported gpu architecture 'compute_75' error: command '/usr/local/cuda-9.0/bin/nvcc' failed with exit status 1
But Im sure my cuda works fine. Can you please tell me where it maybe wrong?System information: Ubuntu 16.04 python3.6 pytorch1.3