AttributeError: module 'torch' has no attribute '__version__'

[kaisark]

I was able to build pytorch using your instruction set, but when testing the installation, I ran into a missing attribute error (see below):

Nvidia TX1 Ubuntu 16 Python 3.5 Cmake 3.7.2

torch==0.1.10+ac9245a torchvision==0.2.2.post3

(cv40py35) nvidia@tegra-ubuntu:\~$ which cmake /usr/local/bin/cmake (cv40py35) nvidia@tegra-ubuntu:\~$ cmake --version cmake version 3.7.2 Cuda 8 Cudnn 5

(cv40py35) nvidia@tegra-ubuntu:\~$ python3 -c 'import sys; print("\n".join(sys.path))'

/home/nvidia/.virtualenvs/cv40py35/lib/python35.zip /home/nvidia/.virtualenvs/cv40py35/lib/python3.5 /home/nvidia/.virtualenvs/cv40py35/lib/python3.5/plat-aarch64-linux-gnu /home/nvidia/.virtualenvs/cv40py35/lib/python3.5/lib-dynload /usr/lib/python3.5 /usr/lib/python3.5/plat-aarch64-linux-gnu /home/nvidia/.virtualenvs/cv40py35/lib/python3.5/site-packages

---

(cv40py35) nvidia@tegra-ubuntu:/~$ python Python 3.5.2 (default, Nov 23 2017, 16:37:01) [GCC 5.4.0 20160609] on linux Type "help", "copyright", "credits" or "license" for more information.

import torch print(torch.version) Traceback (most recent call last): File "", line 1, in AttributeError: module 'torch' has no attribute 'version'

pytorch-install.log

[kaisark]

I was able to reinstall (build from source) and resolve the CUDNN path resolution issues.

Cuda 8 Cudnn 6.0.21

Closing the issue.