I’m trying to get pytorch using the anaconda package manager. I have an Ubuntu 16.04 machine. My CUDA is installed /usr/local/cuda-8.0. Im running the proper command found here. I’ve tried setting CUDA_HOME and LD_LIBRARY_PATH variables. Upon installation, torch.cuda.is_available() returns False.
I also tried compiling from source in which case the log indicates that it found CUDA at the proper location. However, torch.cuda.is_available() still returns False.
Tensorflow is able to access the GPU.
Any help would be appreciated.