torch.cuda.is_available() returns false

Hi. After two days installing every nvidia driver, pytorch (from package and compiling), cuda toolkit version, etc, torch.cuda.is_available() returns false

I’d appreciate very much any help with this issue.

I have 2 GTX 1070 GPUs. Linuxmint 18.3.

Thanks.

It finally works. I don’t really know why.

I executed: sudo systemctl isolate multi-user.target
I had to restart and suddenly it’s working.

Extremely painful this cuda thing.