I have pulled pytorch/pytroch:1.0.1-cuda10.0-cudnn7-devel image from docker hub.
but when I create one container by this images, I found that torch.cuda.is_available() is always False.
this is the command I used to create a container:
NV_GPU=0 nvidia-docker run -it --rm pytorch/pytorch:1.0.1-cuda10.0-cudnn7-devel
And I can find Nvidia device in /dev/ and cuda files in /usr/local/