CUDA not found in docker image

[mlong93]

I pulled the official docker image but after installing all the dependencies, I'm getting torch.cuda.is_available() = False.

Do you have further instructions to use the Docker image?

[XZYW]

I encountered the same problem and the command nvidia-smi didn’t work in docker. I solved this by installing the NVIDIA Container Toolkit and adding the parameter --gpus all when running docker. If you haven’t installed it, you can refer to this issue, and the installation guide