Pytorch 1.9 with CUDA 11.0?

3
  1. Yes, you should be able to use a CUDA10.2 docker container, as its driver requirement would be met by the newer CUDA11 driver.
  2. I don’t know what a Google machine image is and how hard it would be to build it.
  3. You could reuse the Dockerfile from the PyTorch repository.

Also, did you try to install the CUDA11.1 Pytorch 1.9.0 binaries? I don’t know which driver is installed on the node, but the CUDA enhanced compatibility might be used.