GPU usage from the container

Dear organizers,

What should we include in our process.py script in order to enable the GPU from the virtual machine? Doing torch.device("cuda:0") does not work, it says there is no such device.

Thanks

Hi,

Simply specify when you register the module that it requires GPU. Hope we have been able to help you.

The container is set to require GPU

Hi,

The jobs are run with a GPU attached but seems that your code is not detecting it. You should check that the versions of Pytourch, CUDA etc are up to date. Here is the info on the current driver versions used by SageMaker https://grand-challenge.org/forums/forum/mediastinal-lymph-node-quantification-lnq-segmentation-690/topic/docker-submission-nodes-1854/

Hope this was helpful for you.

Probably you (as the host) provides us a simple docker solution (with same sample random prediction code) that supports pytorch and gpu?

Hello, in case it helps someone: it seems to be working with PyTorch 1.13.1 and CUDA 11.7

"Hello, in case it helps someone: it seems to be working with PyTorch 1.13.1 and CUDA 11.7".

Many thanks. I finally use the big tank - Kaggle docker. However the docker image is quite big.