I built both the pytorch-1.2.0-gpu-py3 and pytorch-1.1.0-gpu-py3 images and tried to deploy a model in local mode using PytorchModel. I always get the error:
ModuleNotFoundError: No module named 'sagemaker_pytorch_container.serving'
Which makes sense, since when looking both inside the docker image and inside the src/sagemaker_pytorch_container, I don't see any serving.py script, only training.py.
Did I miss something ? How can I build a pytorch image able to do both training and serving from this repo ?