hvd.size() gave 1 even if I use mutiple GPUs instance

Hi,

I am running example [tensorflow_keras_CIFAR10.ipynb](https://github.com/aws/amazon-sagemaker-examples/blob/master/sagemaker-python-sdk/keras_script_mode_pipe_mode_horovod/tensorflow_keras_CIFAR10.ipynb).
I set the distributions to enable mpi and use ml.p3.8xlarge instance.
But hvd.size() always return 1 for me.

Start with tensorflow_keras_CIFAR10.ipynb and inspect the SageMaker distribution settings for MPI and the ml.p3.8xlarge instance. Run the example while checking the Horovod size reported by the training process; done means the example documents or produces the expected multi-GPU world size instead of 1.

Written by the indexing model from the issue text.