Create model always use estimator's entrypoint

[verdimrc]

Hi, I split train & inference logics into two separate entrypoint files. However, when create a model, the model always carries-over the train entrypoint, instead of using what's specified in create_model(entry_point='inference.py').

Kindly see the behavior shown in the screenshot, which was taken from a SageMaker notebook instance running in us-east-1.