It is very different than sagemaker SDK - BentoML is a flexible and all-in-one solution for model serving. You package the model once and can easily test it, do batch/offline serving and online API serving. And even if you just package your model with BentoML and deploy it to Sagemaker, you get a 10-20x performance improvement out-of-the-box comparing to doing it with the sagemaker sdk.