https://platoaistream.com/plato-data/run-multiple-generative-ai-models-on-gpu-using-amazon-sagemaker-multi-model-endpoints-with-torchserve-and-save-up-to-75-in-inference-costs-amazon-web-services/