https://platoblockchain.com/plato-data/fine-tune-and-deploy-llama-2-models-cost-effectively-in-amazon-sagemaker-jumpstart-with-aws-inferentia-and-aws-trainium-amazon-web-services/