A Better Way to Run Fine-Tuned LLMs on AWS