Abstracting Ai Evaluation Environments With Virtual Machines