Start with the workload

Run open models through an API, choose Compute when you need control over the stack, or tune an open model on your task.