Attempt a CPU-only server instance

[jeffbl]

A CPU-only server instance would be much less expensive than a GPU instance on EC2, or buying an expensive video card.

For the LLM, we're currently using the OpenAPI format, so it can easily be run remotely.

However, there are a number of models we currently run locally, including:
espnet TTS English
espnet TTS French
semseg (soon to be SAM, which likely has a remote endpoint)
YOLO (being replaced by LLM)

In theory, all of these can be run on CPU, but performance and memory may be bottlenecks.

Goal is to try this and see where bottlenecks/issues are, on path to having a realistic CPU-only EC2 instance (cheap!) and then only have to pay for API usage.