Any way to pre-load the model before running llama-server?

[samsymbia]

I would like a way to pre-load the model before running the llama-server. Today I start the server, poll the /health endpoint in a loop until I get a 200 OK response, but that not very effective for the way I use it within a farm of servers. I also want to be able to snapshot the memory once everything is ready but before the first inference is processed. Having a way to do it in two manual steps would be great for my use-case.

[samteezy]

llama-swap has functionality that will do this for you.