Add an /unload endpoint.

[FGDumitru]

I think a specific unloading endpoint would be beneficial when one does not want to wait for the TTL to expire or kill the entire llama-swap process. This would beneficial for pipelines that have multiple processes using VRAM without requiring the use to forcefully kill a specific process.

Right now I'm emulating it with a 'null' model:

"unload": cmd: ls ttl: 1

which I call by '/upstream/unload/' (a little barbaric but I couldn't find any other way of cleanly unload a model before it's expiration time).

Otherwise, great project and thank you for publishing it.

[mostlygeek]

Thanks for the suggestion. I added the /unload endpoint that will unload all loaded models. I'm always surprised at unintended and interesting use cases for llama-swap :)

Release in v91

[FGDumitru]

That was fast! Thank you and have a great day going forward!