Deleted functionality like reuse_model_from_already_loaded, load_model

[vaskers5]

For my purpose - I need support for fast checkpoing changing via API. But this functionality is basicly deleted:

Here is original implementation from a1111.

Currently it's just unload all models(code is here), without keeping them in RAM or VRAM, or somewhere for fast checkpoint changing. I think we can add new argument, like --enable_checkpointing_in_vram and guys who needs this functionality can just activate that

[fishmongr]

Ran into this recently too. Isn't this pretty critical to making switching models happen not painfully slow? Or am I missing something.