Different cuda for different models

[Ferniclestix]

Would love an option to define which cuda is running which model/chkpoint loader, that way we could actually make use of multi gpu for additional models so we dont hit so many memory blocks when running SDXL style setups with multiple models.