I finally ran a hardware accelerated ML container for immich.

My immich is in a VPS and my GPU is in my home server. Cheaper to run that than to rent GPU compute.