>>105975322
Presumably, whatever these backends support. The question is, why use that instead of the specific backends directly?
I know llama.cpp supports gemma just fine.