That's an interesting question. LLMs perform system approximation, so two different systems might create a problem.. You may want to create a AutoGPT agent that has two transformers and chooses the right one as required.