OP is probably using an AI accelerator like this: https://coral.ai/products/accelerator which works great on a PI and uses very little power. It will do the Yolo part, but you can't really expect it to do the multimodal LLM part, although you could try to run Florence directly on the PI too.