[MattPalmer1086]

The software that powers LLM inference is very small, and is the same no matter what task you ask it to perform. LLMs are really the neural architecture and model weights used.