How is running a LLM related to stable diffusion? Your contention is that models will run on devices; but latent diffusion models have lower memory footprints

How is running a LLM related to stable diffusion? Your contention is that models will run on devices; but latent diffusion models have lower memory footprints (see: latent). The hardware you need to run a good LLM is what, 10x more than a latent diffusion one? They are not comparable.