I haven't considered that aspect before, as I haven't utilized any models specifically for lightweight inference. While I've deployed models on CPUs before, I've typically done a normal installation, which has worked well for my needs.