Yes. Starting with the Facebook versions of LLaMA-7B you just quantize the model to 4bit on your desktop (since it takes 14GB of RAM) and then move it to your p

Yes. Starting with the Facebook versions of LLaMA-7B you just quantize the model to 4bit on your desktop (since it takes 14GB of RAM) and then move it to your phone and follow the Android instructions in the repo. https://github.com/ggerganov/llama.cpp/#android https://github.com/ggerganov/llama.cpp/#android I've seen dozens of screenshots of it running in termux on androids by now at completely usable speeds.