Does this mean that with big enough compute capacity - say, Petals https://github.com/bigscience-workshop/petals which distributes the model over the internet over GPUs - we can run LLAMA?