The sample sounds impressive, but based on their claim -- 'Streaming inference is faster than playback even on an A100 40GB for the 3 billion parameter model' -- I don't think this could run on a standard laptop.