you can actually run the model, but it just goes very slowly… i run the 70b model on my m1 mbp, and it technically “requires” 128gb or VRAM - it still runs, just not super fast (though i’d say it’s useable in this case a about 1 word per ~300ms)