4 ms·

Depends on your m1 specs, but should definitely be able to run a 7b model (at least with some quantization).