It's 1 to 3 screen frames. Remember that the phase of the autorepeat is not aligned with anything. ASCII timing diagram showing one possibility:

  Key   : !___!___!___!___!__
  Term  : _!_!___!_!___!_!___
  Screen: _!_!_!_!_!_!_!_!_!_

First key is shown for 3 screen frames, second for 1, third for 2, then the cycle repeats.