advice: in lm stidio, model parameters set context overflow as rolling window, so ai will forget older data, and not data in the middle as on truncate middle, context of model set to 8092 at least
