Adopting exponential moving average (EMA) for PL pipeline

Hi, why not at the end of training_step?