when i have trained the model on chinese datasets, i predictor and its character_index alaways == 4625(num__char -1),so this is why