I see. Thanks for the info. Now I know where those models (with missing inverse bind matrices) are coming from.
