Skip to content
support float data_type when running in gpu