Hi guys. Does anyone know if it is possible to use adaptive batching in a pytorch model but considering two inputs that would be "input_ids" and "attention_mask"??