Overcoming the obstacles of fitting the vision-language model for the generalization of OOD 04/27/2024