fix multicontrolnet `save_pretrained` logic for compatibility (#7821)

fix multicontrolnet save_pretrained logic for compatibility
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

src/diffusers/pipelines/controlnet/multicontrolnet.py

@@ -100,20 +100,16 @@ class MultiControlNetModel(ModelMixin):
             variant (`str`, *optional*):
                 If specified, weights are saved in the format pytorch_model.<variant>.bin.
         """
-        idx = 0
-        model_path_to_save = save_directory
-        for controlnet in self.nets:
+        for idx, controlnet in enumerate(self.nets):
+            suffix = "" if idx == 0 else f"_{idx}"
             controlnet.save_pretrained(
-                model_path_to_save,
+                save_directory + suffix,
                 is_main_process=is_main_process,
                 save_function=save_function,
                 safe_serialization=safe_serialization,
                 variant=variant,
             )
 
-            idx += 1
-            model_path_to_save = model_path_to_save + f"_{idx}"
-
     @classmethod
     def from_pretrained(cls, pretrained_model_path: Optional[Union[str, os.PathLike]], **kwargs):
         r"""