[video processor] fix slow tests (#38881)

* we need to check against mapping to be safe

* need to check only when inferring from image type, otherwise messes custom code

---------

Co-authored-by: Yih-Dar <[email protected]>

Author: zucchini-nlp

src/transformers/models/auto/video_processing_auto.py

@@ -311,7 +311,12 @@ def from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs):
         if video_processor_class is None and video_processor_auto_map is None:
             image_processor_class = config_dict.pop("image_processor_type", None)
             if image_processor_class is not None:
-                video_processor_class = image_processor_class.replace("ImageProcessor", "VideoProcessor")
+                video_processor_class_inferred = image_processor_class.replace("ImageProcessor", "VideoProcessor")
+
+                # Some models have different image processors, e.g. InternVL uses GotOCRImageProcessor
+                # We cannot use GotOCRVideoProcessor when falling back for BC and should try to infer from config later on
+                if video_processor_class_inferred in VIDEO_PROCESSOR_MAPPING_NAMES.values():
+                    video_processor_class = video_processor_class_inferred
             if "AutoImageProcessor" in config_dict.get("auto_map", {}):
                 image_processor_auto_map = config_dict["auto_map"]["AutoImageProcessor"]
                 video_processor_auto_map = image_processor_auto_map.replace("ImageProcessor", "VideoProcessor")