Fix Qwen2.5-VL Video Processor (#38366)

* Update processing_qwen2_5_vl.py

* Update processing_qwen2_5_vl.py

* Update modular_qwen2_5_vl.py

* Fix CI

* Update modular_qwen2_5_vl.py

* Update processing_qwen2_5_vl.py

* Update video_processing_utils.py

src/transformers/models/qwen2_5_vl/modular_qwen2_5_vl.py

@@ -1013,10 +1013,12 @@ class Qwen2_5_VLProcessor(Qwen2VLProcessor):
             image_grid_thw = image_inputs["image_grid_thw"]
 
         if videos is not None:
+            # pop fps in advance for passing kwargs validation
+            fps = output_kwargs["videos_kwargs"].pop("fps", 2.0)
+
             videos_inputs = self.video_processor(videos=videos, **output_kwargs["videos_kwargs"])
             video_grid_thw = videos_inputs["video_grid_thw"]
 
-            fps = output_kwargs["videos_kwargs"].pop("fps", 2.0)
             if isinstance(fps, (int, float)):
                 second_per_grid_ts = [self.video_processor.temporal_patch_size / fps] * len(video_grid_thw)
             elif hasattr(fps, "__len__") and len(fps) == len(video_grid_thw):

src/transformers/models/qwen2_5_vl/processing_qwen2_5_vl.py

@@ -152,10 +152,12 @@ class Qwen2_5_VLProcessor(ProcessorMixin):
             image_grid_thw = image_inputs["image_grid_thw"]
 
         if videos is not None:
+            # pop fps in advance for passing kwargs validation
+            fps = output_kwargs["videos_kwargs"].pop("fps", 2.0)
+
             videos_inputs = self.video_processor(videos=videos, **output_kwargs["videos_kwargs"])
             video_grid_thw = videos_inputs["video_grid_thw"]
 
-            fps = output_kwargs["videos_kwargs"].pop("fps", 2.0)
             if isinstance(fps, (int, float)):
                 second_per_grid_ts = [self.video_processor.temporal_patch_size / fps] * len(video_grid_thw)
             elif hasattr(fps, "__len__") and len(fps) == len(video_grid_thw):

src/transformers/video_processing_utils.py

@@ -250,7 +250,10 @@ class BaseVideoProcessor(BaseImageProcessorFast):
         videos: VideoInput,
         **kwargs: Unpack[VideosKwargs],
     ) -> BatchFeature:
-        validate_kwargs(captured_kwargs=kwargs.keys(), valid_processor_keys=self.valid_kwargs.__annotations__.keys())
+        validate_kwargs(
+            captured_kwargs=kwargs.keys(),
+            valid_processor_keys=list(self.valid_kwargs.__annotations__.keys()) + ["return_tensors"],
+        )
         # Set default kwargs from self. This ensures that if a kwarg is not provided
         # by the user, it gets its default value from the instance, or is set to None.
         for kwarg_name in self.valid_kwargs.__annotations__: