Update handler.py

handler.py

@@ -89,19 +89,18 @@ class EndpointHandler:
         if clip is None or prompt is None:
             return [{"error": "Missing 'clip' or 'prompt' in input data"}]
 
-        # Prepare the inputs for the model
-        inputs_video = self.processor(
-            text=prompt, 
-            videos=clip, 
-            padding=True, 
-            return_tensors="pt"
-        ).to(self.model.device)
+        # Ensure clip_bytes is converted properly to the expected format by the model
+        inputs_video = ml.processor(text=prompt, videos=clip, padding=True, return_tensors="pt").to(ml.model.device)
         
-        # Debug: Check specific components
+        # Debug: Print the entire inputs_video structure
+        print(f"Keys in inputs_video: {inputs_video.keys()}")
+
+        # Rename pixel_values_videos to pixel_values if it exists
         if 'pixel_values_videos' in inputs_video:
-            print(f"pixel_values_videos shape: {inputs_video['pixel_values_videos'].shape}")
+            inputs_video['pixel_values'] = inputs_video.pop('pixel_values_videos')
+            print(f"Renamed pixel_values_videos to pixel_values. New shape: {inputs_video['pixel_values'].shape}")
         else:
-            print("pixel_values not found in inputs_video")
+            print("pixel_values_videos not found in inputs_video")
 
         if 'input_ids' in inputs_video:
             print(f"input_ids shape: {inputs_video['input_ids'].shape}")