Update handler.py

handler.py

@@ -1,9 +1,8 @@
 import json
 import torch
-from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, pipeline
+from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, pipeline, PreTrainedImageProcessor
 from qwen_vl_utils import process_vision_info
 
-
 class EndpointHandler:
     def __init__(self, model_dir):
         # Configure device settings
@@ -28,10 +27,11 @@ class EndpointHandler:
             print(f"Error loading processor: {e}")
             raise
 
-        # Define a VQA pipeline
+        # Define a VQA pipeline with explicitly provided processor
         self.vqa_pipeline = pipeline(
             task="visual-question-answering",
             model=self.model,
+            image_processor=self.processor, # Explicitly pass the image processor
             device=0 if torch.cuda.is_available() else -1  # Use first GPU or CPU
         )
 
@@ -87,34 +87,4 @@ class EndpointHandler:
         except Exception as e:
             error_message = f"Error: {str(e)}"
             print(error_message)
-            return json.dumps({"error": error_message})
-
-
-# Example instantiation and call
-if __name__ == "__main__":
-    # Assuming model directory is set in your context/environment
-    model_directory = "your_model_directory"
-    handler = EndpointHandler(model_directory)
-
-    # Sample request data
-    sample_request = json.dumps({
-        "messages": [
-            {
-                "role": "user",
-                "content": [
-                    {
-                        "type": "image",
-                        "image": "path_or_url_to_your_image"
-                    },
-                    {
-                        "type": "text",
-                        "text": "Describe what you see in the image."
-                    }
-                ]
-            }
-        ]
-    })
-
-    # Process the request
-    output = handler(sample_request)
-    print(output)
+            return json.dumps({"error": error_message})