Update handler.py

handler.py

@@ -1,5 +1,6 @@
 from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
-
+from typing import Dict, List, Any
+import torch
 
 class EndpointHandler:
     def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]):
@@ -24,4 +25,4 @@ class EndpointHandler:
 
         generated_ids = model.generate(**inputs, max_new_tokens=128)
         output_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
-        return output_text[0])
+        return output_text[0]