added text inputs to enable customization for prompting

handler.py

@@ -192,8 +192,11 @@ class EndpointHandler():
 
         # This might give different results from the original argument `image_input`
         image_input = Image.open(user_image_path)
-        text_input = "<grounding>Describe this image in detail:"
-        #text_input = f"<grounding>{text_input}"
+        text_input = data.pop("inputs", None)
+        if not text_input:
+            text_input = "<grounding>Describe this image in detail:"
+        else:
+            text_input = f"<grounding>{text_input}"
 
         inputs = self.processor(text=text_input, images=image_input, return_tensors="pt")