Update app.py

app.py

@@ -5,6 +5,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 import cv2
 import numpy as np
 import ast
+from collections import Counter
 
 
 # # Ensure GPU usage if available
@@ -98,10 +99,23 @@ def predict_answer(video, image, question):
             
             answer = tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip()
             answers.append(answer)
-        return ast.literal_eval(answers[0])
+
+        # Modify this logic based on your specific needs
+        most_common_answer = Counter(answers).most_common(1)[0][0]
+        
+        # Safely evaluate the most common answer assuming it's a string representation of a Python literal
+        try:
+            evaluated_answer = ast.literal_eval(most_common_answer)
+        except (ValueError, SyntaxError):
+            # Handle malformed answer string
+            evaluated_answer = f"Error evaluating answer: {most_common_answer}"
+        
+        return evaluated_answer
+    
+    #     return ast.literal_eval(answers[0])
         
-    else:
-        return "Unsupported file type. Please upload an image or video."
+    # else:
+    #     return "Unsupported file type. Please upload an image or video."
         
 promt_cat_dog = """
     Annotate this image with this schema: