Update app.py

app.py

@@ -39,7 +39,7 @@ class LLMGAIAAgent:
             self.tokenizer = None
             self.model_name = None
     
-    def __call__(self, question: str) -> str:
+    def __call__(self, question: str, task_id: str = None) -> str:
         """Process a question and return an answer using the language model."""
         print(f"Processing question: {question}")
         
@@ -152,7 +152,7 @@ class GAIAAgent:
         }
         print("GAIAAgent initialized with specialized question handlers.")
     
-    def __call__(self, question: str) -> str:
+    def __call__(self, question: str, task_id: str = None) -> str:
         """Process a question and return an appropriate answer."""
         print(f"Processing question: {question}")
         
@@ -200,20 +200,20 @@ class GAIAAgent:
             # Determine operation type
             if any(op in question_lower for op in ["sum", "add", "plus", "+"]):
                 result = sum(int(num) for num in numbers)
-                return f"The sum of the numbers is {result}"
+                return f"{result}"
                 
             elif any(op in question_lower for op in ["difference", "subtract", "minus", "-"]):
                 result = int(numbers[0]) - int(numbers[1])
-                return f"The difference between {numbers[0]} and {numbers[1]} is {result}"
+                return f"{result}"
                 
             elif any(op in question_lower for op in ["product", "multiply", "times", "*"]):
                 result = int(numbers[0]) * int(numbers[1])
-                return f"The product of {numbers[0]} and {numbers[1]} is {result}"
+                return f"{result}"
                 
             elif any(op in question_lower for op in ["divide", "division", "/"]):
                 if int(numbers[1]) != 0:
                     result = int(numbers[0]) / int(numbers[1])
-                    return f"The result of dividing {numbers[0]} by {numbers[1]} is {result}"
+                    return f"{result}"
                 else:
                     return "Cannot divide by zero"
         
@@ -338,7 +338,12 @@ class EvaluationRunner:
                 continue
             
             try:
-                submitted_answer = agent(question_text)
+                # Call agent with task_id parameter if supported
+                if hasattr(agent, '__code__') and 'task_id' in agent.__code__.co_varnames:
+                    submitted_answer = agent(question_text, task_id)
+                else:
+                    submitted_answer = agent(question_text)
+                
                 answers_payload.append({
                     "task_id": task_id, 
                     "submitted_answer": submitted_answer