Update app.py

app.py

@@ -1,16 +1,15 @@
-# app.py
-
 import os
 import gradio as gr
 import requests
 import pandas as pd
-
 from smolagents import (
     CodeAgent,
     DuckDuckGoSearchTool,
     PythonREPLTool,
     OpenAIServerModel,
 )
+from smolagents.tools.python_repl import PythonREPL
+import traceback # Import traceback for detailed error logging
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
@@ -21,12 +20,12 @@ class GaiaAgent:
         self.openai_key = openai_key
         # 1) Initialize the LLM-backed model
         self.model = OpenAIServerModel(
-            model_id="gpt-4",        # or "gpt-3.5-turbo" if you prefer
+            model_id="gpt-4",  # or "gpt-3.5-turbo" if you prefer
             api_key=self.openai_key,
         )
         # 2) Define the tools
         self.search_tool = DuckDuckGoSearchTool()
-        self.python_tool = PythonREPLTool()
+        self.python_tool = PythonREPLTool(timeout=10)  # Initialize PythonREPLTool
         # 3) Create the CodeAgent
         self.agent = CodeAgent(
             model=self.model,
@@ -38,23 +37,31 @@ class GaiaAgent:
                 "Always think in Python code using the available tools.  "
                 "Never answer without executing or checking with a tool.  "
                 "Use DuckDuckGoSearchTool for lookups, PythonREPLTool for "
-                "calculations, string or list manipulations."
+                "calculations, string or list manipulations.  "
+                "Respond with the final answer only. Do not include any extra explanation. "
             )
         )
 
     def __call__(self, question: str) -> str:
-        return self.agent.run(question)
+        try:
+            return self.agent.run(question)
+        except Exception as e:
+            error_message = f"Agent execution failed: {e}\n{traceback.format_exc()}"
+            print(error_message)  # Log the error for debugging
+            return "ERROR: Agent failed to answer." # Return a string, not an exception
+
 
 def run_and_submit_all(profile: gr.OAuthProfile | None, openai_key: str):
     # --- Login & Setup ---
     if not profile:
         return "Please log in to Hugging Face to submit your score.", None
     username = profile.username.strip()
-
     # 1) Instantiate our improved agent
     try:
         agent = GaiaAgent(openai_key)
     except Exception as e:
+        error_message = f"Error initializing agent: {e}\n{traceback.format_exc()}"
+        print(error_message)
         return f"Error initializing agent: {e}", None
 
     # 2) Fetch the GAIA questions
@@ -64,6 +71,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, openai_key: str):
         resp.raise_for_status()
         questions = resp.json()
     except Exception as e:
+        error_message = f"Error fetching questions: {e}\n{traceback.format_exc()}"
+        print(error_message)
         return f"Error fetching questions: {e}", None
 
     # 3) Run the agent on each question
@@ -75,6 +84,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, openai_key: str):
         try:
             ans = agent(q)
         except Exception as e:
+            error_message = f"Error processing question {tid}: {e}\n{traceback.format_exc()}"
+            print(error_message) # Print full traceback
             ans = f"ERROR: {e}"
         answers.append({"task_id": tid, "submitted_answer": ans})
         log.append({"Task ID": tid, "Question": q, "Answer": ans})
@@ -94,13 +105,15 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, openai_key: str):
             f"✅ Submission Successful!\n"
             f"User: {data['username']}\n"
             f"Score: {data['score']}% ({data['correct_count']}/{data['total_attempted']})\n"
-            f"Message: {data.get('message','')}"
+            f"Message: {data.get('message', '')}"
         )
     except Exception as e:
+        error_message = f"Submission failed: {e}\n{traceback.format_exc()}"
+        print(error_message)
         status = f"Submission failed: {e}"
-
     return status, pd.DataFrame(log)
 
+
 # --- Gradio UI ---
 with gr.Blocks() as demo:
     gr.Markdown("# GAIA Benchmark Runner")
@@ -115,7 +128,6 @@ with gr.Blocks() as demo:
     run_btn = gr.Button("Run & Submit")
     out_status = gr.Textbox(label="Status", lines=4)
     out_table = gr.DataFrame(label="Questions & Answers")
-
     run_btn.click(fn=run_and_submit_all, inputs=[login, key_in], outputs=[out_status, out_table])
 
 if __name__ == "__main__":