Update app.py

app.py

@@ -1,85 +1,78 @@
 import os
-import subprocess
-import sys
-
-try:
-    import duckduckgo_search
-except ImportError:
-    subprocess.check_call([sys.executable, "-m", "pip", "install", "duckduckgo-search==6.3.2"])
-
 import gradio as gr
 import requests
 import pandas as pd
 from smolagents import CodeAgent, HfApiModel
 
-try:
-    from smolagents import DuckDuckGoSearchTool
-    HAS_SEARCH = True
-except ImportError:
-    HAS_SEARCH = False
-
+# --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 class BasicAgent:
     def __init__(self):
+        # HfApiModel is native and fast within the HF ecosystem
+        # It uses your HF_TOKEN automatically if set in Secrets
         self.model = HfApiModel(model_id="Qwen/Qwen2.5-72B-Instruct")
         
-        tools = []
-        if HAS_SEARCH:
-            try:
-                tools.append(DuckDuckGoSearchTool())
-            except Exception as e:
-                print(f"Could not init search tool: {e}")
-
+        # We use CodeAgent WITHOUT external tools like Search.
+        # add_base_tools=True provides the Python Interpreter for logic/math.
         self.agent = CodeAgent(
-            tools=tools, 
+            tools=[], 
             model=self.model,
-            add_base_tools=True # Python interpreter for math/logic
+            add_base_tools=True 
         )
-        print(f"Agent initialized. Search available: {HAS_SEARCH}")
+        print("Clean Agent initialized (Python Interpreter enabled).")
 
     def __call__(self, question: str) -> str:
+        # Prompt specifically designed for GAIA Exact Match scoring
         clean_prompt = (
-            f"Question: {question}\n\n"
-            "Instructions: Solve the question above. Provide ONLY the final answer "
-            "value. No explanation, no 'The answer is...', just the value."
+            f"Solve this task: {question}\n\n"
+            "Final Answer Requirement: Provide ONLY the numeric or text value. "
+            "Do not include units, symbols, or conversational filler. "
+            "No 'The answer is...', no 'FINAL ANSWER' text. Just the raw value."
         )
+        
         try:
+            # The agent will write code to solve if the question is complex
             result = self.agent.run(clean_prompt)
             return str(result).strip()
         except Exception as e:
-            return f"Error: {e}"
+            print(f"Agent Error: {e}")
+            return "Error"
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     space_id = os.getenv("SPACE_ID")
+
     if not profile:
         return "Please Login to Hugging Face with the button.", None
 
-    username = f"{profile.username}"
-    # Define agent code link
-    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
+    username = profile.username
+    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else ""
 
-    # 1. Fetch Questions
+    # 1. Fetch
     try:
         response = requests.get(f"{DEFAULT_API_URL}/questions", timeout=15)
         questions_data = response.json()
     except Exception as e:
-        return f"Error fetching questions: {e}", None
+        return f"Fetch Error: {e}", None
 
-    # 2. Run Agent
+    # 2. Run
     agent = BasicAgent()
     answers_payload = []
     results_log = []
     
-    # Process all questions
+    print(f"Starting evaluation for {len(questions_data)} questions...")
+    
     for item in questions_data:
         task_id = item.get("task_id")
         question_text = item.get("question")
+        
         if not task_id: continue
 
+        # Run agent
         submitted_answer = agent(question_text)
+        
         answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
-        results_log.append({"Task ID": task_id, "Answer": submitted_answer})
+        results_log.append({"Task ID": task_id, "Question": question_text[:100], "Answer": submitted_answer})
 
     # 3. Submit
     submission_data = {
@@ -91,20 +84,31 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     try:
         response = requests.post(f"{DEFAULT_API_URL}/submit", json=submission_data, timeout=60)
         result_data = response.json()
-        status = f"Success! Score: {result_data.get('score')}% ({result_data.get('correct_count')}/{result_data.get('total_attempted')})"
-        return status, pd.DataFrame(results_log)
+        
+        final_status = (
+            f"Submission Successful!\n"
+            f"Score: {result_data.get('score')}% "
+            f"({result_data.get('correct_count')}/{result_data.get('total_attempted')} correct)"
+        )
+        return final_status, pd.DataFrame(results_log)
     except Exception as e:
         return f"Submission Failed: {e}", pd.DataFrame(results_log)
 
 # --- Gradio Interface ---
 with gr.Blocks() as demo:
-    gr.Markdown("# GAIA Final Solver")
+    gr.Markdown("# GAIA Agent Evaluation Runner (Clean Version)")
+    gr.Markdown("Uses `smolagents` with a built-in Python Interpreter to solve tasks.")
+
     gr.LoginButton()
-    run_button = gr.Button("Run Evaluation & Submit", variant="primary")
-    status_output = gr.Textbox(label="Status")
-    results_table = gr.DataFrame(label="Answers Generated")
+    run_button = gr.Button("Run Evaluation & Submit All", variant="primary")
+
+    status_output = gr.Textbox(label="Status", lines=4)
+    results_table = gr.DataFrame(label="Generated Answers", wrap=True)
 
-    run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
+    run_button.click(
+        fn=run_and_submit_all,
+        outputs=[status_output, results_table]
+    )
 
 if __name__ == "__main__":
     demo.launch()