Modified files

agent.py

@@ -5,10 +5,11 @@ import re
 from functools import lru_cache, wraps
 from typing import Optional, Dict
 
-from youtube_transcript_api import YouTubeTranscriptApi, TranscriptsDisabled, NoTranscriptFound
-from dotenv import load_dotenv
+# The python-dotenv library is no longer needed as keys are read from the environment (HF Secrets)
+# from dotenv import load_dotenv 
 from requests.exceptions import RequestException
 import wikipedia
+from youtube_transcript_api import YouTubeTranscriptApi, TranscriptsDisabled, NoTranscriptFound
 from llama_index.readers.web import BeautifulSoupWebReader
 
 from smolagents import (
@@ -24,11 +25,17 @@ def configure_logging():
     """Sets up detailed logging configuration."""
     logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(name)s: %(message)s")
 
-def load_api_keys() -> Dict[str, Optional[str]]:
-    """Loads API keys from environment variables."""
-    load_dotenv()
-    keys = {'together': os.getenv('TOGETHER_API_KEY'), 'serpapi': os.getenv('SERPAPI_API_KEY')}
-    if not keys['together']: raise ValueError("TOGETHER_API_KEY is required but not found.")
+def get_api_keys_from_env() -> Dict[str, Optional[str]]:
+    """
+    Retrieves API keys directly from environment variables (e.g., Hugging Face Secrets).
+    """
+    # load_dotenv() is removed. We expect the platform to provide the environment variables.
+    keys = {
+        'together': os.getenv('TOGETHER_API_KEY'), 
+        'serpapi': os.getenv('SERPAPI_API_KEY')
+    }
+    if not keys['together']: 
+        raise ValueError("TOGETHER_API_KEY is required but not found in environment variables.")
     return keys
 
 # --- Custom Exceptions ---
@@ -104,7 +111,7 @@ def initialize_agent():
     logging.info("🚀 Starting GAIA agent initialization...")
     
     try:
-        api_keys = load_api_keys()
+        api_keys = get_api_keys_from_env()
     except ValueError as e:
         logging.error(f"FATAL: {e}")
         return None
@@ -243,7 +250,12 @@ def initialize_agent():
 # --- Main Execution Block for Local Testing ---
 
 def main():
-    """Tests the agent with sample GAIA-style questions."""
+    """
+    Tests the agent with sample GAIA-style questions.
+    For local testing, ensure you have set the required environment variables:
+    export TOGETHER_API_KEY="your_key"
+    export SERPAPI_API_KEY="your_key"
+    """
     configure_logging()
     logging.info("🧪 Starting local agent testing...")
     

app.py

@@ -9,37 +9,21 @@ from agent import initialize_agent # Import the agent initialization function
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
-# Configure logging
-logging.basicConfig(level=logging.INFO)
+# --- Logging Configuration ---
+logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(name)s: %(message)s")
 logger = logging.getLogger(__name__)
 
-# --- Helper Functions ---
+# --- Global Agent Initialization ---
+# The agent is initialized once when the Space starts up.
+# This is critical for performance and to avoid reloading the model on every request.
+logger.info("🚀 Application starting up! Initializing the GAIA agent...")
+AGENT = initialize_agent()
+if AGENT is None:
+    logger.error("💥 FATAL: Agent initialization failed. The application will not be able to process questions.")
+else:
+    logger.info("✅ Agent initialized successfully.")
 
-def extract_final_answer_from_response(response: str) -> str:
-    """
-    Extract the final answer from agent response following GAIA format.
-    The agent should return responses ending with 'FINAL ANSWER: [answer]'
-    """
-    if not response:
-        return ""
-    
-    # The agent wrapper should already return just the final answer
-    # but this is a safety check in case the format isn't perfect
-    if isinstance(response, str):
-        # Look for FINAL ANSWER pattern
-        final_answer_pattern = re.compile(r'FINAL\s+ANSWER\s*:\s*(.+?)(?:\n|$)', re.IGNORECASE | re.DOTALL)
-        match = final_answer_pattern.search(response)
-        
-        if match:
-            answer = match.group(1).strip()
-            # Clean up the answer
-            answer = re.sub(r'\s+', ' ', answer)
-            answer = answer.rstrip('.')
-            return answer
-    
-    # If no FINAL ANSWER pattern found, return the response as is
-    # (the agent wrapper should have already cleaned it)
-    return str(response).strip()
+# --- Helper Functions ---
 
 def _fetch_questions(api_url: str) -> list:
     """Fetches evaluation questions from the API."""
@@ -76,15 +60,10 @@ def _run_agent_on_questions(agent, questions_data: list) -> tuple[list, list]:
         try:
             logger.info(f"Processing task {task_id}: {question_text[:100]}...")
             
-            # The agent is now wrapped to return GAIA-compliant format
-            raw_response = agent(question_text)
-            
-            # Extract the final answer (should already be clean from wrapper)
-            submitted_answer = extract_final_answer_from_response(raw_response)
+            # The agent wrapper returns the final, normalized answer directly.
+            submitted_answer = agent(question_text)
             
-            # Log the full interaction for debugging
-            logger.info(f"Task {task_id} - Raw response: {raw_response}")
-            logger.info(f"Task {task_id} - Final answer: {submitted_answer}")
+            logger.info(f"Task {task_id} - Final answer from agent: {submitted_answer}")
             
             answers_payload.append({
                 "task_id": task_id, 
@@ -94,13 +73,12 @@ def _run_agent_on_questions(agent, questions_data: list) -> tuple[list, list]:
             results_log.append({
                 "Task ID": task_id, 
                 "Question": question_text, 
-                "Raw Response": raw_response,
                 "Final Answer": submitted_answer
             })
             
         except Exception as e:
             error_msg = f"AGENT ERROR: {e}"
-            logger.error(f"Error running agent on task {task_id}: {e}")
+            logger.error(f"Error running agent on task {task_id}: {e}", exc_info=True)
             
             answers_payload.append({
                 "task_id": task_id, 
@@ -110,7 +88,6 @@ def _run_agent_on_questions(agent, questions_data: list) -> tuple[list, list]:
             results_log.append({
                 "Task ID": task_id, 
                 "Question": question_text, 
-                "Raw Response": error_msg,
                 "Final Answer": error_msg
             })
     
@@ -151,44 +128,37 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     """
     Orchestrates the fetching of questions, running the agent, and submitting answers.
     """
-    username = None
-    if profile:
-        username = profile.username
-        logger.info(f"User logged in: {username}")
-    else:
-        logger.info("User not logged in.")
-        return "Please Login to Hugging Face with the button.", None
+    if not profile:
+        logger.warning("Attempted to run evaluation without being logged in.")
+        return "Please Login to Hugging Face with the button above.", None
+    
+    username = profile.username
+    logger.info(f"User '{username}' initiated evaluation.")
 
-    if not username:
-        return "Hugging Face username not found. Please ensure you are logged in.", None
+    if AGENT is None:
+        return "❌ Error: The agent failed to initialize on startup. Please check the Space logs for details.", None
 
     space_id = os.getenv("SPACE_ID")
     if not space_id:
         logger.error("SPACE_ID environment variable not found. Cannot determine agent_code URL.")
-        return "Error: SPACE_ID not set. Cannot determine agent_code URL.", None
+        return "❌ Error: SPACE_ID not set. This is required for submission.", None
     agent_code_url = f"https://huggingface.co/spaces/{space_id}/tree/main"
 
     status_message = ""
     results_df = pd.DataFrame()
+    results_log = []
 
     try:
-        # 1. Instantiate Agent
-        logger.info("Initializing agent...")
-        agent = initialize_agent()
-        if agent is None:
-            raise RuntimeError("Agent initialization failed. Check agent.py for details.")
-        logger.info("Agent initialized successfully.")
-
-        # 2. Fetch Questions
+        # 1. Fetch Questions
         questions_data = _fetch_questions(DEFAULT_API_URL)
 
-        # 3. Run Agent on Questions
-        answers_payload, results_log = _run_agent_on_questions(agent, questions_data)
+        # 2. Run Agent on Questions (using the pre-initialized global agent)
+        answers_payload, results_log = _run_agent_on_questions(AGENT, questions_data)
         if not answers_payload:
             status_message = "Agent did not produce any answers to submit."
             return status_message, pd.DataFrame(results_log)
 
-        # 4. Submit Answers
+        # 3. Submit Answers
         submission_result = _submit_answers(DEFAULT_API_URL, username, agent_code_url, answers_payload)
 
         final_status = (
@@ -205,14 +175,10 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     except RuntimeError as e:
         status_message = f"❌ Operation Failed: {e}"
         logger.error(status_message)
-        # If an error occurs during agent run, results_log might be partially filled
-        if 'results_log' in locals():
-            results_df = pd.DataFrame(results_log)
-        else:
-            results_df = pd.DataFrame([{"Status": "Error", "Details": str(e)}])
+        results_df = pd.DataFrame(results_log) if results_log else pd.DataFrame([{"Status": "Error", "Details": str(e)}])
     except Exception as e:
         status_message = f"💥 Critical Error: {e}"
-        logger.error(status_message)
+        logger.error(status_message, exc_info=True)
         results_df = pd.DataFrame([{"Status": "Critical Error", "Details": str(e)}])
 
     return status_message, results_df
@@ -222,36 +188,26 @@ with gr.Blocks(title="GAIA Benchmark Agent", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
     # 🧠 GAIA Benchmark Evaluation Agent
     
-    **Enhanced AI Agent for General AI Assistant (GAIA) Benchmark**
+    **An advanced agent designed to tackle the General AI Assistant (GAIA) benchmark.**
     """)
     
     gr.Markdown("""
     ## 📋 Instructions:
     
-    1. **Setup**: Clone this Space and ensure your `.env` file contains:
-       ```
-       TOGETHER_API_KEY=your_together_api_key
-       SERPAPI_API_KEY=your_serpapi_key
-       ```
+    1.  **Add Secrets**: If you have cloned this Space, go to the **Settings** tab and add your API keys as **Secrets**.
+        *   `TOGETHER_API_KEY`: Your key from Together AI.
+        *   `SERPAPI_API_KEY`: Your key from SerpApi for Google Search (optional but recommended).
     
-    2. **Login**: Use the button below to log in with your Hugging Face account
+    2.  **Login**: Use the button below to log in with your Hugging Face account. Your username is required for submission.
     
-    3. **Run**: Click 'Run Evaluation & Submit' to process all GAIA questions
+    3.  **Run**: Click 'Run Evaluation & Submit' to start the process. The agent will fetch all questions, solve them, and submit the answers automatically.
     
-    4. **Wait**: The process may take several minutes depending on question complexity
+    4.  **Wait**: The process can take several minutes. You can monitor the progress in the status box and see detailed results in the table below.
     
     ---
     
-    ### 🎯 GAIA Format Requirements:
-    - **Numbers**: No commas, no units (unless specified)
-    - **Strings**: No articles (a, an, the), no abbreviations  
-    - **Lists**: Comma-separated values following above rules
-    
-    ### 🔧 Agent Capabilities:
-    - **Web Research**: Google Search, Wikipedia, webpage analysis
-    - **Video Analysis**: YouTube transcript processing
-    - **Mathematical Computing**: Python execution with scientific libraries
-    - **Multi-step Reasoning**: Complex problem decomposition
+    ### 🎯 GAIA Answer Formatting
+    The agent is designed to automatically format answers according to GAIA's strict requirements (e.g., no commas in numbers, no articles in strings).
     """)
 
     with gr.Row():
@@ -269,7 +225,7 @@ with gr.Blocks(title="GAIA Benchmark Agent", theme=gr.themes.Soft()) as demo:
         label="📝 Detailed Question Results", 
         wrap=True,
         interactive=False,
-        column_widths=["10%", "40%", "25%", "25%"]
+        column_widths=["10%", "60%", "30%"]
     )
 
     run_button.click(
@@ -277,44 +233,32 @@ with gr.Blocks(title="GAIA Benchmark Agent", theme=gr.themes.Soft()) as demo:
         outputs=[status_output, results_table]
     )
 
-    gr.Markdown("""
-    ---
-    ### 💡 Tips for Better Performance:
-    - Ensure stable internet connection for web searches
-    - Monitor the status output for real-time progress
-    - Check the detailed results table for individual question analysis
-    - The agent automatically formats answers according to GAIA requirements
-    """)
-
 if __name__ == "__main__":
     print("\n" + "="*70)
-    print("🚀 GAIA BENCHMARK AGENT STARTING")
+    print("🚀 GAIA BENCHMARK AGENT STARTING UP")
     print("="*70)
     
-    # Check environment variables
-    space_host = os.getenv("SPACE_HOST")
+    # Check environment variables loaded from HF Secrets
     space_id = os.getenv("SPACE_ID")
     together_key = os.getenv("TOGETHER_API_KEY")
     serpapi_key = os.getenv("SERPAPI_API_KEY")
 
-    if space_host:
-        print(f"✅ SPACE_HOST: {space_host}")
-        print(f"   🌐 Runtime URL: https://{space_host}.hf.space")
-    else:
-        print("ℹ️  SPACE_HOST not found (local development)")
-
     if space_id:
         print(f"✅ SPACE_ID: {space_id}")
-        print(f"   📂 Repo URL: https://huggingface.co/spaces/{space_id}")
+        print(f"   - Submission URL will be: https://huggingface.co/spaces/{space_id}")
     else:
-        print("⚠️  SPACE_ID not found - submissions may fail")
+        print("⚠️  SPACE_ID not found - submissions will fail. This is normal for local dev.")
+
+    print(f"🔑 API Keys Status (from Secrets):")
+    print(f"   - Together AI: {'✅ Set' if together_key else '❌ Missing - Agent will fail to initialize!'}")
+    print(f"   - SerpAPI:     {'✅ Set' if serpapi_key else '⚠️ Missing - Google Search tool will be disabled.'}")
 
-    print(f"🔑 API Keys Status:")
-    print(f"   Together AI: {'✅ Set' if together_key else '❌ Missing'}")
-    print(f"   SerpAPI: {'✅ Set' if serpapi_key else '⚠️ Missing (optional)'}")
+    if not together_key:
+        print("\n‼️ CRITICAL: TOGETHER_API_KEY is not set in the Space Secrets.")
+        print("   Please add it in the 'Settings' tab of your Space.")
 
     print("="*70)
-    print("🎯 Launching GAIA Benchmark Interface...")
+    print("🎯 Launching Gradio Interface...")
     print("="*70 + "\n")
 
-    demo.launch(debug=True, share=False)
+    demo.launch(debug=False, share=False)