Update app.py

app.py

@@ -2,7 +2,7 @@ import os
 import gradio as gr
 import requests
 import pandas as pd
-from smolagents import CodeAgent, DuckDuckGoSearchTool, InferenceClientModel, VisitWebpageTool
+from smolagents import CodeAgent, DuckDuckGoSearchTool, LiteLLMModel, VisitWebpageTool
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
@@ -10,32 +10,34 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 # --- Agent Definition ---
 class AgentArchitect:
     def __init__(self):
-        hf_token = os.getenv("HF_TOKEN")
+        # SECURE: Fetches the OpenAI API key from your Space Secrets
+        openai_api_key = os.getenv("OPENAI_API_KEY")
         
-        # We use the 7B Coder model. It is almost certainly on the FREE Serverless tier.
-        # This avoids hitting the 'nscale' or other paid Inference Providers.
-        self.model = InferenceClientModel(
-            model_id="Qwen/Qwen2.5-Coder-7B-Instruct", 
-            token=hf_token
+        if not openai_api_key:
+            print("CRITICAL: OPENAI_API_KEY is missing. Please add it to your Space Secrets!")
+
+        # Bypassing Hugging Face billing completely.
+        # We use gpt-4o-mini because it is highly capable at coding and very cost-effective.
+        self.model = LiteLLMModel(
+            model_id="gpt-4o-mini",
+            api_key=openai_api_key
         )
         
-        # Essential tools for GAIA: Search + Deep Scrapping
         self.tools = [DuckDuckGoSearchTool(), VisitWebpageTool()]
         
-        # CodeAgent is the best scaffold for technical tasks
         self.agent = CodeAgent(
             tools=self.tools,
             model=self.model,
-            add_base_tools=True # Gives it Python for math/data processing
+            add_base_tools=True 
         )
 
     def __call__(self, question: str) -> str:
         try:
-            # We add a strict formatting instruction to help with Exact Match scoring
+            # Enforce Exact Match scoring formatting
             prompt = (
                 f"{question}\n\n"
-                "Final Answer Requirement: Use your tools to find the information. "
-                "Provide ONLY the final answer concisely (e.g., just the name, number, or list)."
+                f"Instructions: Think step-by-step. Solve the problem using your tools. "
+                f"Provide ONLY the final, concise answer."
             )
             result = self.agent.run(prompt)
             return str(result)
@@ -55,6 +57,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     try:
         agent_instance = AgentArchitect()
+        
         response = requests.get(questions_url, timeout=15)
         response.raise_for_status()
         questions_data = response.json()
@@ -62,17 +65,22 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         results_log = []
         answers_payload = []
         
-        # The agent will now process the questions using the free tier model
         for item in questions_data:
             task_id = item.get("task_id")
             question_text = item.get("question")
+            
+            # Agent logic with OpenAI's brain
             submitted_answer = agent_instance(question_text)
+            
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
 
-        # Submission Logic
         agent_code_link = f"https://huggingface.co/spaces/{space_id}/tree/main"
-        submission_data = {"username": username.strip(), "agent_code": agent_code_link, "answers": answers_payload}
+        submission_data = {
+            "username": username.strip(), 
+            "agent_code": agent_code_link, 
+            "answers": answers_payload
+        }
         
         submit_response = requests.post(submit_url, json=submission_data, timeout=60)
         submit_response.raise_for_status()
@@ -81,20 +89,20 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         final_status = (
             f"Submission Successful!\n"
             f"User: {result_data.get('username')}\n"
-            f"Score: {result_data.get('score')}% \n"
+            f"Final Score: {result_data.get('score')}% \n"
             f"Message: {result_data.get('message')}"
         )
         return final_status, pd.DataFrame(results_log)
         
     except Exception as e:
-        return f"Error during run: {e}", None
+        return f"Submission Failed: {e}", None
 
 # --- Gradio UI ---
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🚀 Professional Agent Evaluator (Free Tier Optimized)")
+    gr.Markdown("# 🚀 Professional Agent Evaluator (OpenAI Edition)")
     gr.LoginButton()
     run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
-    status_output = gr.Textbox(label="Submission Result", lines=5)
+    status_output = gr.Textbox(label="Leaderboard Status", lines=4)
     results_table = gr.DataFrame(label="Agent Reasoning Trace", wrap=True)
 
     run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])