Update app.py

app.py

@@ -1,10 +1,11 @@
 import os
+import re
 import gradio as gr
 import requests
 import pandas as pd
 
 from smolagents import CodeAgent, DuckDuckGoSearchTool
-from smolagents.models import InferenceClientModel
+from smolagents.models import OpenAIServerModel
 
 # ---------------------------------------------------
 # Constants
@@ -12,46 +13,73 @@ from smolagents.models import InferenceClientModel
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # ---------------------------------------------------
-# Agent Definition
+# Agent
 # ---------------------------------------------------
 class BasicAgent:
 
     def __init__(self):
 
-        print("Initializing Hugging Face Agent...")
+        print("Initializing Groq Agent...")
 
-        # Hugging Face Inference API model
-        model = InferenceClientModel(
-            model_id="Qwen/Qwen2.5-72B-Instruct",
-            token=os.getenv("HF_TOKEN")
+        self.model = OpenAIServerModel(
+            model_id="llama-3.3-70b-versatile",
+            api_base="https://api.groq.com/openai/v1",
+            api_key=os.getenv("GROQ_API_KEY")
         )
 
-        # Build agent
         self.agent = CodeAgent(
             tools=[
                 DuckDuckGoSearchTool()
             ],
-            model=model,
-            max_steps=6,
-            verbosity_level=1
+            model=self.model,
+            max_steps=3,
+            verbosity_level=0
         )
 
-        print("Agent initialized successfully.")
+        print("Groq Agent initialized.")
 
-    def __call__(self, question: str) -> str:
+    # ---------------------------------------------------
+    # Clean outputs for exact-match grading
+    # ---------------------------------------------------
+    def clean_answer(self, text):
+
+        if text is None:
+            return ""
+
+        text = str(text)
 
-        print(f"\nQuestion: {question[:100]}")
+        # remove markdown/code
+        text = text.replace("```", "")
+        text = text.replace("FINAL ANSWER:", "")
+        text = text.replace("Answer:", "")
+
+        # remove excessive whitespace
+        text = re.sub(r"\s+", " ", text)
+
+        # first line only
+        text = text.split("\n")[0]
+
+        # concise
+        text = text.strip()
+
+        return text[:300]
+
+    # ---------------------------------------------------
+    # Run agent
+    # ---------------------------------------------------
+    def __call__(self, question: str) -> str:
 
         prompt = f"""
-You are an expert GAIA benchmark solving agent.
+You are solving a GAIA benchmark question.
 
-Your job:
-- Think step-by-step.
-- Use web search if needed.
-- Solve the task accurately.
+IMPORTANT:
 - Return ONLY the final answer.
-- Keep answers concise.
-- No explanations unless necessary.
+- No reasoning.
+- No explanations.
+- No markdown.
+- No bullet points.
+- Be concise and accurate.
+- Use web search if needed.
 
 Question:
 {question}
@@ -61,20 +89,18 @@ Question:
 
             result = self.agent.run(prompt)
 
-            if result is None:
-                return "Could not determine the answer."
+            cleaned = self.clean_answer(result)
 
-            final_answer = str(result).strip()
+            print(f"\nQUESTION:\n{question}")
+            print(f"\nANSWER:\n{cleaned}")
 
-            print(f"Answer: {final_answer}")
-
-            return final_answer
+            return cleaned
 
         except Exception as e:
 
             print(f"Agent error: {e}")
 
-            return f"Error: {str(e)}"
+            return ""
 
 
 # ---------------------------------------------------
@@ -82,17 +108,15 @@ Question:
 # ---------------------------------------------------
 def run_and_submit_all(profile: gr.OAuthProfile | None):
 
-    # Get Space ID
     space_id = os.getenv("SPACE_ID")
 
-    # Check login
+    # Login check
     if profile:
         username = profile.username
         print(f"Logged in as: {username}")
     else:
-        return "Please login with Hugging Face first.", None
+        return "Please login with Hugging Face.", None
 
-    # API URLs
     api_url = DEFAULT_API_URL
     questions_url = f"{api_url}/questions"
     submit_url = f"{api_url}/submit"
@@ -101,15 +125,16 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     # Initialize Agent
     # ---------------------------------------------------
     try:
+
         agent = BasicAgent()
 
     except Exception as e:
 
-        print(f"Error initializing agent: {e}")
+        print(f"Initialization error: {e}")
 
         return f"Error initializing agent: {e}", None
 
-    # Link to your Space code
+    # Space repo link
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
 
     # ---------------------------------------------------
@@ -132,7 +157,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     except Exception as e:
 
-        print(f"Error fetching questions: {e}")
+        print(f"Question fetch error: {e}")
 
         return f"Error fetching questions: {e}", None
 
@@ -150,7 +175,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         if not task_id or question_text is None:
             continue
 
-        print(f"\nRunning task: {task_id}")
+        print(f"\nRunning task {task_id}")
 
         try:
 
@@ -178,7 +203,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             })
 
     # ---------------------------------------------------
-    # Submit Answers
+    # Submit
     # ---------------------------------------------------
     submission_data = {
         "username": username.strip(),
@@ -186,10 +211,10 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         "answers": answers_payload
     }
 
-    print("Submitting answers...")
-
     try:
 
+        print("Submitting answers...")
+
         response = requests.post(
             submit_url,
             json=submission_data,
@@ -216,15 +241,15 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     except Exception as e:
 
-        error_msg = f"Submission failed: {e}"
+        error_message = f"Submission failed: {e}"
 
-        print(error_msg)
+        print(error_message)
 
-        return error_msg, pd.DataFrame(results_log)
+        return error_message, pd.DataFrame(results_log)
 
 
 # ---------------------------------------------------
-# Gradio Interface
+# UI
 # ---------------------------------------------------
 with gr.Blocks() as demo:
 
@@ -234,22 +259,19 @@ with gr.Blocks() as demo:
         """
 This agent uses:
 
-- Hugging Face Inference API
+- Groq API
+- Llama 3.3 70B
 - smolagents
-- DuckDuckGo web search
-- Qwen2.5-72B-Instruct
+- DuckDuckGo Search
 """
     )
 
-    # HF Login
     gr.LoginButton()
 
-    # Run Button
     run_button = gr.Button(
         "Run Evaluation & Submit"
     )
 
-    # Outputs
     status_output = gr.Textbox(
         label="Submission Result",
         lines=8,
@@ -261,7 +283,6 @@ This agent uses:
         wrap=True
     )
 
-    # Button Action
     run_button.click(
         fn=run_and_submit_all,
         outputs=[
@@ -271,12 +292,12 @@ This agent uses:
     )
 
 # ---------------------------------------------------
-# Launch App
+# Launch
 # ---------------------------------------------------
 if __name__ == "__main__":
 
     print("\n==============================")
-    print("Starting Hugging Face Agent...")
+    print("Starting Groq Agent...")
     print("==============================\n")
 
     demo.launch(