Update app.py

app.py

@@ -1,109 +1,170 @@
 import os
 import requests
 import json
-from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiClient
+import gradio as ui
+from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiModel
 
-# --- CONFIGURATION ---
+# --- SCORING CONFIGURATION ---
 API_BASE_URL = "https://agents-course-unit4-scoring.hf.space"
-# Ensure your space code link looks exactly like this (.../tree/main)
-HF_USERNAME = "your_hf_username"
-CODE_LINK = f"https://huggingface.co/spaces/{HF_USERNAME}/Final_Assignment_Template/tree/main"
 
-# --- ENGINE SETUP ---
+# Initialize core agent dependencies
 search_tool = DuckDuckGoSearchTool()
 agent = CodeAgent(
-    model=HfApiClient(), 
+    model=HfApiModel(),  # Serverless inference engine
     tools=[search_tool],
-    additional_authorized_imports=["pandas", "numpy", "requests", "bs4", "openpyxl", "pypdf"]
+    additional_authorized_imports=["pandas", "numpy", "requests", "bs4", "openpyxl", "pypdf", "PIL"]
 )
 
-def fetch_questions():
-    """Retrieve the 20 filtered evaluation questions."""
-    response = requests.get(f"{API_BASE_URL}/questions")
+def fetch_evaluation_questions():
+    """Fetches the 20 level 1 filtered validation questions from the course API."""
+    url = f"{API_BASE_URL}/questions"
+    response = requests.get(url)
     response.raise_for_status()
-    return response.json() # Returns a list of dicts with task_id, question, etc.
+    return response.json()
 
-def download_file_if_needed(task_id: str, file_name: str):
-    """Downloads an evaluation attachment if present and returns the local path."""
+def download_task_file(task_id: str, file_name: str):
+    """Downloads an evaluation asset if present and maps it to a local workspace path."""
     if not file_name:
         return None
     
-    local_path = f"./downloads/{task_id}_{file_name}"
-    os.makedirs("./downloads", exist_ok=True)
+    local_directory = "./downloads"
+    os.makedirs(local_directory, exist_ok=True)
+    local_file_path = os.path.join(local_directory, f"{task_id}_{file_name}")
     
+    # Don't re-download if file already exists locally
+    if os.path.exists(local_file_path):
+        return local_file_path
+        
     url = f"{API_BASE_URL}/files/{task_id}"
     response = requests.get(url, stream=True)
     
     if response.status_code == 200:
-        with open(local_path, 'wb') as f:
+        with open(local_file_path, 'wb') as file:
             for chunk in response.iter_content(chunk_size=8192):
-                f.write(chunk)
-        print(f"📥 Downloaded asset for Task {task_id}: {local_path}")
-        return local_path
+                file.write(chunk)
+        print(f"📥 Asset downloaded successfully for Task {task_id} -> {local_file_path}")
+        return local_file_path
     else:
-        print(f"⚠️ No downloadable file asset found for Task {task_id} (Status {response.status_code})")
+        print(f"ℹ️ No download asset found or status code error for Task {task_id}: {response.status_code}")
         return None
 
-def run_evaluation_pipeline():
-    questions = fetch_questions()
-    submitted_answers = []
-    
-    # Strictly instruct the agent to avoid conversational text
-    formatting_instructions = (
-        "\n\nCRITICAL OUTCOME INSTRUCTION: Output ONLY the literal final string answer. "
-        "Do NOT prefix with 'FINAL ANSWER:', do NOT use markdown code blocks for text, "
-        "and do NOT include conversational explanations. Your complete response must contain "
-        "only the raw value matching the requested format."
-    )
-    
-    for item in questions:
-        task_id = item['task_id']
-        raw_question = item['question']
-        # Level 1 items occasionally reference attached data sheets
-        file_name = item.get('file_name', None) 
-        
-        print(f"\n🚀 Processing Task ID: {task_id}")
+def process_agent_evaluation(hf_username: str, profile_code_url: str, progress=ui.Progress()):
+    """Iterates through all evaluation steps, manages files, runs agent queries, and returns results."""
+    if not hf_username.strip() or not profile_code_url.strip():
+        return "❌ Error: Please fill in both your Hugging Face Username and your Public Space Code Link before submitting."
+
+    try:
+        progress(0, desc="Fetching GAIA evaluation questions...")
+        questions = fetch_evaluation_questions()
+        submitted_answers = []
         
-        # Pull any related assets down locally before execution
-        local_file_path = download_file_if_needed(task_id, file_name)
+        # Strict context formatting boundary to prevent verbose conversational leaks
+        formatting_instructions = (
+            "\n\nCRITICAL OUTCOME INSTRUCTION: Output ONLY the literal final string answer. "
+            "Do NOT prefix with 'FINAL ANSWER:', do NOT use markdown code blocks for text, "
+            "and do NOT include conversational explanations. Your complete response must contain "
+            "only the raw value matching the requested format."
+        )
         
-        full_prompt = raw_question
-        if local_file_path:
-            full_prompt += f"\nNote: The file provided for this task has been downloaded locally to: {local_file_path}"
+        total_questions = len(questions)
+        for index, item in enumerate(questions):
+            task_id = item['task_id']
+            raw_question = item['question']
+            file_name = item.get('file_name', None)
             
-        full_prompt += formatting_instructions
-        
-        try:
-            # Execute agent reasoning loop
-            raw_output = agent.run(full_prompt)
-            final_answer = str(raw_output).strip()
+            progress((index / total_questions), desc=f"Evaluating Task {index + 1}/{total_questions} (ID: {task_id})...")
             
-            # Safe fallbacks if formatting leaks through
-            if "the answer is" in final_answer.lower():
-                final_answer = final_answer.lower().split("the answer is")[-1].strip(" :.`\"")
-                
-            print(f"🎯 Cleaned Agent Result: '{final_answer}'")
+            # Look up and handle data attachments
+            local_file_path = download_task_file(task_id, file_name)
             
-            submitted_answers.append({
-                "task_id": task_id,
-                "submitted_answer": final_answer
-            })
+            full_prompt = raw_question
+            if local_file_path:
+                full_prompt += f"\nNote: The file provided for this task has been downloaded locally to: {local_file_path}"
+                
+            full_prompt += formatting_instructions
             
-        except Exception as e:
-            print(f"❌ Error processing task {task_id}: {e}")
-            submitted_answers.append({
-                "task_id": task_id,
-                "submitted_answer": "ERROR_FALLBACK"
-            })
+            try:
+                # Fire up the agent loop
+                raw_output = agent.run(full_prompt)
+                final_answer = str(raw_output).strip()
+                
+                # Cleanup parser fallback if agent leaked phrases
+                if "the answer is" in final_answer.lower():
+                    final_answer = final_answer.lower().split("the answer is")[-1].strip(" :.`\"*")
+                
+                submitted_answers.append({
+                    "task_id": task_id,
+                    "submitted_answer": final_answer
+                })
+                print(f"✅ Clean Answer Logged [{task_id}]: {final_answer}")
+                
+            except Exception as e:
+                print(f"❌ Execution crash on task {task_id}: {e}")
+                submitted_answers.append({
+                    "task_id": task_id,
+                    "submitted_answer": "ERROR_FALLBACK"
+                })
+
+        # --- POST SUBMISSION PAYLOAD ---
+        progress(0.95, desc="Transmitting data package to student leaderboard...")
+        payload = {
+            "username": hf_username.strip(),
+            "agent_code": profile_code_url.strip(),
+            "answers": submitted_answers
+        }
+        
+        headers = {"Content-Type": "application/json"}
+        submit_url = f"{API_BASE_URL}/submit"
+        
+        response = requests.post(submit_url, json=payload, headers=headers)
+        
+        if response.status_code == 200:
+            result_data = response.json()
+            score = result_data.get("score", 0)
+            percentage = score * 100
             
-    # --- PAYLOAD GENERATION FOR POST /SUBMIT ---
-    payload = {
-        "username": HF_USERNAME,
-        "agent_code": CODE_LINK,
-        "answers": submitted_answers
-    }
+            status_summary = (
+                f"### 🎉 Evaluation Complete!\n\n"
+                f"**📊 Final Score:** {percentage:.1f}% ({int(score * total_questions)} / {total_questions} Correct)\n\n"
+                f"**💬 API Server Response:** {result_data.get('message', 'Submission saved successfully!')}\n\n"
+                f"Check your standing on the leaderboard right now!"
+            )
+            return status_summary
+        else:
+            return f"❌ Submission Error (HTTP {response.status_code}): {response.text}"
+
+    except Exception as general_error:
+        return f"❌ A structural runtime execution crash occurred: {str(general_error)}"
+
+# --- GRADIO WEB USER INTERFACE ---
+with ui.Blocks(theme=ui.themes.Soft()) as interface:
+    ui.Markdown("# 🏆 Unit 4: GAIA Benchmark Agent Certification Engine")
+    ui.Markdown(
+        "Welcome! This portal coordinates your custom `smolagents` architecture against the "
+        "20 curated Level-1 GAIA tasks. Fill in your identification markers below to run evaluation and sync metrics."
+    )
+    
+    with ui.Row():
+        input_username = ui.Textbox(
+            label="Hugging Face Username", 
+            placeholder="e.g., dev_pioneer",
+            lines=1
+        )
+        input_code_link = ui.Textbox(
+            label="Public Space Repository Link (agent_code URL)", 
+            placeholder="https://huggingface.co/spaces/your_username/your_space/tree/main",
+            lines=1
+        )
+        
+    evaluate_button = ui.Button("🚀 Start Evaluation & Submit Answers", variant="primary")
     
-    print("\n📦 Built Submission Payload successfully.")
-    return payload
+    output_display = ui.Markdown(value="💡 Provide your workspace metrics above and click run to engage the loop.")
+    
+    evaluate_button.click(
+        fn=process_agent_evaluation,
+        inputs=[input_username, input_code_link],
+        outputs=output_display
+    )
 
-# execution_payload = run_evaluation_pipeline()
+if __name__ == "__main__":
+    interface.launch()