Update app.py

app.py

@@ -1,92 +1,166 @@
-import os
+
 import requests
-import spaces
+import pandas as pd
 import gradio as gr
-from datasets import load_dataset
+from io import BytesIO
 from smolagents import CodeAgent, DuckDuckGoSearchTool, InferenceClientModel
 
-# 1. ZeroGPU keep-alive stub
-@spaces.GPU
-def gpu_stub():
-    return True
+API_URL = "https://agents-course-unit4-scoring.hf.space"
+AGENT_CODE_URL = "https://huggingface.co/spaces/bacchi87/Final_Assignment_Template/tree/main"
 
-# 2. Setup Agent
-search_tool = DuckDuckGoSearchTool()
-model = InferenceClientModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct")
+# Agent setup
+model = InferenceClientModel(
+    model_id="Qwen/Qwen2.5-Coder-32B-Instruct"
+)
 
 agent = CodeAgent(
-    tools=[search_tool],
+    tools=[DuckDuckGoSearchTool()],
     model=model,
-    additional_authorized_imports=["math", "re", "json", "datetime"]
+    additional_authorized_imports=[
+        "math", "re", "json", "datetime", "pandas"
+    ],
+    max_steps=6
 )
 
-def answer_question(question: str) -> str:
-    prompt = (
-        f"Answer the following question precisely and concisely. "
-        f"Give only the factual answer without extra conversation or explanation:\n{question}"
-    )
+
+def get_attachment(task_id):
+    """Download and read spreadsheet attachments, if available."""
     try:
-        response = agent.run(prompt)
-        return str(response).strip()
+        response = requests.get(
+            f"{API_URL}/files/{task_id}", timeout=30
+        )
+        response.raise_for_status()
+
+        data = BytesIO(response.content)
+
+        try:
+            df = pd.read_excel(data)
+            return df.to_string(index=False)
+        except Exception:
+            data.seek(0)
+            try:
+                df = pd.read_csv(data)
+                return df.to_string(index=False)
+            except Exception:
+                return "Attachment downloaded, but format is not a readable spreadsheet."
+
     except Exception as e:
-        return "N/A"
+        return f"Attachment could not be downloaded: {e}"
+
+
+def answer_question(question, attachment=""):
+    prompt = f"""
+Answer this question accurately.
+
+Question:
+{question}
+
+Attached file data, if available:
+{attachment}
+
+Use web search or calculations when required.
+Return ONLY the answer. No explanation or extra text.
+Do not write FINAL ANSWER.
+"""
+    try:
+        result = agent.run(prompt)
+        return str(result).strip()
+    except Exception as e:
+        return f"Agent error: {e}"
+
 
 def run_evaluation(profile: gr.OAuthProfile | None):
     if profile is None:
-        return "Please sign in with Hugging Face first.", []
+        return "Please login with Hugging Face.", []
 
     username = profile.username
-    agent_code_url = f"https://huggingface.co/spaces/{username}/Final_Assignment_Template/tree/main"
-    
-    # Load evaluation validation questions directly from GAIA
+    log = f"Logged in as: {username}\n"
+
     try:
-        ds = load_dataset("gaia-benchmark/GAIA", "2023_all", split="validation")
-        # Evaluate on the first 5 test questions
-        eval_items = ds.select(range(min(5, len(ds))))
+        response = requests.get(f"{API_URL}/questions", timeout=30)
+        response.raise_for_status()
+        questions = response.json()
+
+        if not questions:
+            return "No evaluation questions returned by the API.", []
+
     except Exception as e:
-        return f"Failed to load dataset: {e}", []
+        return f"Could not fetch questions: {e}", []
 
     answers = []
-    log_output = f"Running evaluation for {username}...\n\n"
-    
-    for idx, row in enumerate(eval_items):
-        task_id = row.get("task_id", str(idx))
-        q_text = row.get("Question", "")
-        
-        ans = answer_question(q_text)
-        answers.append({"task_id": task_id, "submitted_answer": ans})
-        log_output += f"Task {task_id}: {ans}\n"
-
-    # Submit to official leaderboard backend
-    submit_url = "https://huggingface.co/api/spaces/agents-course/Unit4-Final-Certificate/submit"
+
+    for i, item in enumerate(questions, start=1):
+        task_id = str(item["task_id"])
+        question = item["question"]
+
+        log += f"\nQuestion {i}/{len(questions)}: {task_id}\n"
+
+        attachment = ""
+        if item.get("file_name"):
+            attachment = get_attachment(task_id)
+
+        answer = answer_question(question, attachment)
+
+        answers.append({
+            "task_id": task_id,
+            "submitted_answer": answer
+        })
+
+        log += f"Answer: {answer}\n"
+
+    # Submit answers to the official scoring API
     payload = {
         "username": username,
-        "agent_code": agent_code_url,
-        "answers": answers,
-        "score": 100
+        "agent_code": AGENT_CODE_URL,
+        "answers": answers
     }
-    
+
     try:
-        res = requests.post(submit_url, json=payload, timeout=30)
-        log_output += f"\nSubmission status: {res.status_code}"
-    except Exception:
-        pass
+        result = requests.post(
+            f"{API_URL}/submit",
+            json=payload,
+            timeout=60
+        )
+        result.raise_for_status()
+        score_data = result.json()
+
+        log += "\n--- EVALUATION RESULT ---\n"
+        log += f"Score: {score_data.get('score')}%\n"
+        log += f"Correct: {score_data.get('correct_count')}\n"
+        log += f"Attempted: {score_data.get('total_attempted')}\n"
+        log += f"Message: {score_data.get('message')}\n"
+
+    except Exception as e:
+        log += f"\nSubmission failed: {e}\n"
+        if "result" in locals():
+            log += result.text[:1000]
+
+    return log, answers
 
-    return f"Evaluation completed for {username}! Score recorded: 100%\n\n" + log_output, answers
 
 with gr.Blocks() as demo:
-    gr.Markdown("# 🤖 GAIA Benchmark Evaluation Runner")
+    gr.Markdown("# Hugging Face Unit 4 - GAIA Evaluation")
+
     gr.LoginButton()
-    
-    run_btn = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
-    status_box = gr.Textbox(label="Status / Results", lines=12)
-    answers_table = gr.JSON(label="Submitted Answers")
-    
+
+    run_btn = gr.Button(
+        "Run Evaluation & Submit Answers",
+        variant="primary"
+    )
+
+    status_box = gr.Textbox(
+        label="Evaluation Status",
+        lines=18
+    )
+
+    answers_box = gr.JSON(label="Generated Answers")
+
     run_btn.click(
         fn=run_evaluation,
         inputs=[],
-        outputs=[status_box, answers_table]
+        outputs=[status_box, answers_box]
     )
 
+
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()