app.py

import os
import gradio as gr
import requests
import pandas as pd

import google.generativeai as genai
from smolagents import CodeAgent, DuckDuckGoSearchTool

# System prompt used by the agent
SYSTEM_PROMPT = """You are a general AI assistant. I will ask you a question.
Report your thoughts, and finish your answer with just the answer — no prefixes like "FINAL ANSWER:".
Your answer should be a number OR as few words as possible OR a comma-separated list of numbers and/or strings.
If you're asked for a number, don’t use commas or units like $ or %, unless specified.
If you're asked for a string, don’t use articles or abbreviations (e.g. for cities), and write digits in plain text unless told otherwise."""

DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"

# Generation result wrapper to match smolagents expectations
class GenerationResult:
    def __init__(self, content, token_usage=None, input_tokens=0, output_tokens=0):
        self.content = content
        self.token_usage = token_usage or {}
        self.input_tokens = input_tokens
        self.output_tokens = output_tokens

# Gemini model wrapper
class GeminiFlashModel:
    def __init__(self, model_id="gemini-1.5-flash", api_key=None):
        genai.configure(api_key=api_key or os.getenv("GEMINI_API_KEY"))
        self.model = genai.GenerativeModel(model_id)
        self.system_prompt = SYSTEM_PROMPT

    # Accept stop_sequences explicitly to avoid unexpected kwarg errors
    def generate(self, messages, stop_sequences=None, **kwargs):
        if not isinstance(messages, list) or not all(isinstance(m, dict) for m in messages):
            raise TypeError("Expected 'messages' to be a list of dicts")

        if not any(m.get("role") == "system" for m in messages):
            messages = [{"role": "system", "content": self.system_prompt}] + messages

        prompt = ""
        for m in messages:
            role = m["role"].capitalize()
            content = m["content"]
            prompt += f"{role}: {content}\n"

        try:
            # Note: genai.GenerativeModel.generate_content may not support stop_sequences
            response = self.model.generate_content(prompt)
            return GenerationResult(
                content=response.text.strip(),
                token_usage={},  # you can extend if API provides token info
                input_tokens=0,
                output_tokens=0
            )
        except Exception as e:
            return GenerationResult(
                content=f"GENERATION ERROR: {e}",
                token_usage={},
                input_tokens=0,
                output_tokens=0
            )

# Agent wrapper
class MyAgent:
    def __init__(self):
        self.model = GeminiFlashModel(model_id="gemini-1.5-flash")
        self.agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=self.model)

    def __call__(self, question: str) -> str:
        # The agent.run expects a string answer
        result = self.agent.run(question)
        # If result is GenerationResult or dict-like, convert to string
        if hasattr(result, "content"):
            return result.content
        elif isinstance(result, dict):
            return result.get("content", str(result))
        else:
            return str(result)

# Main evaluation function
def run_and_submit_all(profile: gr.OAuthProfile | None):
    space_id = os.getenv("SPACE_ID")

    if profile:
        username = profile.username
        print(f"User logged in: {username}")
    else:
        print("User not logged in.")
        return "Please login to Hugging Face.", None

    questions_url = f"{DEFAULT_API_URL}/questions"
    submit_url = f"{DEFAULT_API_URL}/submit"

    try:
        agent = MyAgent()
    except Exception as e:
        return f"Error initializing agent: {e}", None

    try:
        response = requests.get(questions_url, timeout=15)
        response.raise_for_status()
        questions_data = response.json()
    except Exception as e:
        return f"Error fetching questions: {e}", None

    results_log = []
    answers_payload = []

    for item in questions_data:
        task_id = item.get("task_id")
        question_text = item.get("question")
        if not task_id or question_text is None:
            continue
        try:
            submitted_answer = agent(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})
        except Exception as e:
            results_log.append({
                "Task ID": task_id,
                "Question": question_text,
                "Submitted Answer": f"AGENT ERROR: {e}"
            })

    if not answers_payload:
        return "Agent did not return any answers.", pd.DataFrame(results_log)

    submission_data = {
        "username": profile.username.strip(),
        "agent_code": f"https://huggingface.co/spaces/{space_id}/tree/main",
        "answers": answers_payload
    }

    try:
        response = requests.post(submit_url, json=submission_data, timeout=60)
        response.raise_for_status()
        result_data = response.json()
        final_status = (
            f"Submission Successful!\n"
            f"User: {result_data.get('username')}\n"
            f"Score: {result_data.get('score', 'N/A')}% "
            f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
            f"Message: {result_data.get('message', 'No message received.')}"
        )
        return final_status, pd.DataFrame(results_log)
    except Exception as e:
        return f"Submission failed: {e}", pd.DataFrame(results_log)

# Gradio UI setup
with gr.Blocks() as demo:
    gr.Markdown("# Basic Agent Evaluation Runner")
    gr.Markdown("""
    **Instructions:**
    1. Clone this space and configure your Gemini API key.
    2. Log in to Hugging Face.
    3. Run your agent on evaluation tasks and submit answers.
    """)

    gr.LoginButton()
    run_button = gr.Button("Run Evaluation & Submit All Answers")
    status_output = gr.Textbox(label="Submission Result", lines=5, interactive=False)
    results_table = gr.DataFrame(label="Results", wrap=True)

    run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])

if __name__ == "__main__":
    print("🔧 App starting...")
    demo.launch(debug=True, share=False)