app.py

import os
import gradio as gr
import requests
import pandas as pd
from transformers import AutoModelForCausalLM, AutoTokenizer
from gradio import get_oauth_profile

# ---------- Imports for Advanced Agent ----------
import re
from langgraph.graph import StateGraph, MessagesState
from langgraph.prebuilt import tools_condition, ToolNode
from langchain_core.messages import SystemMessage, HumanMessage
from langchain_core.tools import tool
from langchain_community.document_loaders import WikipediaLoader, ArxivLoader
from langchain_community.tools.tavily_search import TavilySearchResults
from groq import Groq

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

# ---------- Tools ----------
from langchain_core.tools import tool
from langchain_community.document_loaders import WikipediaLoader, ArxivLoader
from langchain_community.tools.tavily_search import TavilySearchResults

@tool
def wiki_search(query: str) -> str:
    """Search Wikipedia for a given query and return content from up to 2 relevant pages."""
    docs = WikipediaLoader(query=query, load_max_docs=2).load()
    return "\n\n".join([doc.page_content for doc in docs])

@tool
def web_search(query: str) -> str:
    """Search the web using the Tavily API and return content from up to 3 search results."""
    docs = TavilySearchResults(max_results=3).invoke(query)
    return "\n\n".join([doc.page_content for doc in docs])

@tool
def arvix_search(query: str) -> str:
    """Search academic papers on Arxiv for a given query and return up to 3 result summaries."""
    docs = ArxivLoader(query=query, load_max_docs=3).load()
    return "\n\n".join([doc.page_content[:1000] for doc in docs])

# Tool-based LangGraph builder
def build_tool_graph(system_prompt):
    llm = AutoModelForCausalLM.from_pretrained("gpt2")  # Load Hugging Face GPT-2 model
    tokenizer = AutoTokenizer.from_pretrained("gpt2")

    def assistant(state: MessagesState):
        input_text = state["messages"][-1]["content"]
        inputs = tokenizer(input_text, return_tensors="pt")
        outputs = llm.generate(**inputs)
        result = tokenizer.decode(outputs[0], skip_special_tokens=True)
        return {"messages": [{"content": result}]}

    builder = StateGraph(MessagesState)
    builder.add_node("assistant", assistant)
    builder.add_node("tools", ToolNode([wiki_search, web_search, arvix_search]))
    builder.set_entry_point("assistant")
    builder.set_finish_point("assistant")
    builder.add_conditional_edges("assistant", tools_condition)
    builder.add_edge("tools", "assistant")
    return builder.compile()


# --- Advanced BasicAgent Class ---
class BasicAgent:
    def __init__(self):
        print("BasicAgent initialized.")
        self.client = Groq(api_key=os.environ.get("GROQ_API_KEY", ""))
        self.agent_prompt = (
            """You are a general AI assistant. I will ask you a question. Report your thoughts, and
            finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER].
            YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated
            list of numbers and/or strings.
            If you are asked for a number, don't use comma to write your number neither use units such as $
            or percent sign unless specified otherwise.
            If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the
            digits in plain text unless specified otherwise.
            If you are asked for a comma separated list, apply the above rules depending of whether the element
            to be put in the list is a number or a string."""
        )
        self.tool_chain = build_tool_graph(self.agent_prompt)

    def format_final_answer(self, answer: str) -> str:
        cleaned = " ".join(answer.split())
        match = re.search(r"FINAL ANSWER:\s*(.*)", cleaned, re.IGNORECASE)
        return f"FINAL ANSWER: {match.group(1).strip()}" if match else f"FINAL ANSWER: {cleaned}"

    def query_groq(self, question: str) -> str:
        full_prompt = f"{self.agent_prompt}\n\nQuestion: {question}"
        try:
            response = self.client.chat.completions.create(
                model="llama3-8b-8192",
                messages=[{"role": "user", "content": full_prompt}]
            )
            answer = response.choices[0].message.content
            print(f"[Groq Raw Response]: {answer}")
            return self.format_final_answer(answer).upper()
        except Exception as e:
            print(f"[Groq ERROR]: {e}")
            return self.format_final_answer("GROQ_ERROR")

    def query_tools(self, question: str) -> str:
        try:
            input_state = {
                "messages": [
                    SystemMessage(content=self.agent_prompt),
                    HumanMessage(content=question)
                ]
            }
            result = self.tool_chain.invoke(input_state)
            final_msg = result["messages"][-1].content
            print(f"[LangGraph Final Response]: {final_msg}")
            return self.format_final_answer(final_msg)
        except Exception as e:
            print(f"[LangGraph ERROR]: {e}")
            return self.format_final_answer("TOOL_ERROR")

    def __call__(self, question: str) -> str:
        print(f"Received question: {question[:50]}...")
        if "commutative" in question.lower():
            return self.check_commutativity()
        if self.maybe_reversed(question):
            print("Detected likely reversed riddle.")
            return self.solve_riddle(question)
        if "use tools" in question.lower():
            return self.query_tools(question)
        return self.query_groq(question)

    def check_commutativity(self):
        S = ['a', 'b', 'c', 'd', 'e']
        counter_example_elements = set()
        index = {'a': 0, 'b': 1, 'c': 2, 'd': 3, 'e': 4}
        self.operation_table = [
            ['a', 'b', 'c', 'b', 'd'],
            ['b', 'c', 'a', 'e', 'c'],
            ['c', 'a', 'b', 'b', 'a'],
            ['b', 'e', 'b', 'e', 'd'],
            ['d', 'b', 'a', 'd', 'c']
        ]
        for x in S:
            for y in S:
                x_idx = index[x]
                y_idx = index[y]
                if self.operation_table[x_idx][y_idx] != self.operation_table[y_idx][x_idx]:
                    counter_example_elements.add(x)
                    counter_example_elements.add(y)
        return self.format_final_answer(", ".join(sorted(counter_example_elements)))

    def maybe_reversed(self, text: str) -> bool:
        words = text.split()
        reversed_ratio = sum(
            1 for word in words if word[::-1].lower() in {
                "if", "you", "understand", "this", "sentence", "write",
                "opposite", "of", "the", "word", "left", "answer"
            }
        ) / len(words)
        return reversed_ratio > 0.3

    def solve_riddle(self, question: str) -> str:
        question = question[::-1]
        if "opposite of the word" in question:
            match = re.search(r"opposite of the word ['\"](\w+)['\"]", question)
            if match:
                word = match.group(1).lower()
                opposites = {
                    "left": "right", "up": "down", "hot": "cold",
                    "true": "false", "yes": "no", "black": "white"
                }
                opposite = opposites.get(word, f"UNKNOWN_OPPOSITE_OF_{word}")
                return f"FINAL ANSWER: {opposite.upper()}"
        return self.format_final_answer("COULD_NOT_SOLVE")

# --- Evaluation Logic ---
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:
        return "Please Login to Hugging Face with the button.", None

    api_url = DEFAULT_API_URL
    questions_url = f"{api_url}/questions"
    submit_url = f"{api_url}/submit"
    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"

    try:
        agent = BasicAgent()
    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()
        if not questions_data:
            return "Fetched questions list is empty or invalid format.", None
    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 produce any answers to submit.", pd.DataFrame(results_log)

    submission_data = {
        "username": username.strip(),
        "agent_code": agent_code,
        "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"Overall 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 ---
with gr.Blocks() as demo:
    gr.Markdown("# Basic Agent Evaluation Runner")
    gr.Markdown(
        """
        **Instructions:**
        1. Clone and customize your agent logic.
        2. Log in with Hugging Face.
        3. Click the button to run evaluation and submit your answers.
        """
    )
    gr.LoginButton()
    oauth_profile = gr.OAuthProfile()
    test_checkbox = gr.Checkbox(label="Enable Test Mode (Skip Submission)", value=False)
    run_button = gr.Button("Run Evaluation")
    status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
    run_button.click(
        fn=run_and_submit_all,
        inputs=[gr.OAuthProfile(), test_checkbox],
        outputs=[status_output, results_table]
    )

if __name__ == "__main__":
    print("Launching Gradio Interface...")
    demo.launch(debug=True, share=False)