app.py

import os
import gradio as gr
import requests
import inspect
import pandas as pd
import spaces
from langchain_openai import ChatOpenAI
from langchain.agents import create_agent
from langchain.agents.middleware import ToolCallLimitMiddleware
from langchain_tavily import TavilySearch
from langchain_core.messages import SystemMessage

# (Keep Constants as is)
# --- Constants ---
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
@spaces.GPU(duration=1)
def zerogpu_function():
    return "ok"
# --- Basic Agent Definition ---
# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------

# class BasicAgent:
#     def __init__(self):
#         print("BasicAgent initialized.")
#     def __call__(self, question: str) -> str:
#         print(f"Agent received question (first 50 chars): {question[:50]}...")
#         fixed_answer = "This is a default answer."
#         print(f"Agent returning fixed answer: {fixed_answer}")
#         return fixed_answer
from datasets import load_dataset
from huggingface_hub import hf_hub_download
from langchain.tools import tool
from pathlib import Path
import base64
from langfuse import get_client
from langfuse.langchain import CallbackHandler
from openai import OpenAI
langfuse = get_client()
langfuse_handler = CallbackHandler()
gaia_dataset = load_dataset(
    "gaia-benchmark/GAIA",
    "2023_all",
    split="validation",
)

task_map = {
    item["task_id"]: item
    for item in gaia_dataset
}
omni_client = OpenAI(
    api_key=os.environ["DASHSCOPE_API_KEY"],
    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
)
@tool
def read_text_file(file_path: str) -> str:
    """Read a local text or Python file and return its contents."""

    with open(file_path, "r", encoding="utf-8") as f:
        return f.read()
    
@tool
def download_attachment(task_id: str, file_name: str) -> str:
    """
    Download the attachment associated with a GAIA task.

    Use this tool when the question has an attachment and you need
    the local file path in order to inspect the file.

    Args:
        task_id: The GAIA task ID.
        file_name: The name of the file to download.
    Returns:
        The local path of the downloaded attachment.
    """

    item = task_map.get(task_id)

    if item is None:
        return f"ERROR: task_id {task_id} was not found. Do not call this tool again."

    if not file_name or not file_name.strip():
        return "ERROR: This task has no attachment. Do not call this tool again."

    file_name = item.get("file_name")
    repo_file_path = item.get("file_path")

    if not file_name or not repo_file_path:
        return "ERROR: This task has no attachment. Do not call this tool again."

    local_path = hf_hub_download(
        repo_id="gaia-benchmark/GAIA",
        repo_type="dataset",
        filename=repo_file_path,
    )

    return local_path


from pypdf import PdfReader


@tool
def read_pdf(file_path: str) -> str:
    """
    Read a local PDF file and return its extracted text.

    Use this tool for .pdf attachments after download_attachment
    has returned the local file path.
    """

    reader = PdfReader(file_path)

    pages = []

    for page in reader.pages:
        text = page.extract_text()

        if text:
            pages.append(text)

    return "\n\n".join(pages)

@tool
def read_excel(file_path: str) -> str:
    """Read an Excel spreadsheet and return its contents."""

    import pandas as pd

    df = pd.read_excel(file_path)

    return df.to_string()

@tool
def analyze_media(file_path: str) -> str:
    """
    Inspect an image or audio attachment and return the factual
    information contained in it.

    This tool is for perception only.

    For images:
    - Describe visible objects, text, labels, numbers, positions,
      tables, diagrams, or other observable details accurately.
    - Preserve exact text and numbers when possible.
    - Do NOT solve the user's question.
    - Do NOT infer the final answer.
    - Do NOT perform domain reasoning beyond what is necessary
      to describe the media.

    For audio:
    - Transcribe the spoken content accurately.
    - Identify clearly observable speakers or sound events when useful.
    - Do NOT solve the user's question.
    - Do NOT infer the final answer.

    Args:
        file_path: Local path of the media file.

    Returns:
        A factual description or transcription of the media.
    """
    path = Path(file_path)

    if not path.exists():
        return f"ERROR: File does not exist: {file_path}"

    suffix = path.suffix.lower()

    # 读取文件 → Base64
    with open(path, "rb") as f:
        file_base64 = base64.b64encode(f.read()).decode("utf-8")

    # ---------- 图片 ----------
    if suffix in {".png", ".jpg", ".jpeg"}:
        media_prompt = """
                Inspect this image carefully.
        
                Return only the factual information that is directly observable
                in the image.
        
                Preserve exact text, labels, numbers, symbols, spatial relationships,
                and positions when relevant.
        
                Do not answer any external question.
                Do not solve the task.
                Do not infer a final answer.
                Your job is only to convert the visual information into an accurate
                text representation for another reasoning agent.
                """
        mime_type = {
            ".png": "image/png",
            ".jpg": "image/jpeg",
            ".jpeg": "image/jpeg",
        }[suffix]

        content = [
            {
                "type": "image_url",
                "image_url": {
                    "url": f"data:{mime_type};base64,{file_base64}"
                },
            },
            {
                "type": "text",
                "text": media_prompt.strip(),
            },
        ]

    # ---------- 音频 ----------
    elif suffix in {".mp3", ".wav"}:

        audio_format = suffix[1:]   # ".mp3" -> "mp3"
        media_prompt = """
        Listen to this audio carefully.

        Produce an accurate transcription of the spoken content.
        Preserve names, numbers, dates, and other important details.

        Do not answer questions about the audio.
        Do not solve any task.
        Your job is only to convert the audio information into text for
        another reasoning agent.
        """
        content = [
            {
                "type": "input_audio",
                "input_audio": {
                    "data": f"data:;base64,{file_base64}",
                    "format": audio_format,
                },
            },
            {
                "type": "text",
                "text": media_prompt.strip(),
            },
        ]

    else:
        return f"ERROR: Unsupported media type: {suffix}"

    # ---------- 调 Qwen Omni ----------
    completion = omni_client.chat.completions.create(
        model="qwen3.5-omni-flash",
        messages=[
            {
                "role": "user",
                "content": content,
            }
        ],
        modalities=["text"],
        stream=True,
        timeout=20,
    )

    # Omni 是 streaming，把文字拼起来
    result = ""

    for chunk in completion:
        if (
            chunk.choices
            and chunk.choices[0].delta.content
        ):
            result += chunk.choices[0].delta.content

    return result.strip()

search_tool = TavilySearch(
    max_results=3,
    topic="general",
    search_depth="basic",
    # include_domains=["wikipedia.org"],

)
model = ChatOpenAI(
    model="qwen3.5-plus",
    api_key=os.environ["DASHSCOPE_API_KEY"],
    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
    temperature=0,
    extra_body={"enable_thinking": False},
)

from pydantic import BaseModel, Field
from langchain.agents.structured_output import ToolStrategy


class FinalAnswer(BaseModel):
    answer: str = Field(
        description=(
            "Only the exact final answer requested by the question. "
            "No reasoning, no explanation, no prefix such as "
            "'Final answer:', and no extra commentary."
        )
    )

class BasicAgent:
    def __init__(self):
        print("BasicAgent initialized.")

        agent_model = model
        self.search_calls = 0

        @tool("web_search")
        def limited_search_tool(query: str) -> str:
            """Search Wikipedia for facts needed to answer the current question."""
            if self.search_calls >= 3:
                return (
                    "SEARCH_LIMIT_REACHED: You have already used all three allowed "
                    "web searches for this question. Answer using the information "
                    "already available."
                )

            self.search_calls += 1
            return str(search_tool.invoke({"query": query}))

        self.limited_search_tool = limited_search_tool

        self.agent = create_agent(
            model=agent_model,
            tools=[
                download_attachment,
                read_text_file,
                analyze_media,
                read_excel,
                read_pdf,
                self.limited_search_tool,
            ],
            middleware=[
                ToolCallLimitMiddleware(
                    tool_name="web_search",
                    run_limit=3,
                    exit_behavior="end",
                )
            ],
            response_format=ToolStrategy(FinalAnswer),
            system_prompt="""
You are an agent solving GAIA benchmark questions.

Follow this decision process strictly.

1. ATTACHMENT CHECK

Only enter the attachment workflow if the user message exactly includes "There is an attachment for this question.".

- Use download_attachment with the provided task_id.
- After obtaining the local file path, choose the appropriate tool:
  - Text or source-code file -> read_text_file
  - Excel file -> read_excel
  - PDF file -> read_pdf
  - Image or audio file -> analyze_media
- Use the information returned by the file tool to answer the original question.
- Do not invent or guess the contents of the attachment.
- Inspect the attachment before answering if the question depends on it.

2. NO ATTACHMENT

If the user message does NOT explicitly state that an attachment is present:

- Do NOT call download_attachment.
- Do NOT call read_text_file.
- Do NOT call read_excel.
- Do NOT call read_pdf.
- Do NOT call analyze_media.

Then determine whether the question requires web search or external information.

Use web_search only when external or up-to-date information is necessary.
Prefer answering from the available attachment or your own knowledge.
Use at most three web searches per question, and use fewer when possible.

Make each query precise and Wikipedia-focused.
Only make a follow-up search when the earlier Wikipedia results are insufficient.
Do not repeat equivalent or overly broad searches.



3. FINAL ANSWER FORMAT

Return only the exact final answer requested by the question.

Do not include explanations.
Do not include reasoning.
Do not include prefixes such as "Final answer:".
Follow the exact answer format requested by the question.

"""
        )

    def __call__(
    self,
    question: str,
    task_id: str = "",
    file_name: str = ""
    ) -> str:
        self.search_calls = 0

        content = question
        if file_name:
            content += f"""

        There is an attachment for this question.
        File name: {file_name}
        Task ID: {task_id}

        Use an appropriate file tool if you need to inspect the attachment.
        """
        print(f"Agent received question: {content[:100]}...")

        result = self.agent.invoke({
            "messages": [
                {
                    "role": "user",
                    "content": content
                }
            ],
        },
        config={
        "callbacks": [langfuse_handler]
        }
        )

        structured_response = result.get("structured_response")
        if structured_response is None:
            print("Web-search limit reached; generating a final answer without tools.")
            final_message = model.invoke(
                [
                    SystemMessage(
                        content=(
                            "The research phase is complete because the web-search "
                            "limit was reached. Do not call tools or browse. Answer "
                            "the original question using the search results already in "
                            "this conversation and your own knowledge.\n\n"
                            "FINAL ANSWER FORMAT — mandatory:\n"
                            "- Return only the exact final answer requested by the original question.\n"
                            "- Do not include reasoning, explanations, sources, or prefixes.\n"
                            "- Follow the exact requested format, including units, dates, "
                            "capitalization, ordering, and number of items."
                        )
                    ),
                    *result["messages"],
                ]
            )
            answer = str(final_message.content).strip()
            print(f"Agent answer after search limit: {answer}")
            return answer or "GIVE UP"

        answer = structured_response.answer.strip()

        print(f"Agent answer: {answer}")

        return answer







def run_and_submit_all( profile: gr.OAuthProfile | None):
    """
    Fetches all questions, runs the BasicAgent on them, submits all answers,
    and displays the results.
    """
    # --- Determine HF Space Runtime URL and Repo URL ---
    space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code

    if profile:
        username= f"{profile.username}"
        print(f"User logged in: {username}")
    else:
        print("User not logged in.")
        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"

    # 1. Instantiate Agent ( modify this part to create your agent)
    try:
        agent = BasicAgent()
    except Exception as e:
        print(f"Error instantiating agent: {e}")
        return f"Error initializing agent: {e}", None
    # In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
    print(agent_code)

    # 2. Fetch Questions
    print(f"Fetching questions from: {questions_url}")
    try:
        response = requests.get(questions_url, timeout=15)
        response.raise_for_status()
        questions_data = response.json()
        if not questions_data:
             print("Fetched questions list is empty.")
             return "Fetched questions list is empty or invalid format.", None
        print(f"Fetched {len(questions_data)} questions.")
    except requests.exceptions.RequestException as e:
        print(f"Error fetching questions: {e}")
        return f"Error fetching questions: {e}", None
    except requests.exceptions.JSONDecodeError as e:
         print(f"Error decoding JSON response from questions endpoint: {e}")
         print(f"Response text: {response.text[:500]}")
         return f"Error decoding server response for questions: {e}", None
    except Exception as e:
        print(f"An unexpected error occurred fetching questions: {e}")
        return f"An unexpected error occurred fetching questions: {e}", None

    # 3. Run your Agent
    results_log = []
    answers_payload = []
    print(f"Running agent on {len(questions_data)} questions...")
    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:
            print(f"Skipping item with missing task_id or question: {item}")
            continue
        try:
            submitted_answer = agent(question_text, task_id=task_id, file_name=item.get("file_name", ""))
            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:
             print(f"Error running agent on task {task_id}: {e}")
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})

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

    # 4. Prepare Submission 
    submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
    status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
    print(status_update)

    # 5. Submit
    print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
    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.')}"
        )
        print("Submission successful.")
        results_df = pd.DataFrame(results_log)
        return final_status, results_df
    except requests.exceptions.HTTPError as e:
        error_detail = f"Server responded with status {e.response.status_code}."
        try:
            error_json = e.response.json()
            error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
        except requests.exceptions.JSONDecodeError:
            error_detail += f" Response: {e.response.text[:500]}"
        status_message = f"Submission Failed: {error_detail}"
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df
    except requests.exceptions.Timeout:
        status_message = "Submission Failed: The request timed out."
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df
    except requests.exceptions.RequestException as e:
        status_message = f"Submission Failed: Network error - {e}"
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df
    except Exception as e:
        status_message = f"An unexpected error occurred during submission: {e}"
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df


# --- Build Gradio Interface using Blocks ---
with gr.Blocks() as demo:
    gr.Markdown("# Basic Agent Evaluation Runner")
    gr.Markdown(
        """
        **Instructions:**

        1.  Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc ...
        2.  Log in to your Hugging Face account using the button below. This uses your HF username for submission.
        3.  Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.

        ---
        **Disclaimers:**
        Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
        This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance for the delay process of the submit button, a solution could be to cache the answers and submit in a seperate action or even to answer the questions in async.
        """
    )

    gr.LoginButton()

    run_button = gr.Button("Run Evaluation & Submit All Answers")

    status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
    # Removed max_rows=10 from DataFrame constructor
    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)

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

if __name__ == "__main__":
    print("\n" + "-"*30 + " App Starting " + "-"*30)
    # Check for SPACE_HOST and SPACE_ID at startup for information
    space_host_startup = os.getenv("SPACE_HOST")
    space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup

    if space_host_startup:
        print(f"✅ SPACE_HOST found: {space_host_startup}")
        print(f"   Runtime URL should be: https://{space_host_startup}.hf.space")
    else:
        print("ℹ️  SPACE_HOST environment variable not found (running locally?).")

    if space_id_startup: # Print repo URLs if SPACE_ID is found
        print(f"✅ SPACE_ID found: {space_id_startup}")
        print(f"   Repo URL: https://huggingface.co/spaces/{space_id_startup}")
        print(f"   Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
    else:
        print("ℹ️  SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")

    print("-"*(60 + len(" App Starting ")) + "\n")

    print("Launching Gradio Interface for Basic Agent Evaluation...")
    demo.launch(debug=True, share=False)