app.py

import gradio as gr
import json
import os
import requests

from mistral_hf_wrapper import MistralInference
from agent import solve_task, load_tasks

API_URL = os.getenv("HF_MISTRAL_ENDPOINT")
API_TOKEN = os.getenv("HF_TOKEN")
USERNAME = os.getenv("HF_USERNAME")
CODE_LINK = os.getenv("HF_CODE_LINK")

def run_and_submit_all():
    model = MistralInference(api_url=API_URL, api_token=API_TOKEN)
    tasks = load_tasks()
    answers = []

    for task in tasks:
        try:
            answer = solve_task(task, model)
            answers.append({"task_id": task["task_id"], "submitted_answer": answer.strip()})
        except Exception as e:
            answers.append({"task_id": task["task_id"], "submitted_answer": f"ERROR: {str(e)}"})

    res = requests.post(
        "https://agents-course-unit4-scoring.hf.space/submit",
        headers={"Content-Type": "application/json"},
        json={"username": USERNAME, "agent_code": CODE_LINK, "answers": answers},
    )

    if res.ok:
        return json.dumps(res.json(), indent=2)
    return f"Error submitting: {res.status_code} - {res.text}"

gr.Interface(run_and_submit_all, inputs=[], outputs="textbox", title="GAIA Benchmark Agent Submission").launch()