app.py

import gradio as gr
import json
import os
import requests

from agent import solve_task, load_tasks
from mistral_hf_wrapper import MistralInference

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()

    if not tasks:
        return "No tasks loaded from metadata.jsonl. Make sure the file exists and is valid."

    print(f"[INFO] Loaded {len(tasks)} tasks from metadata.jsonl")

    answers = []
    for i, task in enumerate(tasks):
        print(f"[INFO] Solving task {i+1}/{len(tasks)}: {task.get('question_id', 'N/A')}")

        try:
            result = solve_task(task, model)

            if not result.get("submitted_answer"):  # Check empty responses
                print("[WARN] Empty model response detected")
                result["submitted_answer"] = "ERROR: Empty model response"
            else:
                print(f"Answer: {result['submitted_answer'][:100]}...")

            answers.append(result)

        except Exception as e:
            print(f"[ERROR] Task failed: {e}")
            answers.append({
                "question_id": task.get("question_id", "UNKNOWN"),
                "submitted_answer": f"ERROR: {str(e)}"
            })

    if not answers:
        return "No answers generated. Check model response."

    print("[INFO] Submitting answers to GAIA benchmark API...")
    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:
        print(" Submission successful.")
        return json.dumps(res.json(), indent=2)

    print(f"[ERROR] Submission failed: {res.status_code} - {res.text}")
    return f"Error submitting: {res.status_code} - {res.text}"

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