app.py

import gradio as gr
import json
import os
import time
import requests
from tools import TOOLS
import traceback
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")

MAX_RETRIES = 3
INFERENCE_TIMEOUT = 45  # seconds

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

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

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

    answers = []
    for i, task in enumerate(tasks):
        task_id = task.get("task_id", f"UNKNOWN-{i}")
        print(f"[INFO] Solving task {i+1}/{len(tasks)}: {task_id}")
        
        attempt = 0
        success = False
        while attempt < MAX_RETRIES and not success:
            attempt += 1
            try:
                start = time.time()
                result = solve_task(task, tools=TOOLS)
                duration = time.time() - start

                # Handle empty responses
                if not result.get("submitted_answer"):
                    raise ValueError("Empty model response")

                print(f"[INFO] Answer in {duration:.1f}s: {result['submitted_answer'][:100]}...")
                answers.append({
                    "task_id": task_id,
                    "submitted_answer": result["submitted_answer"].strip()
                })
                success = True

            except Exception as e:
                print(f"[ERROR] ❌ Task {task_id} failed after {attempt} attempts")
                traceback.print_exc()  # Shows full stack trace
                answers.append({
                    "task_id": task_id,
                    "submitted_answer": f"ERROR: {str(e)}"
                })

    with open("partial_answers.json", "w", encoding="utf-8") as f:
        json.dump(answers, f, indent=2)

    print("[INFO] Submitting answers to leaderboard...")

    try:
        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
            },
            timeout=60  # Set submission timeout
        )
        if res.ok:
            print("[INFO]  Submission successful")
            return json.dumps(res.json(), indent=2)
        else:
            return f"Error submitting: {res.status_code} - {res.text}"

    except requests.exceptions.Timeout:
        return " Submission timed out. Try again later."
    except Exception as e:
        return f" Submission failed with exception: {e}"

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