api_server.py

from fastapi import FastAPI
from pydantic import BaseModel
from typing import List, Dict, Any, Union
from datasets import load_dataset
import random
import os

app = FastAPI()

# Carga y filtra nivel 1 GAIA (validation split)
ds = load_dataset("gaia-benchmark/GAIA", "2023_level1", split="validation", trust_remote_code=True)
QUESTIONS = []
GROUND_TRUTH: Dict[str, str] = {}

for item in ds:
    task_id = str(item["task_id"])
    QUESTIONS.append({
        "task_id": task_id,
        "question": item["Question"]
    })
    GROUND_TRUTH[task_id] = str(item["Final answer"])

class AnswerItem(BaseModel):
    task_id: str
    submitted_answer: Union[str, int, float]

class Submission(BaseModel):
    username: str
    agent_code: str
    answers: List[AnswerItem]

class ScoreResponse(BaseModel):
    username: str
    score: float
    correct_count: int
    total_attempted: int
    message: str

@app.get("/questions")
def get_questions():
    # Devuelve las 20 preguntas aleatorias de nivel 1 cada vez
    chosen = random.sample(QUESTIONS, k=min(20, len(QUESTIONS)))
    return chosen

@app.post("/submit")
def submit(sub: Submission):
    correct = sum(
        1 for ans in sub.answers
        if GROUND_TRUTH.get(ans.task_id, "") == str(ans.submitted_answer).strip()
    )
    total = len(sub.answers)
    score = correct / total * 100 if total > 0 else 0.0
    return ScoreResponse(
        username=sub.username,
        score=score,
        correct_count=correct,
        total_attempted=total,
        message=f"Puntuación: {correct}/{total} = {score:.1f}%"
    )