quick_test.py

import os
import requests
from langchain_core.messages import HumanMessage
from agent import build_graph
from huggingface_hub import hf_hub_download
import pyarrow.parquet as pq
from dotenv import load_dotenv

load_dotenv(override=True)

DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"

# Initialize agent
graph = build_graph()

# Fetch 1 question
resp = requests.get(f"{DEFAULT_API_URL}/questions")
questions = resp.json()[:1]

# Load ground truth
token = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACEHUB_API_TOKEN")
path = hf_hub_download(repo_id='gaia-benchmark/GAIA', filename='2023/validation/metadata.parquet', repo_type='dataset', token=token)
df = pq.read_table(path).to_pandas()
answer_map = dict(zip(df['task_id'], df['Final answer']))

# Test
q = questions[0]
task_id = q['task_id']
question = q['question']
ground_truth = answer_map.get(task_id, "NOT FOUND")

print(f"Question: {question[:100]}...")
print(f"Ground Truth: {ground_truth}")
print("-" * 40)

result = graph.invoke({"messages": [HumanMessage(content=question)]})
answer = result['messages'][-1].content
print(f"Agent Answer: {answer}")
print("-" * 40)

is_correct = answer.strip().lower() == str(ground_truth).strip().lower()
print(f"Correct: {is_correct}")