import os from pathlib import Path import pytest import requests from dotenv import load_dotenv load_dotenv(Path(__file__).resolve().with_name(".env")) pytestmark = pytest.mark.live_api def test_hf(): token = os.getenv("HUGGINGFACE_TOKEN") if not token: pytest.skip("HUGGINGFACE_TOKEN not configured") model_id = "meta-llama/Llama-3.1-8B-Instruct" response = requests.post( f"https://api-inference.huggingface.co/models/{model_id}", headers={"Authorization": f"Bearer {token}"}, json={ "inputs": "Hello", "parameters": {"max_new_tokens": 10}, }, timeout=30, ) assert response.status_code == 200, response.text