# pip install requests; export BAY_RUN_API_KEY=bay_live_...
import os
import requests

documents = ["GPU image generation", "OpenAI-compatible CPU embeddings", "Weather forecast"]
# The actual OpenAI-compatible route is POST /v1/rerank.
response = requests.post(f"{os.environ.get('BAY_RUN_API_BASE_URL', 'https://run.huggingbay.xyz/v1')}/rerank", headers={"Authorization": f"Bearer {os.environ['BAY_RUN_API_KEY']}"}, json={"model": "cross-encoder/ms-marco-TinyBERT-L2-v2", "query": "free embeddings API", "documents": documents, "top_n": 3}, timeout=30)
response.raise_for_status()
# relevance_score is a sigmoid-transformed ranking score, not a calibrated probability.
for result in response.json()["results"]:
    print(f"{result['relevance_score']:.3f}  {result['text']}")