# pip install openai; export BAY_RUN_API_KEY=bay_live_...
import os
from openai import OpenAI

MODEL = "BAAI/bge-small-en-v1.5"
client = OpenAI(api_key=os.environ["BAY_RUN_API_KEY"], base_url="https://run.huggingbay.xyz/v1")
docs = ["Bay Run serves open models.", "Receipts bind result hashes.", "GPUs render images."]
query = "How are inference results verified?"

response = client.embeddings.create(model=MODEL, input=[query, *docs])
vectors = [item.embedding for item in response.data]
query_vector, *doc_vectors = vectors

# Bay Run normalizes vectors, so cosine similarity is their dot product.
def cosine(left, right):
    return sum(a * b for a, b in zip(left, right, strict=True))

ranked = sorted(((cosine(query_vector, vector), doc) for vector, doc in zip(doc_vectors, docs, strict=True)), reverse=True)
for score, document in ranked:
    print(f"{score:.3f}  {document}")