Qdrant Vector Store: как сделать быстрый RAG-поиск и хранить embeddings

import os
from openai import OpenAI
from qdrant_client import QdrantClient
from qdrant_client.models import Distance, PointStruct, VectorParams

openai = OpenAI(
    base_url="https://speshu.ai/api/v1",
    api_key=os.environ["SPESHU_AI_API_KEY"],
)

texts = [
    "Qdrant хранит embeddings и payload для vector search.",
    "RAG использует поиск по документам перед ответом LLM.",
]

vectors = openai.embeddings.create(
    model="text-embedding-3-large",
    input=texts,
).data

client = QdrantClient(":memory:")
client.create_collection(
    collection_name="docs",
    vectors_config=VectorParams(size=3072, distance=Distance.COSINE),
)

client.upsert(
    collection_name="docs",
    points=[
        PointStruct(
            id=i,
            vector=item.embedding,
            payload={"text": text},
        )
        for i, (item, text) in enumerate(zip(vectors, texts))
    ],
)

query_vector = openai.embeddings.create(
    model="text-embedding-3-large",
    input="Как хранить embeddings для RAG?",
).data[0].embedding

hits = client.search(
    collection_name="docs",
    query_vector=query_vector,
    limit=2,
)

for hit in hits:
    print(hit.payload["text"], hit.score)