创建 Qdrant 客户端

%pip install llama-index-vector-stores-qdrant llama-index-readers-file llama-index-embeddings-fastembed llama-index-llms-openai

import logging
import sys
import os

import qdrant_client
from IPython.display import Markdown, display
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.core import StorageContext
from llama_index.vector_stores.qdrant import QdrantVectorStore
from llama_index.embeddings.fastembed import FastEmbedEmbedding
from llama_index.core import Settings

Settings.embed_model = FastEmbedEmbedding(model_name="BAAI/bge-base-en-v1.5")

如果是首次运行，请使用以下命令安装依赖项：

!pip install -U qdrant_client fastembed

设置 OpenAI 密钥以验证 LLM

按照以下步骤将 OpenAI API 密钥设置到 OPENAI_API_KEY 环境变量——

1. 使用终端

export OPENAI_API_KEY=your_api_key_here

2. 在 Jupyter Notebook 中使用 IPython 魔法命令

%env OPENAI_API_KEY=<YOUR_OPENAI_API_KEY>

3. 使用 Python 脚本

import os

os.environ["OPENAI_API_KEY"] = "your_api_key_here"

注意：通常建议将 API 密钥等敏感信息设置为环境变量，而不是硬编码到脚本中。

logging.basicConfig(stream=sys.stdout, level=logging.INFO)
logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))

下载数据

!mkdir -p 'data/paul_graham/'
!wget 'https://raw.githubusercontent.com/run-llama/llama_index/main/docs/examples/data/paul_graham/paul_graham_essay.txt' -O 'data/paul_graham/paul_graham_essay.txt'

加载文档

# load documents
documents = SimpleDirectoryReader("./data/paul_graham/").load_data()

构建 VectorStoreIndex

client = qdrant_client.QdrantClient(
    # you can use :memory: mode for fast and light-weight experiments,
    # it does not require to have Qdrant deployed anywhere
    # but requires qdrant-client >= 1.1.1
    # location=":memory:"
    # otherwise set Qdrant instance address with:
    # url="http://<host>:<port>"
    # otherwise set Qdrant instance with host and port:
    host="localhost",
    port=6333
    # set API KEY for Qdrant Cloud
    # api_key="<qdrant-api-key>",
)

vector_store = QdrantVectorStore(client=client, collection_name="paul_graham")
storage_context = StorageContext.from_defaults(vector_store=vector_store)
index = VectorStoreIndex.from_documents(
    documents,
    storage_context=storage_context,
)

查询索引

# set Logging to DEBUG for more detailed outputs
query_engine = index.as_query_engine()
response = query_engine.query("What did the author do growing up?")

display(Markdown(f"<b>{response}</b>"))

作者在上大学之前从事写作和编程。

# set Logging to DEBUG for more detailed outputs
query_engine = index.as_query_engine()
response = query_engine.query(
    "What did the author do after his time at Viaweb?"
)

display(Markdown(f"<b>{response}</b>"))

作者在 Viaweb 工作之后，安排为一家为客户做项目的团队从事自由职业工作。

异步构建 VectorStoreIndex

# To connect to the same event-loop,
# allows async events to run on notebook

import nest_asyncio

nest_asyncio.apply()

aclient = qdrant_client.AsyncQdrantClient(
    # you can use :memory: mode for fast and light-weight experiments,
    # it does not require to have Qdrant deployed anywhere
    # but requires qdrant-client >= 1.1.1
    location=":memory:"
    # otherwise set Qdrant instance address with:
    # uri="http://<host>:<port>"
    # set API KEY for Qdrant Cloud
    # api_key="<qdrant-api-key>",
)

vector_store = QdrantVectorStore(
    collection_name="paul_graham",
    client=client,
    aclient=aclient,
    prefer_grpc=True,
)
storage_context = StorageContext.from_defaults(vector_store=vector_store)
index = VectorStoreIndex.from_documents(
    documents,
    storage_context=storage_context,
    use_async=True,
)

异步查询索引

query_engine = index.as_query_engine(use_async=True)
response = await query_engine.aquery("What did the author do growing up?")

display(Markdown(f"<b>{response}</b>"))

作者从事短篇小说写作和编程，特别是在九年级时使用早期版本的 Fortran 在 IBM 1401 计算机上编程。后来，作者转向微型计算机，大约在 1980 年从 TRS-80 开始，并编写了简单游戏、程序和文字处理器。

# set Logging to DEBUG for more detailed outputs
query_engine = index.as_query_engine(use_async=True)
response = await query_engine.aquery(
    "What did the author do after his time at Viaweb?"
)

display(Markdown(f"<b>{response}</b>"))

作者在 Viaweb 之后参与共同创立了 Y Combinator。

混合搜索

在创建 Qdrant 索引时可以启用混合搜索。这里，我们使用 Qdrant 的 BM25 能力快速创建稀疏和稠密索引，以进行混合检索。

from qdrant_client import QdrantClient, AsyncQdrantClient
from llama_index.core import VectorStoreIndex
from llama_index.core import StorageContext
from llama_index.vector_stores.qdrant import QdrantVectorStore

client = QdrantClient(host="localhost", port=6333)
aclient = AsyncQdrantClient(host="localhost", port=6333)

vector_store = QdrantVectorStore(
    client=client,
    aclient=aclient,
    collection_name="paul_graham_hybrid",
    enable_hybrid=True,
    fastembed_sparse_model="Qdrant/bm25",
)

index = VectorStoreIndex.from_documents(
    documents,
    storage_context=StorageContext.from_defaults(vector_store=vector_store),
)

# retrieve 2 sparse, 2 dense, and filter down to 3 total hybrid results
query_engine = index.as_query_engine(
    vector_store_query_mode="hybrid",
    sparse_top_k=2,
    similarity_top_k=2,
    hybrid_top_k=3,
)

response = query_engine.query("What did the author do growing up?")
display(Markdown(f"<b>{response}</b>"))

保存与加载

要恢复索引，在大多数情况下，只需使用向量存储对象本身即可恢复。索引会由 Qdrant 自动保存。

loaded_index = VectorStoreIndex.from_vector_store(
    vector_store,
    # Embedding model should match the original embedding model
    # embed_model=Settings.embed_model
)