Hnswlib 是一个快速的近似最近邻搜索索引。它是一个轻量级、仅头文件的 C++ HNSW 实现，除 C++11 外没有其他依赖。Hnswlib 提供 Python 绑定。

%pip install llama-index
%pip install llama-index-vector-stores-hnswlib
%pip install llama-index-embeddings-huggingface
%pip install hnswlib

导入包依赖项

from llama_index.vector_stores.hnswlib import HnswlibVectorStore
from llama_index.core import (
    VectorStoreIndex,
    StorageContext,
    SimpleDirectoryReader,
)
from llama_index.embeddings.huggingface import HuggingFaceEmbedding

加载示例数据

!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'

读取数据

documents = SimpleDirectoryReader("./data/paul_graham/").load_data()
print(f"Total documents: {len(documents)}")
print(f"First document, id: {documents[0].doc_id}")
print(f"First document, hash: {documents[0].hash}")
print(
    "First document, text"
    f" ({len(documents[0].text)} characters):\n{'='*20}\n{documents[0].text[:360]} ..."
)

加载嵌入模型

embed_model = HuggingFaceEmbedding(
    model_name="sentence-transformers/all-MiniLM-L6-v2",
    normalize=True,
)

根据 Hnswlib.Index 参数创建 Hnswlib 向量存储对象

hnswlib_vector_store = HnswlibVectorStore.from_params(
    space="ip",
    dimension=embed_model._model.get_sentence_embedding_dimension(),
    max_elements=1000,
)

或者，你也可以自行创建 Hnswlib.Index 对象。

import hnswlib

index = hnswlib.Index(
    "ip", embed_model._model.get_sentence_embedding_dimension()
)
index.init_index(max_elements=1000)

hnswlib_vector_store = HnswlibVectorStore(index)

从文档构建索引

hnswlib_storage_context = StorageContext.from_defaults(
    vector_store=hnswlib_vector_store
)
hnswlib_index = VectorStoreIndex.from_documents(
    documents,
    storage_context=hnswlib_storage_context,
    embed_model=embed_model,
    show_progress=True,
)

查询索引

k = 5
query = "Before college I wrote what begginers should write."
hnswlib_vector_retriever = hnswlib_index.as_retriever(similarity_top_k=k)
nodes_with_scores = nodes_with_scores = hnswlib_vector_retriever.retrieve(
    query
)
for node in nodes_with_scores:
    print(f"Node {node.id_} | Score: {node.score:.3f} - {node.text[:120]}...")