Google Vertex AI Vector Search¶

本 Notebook 演示了如何使用与 Google Cloud Vertex AI Vector Search 向量数据库相关的功能。

Google Vertex AI Vector Search，以前称为 Vertex AI Matching Engine，提供行业领先的高规模、低延迟向量数据库。这些向量数据库通常被称为向量相似性匹配或近似最近邻 (ANN) 服务。

注意：LlamaIndex 期望 Vertex AI Vector Search 端点和已部署的索引已经创建。创建空索引可能需要长达一分钟的时间，将索引部署到端点可能需要长达 30 分钟。

要了解如何创建索引，请参阅创建索引并将其部署到端点部分。
如果您已经部署了索引，请跳到从文本创建 VectorStore。

安装¶

如果您在 Colab 上打开此 Notebook，可能需要安装 LlamaIndex 🦙。

输入 [ ]

! pip install llama-index llama-index-vector-stores-vertexaivectorsearch llama-index-llms-vertex

! pip install llama-index llama-index-vector-stores-vertexaivectorsearch llama-index-llms-vertex

创建索引并将其部署到端点¶

• 本部分演示了创建新索引并将其部署到端点。

输入 [ ]

# TODO : Set values as per your requirements

# Project and Storage Constants
PROJECT_ID = "[your_project_id]"
REGION = "[your_region]"
GCS_BUCKET_NAME = "[your_gcs_bucket]"
GCS_BUCKET_URI = f"gs://{GCS_BUCKET_NAME}"

# The number of dimensions for the textembedding-gecko@003 is 768
# If other embedder is used, the dimensions would probably need to change.
VS_DIMENSIONS = 768

# Vertex AI Vector Search Index configuration
# parameter description here
# https://cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.MatchingEngineIndex#google_cloud_aiplatform_MatchingEngineIndex_create_tree_ah_index
VS_INDEX_NAME = "llamaindex-doc-index"  # @param {type:"string"}
VS_INDEX_ENDPOINT_NAME = "llamaindex-doc-endpoint"  # @param {type:"string"}

# TODO : 根据您的要求设置值 # 项目和存储常量 PROJECT_ID = "[your_project_id]" REGION = "[your_region]" GCS_BUCKET_NAME = "[your_gcs_bucket]" GCS_BUCKET_URI = f"gs://{GCS_BUCKET_NAME}" # textembedding-gecko@003 的维度数量为 768 # 如果使用其他嵌入模型，维度可能需要更改。 VS_DIMENSIONS = 768 # Vertex AI Vector Search 索引配置 # 参数说明在这里 # https://cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.MatchingEngineIndex#google_cloud_aiplatform_MatchingEngineIndex_create_tree_ah_index VS_INDEX_NAME = "llamaindex-doc-index" # @param {type:"string"} VS_INDEX_ENDPOINT_NAME = "llamaindex-doc-endpoint" # @param {type:"string"}

输入 [ ]

from google.cloud import aiplatform

aiplatform.init(project=PROJECT_ID, location=REGION)

from google.cloud import aiplatform aiplatform.init(project=PROJECT_ID, location=REGION)

创建 Cloud Storage 存储桶¶

输入 [ ]

# Create a bucket.
! gsutil mb -l $REGION -p $PROJECT_ID $GCS_BUCKET_URI

# 创建一个存储桶。 ! gsutil mb -l $REGION -p $PROJECT_ID $GCS_BUCKET_URI

创建空索引¶

注意：在创建索引时，应指定 "index_update_method" - BATCH_UPDATE 或 STREAM_UPDATE

批处理索引适用于您想批量更新索引的情况，数据是经过一段时间存储的，例如每周或每月处理的系统。

流式索引适用于您希望在新数据添加到数据存储时立即更新索引的情况，例如，如果您有一家书店，并希望尽快在线显示新库存。

选择哪种类型很重要，因为设置和要求不同。

有关配置索引的更多详细信息，请参阅官方文档和API 参考。

输入 [ ]

# NOTE : This operation can take upto 30 seconds

# check if index exists
index_names = [
    index.resource_name
    for index in aiplatform.MatchingEngineIndex.list(
        filter=f"display_name={VS_INDEX_NAME}"
    )
]

if len(index_names) == 0:
    print(f"Creating Vector Search index {VS_INDEX_NAME} ...")
    vs_index = aiplatform.MatchingEngineIndex.create_tree_ah_index(
        display_name=VS_INDEX_NAME,
        dimensions=VS_DIMENSIONS,
        distance_measure_type="DOT_PRODUCT_DISTANCE",
        shard_size="SHARD_SIZE_SMALL",
        index_update_method="STREAM_UPDATE",  # allowed values BATCH_UPDATE , STREAM_UPDATE
    )
    print(
        f"Vector Search index {vs_index.display_name} created with resource name {vs_index.resource_name}"
    )
else:
    vs_index = aiplatform.MatchingEngineIndex(index_name=index_names[0])
    print(
        f"Vector Search index {vs_index.display_name} exists with resource name {vs_index.resource_name}"
    )

# 注意：此操作可能需要长达 30 秒 # 检查索引是否存在 index_names = [ index.resource_name for index in aiplatform.MatchingEngineIndex.list( filter=f"display_name={VS_INDEX_NAME}" ) ] if len(index_names) == 0: print(f"正在创建 Vector Search 索引 {VS_INDEX_NAME} ...") vs_index = aiplatform.MatchingEngineIndex.create_tree_ah_index( display_name=VS_INDEX_NAME, dimensions=VS_DIMENSIONS, distance_measure_type="DOT_PRODUCT_DISTANCE", shard_size="SHARD_SIZE_SMALL", index_update_method="STREAM_UPDATE", # 允许的值 BATCH_UPDATE , STREAM_UPDATE ) print( f"Vector Search 索引 {vs_index.display_name} 已创建，资源名称为 {vs_index.resource_name}" ) else: vs_index = aiplatform.MatchingEngineIndex(index_name=index_names[0]) print( f"Vector Search 索引 {vs_index.display_name} 已存在，资源名称为 {vs_index.resource_name}" )

创建端点¶

要使用索引，您需要创建一个索引端点。它充当接受索引查询请求的服务器实例。端点可以是公共端点或私有端点。

让我们创建一个公共端点。

输入 [ ]

endpoint_names = [
    endpoint.resource_name
    for endpoint in aiplatform.MatchingEngineIndexEndpoint.list(
        filter=f"display_name={VS_INDEX_ENDPOINT_NAME}"
    )
]

if len(endpoint_names) == 0:
    print(
        f"Creating Vector Search index endpoint {VS_INDEX_ENDPOINT_NAME} ..."
    )
    vs_endpoint = aiplatform.MatchingEngineIndexEndpoint.create(
        display_name=VS_INDEX_ENDPOINT_NAME, public_endpoint_enabled=True
    )
    print(
        f"Vector Search index endpoint {vs_endpoint.display_name} created with resource name {vs_endpoint.resource_name}"
    )
else:
    vs_endpoint = aiplatform.MatchingEngineIndexEndpoint(
        index_endpoint_name=endpoint_names[0]
    )
    print(
        f"Vector Search index endpoint {vs_endpoint.display_name} exists with resource name {vs_endpoint.resource_name}"
    )

endpoint_names = [ endpoint.resource_name for endpoint in aiplatform.MatchingEngineIndexEndpoint.list( filter=f"display_name={VS_INDEX_ENDPOINT_NAME}" ) ] if len(endpoint_names) == 0: print( f"正在创建 Vector Search 索引端点 {VS_INDEX_ENDPOINT_NAME} ..." ) vs_endpoint = aiplatform.MatchingEngineIndexEndpoint.create( display_name=VS_INDEX_ENDPOINT_NAME, public_endpoint_enabled=True ) print( f"Vector Search 索引端点 {vs_endpoint.display_name} 已创建，资源名称为 {vs_endpoint.resource_name}" ) else: vs_endpoint = aiplatform.MatchingEngineIndexEndpoint( index_endpoint_name=endpoint_names[0] ) print( f"Vector Search 索引端点 {vs_endpoint.display_name} 已存在，资源名称为 {vs_endpoint.resource_name}" )

将索引部署到端点¶

使用索引端点，通过指定唯一的已部署索引 ID 来部署索引。

注意：此操作可能需要长达 30 分钟。

输入 [ ]

# check if endpoint exists
index_endpoints = [
    (deployed_index.index_endpoint, deployed_index.deployed_index_id)
    for deployed_index in vs_index.deployed_indexes
]

if len(index_endpoints) == 0:
    print(
        f"Deploying Vector Search index {vs_index.display_name} at endpoint {vs_endpoint.display_name} ..."
    )
    vs_deployed_index = vs_endpoint.deploy_index(
        index=vs_index,
        deployed_index_id=VS_INDEX_NAME,
        display_name=VS_INDEX_NAME,
        machine_type="e2-standard-16",
        min_replica_count=1,
        max_replica_count=1,
    )
    print(
        f"Vector Search index {vs_index.display_name} is deployed at endpoint {vs_deployed_index.display_name}"
    )
else:
    vs_deployed_index = aiplatform.MatchingEngineIndexEndpoint(
        index_endpoint_name=index_endpoints[0][0]
    )
    print(
        f"Vector Search index {vs_index.display_name} is already deployed at endpoint {vs_deployed_index.display_name}"
    )

# 检查端点是否存在 index_endpoints = [ (deployed_index.index_endpoint, deployed_index.deployed_index_id) for deployed_index in vs_index.deployed_indexes ] if len(index_endpoints) == 0: print( f"正在将 Vector Search 索引 {vs_index.display_name} 部署到端点 {vs_endpoint.display_name} ..." ) vs_deployed_index = vs_endpoint.deploy_index( index=vs_index, deployed_index_id=VS_INDEX_NAME, display_name=VS_INDEX_NAME, machine_type="e2-standard-16", min_replica_count=1, max_replica_count=1, ) print( f"Vector Search 索引 {vs_index.display_name} 已部署到端点 {vs_deployed_index.display_name}" ) else: vs_deployed_index = aiplatform.MatchingEngineIndexEndpoint( index_endpoint_name=index_endpoints[0][0] ) print( f"Vector Search 索引 {vs_index.display_name} 已部署到端点 {vs_deployed_index.display_name}" )

从文本创建向量存储¶

注意：如果您有现有的 Vertex AI Vector Search 索引和端点，可以使用以下代码进行分配

输入 [ ]

# TODO : replace 1234567890123456789 with your actual index ID
vs_index = aiplatform.MatchingEngineIndex(index_name="1234567890123456789")

# TODO : replace 1234567890123456789 with your actual endpoint ID
vs_endpoint = aiplatform.MatchingEngineIndexEndpoint(
    index_endpoint_name="1234567890123456789"
)

# TODO : 将 1234567890123456789 替换为您实际的索引 ID vs_index = aiplatform.MatchingEngineIndex(index_name="1234567890123456789") # TODO : 将 1234567890123456789 替换为您实际的端点 ID vs_endpoint = aiplatform.MatchingEngineIndexEndpoint( index_endpoint_name="1234567890123456789" )

输入 [ ]

# import modules needed
from llama_index.core import (
    StorageContext,
    Settings,
    VectorStoreIndex,
    SimpleDirectoryReader,
)
from llama_index.core.schema import TextNode
from llama_index.core.vector_stores.types import (
    MetadataFilters,
    MetadataFilter,
    FilterOperator,
)
from llama_index.llms.vertex import Vertex
from llama_index.embeddings.vertex import VertexTextEmbedding
from llama_index.vector_stores.vertexaivectorsearch import VertexAIVectorStore

# 导入所需模块 from llama_index.core import ( StorageContext, Settings, VectorStoreIndex, SimpleDirectoryReader, ) from llama_index.core.schema import TextNode from llama_index.core.vector_stores.types import ( MetadataFilters, MetadataFilter, FilterOperator, ) from llama_index.llms.vertex import Vertex from llama_index.embeddings.vertex import VertexTextEmbedding from llama_index.vector_stores.vertexaivectorsearch import VertexAIVectorStore

从纯文本创建不带元数据过滤器的简单向量存储¶

输入 [ ]

# setup storage
vector_store = VertexAIVectorStore(
    project_id=PROJECT_ID,
    region=REGION,
    index_id=vs_index.resource_name,
    endpoint_id=vs_endpoint.resource_name,
    gcs_bucket_name=GCS_BUCKET_NAME,
)

# set storage context
storage_context = StorageContext.from_defaults(vector_store=vector_store)

# 设置存储 vector_store = VertexAIVectorStore( project_id=PROJECT_ID, region=REGION, index_id=vs_index.resource_name, endpoint_id=vs_endpoint.resource_name, gcs_bucket_name=GCS_BUCKET_NAME, ) # 设置存储上下文 storage_context = StorageContext.from_defaults(vector_store=vector_store)

使用Vertex AI Embeddings作为嵌入模型¶

输入 [ ]

# configure embedding model
embed_model = VertexTextEmbedding(
    model_name="textembedding-gecko@003",
    project=PROJECT_ID,
    location=REGION,
)

# setup the index/query process, ie the embedding model (and completion if used)
Settings.embed_model = embed_model

# 配置嵌入模型 embed_model = VertexTextEmbedding( model_name="textembedding-gecko@003", project=PROJECT_ID, location=REGION, ) # 设置索引/查询过程，即嵌入模型（以及如果使用的话，补全模型） Settings.embed_model = embed_model

将向量和映射的文本块添加到向量存储¶

输入 [ ]

# Input texts
texts = [
    "The cat sat on",
    "the mat.",
    "I like to",
    "eat pizza for",
    "dinner.",
    "The sun sets",
    "in the west.",
]
nodes = [
    TextNode(text=text, embedding=embed_model.get_text_embedding(text))
    for text in texts
]

vector_store.add(nodes)

# 输入文本 texts = [ "猫坐在", "垫子上。", "我喜欢", "吃披萨做", "晚餐。", "太阳下山", "在西方。", ] nodes = [ TextNode(text=text, embedding=embed_model.get_text_embedding(text)) for text in texts ] vector_store.add(nodes)

运行相似性搜索¶

输入 [ ]

# define index from vector store
index = VectorStoreIndex.from_vector_store(
    vector_store=vector_store, embed_model=embed_model
)
retriever = index.as_retriever()

# 从向量存储定义索引 index = VectorStoreIndex.from_vector_store( vector_store=vector_store, embed_model=embed_model ) retriever = index.as_retriever()

输入 [ ]

response = retriever.retrieve("pizza")
for row in response:
    print(f"Score: {row.get_score():.3f} Text: {row.get_text()}")

response = retriever.retrieve("披萨") for row in response: print(f"得分: {row.get_score():.3f} 文本: {row.get_text()}")

Score: 0.703 Text: eat pizza for
Score: 0.626 Text: dinner.

添加带有元数据属性的文档并使用过滤器¶

输入 [ ]

# Input text with metadata
records = [
    {
        "description": "A versatile pair of dark-wash denim jeans."
        "Made from durable cotton with a classic straight-leg cut, these jeans"
        " transition easily from casual days to dressier occasions.",
        "price": 65.00,
        "color": "blue",
        "season": ["fall", "winter", "spring"],
    },
    {
        "description": "A lightweight linen button-down shirt in a crisp white."
        " Perfect for keeping cool with breathable fabric and a relaxed fit.",
        "price": 34.99,
        "color": "white",
        "season": ["summer", "spring"],
    },
    {
        "description": "A soft, chunky knit sweater in a vibrant forest green. "
        "The oversized fit and cozy wool blend make this ideal for staying warm "
        "when the temperature drops.",
        "price": 89.99,
        "color": "green",
        "season": ["fall", "winter"],
    },
    {
        "description": "A classic crewneck t-shirt in a soft, heathered blue. "
        "Made from comfortable cotton jersey, this t-shirt is a wardrobe essential "
        "that works for every season.",
        "price": 19.99,
        "color": "blue",
        "season": ["fall", "winter", "summer", "spring"],
    },
    {
        "description": "A flowing midi-skirt in a delicate floral print. "
        "Lightweight and airy, this skirt adds a touch of feminine style "
        "to warmer days.",
        "price": 45.00,
        "color": "white",
        "season": ["spring", "summer"],
    },
]

nodes = []
for record in records:
    text = record.pop("description")
    embedding = embed_model.get_text_embedding(text)
    metadata = {**record}
    nodes.append(TextNode(text=text, embedding=embedding, metadata=metadata))

vector_store.add(nodes)

# 输入带元数据的记录 records = [ { "description": "一条多功能深色水洗牛仔裤。" "采用耐磨棉质面料，经典直筒剪裁，轻松从休闲日过渡到更正式的场合。", "price": 65.00, "color": "blue", "season": ["fall", "winter", "spring"], }, { "description": "一件轻质亚麻纽扣衬衫，采用挺括的白色。" "透气面料和宽松剪裁，非常适合保持凉爽。", "price": 34.99, "color": "white", "season": ["summer", "spring"], }, { "description": "一件柔软厚实的针织毛衣，采用充满活力的森林绿色。" "宽松剪裁和舒适羊毛混纺，是气温下降时保暖的理想选择。", "price": 89.99, "color": "green", "season": ["fall", "winter"], }, { "description": "一件经典圆领 T 恤，采用柔软的杂蓝色。" "采用舒适的棉质平纹针织面料，这件 T 恤是衣橱必备品，适合每个季节。", "price": 19.99, "color": "blue", "season": ["fall", "winter", "summer", "spring"], }, { "description": "一条飘逸的中长裙，带有精致的碎花图案。" "轻盈透气，这条裙子为温暖的日子增添一丝女性风格。", "price": 45.00, "color": "white", "season": ["spring", "summer"], }, ] nodes = [] for record in records: text = record.pop("description") embedding = embed_model.get_text_embedding(text) metadata = {**record} nodes.append(TextNode(text=text, embedding=embedding, metadata=metadata)) vector_store.add(nodes)

运行带过滤器的相似性搜索¶

输入 [ ]

# define index from vector store
index = VectorStoreIndex.from_vector_store(
    vector_store=vector_store, embed_model=embed_model
)

# 从向量存储定义索引 index = VectorStoreIndex.from_vector_store( vector_store=vector_store, embed_model=embed_model )

输入 [ ]

# simple similarity search without filter
retriever = index.as_retriever()
response = retriever.retrieve("pants")

for row in response:
    print(f"Text: {row.get_text()}")
    print(f"   Score: {row.get_score():.3f}")
    print(f"   Metadata: {row.metadata}")

# 不带过滤器的简单相似性搜索 retriever = index.as_retriever() response = retriever.retrieve("裤子") for row in response: print(f"文本: {row.get_text()}") print(f" 得分: {row.get_score():.3f}") print(f" 元数据: {row.metadata}")

Text: A pair of well-tailored dress pants in a neutral grey. Made from a wrinkle-resistant blend, these pants look sharp and professional for workwear or formal occasions.
   Score: 0.669
   Metadata: {'price': 69.99, 'color': 'grey', 'season': ['fall', 'winter', 'summer', 'spring']}
Text: A pair of tailored black trousers in a comfortable stretch fabric. Perfect for work or dressier events, these trousers provide a sleek, polished look.
   Score: 0.642
   Metadata: {'price': 59.99, 'color': 'black', 'season': ['fall', 'winter', 'spring']}

输入 [ ]

# similarity search with text filter
filters = MetadataFilters(filters=[MetadataFilter(key="color", value="blue")])
retriever = index.as_retriever(filters=filters, similarity_top_k=3)
response = retriever.retrieve("denims")

for row in response:
    print(f"Text: {row.get_text()}")
    print(f"   Score: {row.get_score():.3f}")
    print(f"   Metadata: {row.metadata}")

# 带文本过滤器的相似性搜索 filters = MetadataFilters(filters=[MetadataFilter(key="color", value="blue")]) retriever = index.as_retriever(filters=filters, similarity_top_k=3) response = retriever.retrieve("牛仔裤") for row in response: print(f"文本: {row.get_text()}") print(f" 得分: {row.get_score():.3f}") print(f" 元数据: {row.metadata}")

Text: A versatile pair of dark-wash denim jeans.Made from durable cotton with a classic straight-leg cut, these jeans transition easily from casual days to dressier occasions.
   Score: 0.704
   Metadata: {'price': 65.0, 'color': 'blue', 'season': ['fall', 'winter', 'spring']}
Text: A denim jacket with a faded wash and distressed details. This wardrobe staple adds a touch of effortless cool to any outfit.
   Score: 0.667
   Metadata: {'price': 79.99, 'color': 'blue', 'season': ['fall', 'spring', 'summer']}

输入 [ ]

# similarity search with text and numeric filter
filters = MetadataFilters(
    filters=[
        MetadataFilter(key="color", value="blue"),
        MetadataFilter(key="price", operator=FilterOperator.GT, value=70.0),
    ]
)
retriever = index.as_retriever(filters=filters, similarity_top_k=3)
response = retriever.retrieve("denims")

for row in response:
    print(f"Text: {row.get_text()}")
    print(f"   Score: {row.get_score():.3f}")
    print(f"   Metadata: {row.metadata}")

# 带文本和数字过滤器的相似性搜索 filters = MetadataFilters( filters=[ MetadataFilter(key="color", value="blue"), MetadataFilter(key="price", operator=FilterOperator.GT, value=70.0), ] ) retriever = index.as_retriever(filters=filters, similarity_top_k=3) response = retriever.retrieve("牛仔裤") for row in response: print(f"文本: {row.get_text()}") print(f" 得分: {row.get_score():.3f}") print(f" 元数据: {row.metadata}")

Text: A denim jacket with a faded wash and distressed details. This wardrobe staple adds a touch of effortless cool to any outfit.
   Score: 0.667
   Metadata: {'price': 79.99, 'color': 'blue', 'season': ['fall', 'spring', 'summer']}

使用 Vertex AI Vector Search 和 Gemini Pro 解析、索引和查询 PDF¶

输入 [ ]

! mkdir -p ./data/arxiv/
! wget 'https://arxiv.org/pdf/1706.03762.pdf' -O ./data/arxiv/test.pdf

! mkdir -p ./data/arxiv/ ! wget 'https://arxiv.org/pdf/1706.03762.pdf' -O ./data/arxiv/test.pdf

E0501 00:56:50.842446801  266241 backup_poller.cc:127]                 Run client channel backup poller: UNKNOWN:pollset_work {created_time:"2024-05-01T00:56:50.841935606+00:00", children:[UNKNOWN:Bad file descriptor {created_time:"2024-05-01T00:56:50.841810434+00:00", errno:9, os_error:"Bad file descriptor", syscall:"epoll_wait"}]}
--2024-05-01 00:56:52--  https://arxiv.org/pdf/1706.03762.pdf
Resolving arxiv.org (arxiv.org)... 151.101.67.42, 151.101.195.42, 151.101.131.42, ...
Connecting to arxiv.org (arxiv.org)|151.101.67.42|:443... connected.
HTTP request sent, awaiting response... 301 Moved Permanently
Location: http://arxiv.org/pdf/1706.03762 [following]
--2024-05-01 00:56:52--  http://arxiv.org/pdf/1706.03762
Connecting to arxiv.org (arxiv.org)|151.101.67.42|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 2215244 (2.1M) [application/pdf]
Saving to: ‘./data/arxiv/test.pdf’

./data/arxiv/test.p 100%[===================>]   2.11M  --.-KB/s    in 0.07s   

2024-05-01 00:56:52 (31.5 MB/s) - ‘./data/arxiv/test.pdf’ saved [2215244/2215244]

输入 [ ]

# load documents
documents = SimpleDirectoryReader("./data/arxiv/").load_data()
print(f"# of documents = {len(documents)}")

# 加载文档 documents = SimpleDirectoryReader("./data/arxiv/").load_data() print(f"文档数量 = {len(documents)}")

# of documents = 15

输入 [ ]

# setup storage
vector_store = VertexAIVectorStore(
    project_id=PROJECT_ID,
    region=REGION,
    index_id=vs_index.resource_name,
    endpoint_id=vs_endpoint.resource_name,
    gcs_bucket_name=GCS_BUCKET_NAME,
)

# set storage context
storage_context = StorageContext.from_defaults(vector_store=vector_store)

# configure embedding model
embed_model = VertexTextEmbedding(
    model_name="textembedding-gecko@003",
    project=PROJECT_ID,
    location=REGION,
)

vertex_gemini = Vertex(
    model="gemini-pro",
    context_window=100000,
    temperature=0,
    additional_kwargs={},
)

# setup the index/query process, ie the embedding model (and completion if used)
Settings.llm = vertex_gemini
Settings.embed_model = embed_model

# 设置存储 vector_store = VertexAIVectorStore( project_id=PROJECT_ID, region=REGION, index_id=vs_index.resource_name, endpoint_id=vs_endpoint.resource_name, gcs_bucket_name=GCS_BUCKET_NAME, ) # 设置存储上下文 storage_context = StorageContext.from_defaults(vector_store=vector_store) # 配置嵌入模型 embed_model = VertexTextEmbedding( model_name="textembedding-gecko@003", project=PROJECT_ID, location=REGION, ) vertex_gemini = Vertex( model="gemini-pro", context_window=100000, temperature=0, additional_kwargs={}, ) # 设置索引/查询过程，即嵌入模型（以及如果使用的话，补全模型） Settings.llm = vertex_gemini Settings.embed_model = embed_model

输入 [ ]

# define index from vector store
index = VectorStoreIndex.from_documents(
    documents, storage_context=storage_context
)

# 从向量存储定义索引 index = VectorStoreIndex.from_documents( documents, storage_context=storage_context )

输入 [ ]

query_engine = index.as_query_engine()

query_engine = index.as_query_engine()

输入 [ ]

response = query_engine.query(
    "who are the authors of paper Attention is All you need?"
)

print(f"Response:")
print("-" * 80)
print(response.response)
print("-" * 80)
print(f"Source Documents:")
print("-" * 80)
for source in response.source_nodes:
    print(f"Sample Text: {source.text[:50]}")
    print(f"Relevance score: {source.get_score():.3f}")
    print(f"File Name: {source.metadata.get('file_name')}")
    print(f"Page #: {source.metadata.get('page_label')}")
    print(f"File Path: {source.metadata.get('file_path')}")
    print("-" * 80)

response = query_engine.query( "论文《Attention is All you need》的作者是谁？" ) print(f"响应:") print("-" * 80) print(response.response) print("-" * 80) print(f"源文档:") print("-" * 80) for source in response.source_nodes: print(f"样本文本: {source.text[:50]}") print(f"相关性得分: {source.get_score():.3f}") print(f"文件名: {source.metadata.get('file_name')}") print(f"页码: {source.metadata.get('page_label')}") print(f"文件路径: {source.metadata.get('file_path')}") print("-" * 80)

Response:
--------------------------------------------------------------------------------
The authors of the paper "Attention Is All You Need" are:

* Ashish Vaswani
* Noam Shazeer
* Niki Parmar
* Jakob Uszkoreit
* Llion Jones
* Aidan N. Gomez
* Łukasz Kaiser
* Illia Polosukhin
--------------------------------------------------------------------------------
Source Documents:
--------------------------------------------------------------------------------
Sample Text: Provided proper attribution is provided, Google he
Relevance score: 0.720
File Name: test.pdf
Page #: 1
File Path: /home/jupyter/llama_index/docs/docs/examples/vector_stores/data/arxiv/test.pdf
--------------------------------------------------------------------------------
Sample Text: length nis smaller than the representation dimensi
Relevance score: 0.678
File Name: test.pdf
Page #: 7
File Path: /home/jupyter/llama_index/docs/docs/examples/vector_stores/data/arxiv/test.pdf
--------------------------------------------------------------------------------

---

清理¶

请在运行实验后删除 Vertex AI Vector Search 索引和索引端点，以避免产生额外费用。请注意，只要端点正在运行，就会收费。

⚠️ 注意：启用 `CLEANUP_RESOURCES` 标志会删除 Vector Search 索引、索引端点和 Cloud Storage 存储桶。请谨慎运行。

输入 [ ]

CLEANUP_RESOURCES = False

CLEANUP_RESOURCES = False

• 取消部署索引并删除索引端点

输入 [ ]

if CLEANUP_RESOURCES:
    print(
        f"Undeploying all indexes and deleting the index endpoint {vs_endpoint.display_name}"
    )
    vs_endpoint.undeploy_all()
    vs_endpoint.delete()

if CLEANUP_RESOURCES: print( f"正在取消部署所有索引并删除索引端点 {vs_endpoint.display_name}" ) vs_endpoint.undeploy_all() vs_endpoint.delete()

• 删除索引

输入 [ ]

if CLEANUP_RESOURCES:
    print(f"Deleting the index {vs_index.display_name}")
    vs_index.delete()

if CLEANUP_RESOURCES: print(f"正在删除索引 {vs_index.display_name}") vs_index.delete()

• 删除 Cloud Storage 存储桶中的内容

输入 [ ]

if CLEANUP_RESOURCES and "GCS_BUCKET_NAME" in globals():
    print(f"Deleting contents from the Cloud Storage bucket {GCS_BUCKET_NAME}")

    shell_output = ! gsutil du -ash gs://$GCS_BUCKET_NAME
    print(shell_output)
    print(
        f"Size of the bucket {GCS_BUCKET_NAME} before deleting = {' '.join(shell_output[0].split()[:2])}"
    )

    # uncomment below line to delete contents of the bucket
    # ! gsutil -m rm -r gs://$GCS_BUCKET_NAME

if CLEANUP_RESOURCES and "GCS_BUCKET_NAME" in globals(): print(f"正在删除 Cloud Storage 存储桶 {GCS_BUCKET_NAME} 中的内容") shell_output = ! gsutil du -ash gs://$GCS_BUCKET_NAME print(shell_output) print( f"删除前存储桶 {GCS_BUCKET_NAME} 的大小 = {' '.join(shell_output[0].split()[:2])}" ) # 取消注释下一行可删除存储桶中的内容 # ! gsutil -m rm -r gs://$GCS_BUCKET_NAME