将 Milvus 与 Jina AI 集成

本指南演示了如何使用 Jina AI 嵌入和 Milvus 来进行相似性搜索和检索任务。

谁是 Jina AI

Jina AI 成立于 2020 年，总部位于柏林，是一家开创性的 AI 公司，专注于通过其搜索基础技术革命人工智能的未来。专门从事多模态 AI，Jina AI 旨在通过其集成的组件套件（包括嵌入、重排序器、提示操作和核心基础设施）帮助企业和开发者利用多模态数据的力量创造价值并节省成本。

Jina AI 的前沿嵌入技术拥有顶级性能，具有 8192 令牌长度模型，非常适合全面的数据表示。提供多语言支持并与 OpenAI 等领先平台无缝集成，这些嵌入促进了跨语言应用。

Milvus 和 Jina AI 的嵌入

为了高效地存储和搜索这些嵌入以实现速度和规模，需要专门为此目的设计的特定基础设施。Milvus 是一个广为人知的先进开源向量数据库，能够处理大规模向量数据。Milvus 根据大量度量实现快速准确的向量（嵌入）搜索。其可扩展性允许无缝处理大量图像数据，即使数据集增长也能确保高性能搜索操作。

示例

Jina 嵌入已经集成到 PyMilvus 模型库中。现在，我们将演示代码示例来展示如何在实际中使用 Jina 嵌入。

在开始之前，我们需要为 PyMilvus 安装模型库。

$ pip install -U pymilvus
$ pip install "pymilvus[model]"

如果您使用 Google Colab，要启用刚刚安装的依赖项，您可能需要重启运行时。（点击屏幕顶部的"Runtime"菜单，从下拉菜单中选择"Restart session"）。

通用嵌入

Jina AI 的核心嵌入模型在理解详细文本方面表现出色，使其非常适合语义搜索、内容分类，从而支持高级情感分析、文本摘要和个性化推荐系统。

from pymilvus.model.dense import JinaEmbeddingFunction

jina_api_key = "&lt;YOUR_JINA_API_KEY&gt;"
ef = JinaEmbeddingFunction(
    "jina-embeddings-v3", 
    jina_api_key,
    task="retrieval.passage",
    dimensions=1024
)

query = "what is information retrieval?"
doc = "Information retrieval is the process of finding relevant information from a large collection of data or documents."

qvecs = ef.encode_queries([query])  # This method uses `retrieval.query` as the task
dvecs = ef.encode_documents([doc])  # This method uses `retrieval.passage` as the task

双语嵌入

Jina AI 的双语模型增强了多语言平台、全球支持和跨语言内容发现。专为德英和中英翻译设计，它们促进不同语言群体之间的理解，简化跨语言交互。

from pymilvus.model.dense import JinaEmbeddingFunction

jina_api_key = "&lt;YOUR_JINA_API_KEY&gt;"
ef = JinaEmbeddingFunction("jina-embeddings-v2-base-de", jina_api_key)

query = "what is information retrieval?"
doc = "Information Retrieval ist der Prozess, relevante Informationen aus einer großen Sammlung von Daten oder Dokumenten zu finden."

qvecs = ef.encode_queries([query])
dvecs = ef.encode_documents([doc])

代码嵌入

Jina AI 的代码嵌入模型提供了通过代码和文档进行搜索的能力。它支持英语和 30 种流行的编程语言，可用于增强代码导航、简化代码审查和自动化文档协助。

from pymilvus.model.dense import JinaEmbeddingFunction

jina_api_key = "&lt;YOUR_JINA_API_KEY&gt;"
ef = JinaEmbeddingFunction("jina-embeddings-v2-base-code", jina_api_key)

# Case1: Enhanced Code Navigation
# query: text description of the functionality
# document: relevant code snippet

query = "function to calculate average in Python."
doc = """
def calculate_average(numbers):
    total = sum(numbers)
    count = len(numbers)
    return total / count
"""

# Case2: Streamlined Code Review
# query: text description of the programming concept
# document: relevante code snippet or PR

query = "pull quest related to Collection"
doc = "fix:[restful v2] parameters of create collection ..."

# Case3: Automatic Documentation Assistance
# query: code snippet you need explanation
# document: relevante document or DocsString

query = "What is Collection in Milvus"
doc = """
In Milvus, you store your vector embeddings in collections. All vector embeddings within a collection share the same dimensionality and distance metric for measuring similarity.
Milvus collections support dynamic fields (i.e., fields not pre-defined in the schema) and automatic incrementation of primary keys.
"""

qvecs = ef.encode_queries([query])
dvecs = ef.encode_documents([doc])

使用 Jina 和 Milvus 进行语义搜索

通过强大的向量嵌入功能，我们可以将使用 Jina AI 模型检索的嵌入与 Milvus Lite 向量数据库相结合来执行语义搜索。

from pymilvus.model.dense import JinaEmbeddingFunction
from pymilvus import MilvusClient

jina_api_key = "&lt;YOUR_JINA_API_KEY&gt;"
DIMENSION = 1024  # `jina-embeddings-v3` supports flexible embedding sizes (32, 64, 128, 256, 512, 768, 1024), allowing for truncating embeddings to fit your application. 
ef = JinaEmbeddingFunction(
    "jina-embeddings-v3", 
    jina_api_key,
    task="retrieval.passage",
    dimensions=DIMENSION,
)


doc = [
    "In 1950, Alan Turing published his seminal paper, 'Computing Machinery and Intelligence,' proposing the Turing Test as a criterion of intelligence, a foundational concept in the philosophy and development of artificial intelligence.",
    "The Dartmouth Conference in 1956 is considered the birthplace of artificial intelligence as a field; here, John McCarthy and others coined the term 'artificial intelligence' and laid out its basic goals.",
    "In 1951, British mathematician and computer scientist Alan Turing also developed the first program designed to play chess, demonstrating an early example of AI in game strategy.",
    "The invention of the Logic Theorist by Allen Newell, Herbert A. Simon, and Cliff Shaw in 1955 marked the creation of the first true AI program, which was capable of solving logic problems, akin to proving mathematical theorems.",
]

dvecs = ef.encode_documents(doc) # This method uses `retrieval.passage` as the task

data = [
    {"id": i, "vector": dvecs[i], "text": doc[i], "subject": "history"}
    for i in range(len(dvecs))
]

milvus_client = MilvusClient("./milvus_jina_demo.db")
COLLECTION_NAME = "demo_collection"  # Milvus collection name
if milvus_client.has_collection(collection_name=COLLECTION_NAME):
    milvus_client.drop_collection(collection_name=COLLECTION_NAME)
milvus_client.create_collection(collection_name=COLLECTION_NAME, dimension=DIMENSION)

res = milvus_client.insert(collection_name=COLLECTION_NAME, data=data)

print(res["insert_count"])

关于 MilvusClient 的参数：

将 uri 设置为本地文件，例如 ./milvus.db，是最方便的方法，因为它会自动利用 Milvus Lite 将所有数据存储在此文件中。
如果您有大规模数据，可以在 docker 或 kubernetes 上设置性能更高的 Milvus 服务器。在这种设置中，请使用服务器 uri，例如 http://localhost:19530，作为您的 uri。
如果您想使用 Zilliz Cloud，Milvus 的完全托管云服务，请调整 uri 和 token，它们对应于 Zilliz Cloud 中的公共端点和 Api 密钥。

将所有数据存储在 Milvus 向量数据库中后，我们现在可以通过为查询生成向量嵌入并进行向量搜索来执行语义搜索。

queries = "What event in 1956 marked the official birth of artificial intelligence as a discipline?"
qvecs = ef.encode_queries([queries]) # This method uses `retrieval.query` as the task

res = milvus_client.search(
    collection_name=COLLECTION_NAME,  # target collection
    data=[qvecs[0]],  # query vectors
    limit=3,  # number of returned entities
    output_fields=["text", "subject"],  # specifies fields to be returned
)[0]

for result in res:
    print(result)

{'id': 1, 'distance': 0.8802614808082581, 'entity': {'text': "The Dartmouth Conference in 1956 is considered the birthplace of artificial intelligence as a field; here, John McCarthy and others coined the term 'artificial intelligence' and laid out its basic goals.", 'subject': 'history'}}

Jina Reranker

Jina AI 还提供重排序器，以在使用嵌入搜索后进一步提高检索质量。

from pymilvus.model.reranker import JinaRerankFunction

jina_api_key = "&lt;YOUR_JINA_API_KEY&gt;"

rf = JinaRerankFunction("jina-reranker-v1-base-en", jina_api_key)

query = "What event in 1956 marked the official birth of artificial intelligence as a discipline?"

documents = [
    "In 1950, Alan Turing published his seminal paper, 'Computing Machinery and Intelligence,' proposing the Turing Test as a criterion of intelligence, a foundational concept in the philosophy and development of artificial intelligence.",
    "The Dartmouth Conference in 1956 is considered the birthplace of artificial intelligence as a field; here, John McCarthy and others coined the term 'artificial intelligence' and laid out its basic goals.",
    "In 1951, British mathematician and computer scientist Alan Turing also developed the first program designed to play chess, demonstrating an early example of AI in game strategy.",
    "The invention of the Logic Theorist by Allen Newell, Herbert A. Simon, and Cliff Shaw in 1955 marked the creation of the first true AI program, which was capable of solving logic problems, akin to proving mathematical theorems.",
]

rf(query, documents)

[RerankResult(text="The Dartmouth Conference in 1956 is considered the birthplace of artificial intelligence as a field; here, John McCarthy and others coined the term 'artificial intelligence' and laid out its basic goals.", score=0.9370958209037781, index=1), RerankResult(text='The invention of the Logic Theorist by Allen Newell, Herbert A. Simon, and Cliff Shaw in 1955 marked the creation of the first true AI program, which was capable of solving logic problems, akin to proving mathematical theorems.', score=0.35420963168144226, index=3), RerankResult(text="In 1950, Alan Turing published his seminal paper, 'Computing Machinery and Intelligence,' proposing the Turing Test as a criterion of intelligence, a foundational concept in the philosophy and development of artificial intelligence.", score=0.3498658835887909, index=0), RerankResult(text='In 1951, British mathematician and computer scientist Alan Turing also developed the first program designed to play chess, demonstrating an early example of AI in game strategy.', score=0.2728956639766693, index=2)]

谁是 Jina AI​

Milvus 和 Jina AI 的嵌入​

示例​

通用嵌入​

双语嵌入​

代码嵌入​

使用 Jina 和 Milvus 进行语义搜索​

Jina Reranker​