使用 Milvus 构建 RAG

在本教程中，我们将向您展示如何使用 Milvus 构建 RAG（检索-增强生成）管道。

RAG 系统结合了检索系统和生成模型，可根据给定提示生成新文本。系统首先使用 Milvus 从语料库中检索相关文档，然后使用生成模型根据检索到的文档生成新文本。

准备工作

依赖和环境

$ pip install --upgrade pymilvus openai requests tqdm

在本例中，我们将使用 OpenAI 作为 LLM。您应将api key OPENAI_API_KEY 作为环境变量。

import os

os.environ["OPENAI_API_KEY"] = "sk-\***\*\*\*\*\*\***"

准备数据

我们使用Milvus Documentation 2.4.x中的常见问题页面作为 RAG 中的私有知识，这对于简单的 RAG 管道来说是一个很好的数据源。

下载 zip 文件并将文档解压缩到milvus_docs 文件夹中。

$ wget https://github.com/milvus-io/milvus-docs/releases/download/v2.4.6-preview/milvus_docs_2.4.x_en.zip
$ unzip -q milvus_docs_2.4.x_en.zip -d milvus_docs

我们从milvus_docs/en/faq 文件夹中加载所有标记文件。对于每个文件，我们只需简单地使用 "#"来分隔文件中的内容，这样就能大致分隔出 markdown 文件中每个主要部分的内容。

from glob import glob

text_lines = []

for file_path in glob("milvus_docs/en/faq/\*.md", recursive=True):
with open(file_path, "r") as file:
file_text = file.read()

    text_lines += file_text.split("# ")

准备嵌入模型

我们对 OpenAI 客户端进行初始化，准备嵌入模型。

from openai import OpenAI

openai*client = OpenAI()

定义一个函数，使用 OpenAI 客户端生成文本嵌入。我们以text-embedding-3-small模型为例。

def emb_text(text):
    return (
        openai_client.embeddings.create(input=text, model="text-embedding-3-small")
        .data[0]
        .embedding
    )

生成一个测试嵌入，并打印其维度和前几个元素。

test_embedding = emb_text("This is a test")
embedding_dim = len(test_embedding)
print(embedding_dim)
print(test_embedding[:10])

1536
[0.00988506618887186, -0.005540902726352215, 0.0068014683201909065, -0.03810417652130127, -0.018254263326525688, -0.041231658309698105, -0.007651153020560741, 0.03220026567578316, 0.01892443746328354, 0.00010708322952268645]

将数据载入 Milvus

创建集合

from pymilvus import MilvusClient

milvus*client = MilvusClient(uri="./milvus_demo.db")

collection_name = "my_rag_collection"

检查集合是否已存在，如果已存在，则将其删除。

if milvus_client.has_collection(collection_name):
    milvus_client.drop_collection(collection_name)

使用指定参数创建新集合。

如果我们没有指定任何字段信息，Milvus 会自动创建一个主键的默认id 字段，以及一个用于存储向量数据的vector 字段。保留的 JSON 字段用于存储非模式定义的字段及其值。

milvus_client.create_collection(
    collection_name=collection_name,
    dimension=embedding_dim,
    metric_type="IP",  # Inner product distance
    consistency_level="Strong",  # Strong consistency level
)

插入数据

遍历文本行，创建嵌入，然后将数据插入 Milvus。

这里有一个新字段text ，它是Collection Schema 中的一个未定义字段。它将被自动添加到保留的 JSON Dynamic Field中，在高层次上可被视为普通字段。

from tqdm import tqdm

data = []

for i, line in enumerate(tqdm(text_lines, desc="Creating embeddings")):
data.append({"id": i, "vector": emb_text(line), "text": line})

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

Creating embeddings: 100%|██████████| 72/72 [00:27<00:00,  2.67it/s]





{'insert_count': 72,
 'ids': [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71],
 'cost': 0}

构建 RAG

为查询检索数据

让我们指定一个关于 Milvus 的常见问题。

question = "How is data stored in milvus?"

在集合中搜索该问题并检索语义前 3 个匹配项。

search_res = milvus_client.search(
    collection_name=collection_name,
    data=[
        emb_text(question)
    ],  # Use the `emb_text` function to convert the question to an embedding vector
    limit=3,  # Return top 3 results
    search_params={"metric_type": "IP", "params": {}},  # Inner product distance
    output_fields=["text"],  # Return the text field
)

让我们看看查询的搜索结果

import json

retrieved_lines_with_distances = [
(res["entity"]["text"], res["distance"]) for res in search_res[0]
]
print(json.dumps(retrieved_lines_with_distances, indent=4))

[
    [
        " Where does Milvus store data?\n\nMilvus deals with two types of data, inserted data and metadata. \n\nInserted data, including vector data, scalar data, and collection-specific schema, are stored in persistent storage as incremental log. Milvus supports multiple object storage backends, including [MinIO](https://min.io/), [AWS S3](https://aws.amazon.com/s3/?nc1=h_ls), [Google Cloud Storage](https://cloud.google.com/storage?hl=en#object-storage-for-companies-of-all-sizes) (GCS), [Azure Blob Storage](https://azure.microsoft.com/en-us/products/storage/blobs), [Alibaba Cloud OSS](https://www.alibabacloud.com/product/object-storage-service), and [Tencent Cloud Object Storage](https://www.tencentcloud.com/products/cos) (COS).\n\nMetadata are generated within Milvus. Each Milvus module has its own metadata that are stored in etcd.\n\n###",
        0.7883545756340027
    ],
    [
        "How does Milvus handle vector data types and precision?\n\nMilvus supports Binary, Float32, Float16, and BFloat16 vector types.\n\n- Binary vectors: Store binary data as sequences of 0s and 1s, used in image processing and information retrieval.\n- Float32 vectors: Default storage with a precision of about 7 decimal digits. Even Float64 values are stored with Float32 precision, leading to potential precision loss upon retrieval.\n- Float16 and BFloat16 vectors: Offer reduced precision and memory usage. Float16 is suitable for applications with limited bandwidth and storage, while BFloat16 balances range and efficiency, commonly used in deep learning to reduce computational requirements without significantly impacting accuracy.\n\n###",
        0.6757288575172424
    ],
    [
        "How much does Milvus cost?\n\nMilvus is a 100% free open-source project.\n\nPlease adhere to [Apache License 2.0](http://www.apache.org/licenses/LICENSE-2.0) when using Milvus for production or distribution purposes.\n\nZilliz, the company behind Milvus, also offers a fully managed cloud version of the platform for those that don't want to build and maintain their own distributed instance. [Zilliz Cloud](https://zilliz.com/cloud) automatically maintains data reliability and allows users to pay only for what they use.\n\n###",
        0.6421123147010803
    ]
]

使用 LLM 获取 RAG 响应

将检索到的文档转换为字符串格式。

context = "\n".join(
    [line_with_distance[0] for line_with_distance in retrieved_lines_with_distances]
)

为 Lanage 模型定义系统和用户提示。该提示与从 Milvus 检索到的文档组装在一起。

SYSTEM_PROMPT = """
Human: You are an AI assistant. You are able to find answers to the questions from the contextual passage snippets provided.
"""
USER_PROMPT = f"""
Use the following pieces of information enclosed in <context> tags to provide an answer to the question enclosed in <question> tags.
<context>
{context}
</context>
<question>
{question}
</question>
"""

使用 OpenAI ChatGPT 根据提示生成响应。

response = openai_client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages=[
        {"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user", "content": USER_PROMPT},
    ],
)
print(response.choices[0].message.content)

Milvus stores data in persistent storage as incremental logs, including inserted data (vector data, scalar data, and collection-specific schema) and metadata. Inserted data is stored in various object storage backends like MinIO, AWS S3, Google Cloud Storage, Azure Blob Storage, Alibaba Cloud OSS, and Tencent Cloud Object Storage. Metadata generated within Milvus is stored in etcd.

快速部署

要了解如何使用本教程启动在线演示，请参阅示例应用程序。