目录
[1 安装](#1 安装)
[2 使用案例](#2 使用案例)
[2.1 文档搜索比如rag](#2.1 文档搜索比如rag)
[2.2 图片检索以图搜图功能](#2.2 图片检索以图搜图功能)
[3 集成](#3 集成)
Milvus 提供强大的数据建模功能，使您能够将非结构化或多模式数据组织成结构化的 Collections。它支持多种数据类型，适用于不同的属性模型，包括常见的数字和字符类型、各种向量类型、数组、集合和 JSON
1 安装
https://milvus.io/docs/zh/install_standalone-docker-compose.md
curl -SL https://github.com/docker/compose/releases/download/v2.30.3/docker-compose-linux-x86_64 -o /usr/local/bin/docker-compose
#将可执行权限赋予安装目标路径中的独立二进制文件
sudo chmod +x /usr/local/bin/docker-compose
sudo ln -s /usr/local/bin/docker-compose /usr/bin/docker-compose
wget https://github.com/milvus-io/milvus/releases/download/v2.6.9/milvus-standalone-docker-compose.yml -O docker-compose.yml
sudo docker compose up -d
Creating milvus-etcd ... done
Creating milvus-minio ... done
Creating milvus-standalone ... done
如何添加密码功能
#添加密码https://milvus.io/docs/zh/authenticate.md?tab=docker
...
common:
...
security:
authorizationEnabled: true
...
#默认密码root:Milvus
#pip install -U pymilvus
from pymilvus import MilvusClient
client = MilvusClient(
uri='http://localhost:19530', # replace with your own Milvus server address
token="root:Milvus"
) 2 使用案例
2.1 文档搜索比如rag
这里以云上模型为例，自建embed的话可以参考我之前的文章
python
import dashscope
from dashscope import TextEmbedding
from pymilvus import MilvusClient
dashscope.api_key=''
client = MilvusClient("http://127.0.0.1:19530")
def create_collection(collection_name):
if client.has_collection(collection_name="demo_collection"):
client.drop_collection(collection_name="demo_collection")
client.create_collection(
collection_name=collection_name,
dimension=1024, # The vectors we will use in this demo has 768 dimensions
metric_type="IP", # Inner product distance
consistency_level="Bounded", # Supported values are (`"Strong"`, `"Session"`, `"Bounded"`, `"Eventually"`). See https://milvus.io/docs/consistency.md#Consistency-Level for more details.
)
print(client.list_collections())
print(client.describe_collection(collection_name=client.list_collections()[0]))
def emb_text(text):
return (
TextEmbedding.call(
model="text-embedding-v4",
input=text,
dimension=1024
)
).output['embeddings']
def insert_data(collection_name, data):
res = client.insert(collection_name=collection_name, data=data)
return res
def search_data(collection_name):
query_vectors = emb_text(["深度学习"])[0]['embedding']
res = client.search(
collection_name=collection_name, # target collection
data=[query_vectors], # query vectors
limit=2, # number of returned entities
output_fields=["text", "subject"], # specifies fields to be returned
)
print(res)
if __name__ == '__main__':
# 创建集合
create_collection("demo_collection")
documents = [
"人工智能是计算机科学的一个分支",
"机器学习是实现人工智能的重要方法",
"深度学习是机器学习的一个子领域"
]
# test_embedding = emb_text("This is a test")
test_embedding = emb_text(documents)
embedding_dim = len(test_embedding)
# print(embedding_dim) 索引长度也就是定义的维度1024
# print(test_embedding) 索引内容
data = [
{"id": i, "vector": test_embedding[i]['embedding'], "text": documents[i], "subject": "demo"}
for i in range(len(documents))
]
print("Data has", len(data), "entities, each with fields: ", data[0].keys())
print("Vector dim:", len(data[0]["vector"]))
# 插入数据
res = insert_data("demo_collection", data)
print(res)
# 查询数据
search_data("demo_collection")可以通过相似值检索出相关内容
2.2 图片检索以图搜图功能
这里有两张柯基的图片一张金毛的图片
python
import base64
import os
import dashscope
from milvus.demo import client
dashscope.api_key=''
image = "https://dashscope.oss-cn-beijing.aliyuncs.com/images/256_1.png"
def create_collection(collection_name):
if client.has_collection(collection_name=collection_name):
client.drop_collection(collection_name=collection_name)
client.create_collection(
collection_name=collection_name,
auto_id=True,
vector_field_name="vector",
dimension=1152, # The vectors we will use in this demo has 768 dimensions
metric_type="IP", # Inner product distance
consistency_level="Bounded", # Supported values are (`"Strong"`, `"Session"`, `"Bounded"`, `"Eventually"`). See https://milvus.io/docs/consistency.md#Consistency-Level for more details.
)
print(client.list_collections())
print(client.describe_collection(collection_name=client.list_collections()[0]))
# def insert_data(collection_name, data):
# res = client.insert(collection_name=collection_name, data=data)
# return res
def image_to_base64(image_path):
with open(image_path, "rb") as image_file:
# 读取文件并转换为Base64
base64_image = base64.b64encode(image_file.read()).decode('utf-8')
# 设置图像格式
image_format = "png" # 根据实际情况修改，比如jpg、bmp 等
image_data = f"data:image/{image_format};base64,{base64_image}"
# 输入数据
input = [{'image': image_data}]
return input
#input = [{'image': image}]
def emb_text(input):
# 调用模型接口
resp = dashscope.MultiModalEmbedding.call(
model="tongyi-embedding-vision-plus",
input=input
).output['embeddings'][0]['embedding']
# print(resp)
# print(len(resp))
return resp
def search_data(input,collection_name):
query_vectors = emb_text(input)
res = client.search(
collection_name=collection_name, # target collection
data=[query_vectors], # query vectors
limit=2, # number of returned entities
output_fields=["filename"], # specifies fields to be returned
)
print(res)
if __name__ == '__main__':
create_collection("image_embeddings")
for file in os.listdir("../data"):
if file.endswith(".png"):
input = image_to_base64("../data/" + file)
image_embedding = emb_text(input)
res = client.insert(
"image_embeddings",
{"vector": image_embedding, "filename": file},
)
print(res)
search_data(image_to_base64("../data/柯基1.png"),"image_embeddings")
#emb_text(input)这里通过filename做演示，发现搜索柯基图片的时候返回的也是柯基，实际业务可以将图片地址返回前端使用以图搜相似图片
3 集成
后续更新
https://docs.llamaindex.org.cn/en/stable/examples/vector_stores/MilvusIndexDemo/