【自然语言处理与大模型】向量数据库：Milvus使用指南

Milvus 是一个开源的向量数据库，Milvus Lite 是 Milvus 向量数据库的轻量级版本，能为 AI 应用提供向量相似性搜索功能。它非常适合用于快速原型开发、资源有限的环境。

安装

bash 复制代码

# 安装 milvus 要求Python 3.8+
pip install "pymilvus[model]"

使用

python 复制代码

from pymilvus import MilvusClient
from pymilvus import model



# 创建一个客户端
client = MilvusClient("./milvus_demo.db")  # 指定一个存储所有数据的文件路径

# 加载本地的词嵌入向量模型
"""
如果你安装了，model依赖则默认值为all-MiniLM-L6-v2
"""
sentence_transformer_ef = model.dense.SentenceTransformerEmbeddingFunction(
    model_name='all-MiniLM-L6-v2', # 指定模型路径
    device='cpu' # 指定要使用的设备，例如"cpu"或"cuda:0"
)


# 创建一个集合
# 检查名为"demo_collection"的集合是否存在
if client.has_collection(collection_name="demo_collection"):
    # 如果存在则删除该集合
    client.drop_collection(collection_name="demo_collection")

client.create_collection(
    collection_name="demo_collection",
    dimension=768,  # 指定向量的维度
)

docs = [
    "Artificial intelligence was founded as an academic discipline in 1956.",
    "Alan Turing was the first person to conduct substantial research in AI.",
    "Born in Maida Vale, London, Turing was raised in southern England.",
]

# 将文档向量化
vectors = sentence_transformer_ef.encode_documents(docs)

# 打印embedding后的文档
print("Embeddings:", vectors)
print("Dim:", sentence_transformer_ef.dim, vectors[0].shape)

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




# 插：将数据插入向量数据库
client.insert("demo_collection", data)



# 查：search相似度搜索 或 query关键字匹配
res = client.search(
    collection_name="demo_collection",
    data=[vectors[0]],
    filter="subject == 'history'",   # 过滤条件
    limit=2,  # 最相似的2条
    output_fields=["text", "subject"]
)
print(res)

res = client.query(
    collection_name="demo_collection",
    filter="subject == 'history'",     # 过滤条件
    output_fields=["text", "subject"]  # 只展示的字段
)
print(res)




# 改：修改其中id=1文档
update_docs = ["Artificial intelligence research began in mid-20th century"]
update_vectors = sentence_transformer_ef.encode_documents(update_docs)
update_data = [{
    "id": 1,  # 指定要更新的文档ID
    "text": "Artificial intelligence research began in mid-20th century",  # 新文本
    "vector": update_vectors[0],   # 新向量
    "subject": "computer_science"  # 新分类
}]
res = client.upsert(
    collection_name="demo_collection",
    data=[update_data]  # 注意数据需要是列表格式
)
print(res)




# 删：删除一个文档
res = client.delete(
    collection_name="demo_collection",
    filter="subject == 'history'"  # 过滤条件
)
print(res)