OceanBase 向量存储

OceanBase 数据库是一个分布式关系型数据库。它完全由蚂蚁集团开发。OceanBase 数据库构建在普通服务器集群之上。基于 Paxos 协议及其分布式架构，OceanBase 数据库提供高可用性和线性可扩展性。OceanBase 数据库不依赖于特定的硬件架构。

本笔记本详细介绍了如何在LlamaIndex中使用OceanBase向量存储功能。

设置

%pip install llama-index-vector-stores-oceanbase
%pip install llama-index
# choose dashscope as embedding and llm model, your can also use default openai or other model to test
%pip install llama-index-embeddings-dashscope
%pip install llama-index-llms-dashscope

使用Docker部署独立的OceanBase服务器

%docker run --name=ob433 -e MODE=slim -p 2881:2881 -d oceanbase/oceanbase-ce:4.3.3.0-100000142024101215

创建 ObVecClient

from pyobvector import ObVecClient

client = ObVecClient()
client.perform_raw_text_sql(
    "ALTER SYSTEM ob_vector_memory_limit_percentage = 30"
)

配置dashscope嵌入模型和LLM。

# set Embbeding model
import os
from llama_index.core import Settings
from llama_index.embeddings.dashscope import DashScopeEmbedding

# Global Settings
Settings.embed_model = DashScopeEmbedding()

# config llm model
from llama_index.llms.dashscope import DashScope, DashScopeGenerationModels

dashscope_llm = DashScope(
    model_name=DashScopeGenerationModels.QWEN_MAX,
    api_key=os.environ.get("DASHSCOPE_API_KEY", ""),
)

加载文档

from llama_index.core import (
    SimpleDirectoryReader,
    load_index_from_storage,
    VectorStoreIndex,
    StorageContext,
)
from llama_index.vector_stores.oceanbase import OceanBaseVectorStore

下载数据 & 加载数据

!mkdir -p 'data/paul_graham/'
!wget 'https://raw.githubusercontent.com/run-llama/llama_index/main/docs/examples/data/paul_graham/paul_graham_essay.txt' -O 'data/paul_graham/paul_graham_essay.txt'

# load documents
documents = SimpleDirectoryReader("./data/paul_graham/").load_data()

oceanbase = OceanBaseVectorStore(
    client=client,
    dim=1536,
    drop_old=True,
    normalize=True,
)

storage_context = StorageContext.from_defaults(vector_store=oceanbase)
index = VectorStoreIndex.from_documents(
    documents, storage_context=storage_context
)

查询索引

# set Logging to DEBUG for more detailed outputs
query_engine = index.as_query_engine(llm=dashscope_llm)
res = query_engine.query("What did the author do growing up?")
res.response

'Growing up, the author worked on two main activities outside of school: writing and programming. They wrote short stories, which they admits were not particularly good, lacking plot but containing characters with strong emotions. They also started programming at a young age, initially on an IBM 1401 computer using an early version of Fortran, though they found it challenging due to the limitations of punch card input and their lack of data to process. Their programming journey真正 took off when microcomputers became available, allowing them to write more interactive programs such as games, a rocket flight predictor, and a simple word processor.'

元数据过滤

OceanBase 向量存储支持在查询时以 =、>、<、!=、>=、<=、in、not in、like、IS NULL 的形式进行元数据过滤。

from llama_index.core.vector_stores import (
    MetadataFilters,
    MetadataFilter,
)

query_engine = index.as_query_engine(
    llm=dashscope_llm,
    filters=MetadataFilters(
        filters=[
            MetadataFilter(key="book", value="paul_graham", operator="!="),
        ]
    ),
    similarity_top_k=10,
)

res = query_engine.query("What did the author learn?")
res.response

'Empty Response'

删除文档

oceanbase.delete(documents[0].doc_id)

query_engine = index.as_query_engine(llm=dashscope_llm)
res = query_engine.query("What did the author do growing up?")
res.response

'Empty Response'