semantic ai

语义ai lib

检索调查系统（RAG）的开源框架使用语义搜索来检索预期的结果并借助LLM（大语言模型）生成人类可读的对话响应。

语义AI库文档在这里文档

要求

Python 3.10+异步

安装

 # Using pip
$ python -m pip install semantic-ai

# Manual install
$ python -m pip install .

设置环境变量

在.env文件中设置凭据。仅给出一个连接器，一个索引器和一个LLM模型配置的凭据。其他字段放为空

 # Default
FILE_DOWNLOAD_DIR_PATH= # default directory name 'download_file_dir'
EXTRACTED_DIR_PATH= # default directory name 'extracted_dir'

# Connector (SharePoint, S3, GCP Bucket, GDrive, Confluence etc.,)
CONNECTOR_TYPE= " connector_name " # sharepoint
SHAREPOINT_CLIENT_ID= " client_id "
SHAREPOINT_CLIENT_SECRET= " client_secret "
SHAREPOINT_TENANT_ID= " tenant_id "
SHAREPOINT_HOST_NAME= ' <tenant_name>.sharepoint.com '
SHAREPOINT_SCOPE= ' https://graph.microsoft.com/.default '
SHAREPOINT_SITE_ID= " site_id "
SHAREPOINT_DRIVE_ID= " drive_id "
SHAREPOINT_FOLDER_URL= " folder_url " # /My_folder/child_folder/

# Indexer
INDEXER_TYPE= " <vector_db_name> " # elasticsearch, qdrant, opensearch
ELASTICSEARCH_URL= " <elasticsearch_url> " # give valid url
ELASTICSEARCH_USER= " <elasticsearch_user> " # give valid user
ELASTICSEARCH_PASSWORD= " <elasticsearch_password> " # give valid password
ELASTICSEARCH_INDEX_NAME= " <index_name> "
ELASTICSEARCH_SSL_VERIFY= " <ssl_verify> " # True or False

# Qdrant
QDRANT_URL= " <qdrant_url> "
QDRANT_INDEX_NAME= " <index_name> "
QDRANT_API_KEY= " <apikey> "

# Opensearch
OPENSEARCH_URL= " <opensearch_url> "
OPENSEARCH_USER= " <opensearch_user> "
OPENSEARCH_PASSWORD= " <opensearch_password> "
OPENSEARCH_INDEX_NAME= " <index_name> "

# LLM
LLM_MODEL= " <llm_model> " # llama, openai
LLM_MODEL_NAME_OR_PATH= " " # model name
OPENAI_API_KEY= " <openai_api_key> " # if using openai

# SQL
SQLITE_SQL_PATH= " <database_path> " # sqlit db path

# MYSQL
MYSQL_HOST= " <host_name> " # localhost or Ip Address
MYSQL_USER= " <user_name> "
MYSQL_PASSWORD= " <password> "
MYSQL_DATABASE= " <database_name> "
MYSQL_PORT= " <port> " # default port is 3306

方法1：加载.env文件。 env文件应具有凭据

%load_ext dotenv
%dotenv
%dotenv relative/or/absolute/path/to/.env

(or)

dotenv -f .env run -- python

方法2：

 from semantic_ai . config import Settings
settings = Settings ()

义务

1。导入模块

 import asyncio
import semantic_ai

2。要从给定源下载文件，请从下载的文件中提取内容，然后在给定向量数据库中索引提取的数据。

 await semantic_ai . download ()
await semantic_ai . extract ()
await semantic_ai . index ()

下载，提取和索引完成后，我们可以从索引矢量数据库中生成答案。该代码下面给出。

3。使用检索LLM模型从索引向量db产生答案。

 search_obj = await semantic_ai . search ()
query = ""
search = await search_obj . generate ( query )

假设作业运行很长时间，我们可以观看处理的文件数，文件数量失败以及存储在“ extract_dir_path/meta”目录中的文本文件中的文件名。

例子

连接源并获取连接对象。我们可以在示例文件夹中看到这一点。示例：SharePoint连接器

 from semantic_ai . connectors import Sharepoint

CLIENT_ID = '<client_id>'  # sharepoint client id
CLIENT_SECRET = '<client_secret>'  # sharepoint client seceret
TENANT_ID = '<tenant_id>'  # sharepoint tenant id
SCOPE = 'https://graph.microsoft.com/.default'  # scope
HOST_NAME = "<tenant_name>.sharepoint.com"  # for example 'contoso.sharepoint.com'

# Sharepoint object creation
connection = Sharepoint (
    client_id = CLIENT_ID ,
    client_secret = CLIENT_SECRET ,
    tenant_id = TENANT_ID ,
    host_name = HOST_NAME ,
    scope = SCOPE
)

结构

1。导入模块

 import asyncio
import semantic_ai

2。数据库连接

sqlite：

 from semantic_ai . connectors import Sqlite

file_path = < database_file_path >

sql = Sqlite ( sql_path = file_path )

mysql：

 from semantic_ai . connectors import Mysql

sql = Mysql (
    host = < host_name > ,
    user = < user_name > ,
    password = < password > ,
    database = < database > ,
    port = < port_number >  # 3306 is default port
)

3。使用检索LLM模型从数据库中生成答案。

 query = ""
search_obj = await semantic_ai . db_search ( query = query )

在服务器中运行

$ semantic_ai serve -f .env

INFO:     Loading environment from ' .env '
INFO:     Started server process [43973]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)

在http://127.0.0.1:8000/semantic-ai上打开浏览器

交互式API文档

现在访问http://127.0.0.1:8000/docs。您将看到自动交互式API文档（由Swagger UI提供）：