基于Langchain的SQL交互Agent

Github链接

代码已开源，点点❤家人们，顶着一堆作业写的QAQ

https://github.com/SYuan03/llm4db

参考链接

官网

DashScope（阿里云）API

https://dashscope.console.aliyun.com/dashboard

好多地方都用了gpt-3.5-turbo的模型，4o又太贵，所以体验下Qwen的模型，体验额度还挺大的

Process

以下记录几个问题点

处理密码特殊字符

这步有点气人咳咳

了解到密码中含有@等特殊词汇需要使用

from urllib.parse import quote_plus

然后用了发现还是连不上，OK排查了一番后发现填成服务器密码了QAQ

改了之后终于连好了

StrOutputParser

OutputParser that parses LLMResult into the top likely string.

Create a new model by parsing and validating input data from keyword arguments.

Raises ValidationError if the input data cannot be parsed to form a valid model.

demo

db = SQLDatabase.from_uri(uri)

llm = Tongyi(model='qwen-max-1201', temperature=0)

# 定义回答问题的模板，把前面的问题、查询和结果传给LLM，然后解析输出
answer_prompt = PromptTemplate.from_template(
    """
    根据以下用户问题、对应的SQL查询以及SQL查询结果，回答用户的问题。
    问题: {question}
    SQL 查询: {query}
    SQL 查询结果: {result}
    答案:
    """
)
# 创建SQL查询链
write_query = create_sql_query_chain(llm, db)
# 创建SQL查询工具
execute_query = QuerySQLDataBaseTool(db=db)

# 定义链再传给LLM，然后使用StrOutputParser解析输出
answer = answer_prompt | llm | StrOutputParser()
chain = (
        RunnablePassthrough.assign(query=write_query).assign(
            result=itemgetter("query") | execute_query
        )
        | answer
)

res = chain.invoke({"question": "一共有多少仓库？有多少trending？"})
print(res)

There are 703 repositories.

Cool，验证下

想看下用了多少token的，结果好像数据没刷新

NextSteps

只试了下few shot

Prompting strategies

https://python.langchain.com/v0.1/docs/use_cases/sql/prompting/

看了下可以把表结构也一起传下，增强一下，加点描述之类的

Few-shot examples

https://python.langchain.com/v0.1/docs/use_cases/sql/prompting/#few-shot-examples

还可以搞点few shot

补充：StreamForHttps

部署接口在服务器上的时候发现

http://ip:8000/dbchat/stream能正常流式响应

但是经过nginx反向代理的https的https://ip:8001/dbchat/stream没法正常流式响应，直接~~依大托~~出来

之前的nginx配置如下

server {
    listen 8001 ssl;
    server_name <ip>;

    ssl_certificate /etc/nginx/ssl/certificate.crt;
    ssl_certificate_key /etc/nginx/ssl/private.key;
    ssl_trusted_certificate /etc/nginx/ssl/ca_bundle.crt;

    ssl_protocols TLSv1.2 TLSv1.3;
    ssl_prefer_server_ciphers on;
    ssl_ciphers "EECDH+AESGCM:EDH+AESGCM:AES256+EECDH:AES256+EDH";

    location / {
        proxy_pass http://localhost:8000;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;

    }

}

关闭缓存后修改如下

server {
    listen 8001 ssl;
    server_name <ip>;

    ssl_certificate /etc/nginx/ssl/certificate.crt;
    ssl_certificate_key /etc/nginx/ssl/private.key;
    ssl_trusted_certificate /etc/nginx/ssl/ca_bundle.crt;

    ssl_protocols TLSv1.2 TLSv1.3;
    ssl_prefer_server_ciphers on;
    ssl_ciphers "EECDH+AESGCM:EDH+AESGCM:AES256+EECDH:AES256+EDH";

    location /stream {
        proxy_pass http://localhost:8000/stream;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;

        proxy_buffering off;  # 关键配置，确保禁用代理缓冲
        proxy_http_version 1.1;
        chunked_transfer_encoding on;
    }

}

发现解决了问题

致谢

Paying tribute to the legendary software engineering warrior ——SEBugMaker！

And SYuan03 as well. (Ahem, paying tribute to oneself might seem a bit odd)

ChrisDing's bblog