使用Chainlit接入通义千问快速实现一个自然语言转sql语言的智能体

news2024/11/15 11:15:46

文本到 SQL

让我们构建一个简单的应用程序，帮助用户使用自然语言创建 SQL 查询。

最终结果预览

先决条件

此示例有额外的依赖项。你可以使用以下命令安装它们：

pip install chainlit openai

导入

应用程序

from openai import AsyncOpenAI

import chainlit as cl
cl.instrument_openai()
client = AsyncOpenAI(api_key="YOUR_OPENAI_API_KEY")

定义提示模板和 LLM 设置

代码

template = """SQL tables (and columns):
* Customers(customer_id, signup_date)
* Streaming(customer_id, video_id, watch_date, watch_minutes)

A well-written SQL query that {input}:
```"""


settings = {
    "model": "gpt-3.5-turbo",
    "temperature": 0,
    "max_tokens": 500,
    "top_p": 1,
    "frequency_penalty": 0,
    "presence_penalty": 0,
    "stop": ["```"],
}

添加辅助逻辑

在这里，我们用@on_message main装饰器装饰该函数，以告诉 Chainlit在每次用户发送消息时运行该main函数。

然后，我们在步骤中将文本包装到 SQL 逻辑中。

应用程序

@cl.set_starters
async def starters():
    return [
       cl.Starter(
           label=">50 minutes watched",
           message="Compute the number of customers who watched more than 50 minutes of video this month."
       )
    ]

@cl.on_message
async def main(message: cl.Message):
    stream = await client.chat.completions.create(
        messages=[
            {
                "role": "user",
                "content": template.format(input=message.content),
            }
        ], stream=True, **settings
    )

    msg = await cl.Message(content="", language="sql").send()

    async for part in stream:
        if token := part.choices[0].delta.content or "":
            await msg.stream_token(token)

    await msg.update()

完整代码如下：

text2sql.py

import base64
from io import BytesIO
from pathlib import Path

import chainlit as cl
from chainlit.element import ElementBased
from chainlit.input_widget import Select, Slider, Switch, TextInput
from openai import AsyncOpenAI

client = AsyncOpenAI()

author = "Tarzan"

template = """SQL tables (and columns):
* Customers(customer_id, signup_date)
* Streaming(customer_id, video_id, watch_date, watch_minutes)

A well-written SQL query that {input}:
```"""


def encode_image(image_path):
    with open(image_path, "rb") as image_file:
        return base64.b64encode(image_file.read()).decode("utf-8")


@cl.on_settings_update
async def on_settings_update(settings: cl.chat_settings):
    cl.user_session.set("settings", settings)


@cl.step(type="tool")
async def tool():
    # Simulate a running task
    await cl.sleep(2)

    return "Response from the tool!"


@cl.on_chat_start
async def start_chat():
    settings = await cl.ChatSettings(
        [TextInput(id="SystemPrompt", label="System Prompt", initial="You are a helpful assistant."),
         Select(
             id="Model",
             label="Model",
             values=["qwen-turbo", "qwen-plus", "qwen-max", "qwen-long"],
             initial_index=0,
         ),
         Slider(
             id="Temperature",
             label="Temperature",
             initial=1,
             min=0,
             max=2,
             step=0.1,
         ),
         Slider(
             id="MaxTokens",
             label="MaxTokens",
             initial=1000,
             min=1000,
             max=3000,
             step=100,
         ),
         Switch(id="Streaming", label="Stream Tokens", initial=True),
         ]
    ).send()
    cl.user_session.set("settings", settings)
    cl.user_session.set(
        "message_history",
        [{"role": "system", "content": settings["SystemPrompt"]}],
    )
    content = "你好，我是泰山AI智能客服，有什么可以帮助您吗？"
    msg = cl.Message(content=content, author=author)
    await msg.send()


@cl.on_message
async def on_message(message: cl.Message):
    settings = cl.user_session.get("settings")
    print('settings', settings)
    streaming = settings['Streaming']
    model = settings['Model']
    images = [file for file in message.elements if "image" in file.mime]
    files = [file for file in message.elements if "application" in file.mime]
    messages = cl.user_session.get("message_history")
    if files:
        files = files[:3]
        file_ids = []
        for file in files:
            file_object = await client.files.create(file=Path(file.path), purpose="file-extract")
            file_ids.append(f"fileid://{file_object.id}")
        flies_content = {
            "role": "system",
            "content": ",".join(file_ids)
        }
        messages.append(flies_content)
    if images and model in ["qwen-plus", "qwen-max"]:
        # Only process the first 3 images
        images = images[:3]
        images_content = [
            {
                "type": "image_url",
                "image_url": {
                    "url": f"data:{image.mime};base64,{encode_image(image.path)}"
                },
            }
            for image in images
        ]
        model = "qwen-vl" + model[4:]
        img_message = [
            {
                "role": "user",
                "content": [{"type": "text", "text": message.content}, *images_content],
            }
        ]
        messages = messages + img_message
    msg = cl.Message(content="", author=author)
    await msg.send()
    # Call the tool
    # tool_res = await tool
    messages.append({"role": "user", "content": template.format(input=message.content)})
    print('messages', messages)
    response = await client.chat.completions.create(
        model=model,
        messages=messages,
        temperature=settings['Temperature'],
        max_tokens=int(settings['MaxTokens']),
        stream=streaming
    )
    if streaming:
        async for part in response:
            if token := part.choices[0].delta.content or "":
                await msg.stream_token(token)
    else:
        if token := response.choices[0].message.content or "":
            await msg.stream_token(token)
    print('messages', messages)
    messages.append({"role": "assistant", "content": msg.content})
    cl.user_session.set("message_history", messages)
    await msg.update()