使用Azure OpenAI服务API - 通过函数调用进行流媒体处理

使用Azure OpenAI进行流式处理的功能调用

介绍

功能调用使您能够将模型（如GPT-4）与外部工具和系统进行链接。这对各种目的都是有益的，包括用额外功能增强人工智能助理，并在您的应用程序和模型之间创建稳健的集成。

函数调用在聊天完成 API、助手 API 和批处理 API 中均受支持。然而，这份简短指南侧重于使用聊天完成 API 进行函数调用，并使用 Azure OpenAI 服务进行流式处理。

单个工具/功能调用示例

在这个指南中，我们创建了一个样本函数，用来调用雅虎金融API，来检索给定股票代码的股票价格。

在高层次上，使用函数可以分为三个步骤：

使用聊天完成API与您的函数和用户的输入。
利用模型的响应来调用您的API或函数。
再次调用聊天完成API，将您的函数的响应加入其中以获取最终的回复。

注意：创建Python虚拟环境或Conda环境超出本文档范围。请确保已安装所需的库使用pip.

注意：确保提供Azure OpenAI端点、API密钥、API版本和模型部署名称。

有打印语句和注释的打印语句可以帮助理解代码。如果需要深入了解，取消注释这些打印语句。例如，分块中的增量是什么样子。

import os
import openai
from dotenv import load_dotenv
import json
import yfinance as yf  
from datetime import datetime

# load environment variable.
load_dotenv()

# Load OpenAI variables. Make sure to provide those variables as per your environment.
azure_endpoint: str = os.environ["AZURE_OPENAI_ENDPOINT"] #PLACEHOLDER FOR YOUR AZURE OPENAI ENDPOINT
azure_openai_api_key: str = os.environ["AZURE_OPENAI_API_KEY"] #PLACEHOLDER FOR YOUR AZURE OPENAI KEY
azure_openai_api_version: str = os.environ["AZURE_OPENAI_API_VERSION"] # API VERSION PLACE HOLDER
azure_deployment: str = os.environ["OPENAI_DEPLOYMENT_NAME_GPT4o"] # AZURE OPENAI DEPLOYMENT NAME


# Define the function for the model
tools = [
        {
            "type": "function",
            "function": {
                "name": "get_stock_price",
                "description": "Retrieve lowest and highest stock price for a given stock symbol and date",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "symbol": {
                            "type": "string",
                            "description": "Stock symbol, for example MSFT for Microsoft"
                        }
                },
                    "required": ["symbol"],
                },
            }   
        }
    ]

# Initialize Azure OpenAI client with key-based authentication
client = openai.AzureOpenAI(
    azure_endpoint = azure_endpoint,
    api_key = azure_openai_api_key,
    api_version = azure_openai_api_version,
)

# create a funtion
def get_stock_price(symbol):
    print(f"Getting stock price for {symbol}")
    
    ticker = yf.Ticker(symbol)
    info = ticker.info
   
    date = datetime.today().strftime('%Y-%m-%d')  
    return json.dumps({  
        "symbol": symbol,  
        "date": date,
        "current price": info['currentPrice'], 
        "low price": info['dayLow'],  
        "high price": info['dayHigh'],
        "opening price": info['open'],
        "link": f"https://finance.yahoo.com/quote/{symbol}/"
    }
    )
    
# initial user message    
messages = []
tool_call_accumulator = ""
messages.append({"role": "user", "content": "what is the price of tesla stock"}) # single function call


# First API call: Ask the model to use the function
response = client.chat.completions.create(
    model= azure_deployment,
    messages=messages,
    tools=tools,
    tool_choice="auto",
    stream=True  # this time, we set stream=True
)

# process the model 
for chunk in response:
    if len(chunk.choices) > 0:
        delta = chunk.choices[0].delta
        #print(f"delta: {delta}") # print delta from the chunk for learning
        
        if delta.tool_calls:
                if(delta.tool_calls[0].function.name):
                    tool_name = chunk.choices[0].delta.tool_calls[0].function.name
                    tool_id = chunk.choices[0].delta.tool_calls[0].id
                    messages.append(delta)
                
                if(chunk.choices[0].delta.tool_calls[0].function.arguments):    
                    tool_call_accumulator+=delta.tool_calls[0].function.arguments

# print function related outputs                    
print(f"tool_name:{tool_name}")   
print(f"tool_id:{tool_id}")    
print(f"tool_call_accumulator:{tool_call_accumulator}")   



# handle the function call
if tool_name == "get_stock_price":
    function_args=json.loads(tool_call_accumulator)
    #print(function_args.get("symbol"))
    stock_price_data = get_stock_price(symbol=function_args.get("symbol"))
    #print(f"stock_price_data: {stock_price_data}")
    messages.append({
                    "tool_call_id": tool_id,
                    "role": "tool",
                    "name": tool_name,
                    "content": stock_price_data,
                    })

#print(f"messages: {messages}") 
print("=============================================")  
    
#  second API call: Get the final response with stream
response = client.chat.completions.create(
    model= azure_deployment,
    messages=messages,
    temperature=0.7,
    stream=True  # this time, we set stream=True
)


for chunk in response:
    if len(chunk.choices) > 0:
        delta = chunk.choices[0].delta
        #print(f"delta: {delta}")
        
       
        if delta.content:
            print(delta.content, end="", flush=True)

输出：

PS C:\Development\openai\chatgpt-demo> & C:/Users/muafzal/Miniconda3/envs/.conda39/python.exe c:/Development/openai/chatgpt-demo/Azure_OpenAI_FunctionCall_With_Stream.py
tool_name:get_stock_price
tool_id:call_grJz8BFcBcfIQD8HAdDuuQ6U
tool_call_accumulator:{"symbol":"TSLA"}
Getting stock price for TSLA
=============================================
As of the latest data on October 6, 2024, the price of Tesla (TSLA) stock is $250.08. Here are some additional details:

- **Low Price:** $244.58
- **High Price:** $250.96
- **Opening Price:** $246.7

For more information, you can visit [Tesla on Yahoo Finance](https://finance.yahoo.com/quote/TSLA/).
PS C:\Development\openai\chatgpt-demo>