OpenAI remove key access while using AAD authentication

题意:"OpenAI 在使用 AAD 认证时移除了密钥访问权限"

问题背景:

I am calling Azure OpenAI API in my python code. To set it up, we need to provide a few parameters, one of which is openai.api_key. There are 2 options to get this value -

"我在 Python 代码中调用 Azure OpenAI API。要进行设置,我们需要提供一些参数,其中之一是 `openai.api_key`。获取此值有两种选择:"

  1. Static key
  2. AAD Token

I am using AAD token (refer this - https://github.com/openai/openai-python#microsoft-azure-active-directory-authentication) but still I can access it using the static key. I want to remove the static key access.

"我正在使用 AAD 令牌(参考这里 - https://github.com/openai/openai-python#microsoft-azure-active-directory-authentication),但我仍然可以使用静态密钥进行访问。我想移除静态密钥的访问权限。"

Is there any way to disable the static key and only authenticate using AAD authentication?

"有没有办法禁用静态密钥,并仅使用 AAD 认证进行身份验证?"

问题解决:

Is there any way to disable the static key and only authenticate using AAD authentication?

If you need to authenticate using only AAD authentication without using api.key you can use the below code with request method using Python.

"如果你需要仅使用 AAD 认证而不使用 `api.key` 进行身份验证,你可以使用以下代码,通过 Python 中的 `request` 方法实现。"

First, create an application and assign Cognitive Services User role to your application.

"首先,创建一个应用程序并为该应用程序分配 `Cognitive Services User` 角色。"

Code: 代码

python 复制代码
import requests
import json


tenantid="xxxxx"
clientid="xxxxx"
clientsecret="xxxxxx"

auth_url = f'https://login.microsoftonline.com/{tenantid}/oauth2/v2.0/token'
data = {
    'grant_type': 'client_credentials',
    'client_id': clientid,
    'client_secret': clientsecret,
    'scope': 'https://cognitiveservices.azure.com/.default'
}
response = requests.post(auth_url, data=data).json()
access_token=response['access_token']

url="https://YOUR_RESOURCE_NAME.openai.azure.com/openai/deployments/YOUR_DEPLOYMENT_NAME/completions?api-version=2023-05-15"

h1 = {
    "Authorization": f"Bearer {access_token}",
    "Content-Type":"application/json"
}

body={
    "prompt":"MS Dhoni is greatest captain",
    "max_tokens": 100
}
response = requests.post(url,headers=h1,data=json.dumps(body))
s=json.loads(response.text)
d=json.dumps(s,indent=3)
print(d)

Output: 输出

python 复制代码
 {
   "id": "cmpl-880wqMQxxxxxxxPl",
   "object": "text_completion",
   "created": 1696xxxx0,
   "model": "text-davinci-002",
   "choices": [
      {
         "text": " India has had, he is motivation: Ravichandran AshwinR Ashwin tweeted saying MS Dhoni has given all he had to India cricket for over 15 years and has been a motivation for everybody. Michael holding reveals his best captain story and it involves Kapil and Viv. The Indians were taken by surprise and after the initial shock it took off from there.New Zealand lost by 15 runs. It is a catch that has a contenders for the most iconic cricketing pictures.\n\nIndia",
         "index": 0,
         "finish_reason": "length",
         "logprobs": null
      }
   ],
   "usage": {
      "completion_tokens": 100,
      "prompt_tokens": 6,
      "total_tokens": 106
   }
}

Reference: 引用

How to configure Azure OpenAI Service with managed identities - Azure OpenAI | Microsoft Learn

相关推荐
captain3762 分钟前
MyBatis操作数据库(入门2)
java·数据库·spring boot·后端·spring·mybatis
wflynn15 分钟前
GitHub 今日推荐|Edge0:让大号 MoE 模型在一台 Mac 上流式跑起来
python·开源·github
2601_9679801132 分钟前
python之用scapy分层解析pcap报文(Ethernet帧、IP数据包、TCP数据包、UDP数据包、Raw数据包)
python·scapy·pcap·协议分析·网络数据包
心疼你的一切39 分钟前
Python自动发邮件实战定时发报表
大数据·开发语言·python
IT_陈寒1 小时前
Java序列化踩坑记:对象循环引用让我加班两小时
前端·人工智能·后端
泡海椒2 小时前
05-雷达图多维画像-信用评级多维画像
后端
王中阳Go2 小时前
5.27 分钟干完后端和全部测试:一场三小时笔试的五个失分点,四个不在代码上
大数据·数据库·人工智能·后端·大语言模型
大哥43092 小时前
Modbus 的"字序":一个配错就变成天文数字,或者干脆变成 0
后端
岩钧2 小时前
New API 源码解析:一个生产级 AI 网关的架构与实现
后端
学到老活到老2 小时前
第 24 集:配置文件写了 80,Spring Boot 为什么读到 90?
java·后端·spring