Azure AI Model Inference 与 Lobe Chat 集成指南 | Guide to Integrating Azure AI Model Inference with Lobe Chat

[R3pl4c3r]

配置 Azure AI Model Inference 与 Lobe Chat 的集成指南

1️⃣ 鉴权方式差异

当使用 Azure AI Model Inference API 时，需注意其鉴权方式与 OpenAI 不同：

• 请求头需添加 api-key: <your-api-key>（而非 OpenAI 的 Authorization: Bearer 格式）
• 官方文档：Azure AI Model Inference 鉴权说明

2️⃣ 配置 Azure 权限

1. 进入目标资源的 Access control (IAM) 面板
2. 在 Role assignments 中为账户分配 Cognitive Services User 角色
3. 官方指引：Azure Entra ID 配置文档

3️⃣ 获取访问令牌

通过 Azure Cloud Shell 执行：

# 登录认证
az login --scope https://cognitiveservices.azure.com/.default

# 获取访问令牌
az account get-access-token --resource https://cognitiveservices.azure.com --query "accessToken" --output tsv

4️⃣ 部署并配置 one-api

1. 本地部署 one-api 服务
2. 使用 root 账户登录后：
   • 在 渠道管理 添加：
     • 令牌：上一步获取的访问令牌
     • 终结点：AI Foundry 显示的 endpoint
     • 模型名称：AI Foundry 显示的 "model=xxx" 的这个 xxx
   • 在 令牌管理 创建新访问令牌

5️⃣ 集成到 Lobe Chat

1. 在 Lobe Chat 自定义提供商设置中填写：
   • API 终结点：http://localhost:3000/v1（本地 one-api 地址）
   • 模型名称：与上一步对应 Azure 部署的模型名称
   • 鉴权令牌：one-api 生成的访问令牌

💡 提示：确保本地服务端口与防火墙设置匹配，若部署在服务器需配置域名和 HTTPS 证书。

---

Guide to Integrating Azure AI Model Inference with Lobe Chat

1️⃣ Differences in Authentication Methods

When using the Azure AI Model Inference API, note that its authentication method differs from OpenAI:

• The request header requires api-key: <your-api-key> (instead of OpenAI's Authorization: Bearer format)
• Official documentation: Azure AI Model Inference Authentication Guide

2️⃣ Configuring Azure Permissions

1. Navigate to the Access control (IAM) panel of the target resource
2. Assign the Cognitive Services User role to your account in the Role assignments section
3. Official guide: Azure Entra ID Configuration Documentation

3️⃣ Obtaining Access Tokens

Execute the following commands via Azure Cloud Shell:

# Login authentication
az login --scope https://cognitiveservices.azure.com/.default

# Get access token
az account get-access-token --resource https://cognitiveservices.azure.com --query "accessToken" --output tsv

4️⃣ Deploying and Configuring one-api

1. Locally deploy the one-api service
2. After logging in with the root account:
   • Add in Channel tab:
     • Token: The access token obtained in the previous step
     • Endpoint: The endpoint displayed in AI Foundry
     • Model Name: The xxx in "model=xxx" displayed in AI Foundry
   • Create a new access token in Token tab

5️⃣ Integrating with Lobe Chat

1. Fill in the custom provider settings in Lobe Chat:
   • API Endpoint: http://localhost:3000/v1 (local one-api address)
   • Model Name: Corresponding to the model name deployed in Azure from the previous step
   • Authentication Token: Access token generated by one-api

💡 Tip: Ensure that the local service port matches the firewall settings. If deployed on a server, configure the domain name and HTTPS certificate.

[Mr-win7]

POST https://.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview
Authorization: Bearer
Content-Type: application/json

官方文档里用的就是Authorization header的版本