Getting Started with Azure AI Phi-3 Chat API, GPT-4 Assistants API and Azure Translator

Overview

This is a lightweight python project to demonistrate how-to

1. Create and manage GPT-4 assistants and threads.
2. Create Phi-3-mini chat.
3. Use Azure Translator to complement the model translation capabilities to improve the low resource languages translation quality.
4. Use fast python async calls with Quart and Uvicorn.

Requirements

• You created AIServices resource/project (https://ai.azure.com) and deployed GPT-4 and Phi-3-mini models. You will need this information to modify ".env_template."
• [optional] Install VS code
• Install node.js and npm using https://nodejs.org/dist/v22.1.0/node-v22.1.0-x64.msi
• Clone this repo
• Launch VS code
• Select the project folder
• Start VS code terminal and run the following install commands:
  • npm install
  • npm install @mui/material @emotion/react @chatscope/chat-ui-kit-react
  • npm install react-select
  • python -m pip install -r requirements.txt

.env must be added to the root directory:

• modify ".env_template" and rename to ".env"

Launch the app

1. Start VS code and open the app folder from "File" tab
2. Create 2 terminals
3. In terminal 1 run: python -m uvicorn app:app --port 5000 --reload
   • Wait for Application startup complete.
4. In terminal 2 run: npm start
   • You should see:

You can:

1. Type query in Ask anything.
2. Use slider switch for Azure AI Translator.
3. Select deployment model (phi-3-mini-4k-instruct as an example for chat and GPT-4 for assistants).
4. Change query paramters as needed.
5. Clear chat to remove session chat history.

NOTE:

• When Azure Traslator is enabled, this repo supports translation from source language to French, German, Chinese, Spanish and English.
• You can add more target lanaguages by modifying get_target_language() found in backend/gpt_2_assistant.py and backend/phi_3_assistant.py