Prompt Optimization

This repository demonstrates prompt optimization using both Local LLMs and an Azure AD Service Principal.

Prerequisites

Dataset Creation

A dataset was created to identify Personally Identifiable Information (PII) using ChatGPT with the following prompt:

Create an Excel file that contains three columns: 'id', 'text', and 'contains_pii'. 
Add around 1,000 rows of text in the 'text' column, ensuring some rows contain PII and others do not. 
Fill the 'contains_pii' column with 'yes' or 'no' accordingly. Be consistent.

The generated dataset is saved in the data folder.

Objective

Using this dataset, we aim to optimize prompts by:

1. Running a local LLM.
2. Utilizing a GPT-4o endpoint with an Azure AD Service Principal.

---

Notebook Overview

Part 1: Optimization Using a Local LLM

This section demonstrates how to optimize prompts using a locally hosted LLM. The deepseek-r1:8b model has already been downloaded.

Steps:

1. Start the Ollama server:

   ollama serve
   

2. Run the first part of the notebook.

---

Part 2: Optimization Using GPT-4o Endpoint

This section demonstrates how to optimize prompts using a GPT-4o endpoint with an Azure AD Service Principal.

Required Values:

• TENANT_ID: Your Azure tenant ID.
• CLIENT_ID: Your Azure client ID.
• CLIENT_SECRET: Your Azure client secret.
• AZURE_OPENAI_ENDPOINT: The endpoint URL, e.g., https://gpt4o....
• AZURE_OPENAI_DEPLOYMENT: The deployment name (not the raw model name), e.g., gpt-4o.
• AZURE_API_VERSION: The API version, e.g., 2023-05-15.

---

Folder Structure

prompt_optimization/
├── data/                # Contains the PII dataset
├── notebooks/           # Jupyter notebooks for optimization
├── README.md            # Project documentation

---

Getting Started

1. Ensure the dataset is available in the data folder.
2. Follow the steps in the notebook to perform prompt optimization.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.