Skip to content

Getting Started with Reasoning Models

This workshop covers uses of Reasoning Models. The workshop includes several example use cases demonstrating how to implement reasoning-based solutions for various business scenarios.

Setup Instructions

To get started with this workshop:

  1. Visit the GitHub repository
  2. Follow the setup instructions in the README to:
    • Clone the repository
    • Install required dependencies
    • Set up your Azure OpenAI endpoint and API key
    • Configure your environment

Prerequisites

Before starting:

  1. Ensure you have an Azure account with OpenAI access
  2. Install required Python packages:
    Bash
    1
    pip install openai azure-identity python-dotenv
  3. Set up your .env file with:
    Text Only
    1
2
    AZURE_OPENAI_ENDPOINT=your_endpoint
AZURE_OPENAI_API_KEY=your_key

Workshop Notebooks

1. Setup & Environment (00-setup-aoai.ipynb)

Learn how to set up your environment for working with reasoning models:

Overview

  • Installing required Python packages
  • Setting up environment variables
  • Testing your Azure OpenAI connection
  • Understanding token usage and costs

Key Concepts

  • Azure OpenAI endpoint and key configuration
  • Python package requirements
  • Environment variable management
  • Basic API connectivity testing
  • Token usage monitoring
  • Content filter results understanding

2. Basic Text Reasoning (01-text-reasoning.ipynb)

Introduction to fundamental concepts through text-based examples:

Key Concepts

  • Reasoning Levels
    • Low: Quick, straightforward responses
    • Medium: Balanced analysis and explanation
    • High: Deep, thorough analysis with detailed rationale
  • Model Interaction
    • Setting up chat completions
    • Using developer messages
    • Handling responses
    • Managing tokens
  • Response Analysis
    • Understanding reasoning tokens
    • Evaluating response quality
    • Measuring performance
    • Analyzing token usage

Best Practices

  • Clear and specific prompt engineering
  • Proper response handling and validation
  • Effective error management

3. Advanced Features (02-advanced-reasoning.ipynb)

Explore advanced capabilities including:

Key Features

  • Developer Messages
    • Setting context and goals
    • Defining model behavior
    • Controlling output format
  • Structured Outputs
    • JSON schema definition
    • Type validation
    • Response formatting
  • Function Calling
    • Function definition
    • Parameter handling
    • Response parsing
  • Vision Support
    • Image analysis
    • Multi-modal reasoning
    • Visual context integration

Use Cases

  • Data extraction and processing
  • Visual analysis and understanding
  • System integration and automation

4. Model Comparison (03-model-comparison.ipynb)

Compare GPT and Reasoning models across various tasks:

Evaluation Criteria

  • Clarity: Organization, structure, and flow
  • Accuracy: Factual accuracy and precision
  • Completeness: Coverage and depth of analysis
  • Analytical Depth: Reasoning quality and insights
  • Multi-Dataset Synthesis: Data integration and patterns

Usage Guidelines

  • Choose GPT Models for:
    • General text generation
    • Creative tasks
    • Simple Q&A
  • Choose Reasoning Models for:
    • Complex analysis
    • Multi-step reasoning
    • Structured outputs

Use Case Model Evaluation

When working with these use cases:

  1. Start with the README.md in each directory
  2. Review the prompt.md file to understand the input format
  3. Study the in-depth analysis for detailed Analysis
  4. Use Model Comparison (03-model-comparison.ipynb) to compare reasoning model and GPT model outputs using multi dimension evaluation criteria

Contributing

Want to add a new use case? Follow these steps:

  1. Create a new directory under usecases/
  2. Include:
    • README.md with overview
    • prompt.md with example prompts
    • in-depth analysis document
  3. Submit a pull request

Getting Help

If you encounter issues: - Check the Azure OpenAI documentation - Review the Reasoning Models guide - Submit issues on our GitHub repository