Boost your Prompt Engineering Efficiency with Prompty and VS Code!

Hello! I’m Wada (@cognac_n), an AI Evangelist at KINTO Technologies.
How do you manage your prompts? Today, I will introduce Prompty, a tool that simplifies creating/editing, testing, implementing, and organizing prompts!

1. What is Prompty?

Prompty is a tool designed to streamline prompt development for large language models (LLMs). This enables centralization of prompts and parameters in YAML format, making it ideal for version control on GitHub and improving collaboration in team environments. Using the Visual Studio Code (VS Code) extension can greatly improve the efficiency of prompt engineering.
Benefits of Introducing Prompty
Although integration with Azure AI Studio and Prompt Flow offers benefits, this article will focus on the integration with VS Code.

Who should consider using Prompty:

• Those looking to speed up prompt development
• Developers who need version control for prompts
• Teams collaborating on prompt creation
• Anyone wanting to simplify prompt execution on the application side
  https://github.com/microsoft/prompty

2. Prerequisites

Requirements (at the time of writing)

• Python 3.9 or higher
• Vs Code (if using the extension)
• OpenAI API Key or Azure OpenAI Endpoint (depending on the LLM in use)

Installation and initial setup

Install the VS Code extension

https://marketplace.visualstudio.com/items?itemName=ms-toolsai.prompty

Use pip or other package managers to install the necessary libraries

pip install prompty

https://pypi.org/project/prompty/

3. Try It Out

3-1. Create a New Prompty File

Right-click in the Explorer tab and select "New Prompty" to create a template.
New Prompty
The created template is as follows:

---
name: ExamplePrompt
description: A prompt that uses context to ground an incoming question
authors:
  - Seth Juarez
model:
  api: chat
  configuration:
    type: azure_openai
    azure_endpoint: ${env:AZURE_OPENAI_ENDPOINT}
    azure_deployment: <your-deployment>
  parameters:
    max_tokens: 3000
sample:
  firstName: Seth
  context: >
    The Alpine Explorer Tent boasts a detachable divider for privacy, 
    numerous mesh windows and adjustable vents for ventilation, and 
    a waterproof design. It even has a built-in gear loft for storing 
    your outdoor essentials. In short, it's a blend of privacy, comfort, 
    and convenience, making it your second home in the heart of nature!
  question: What can you tell me about your tents?
---
system:
You are an AI assistant who helps people find information. As the assistant, 
you answer questions briefly, succinctly, and in a personable manner using 
markdown and even add some personal flair with appropriate emojis.
# Customer
You are helping {{firstName}} to find answers to their questions.
Use their name to address them in your responses.
# Context
Use the following context to provide a more personalized response to {{firstName}}:
{{context}}
user:
{{question}}

In the area enclosed by ---, specify parameters. Below this section, add the main content of the prompt. You can define roles using system: or user:.

Basic Parameter Overview

3-2. Configuring API Keys and Parameters

There are several ways to set the required API keys, endpoint information, and execution parameters.

[Option 1] Specifying in the .prompty file

This involves directly adding these details to the .prompty file.

model:
  api: chat
  configuration:
    type: azure_openai
    azure_endpoint: ${env:AZURE_OPENAI_ENDPOINT}
    azure_deployment: <your-deployment>
  parameters:
    max_tokens: 3000

You can also reference environment variables, such as ${env:AZURE_OPENAI_ENDPOINT}. However, please note that azure_openai_api_key cannot be configured in this way. ![azure_openai_api_key cannot be written directly in the .prompty file] (/assets/blog/authors/s.wada/20240821/image_3.png =750x) azure_openai_api_key cannot be written directly in the .prompty file

[Option 2] Configuring with settings.json

Another approach is to use VS Code’s settings.json. If the settings are incomplete and you click the play button in the upper-right corner, you will be prompted to edit settings.json. You can create multiple configurations beyond the default definition and switch between them during testing. When type is set to azure_openai and api_key is left empty, the process will direct you to authenticate using Azure Entra ID, as explained later.

{
  "prompty.modelConfigurations": [
    {
      "name": "default",
      "type": "azure_openai",
      "api_version": "2023-12-01-preview",
      "azure_endpoint": "${env:AZURE_OPENAI_ENDPOINT}",
      "azure_deployment": "",
      "api_key": "${env:AZURE_OPENAI_API_KEY}"
    },
    {
      "name": "gpt-3.5-turbo",
      "type": "openai",
      "api_key": "${env:OPENAI_API_KEY}",
      "organization": "${env:OPENAI_ORG_ID}",
      "base_url": "${env:OPENAI_BASE_URL}"
    }
  ]
}

[Option 3] Configuration with a .env file

By creating a .env file, environment variables can be read directly from it. Note that the .env file must be located in the same directory as the .prompty file you are using. This setup is especially convenient for local testing.

AZURE_OPENAI_API_KEY=YOUR_AZURE_OPENAI_API_KEY
AZURE_OPENAI_ENDPOINT=YOUR_AZURE_OPENAI_ENDPOINT
AZURE_OPENAI_API_VERSION=YOUR_AZURE_OPENAI_API_VERSION

[Option 4] Configuring with Azure Entra ID

By signing in with an Azure Entra ID that has the appropriate permissions, you can access the API.
I haven’t tested this option yet

3-3. Running Prompts in VS Code

You can easily execute prompts by clicking the play button in the upper right corner. The results are displayed in the OUTPUT section. To view raw data, including placeholder substitution and token usage, select "Prompty Output(Verbose)" from the dropdown in the OUTPUT panel. This option is useful for checking detailed information.
Use the Play button in the upper right to run the prompt
Results can be seen in the OUTPUT section

3-4. Other Parameters

Various parameters are introduced on the following page.
Defining options like inputs and outputs, especially when using json mode, improves prompt visibility, so be sure to set them.

inputs:
  firstName:
    type: str
    description: The first name of the person asking the question.
  context:
    type: str
    description: The context or description of the item or topic being discussed.
  question:
    type: str
    description: The specific question being asked.

3-5. Integrating with an Application

The syntax for integration may vary depending on the library used in your application. As Prompty is frequently updated, be sure to check the latest documentation regularly. Here’s an example code snippet demonstrating the use of Prompty with Prompt Flow. This allows for simple prompt execution.

from promptflow.core import Prompty, AzureOpenAIModelConfiguration
# Set up configuration to load Prompty using AzureOpenAIModelConfiguration
configuration = AzureOpenAIModelConfiguration(
    azure_deployment="gpt-4o", # Specify the deployment name for Azure OpenAI
    api_key="${env:AZURE_OPENAI_API_KEY}", # Retrieve the API key from environment variables
    api_version="${env:AZURE_OPENAI_API_VERSION}", # Retrieve the API version from environment variables
    azure_endpoint="${env:AZURE_OPENAI_ENDPOINT}", # Retrieve the Azure endpoint from environment variables
)
# Configure to override model parameters
# Here, max_tokens is overridden as an example
override_model = {"configuration": configuration, "max_tokens": 2048}
# Load Prompty with the overridden model settings
prompty = Prompty.load(
    source="to_your_prompty_file_path", # Specify the Prompty file to use
    model=override_model # Apply the overridden model settings
)
# Execute prompty
result = prompty(
    firstName=first_name, context=context, question=question
# Execute Prompty based on the provided text and obtain the result

4. Summary

Prompty is a powerful tool that can significantly streamline prompt engineering tasks. In particular, the development environment integrated with VS Code allows for seamless creation, testing, implementation, and management of prompts, making it highly user-friendly. Mastering Prompty can greatly enhance the efficiency and quality of prompt engineering. I encourage everyone to give it a try!
Benefits of Introducing Prompty (Repost)

We Are Hiring！

At KINTO Technologies, we are seeking colleagues to help drive the adoption of generative AI in our business. We are open to casual interviews, so if you’re even slightly interested, please contact us via the link below or through X DM . We look forward to hearing from you! https://hrmos.co/pages/kinto-technologies/jobs/1955878275904303115 Learn more about how we work with generative AI here. https://blog.kinto-technologies.com/posts/2024-01-26-GenerativeAIDevelopProject/
Thank you for reading this far.