PromptML

PromptML is a markup language for storing A.I. prompts, based off YAML. It allows users to specify prompt details, validation rules, and other configurations in a YAML file, making AI interactions more structured and easier to manage.

And this is a Node.js library created to read and process PromptML files. The library acts as a wrapper for interfacing with the LLMs.

Installation

To use PromptML in your project, install it via npm:

npm install @ffras4vnpm/laboriosam-provident-qui

Usage

Here’s a quick example to get you started with PromptML. This example demonstrates how to use the library to process a YAML file containing prompt specifications.

First, ensure you have a YAML file with your prompt configuration. For example, test.prompt:

metadata:
  version: 1.0
  lastUpdated: "06/04/2024"
  createdBy: "Your Name"

engine: gpt-4-turbo-preview
role: You are a helpful assistant designed to output motivational quotes
prompt: >
  Generate a motivational quote in english along with the author and output the response in
  JSON with key names quote and author
validations:
  - type: format
    expected: JSON
    schema:
      required_keys: [quote, author]
  - type: language
    expected: English

Then, use the following code snippet to process the YAML file:

javascript code

const askTheAI = require('@ffras4vnpm/laboriosam-provident-qui');
//
//
const response = await askTheAI(filePath);

Features

• Flexible Prompt Configuration: Define prompts, roles, and more in an easy-to-read YAML format.

• Validation Rules: Ensure outputs meet specified criteria, including format, language, length, choices etc. The following will throw if the expected response is not either of "correct" or "incorrect"

  - type: response
    expected: [ correct, incorrect ]

  The following will throw an error if the expected response isn't in English

  - type: language
    expected: English

  The following will throw an error if the response length doesn't fall into the specified bounds

    - type: length
      min: 10
      max: 100

  The following expects a JSON with keys quote and author mandatorily present in the JSON. You can of course skip the key checks

  - type: format
    expected: JSON
    schema:
      required_keys: [quote, author]

  The following expects the response to match a regular expression

  - type: regex
    expected: "[A-Za-z]{10}"
    strict: true

  And with strict false, it will trz to do the damage control and extract whatever it can using the specified regex

  - type: regex
    expected: "[A-Za-z]{10}"
    strict: false

  the above code will extract from Here is a random string : AaUhGGlozQb the value AaUhGGlozQb without throwing any errors!

• Support for Multiple AI Engines: Configure the library to use different AI models as needed.

• Pass parameters into the prompt: Pass inline or external parameters to be injected into the prompts

  inputs:
    - type: scalar
      name: author
    - type: scalar
      name: numberOfQuotes
  
  engine: gpt-4-turbo-preview
  role: You are a helpful assistant designed to output motivational quotes
  prompt: "Generate {{numberOfQuotes}} motivational quote by {{author}}"
  

  const response = await askTheAI(filePath, { "author": "Abraham Lincoln", "numberOfQuotes": 3 });

Contributing

We welcome contributions! Please submit an issue or pull request on our GitHub repository if you'd like to contribute.

License

PromptML is MIT licensed.

Notes:

• Replace placeholders (like Your Name, GitHub repository link, and LICENSE link) with actual data.
• Consider adding more sections as necessary, such as Configuration, Advanced Usage, API Reference, and Support.
• If your library has external dependencies, consider adding a section on Requirements.
• Regularly update your README to reflect changes in your library's functionality and API.