aiverify-moonshot 0.7.6

AI Verify advances Gen AI testing with Project Moonshot.

Version 0.7.6

A simple and modular tool to evaluate any LLM-based AI systems.

🎯 Motivation

Developed by the AI Verify Foundation, Moonshot is a tool to bring Benchmarking and Red-Teaming together to help AI developers, compliance teams evaluate LLM-based Apps and LLMs.

🚀 Why Moonshot

In the rapidly evolving landscape of Generative AI, ensuring safety, reliability, and performance of LLM applications is paramount. Moonshot addresses this critical need by providing a unified platform for:

• Benchmark Tests: Systematically test LLM Apps or LLMs across critical trust & safety risks using a wide array of open-source benchmark dataset and metrics, including guided workflows to implement IMDA's Starter Kit for LLM-based App Testing.
• Red Team Attacks: Proactively identify vulnerabilities and potential misuse scenarios in your LLM applications through streamlined adversarial prompting.

🔑 Key Features

• User-friendly Interfaces: Interact with Moonshot via an intuitive Web UI for visual insights, and an interactive Command Line Interface (CLI) for quick operations.
• Comprehensive Benchmarking:
  • View list of available datasets available
  • Test for Performance (e.g., accuracy, BLEU)
  • Ensure Trust & Safety e.g., bias, toxicity, hallucination)
  • Utilize built-in workflow to implement IMDA's Starter Kit for LLM-based App Testing. View available pre-built Cookbooks
• Powerful Red-Teaming:
  • View list of available attack modules
  • Simplify adversarial prompt generation using algorithmic strategies or generative LLM to uncover potential misuse.
  • Leverage prompt templates, context strategies, and automated attack modules.
• Customizable Recipes: Build your own benchmark tests with custom datasets (input-target pairs), prompt templates (optional), evaluation metric, and grading scales. View available pre-built Recipes
• Insightful Reporting: Use our HTML reports with interactive charts for clear visualization of test results, and download detailed raw JSON results for deeper programmatic analysis.
• Extensible & Modular: Designed for easy extension and integration with new LLM applications, benchmarks, and attack techniques.

Getting Started

Moonshot can be used through several interfaces:

• User-friendly Web UI - Web UI User Guide
• Interactive Command Line Interface - CLI User Guide
• Seamless Integration into your MLOps workflow via Moonshot Library APIs or Moonshot Web APIs - Notebook Examples, Web API Docs

💻 Let's Go!

This section will guide you through getting Moonshot up and running.

✅ Prerequisites

1. Python: Version 3.11 is required.

2. Git Version Control: Git is essential for cloning the repository.

3. (Optional) Virtual Environment: Highly recommended to manage dependencies.

   # Create a virtual environment
   python -m venv venv
   
   # Activate the virtual environment
   source venv/bin/activate
   

4. If you plan to install our Web UI, you will also need Node.js version 20.11.1 LTS and above

⬇️ Installation

You can install Moonshot in various ways depending on your needs

1. Using pip (Recommended for most users)

# Install Project Moonshot's Python Library, which includes Moonshot's full functionalities (Library APIs, CLI and Web APIs)
pip install "aiverify-moonshot[all]"

# Clone and install test assets and Web UI
python -m moonshot -i moonshot-data -i moonshot-ui

⚠️ You will need to have test assets from moonshot-data before you can run any tests.

🖼️ If you plan to install our Web UI, you will also need moonshot-ui

Check out our Installation Guide for more details.

2. From Source Code (For developers and contributors)

# To install from source code (Full functionalities)
git clone [e-mail]:aiverify-foundation/moonshot.git
cd moonshot
pip install -r requirements.txt

If you have installation issues, please take a look at the Troubleshooting Guide.

Other installation options

Here's a summary of other installation commands available:

# To install Moonshot library APIs only
pip install aiverify-moonshot

# To install Moonshot library APIs and Web APIs only
pip install "aiverify-moonshot[web-api]"

# To install Moonshot library APIs and CLI only
pip install "aiverify-moonshot[cli]"

Check out our Installation Guide for more details.

🏃‍♀️ Run Moonshot

Running the Web UI

python -m moonshot web

Open http://localhost:3000/ in a browser and you should see this homepage:

Refer to this guide to discover the rich features available in Moonshot Web UI

Running the Interactive CLI

python -m moonshot cli interactive

Refer to this Command List to discover the list of CLI commands for Moonshot

📚 Documentation & User Guides

For detailed information on configuring, using, and extending Moonshot, please refer to our comprehensive documentation:

Guides for Moonshot Web UI

• Getting Started with Moonshot Web UI
• Creating Your Custom Cookbook via Moonshot Web UI
• Creating Your Custom Connector Endpoint via Moonshot Web UI
• Running Benchmark Test on Moonshot Web UI
• Running Red Teaming on Moonshot Web UI

Guides for Moonshot Interactive CLI

• Getting Started with Moonshot Interactive CLI
• Creating Your Custom Benchmark Tests for Your RAG Apps via Moonshot Interactive CLI
• Creating Your Custom Connector Endpoint via Moonshot Interactive CLI
• Running Benchmark Test on Moonshot Interactive CLI
• Running Red Teaming on Moonshot Interactive CLI

For Users Interested in trying Out Moonshot using Jupyter Notebook

• Moonshot Library Python Notebook Examples

🤝 Contribution

Moonshot is an open-source project, and we welcome contributions from the community! Whether fixing a bug, adding a new feature, improving documentation, or suggesting an enhancement, your efforts are highly valued.

Please refer to our Contributor Guide for details on how to get started.

✨ Project Status

Moonshot is currently in beta. We are actively developing new features, improving existing ones, and enhancing stability. We encourage you to try it out and provide feedback!

📜 License

Moonshot is released under the Apache Software License 2.0