SAMMO is a Python library for prompt engineering and optimization, targeting researchers and developers who need to systematically improve LLM prompt performance. It offers a structure-aware approach to prompt modification and execution, enabling efficient data labeling, prompt prototyping, and instruction optimization.

How It Works

SAMMO treats prompts as programs, allowing manipulation via CSS-like selectors and Markdown parsing. This structure-aware approach facilitates multi-objective optimization, such as balancing prompt length and accuracy, using techniques like beam search. It supports parallel execution and rate limiting for large-scale LLM interactions.

Quick Start & Requirements

• Install: pip install sammo jupyter
• Prerequisites: Python 3.9+
• To run tutorials: Clone the repository (git clone https://github.com/microsoft/sammo.git), cd sammo, and launch Jupyter Notebook (jupyter notebook --notebook-dir docs/tutorials).
• Live demo available on Binder.

Highlighted Details

• Supports prompt optimization via structure-aware metaprompting.
• Enables efficient data labeling through minibatching.
• Offers parallelization and rate-limiting for large-scale prompt execution.
• Features recent updates for Markdown-to-program conversion and CSS selector-based prompt querying.

Maintenance & Community

The project is authored by Tobias Schnabel and welcomes contributions, requiring a Contributor License Agreement (CLA). It adheres to the Microsoft Open Source Code of Conduct.

Licensing & Compatibility

Licensed under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

SAMMO is positioned as a tool for prompt optimization and large-scale execution, explicitly noting it is less suitable for interactive, agent-based or production-ready LLM applications, recommending AutoGen and LangChain respectively for those use cases.