microsoft/sammo

Supports structured outputs and enables prompt optimization through symbolic program search with mutation operators (paraphrasing, rewriting, text generation) applied at compile-time rather than runtime. Handles production-scale LLM execution with built-in parallelization, rate-limiting, and batch packing for efficient data labeling across multiple LLM providers via a unified runner interface.

753 stars and 177 monthly downloads. No commits in the last 6 months. Available on PyPI.