Guidance

Guidance is an efficient programming paradigm designed to steer and control large language models (LLMs). It provides a Pythonic interface that allows developers to interact with LLMs using familiar Python idioms, enabling seamless interleaving of control flow (conditionals, loops, tool use) and text generation. This approach helps users structure LLM output, achieve high-quality results, and potentially reduce latency and cost compared to traditional prompting or fine-tuning methods. The tool is particularly useful for developers and researchers working with LLMs who need fine-grained control over the generation process. It supports constraining output using regular expressions and context-free grammars, ensuring that the generated text adheres to specific formats or choices. Guidance also facilitates the creation of custom functions that encapsulate complex interactions with language models, making it easier to build sophisticated AI applications.