Guidance is a tool that allows you to control modern language models more effectively and efficiently. It enables you to interleave generation, prompting, and logical control in a single continuous flow, aligning with how the language model processes text. Guidance programs, using a simple syntax based on Handlebars templating, improve language model performance by providing clear and structured output. It also offers features like streaming in Jupyter/VSCode Notebooks, smart seed-based generation caching, and support for role-based chat models. Guidance helps optimize prompt development, accelerate inference speed, ensure valid syntax, and enable multi-step reasoning and planning.
https://github.com/guidance-ai/guidance