guidance

You can constrain LLM output to valid JSON and XML using regex constraints and grammars. This enforces structured generation by restricting the model's output space to specific syntax patterns during decoding.

Build multi-step workflows with Pythonic control flow and structured output constraints to chain LLM calls. This allows you to extract data from a JSON document, validate it, and then transform it into XML format sequentially.

You need the guidance library to enforce structured output, with optional support for transformer backends like Hugging Face models or llama.cpp. These backends provide the underlying model inference for the constrained generation framework.

The best way to generate structured data like dates and emails is by applying structured output constraints with regex. This guarantees the language model formats specific data types exactly as required without parsing errors.

Yes, constrained generation applies to automated code generation by using regex and grammars to enforce syntax rules. This ensures the LLM generates valid code snippets that conform to specific language constraints.

LLM outputs become invalid without constrained generation. Applying regex constraints and grammars during decoding forces the model to follow exact syntax rules, preventing malformed structured data and format violations.