Automata-based constraints for language model decoding

TL;DR

This paper introduces an automata-based method to constrain language model decoding, ensuring outputs adhere to formal languages like JSON or YAML efficiently and correctly, addressing tokenization issues and enabling scalable, task-specific applications.

Contribution

The authors develop a novel automata-based framework for constraining language model outputs to formal languages, offering a faster, correct, and extensible alternative to previous bespoke solutions.

Findings

System compiles constraints approximately 7,000 times faster

Ensures provable correctness of constrained decoding

Extensible to deterministic context-free languages

Abstract

Language models (LMs) are often expected to generate strings in some formal language; for example, structured data, API calls, or code snippets. Although LMs can be tuned to improve their adherence to formal syntax, this does not guarantee conformance, especially with smaller LMs suitable for large-scale deployment. In addition, tuning requires significant resources, making it impractical for uncommon or task-specific formats. To prevent downstream parsing errors we would ideally constrain the LM to only produce valid output, but this is severely complicated by tokenization, which is typically both ambiguous and misaligned with the formal grammar. We solve these issues through the application of automata theory, deriving an efficient closed-form solution for the regular languages, a broad class of formal languages with many practical applications, including API calls or schema-guided JSON and YAML. We also discuss pragmatic extensions for coping with the issue of high branching factor, and extend our techniques to deterministic context-free languages, which similarly admit an efficient closed-form solution. Previous work on this topic (Willard and Louf, 2023) layers bespoke solutions onto automata, leading to problems with speed, correctness, and extensibility. Instead, we reformulate the entire task in terms of automata so we can leverage well-studied and well-optimized algorithms. Our system compiles constraints ~7,000x faster, is provably correct, and can be extended in a modular fashion.