Grammar-Constrained Decoding Formalization
What it is
A Lean 4 formal verification of a grammar-constrained decoding algorithm. It proves the correctness of a system that combines a detokenizing lexer FST with a pushdown parser to ensure LLM outputs adhere to formal grammars like JSON or SQL.
The problem it solves
Ensuring that complex grammar constraints (like nested JSON or SQL) are correctly and safely enforced during LLM inference without excluding valid outputs or allowing invalid ones.
What you could use it for
- Verifying the correctness of LLM output constraints for JSON/SQL generation
- Ensuring sound and complete token masking during inference
- Formal verification of grammar-based decoding algorithms
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.