LM Format Enforcer for decoder constraints

How it works

The library matches characters accepted by a parser to token prefixes in a tokenizer. Integrations use the resulting masks during generation, leaving room for multiple valid ways to produce the same structure.

What you need

Choose a tokenizer and supported decoding integration before working on the schema. Exact accepted tokens depend on how that tokenizer represents the characters the parser expects.

What to try first

Start with a schema that has a nested object and an optional field. Test outputs with whitespace and escaped characters, then check whether the integration returns complete data at the token limit.

If using streaming, feed each emitted fragment to the client parser. A correct final object can still arrive in fragments that the consumer must buffer.

Where to look for failures

The caller needs access to a compatible decoding integration. A model may still invent data while following the schema, and interruption can leave a partial object.

Related Franklin pages

These approaches control generation at different points. Run the same nested schema and check compatibility with the model backend before comparing speed or accepted outputs.

• Generate the variable parts of a JSON document while code supplies the fixed syntax. Jsonformer for schema-shaped local extraction
• Apply structured-output constraints inside a model inference engine. XGrammar for grammar-constrained decoding

For the distinction between requesting structured output and enforcing its format, see Franklin’s glossary. Structured output

Project documentation

• Current project README
• README at the reviewed revision