8 releases

0.1.7 Jan 7, 2024
0.1.6 Jan 7, 2024

#78 in Machine learning

Download history 36/week @ 2024-01-01 18/week @ 2024-01-08 3/week @ 2024-01-22 4/week @ 2024-01-29 1/week @ 2024-02-19 6/week @ 2024-02-26 17/week @ 2024-03-11 52/week @ 2024-04-01

69 downloads per month
Used in 2 crates

MIT license

145KB
3K SLoC

GBNF-rs

A library for working with llama.cpp GBNF files. GBNF files represent grammar of AI output. This project is meant to help make it easier to constrain and guide GBNF driven AIs like llama.cpp using JSON schema ( a way to define the shape of JSON data ). The hope is make more useful outputs when combined with system prompting (that is hopefully also aware of JSON schema to some degree).

  • Data structures for representing GBNF
  • Rendering of a GBNF file from data structures
  • Conversion of a useful subset of JSON schema to GBNF grammar
  • Easily installable CLI converter jsonschema2gbnf that uses library
  • MIT licensed

This library was primarily built for it's sister project, an LLM API epistemology.

Screenshot 2024-01-06 at 11 00 42 PM

Installing

cargo add gnbf

JSON schema support

Currently this library can convert a limited but very useful subset of JSON schema:

  • boolean, number, string
  • object with all required properties
  • enum
  • oneOf
  • property order of objects is preserved
  • array support

Known issues:

  • objects with property names with underscores don't translate well right now
  • no minimum, maximums, fixed lengths, etc. (though putting the schema in system prompt may help depending on model)

Here's one of the most complext JSON schemas that can be handled right now:

{
    "$schema": "https://json-schema.org/draft/2019-09/schema",
    "type": "object",
    "properties": {
        "name": {
            "description": "name of a computer user",
            "type": "string"
        },
        "age": {
            "description": "age of a computer user",
            "type": "number"
        },
        "usesAI": {
            "description": "do they use AI",
            "type": "boolean"
        },
        "favoriteAnimal": {
            "description": "favorite animal",
            "enum": [
                "dog",
                "cat",
                "none"
            ]
        },
        "currentAIModel": {
            "oneOf": [
                {
                    "type": "object",
                    "properties": {
                        "type": {
                            "value": "hugging_face"
                        },
                        "name": {
                            "description": "name of hugging face model",
                            "type": "string"
                        }
                    }
                },
                {
                    "type": "object",
                    "properties": {
                        "type": {
                            "value": "openai"
                        }
                    }
                }
            ]
        },
        "favoriteColors": {
            "type": "array",
            "items": {
                "type": "string"
            }
        }
    }
}

JSON-Schema Converting to AI Grammar

fn simple_json_schema_basic_object_example() {
        let schema = r#"
    {
        "$id": "https://example.com/enumerated-values.schema.json",
        "$schema": "https://json-schema.org/draft/2020-12/schema",
        "title": "Enumerated Values",
        "type": "object",
        "properties": {
            "a": {
                "type": "boolean"
            },
            "b": {
                "type": "number"
            },
            "c": {
                "type": "string"
            }
        }
    }
            "#;
        let g = Grammar::from_json_schema(schema).unwrap();
        let s = g.to_string();
        pretty_assertions::assert_eq!(
            s,
            r#"################################################
# DYNAMICALLY GENERATED JSON-SCHEMA GRAMMAR
# $id: https://example.com/enumerated-values.schema.json
# $schema: https://json-schema.org/draft/2020-12/schema
# title: Enumerated Values
################################################

symbol1-a-value ::= boolean ws
symbol2-b-value ::= number ws
symbol3-c-value ::= string ws
root ::= "{" ws 
 "a" ws ":" ws symbol1-a-value 
 "b" ws ":" ws symbol2-b-value "," ws 
 "c" ws ":" ws symbol3-c-value "," ws 
 "}" ws

###############################
# Primitive value type symbols
###############################
null ::= "null" ws
boolean ::= "true" | "false" ws
string ::= "\"" ([^"\\] | "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]))* "\"" ws
number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
ws ::= [ ]
"#
        )
    }

Future goals of this project:

  • offer standard grammars
  • handle useful conversion of JSON schema to GBNF (we won't be able to handle it all probably).
  • parsing of GBNF files using Nom 7.

I'm totally down for contributors, please add tests.

See the documentation.

Attribution

Multiple MIT licensed examples of GBNF were used from the llama.cpp examples for grammar for automated tests for compliance and general inspiration for this project from python JSON schema converter. Thank you.

Dependencies

~2.4–3.5MB
~66K SLoC