• GitHub logo
  • Discord logo
Loading ...
BespokeLabs
Bespoke MiniCheck
5
See on Github
Validates that the LLM-generated text is supported by the provided context using BespokeLabs.AI's MiniCheck API.
en
string
ML
Brand risk
Factuality
Chatbots
Customer Support

Overview

updated 2 years
Developed by:
Guardrails AI
Date of development:
September 2, 2024
Validator type:
Format
Blog:
License:
Apache 2
Input/Output:
Output

Playground

Description
Intended Use

This validator uses BespokeLabs.AI's minicheck API to evaluate LLM generated text against the provided context. It checks if the claims in the generated text are supported by the given context, using a configurable threshold for determining support. The validator can process the input as a whole or split it into individual sentences for more granular evaluation.

Requirements

  • Dependencies:

    • guardrails-ai>=0.4.0
    • nltk
    • bespokelabs
    • tenacity

  • Foundation model access keys:

Installation
$ guardrails hub install hub://bespokelabs/bespoke_minicheck
Usage Examples
Validating string output via Python

In this example, we apply the validator to a string output generated by an LLM.

# Import Guard and Validator
from guardrails.hub import BespokeMiniCheck
from guardrails import Guard

# Setup Guard
guard = Guard().use(
    BespokeMiniCheck,
    split_sentences=True,
    threshold=0.5,
)

guard.validate("Alex likes cats.", metadata={"context": "Alex likes cats and dogs"}) # validation passes
guard.validate("Alex likes cats.", metadata={"context": "Alex likes dogs, but not cats."})  # validation fails
API Reference

__init__(self, threshold: float = 0.5, split_sentences: bool = True, on_fail: Optional[Callable] = None)

Initializes a new instance of the BespokeMiniCheck class.

Parameters

  • threshold (float, optional): The minimum score for a claim to be considered supported. Defaults to 0.5.
  • split_sentences (bool, optional): Whether to split the input into sentences for individual evaluation. Defaults to True.
  • on_fail (Optional[Callable], optional): A callable to execute when the validation fails. Defaults to None.

validate(self, value: Any, metadata: Dict = {}) -> ValidationResult

Validates the given value using the rules defined in this validator, relying on the metadata provided to customize the validation process. This method is automatically invoked by guard.parse(...), ensuring the validation logic is applied to the input data.

Note:

  1. This method should not be called directly by the user. Instead, invoke guard.parse(...) where this method will be called internally for each associated Validator.
  2. When invoking guard.parse(...), ensure to pass the appropriate metadata dictionary that includes keys and values required by this validator. If guard is associated with multiple validators, combine all necessary metadata into a single dictionary.

Parameters

  • value (Any): The input value to validate.

  • metadata (Dict): A dictionary containing metadata required for validation. Keys and values must match the expectations of this validator.

    KeyTypeDescriptionDefault
    thresholdfloatThe minimum score for a claim to be considered supported.Value set during initialization
    split_sentencesboolWhether to split the input into sentences for individual evaluation.Value set during initialization
    contextsList[str]A list of context strings to validate claims against.[]

Returns

  • ValidationResult: A PassResult if all claims are supported, or a FailResult with an error message and a fix value containing only supported claims.

Raises

  • ValueError: If the contexts provided are not a non-empty list of strings.