nemoguardrails.library.hallucination.actions | NVIDIA NeMo Guardrails Library Developer Guide

Module Contents

Functions

Name	Description
`_hallucination_outcome`	-
`self_check_hallucination`	Checks if the last bot response is a hallucination by checking multiple completions for self-consistency.

Data

HALLUCINATION_NUM_EXTRA_RESPONSES

log

API

nemoguardrails.library.hallucination.actions._hallucination_outcome(
    is_hallucination: bool
) -> nemoguardrails.actions.rail_outcome.RailOutcome

nemoguardrails.library.hallucination.actions.self_check_hallucination(
    llm: nemoguardrails.types.LLMModel,
    llm_task_manager: nemoguardrails.llm.taskmanager.LLMTaskManager,
    context: typing.Optional[dict] = None,
    use_llm_checking: bool = True,
    config: typing.Optional[nemoguardrails.RailsConfig] = None,
    bot_message: typing.Optional[str] = None,
    last_bot_prompt: typing.Optional[str] = None,
    kwargs = {}
) -> nemoguardrails.actions.rail_outcome.RailOutcome

async

Checks if the last bot response is a hallucination by checking multiple completions for self-consistency.

:return: A blocking outcome if hallucination is detected, otherwise an allowing outcome.

nemoguardrails.library.hallucination.actions.HALLUCINATION_NUM_EXTRA_RESPONSES = 2

nemoguardrails.library.hallucination.actions.log = logging.getLogger(__name__)