Integrity

Integrity methods evaluate query strings based on some available context to check for possible hallucinations and ungrounded or incorrect conclusions. Integrity methods will typically only apply to output guards in RAG-based applications.

All integrity models support context as a parameter where the context for the detector can be initialized.

Warning

Integrity methods are currently experimental! We are in the process of experimentally validating them and integrating them into Dome.

The table below lists the integrity methods we currently support. The ID column should be used to use the detection method in a config.

Name

ID

Description

HHEM

hhem-hallucination

Classifier to detect hallucinations

Hallucination Prompt Engineering

hallucination-llm

Detect hallucinations via LLM prompt engineering

Fact-Check Classifier

fact-check-roberta

Roberta Model for fact-checking

Fact-Check Prompt Engineering

fact-check-llm

Fact-checking via LLM prompt engineering

HHEM (hhem-hallucination)

Uses the HHEM Model by Vectara to determine if there might be possible model hallucinations.

Parameters

  • context (optional str): Sets the initial context.

  • factual_consistency_score_threshold (optional float): The factual consistency score threshold. Important: any input where the factual consistency score is lower than the threshold is classified as a possible hallucination. Default value is 0.5.

Hallucination Prompt Engineering (hallucination-llm)

Uses a prompt template outlined in NeMo Guardrails to detect hallucinations given a context and hypothesis.

Parameters

  • context (optional str): Sets the initial context.

  • hub_name (optional str): The hub that hosts the model you want to use. Currently supports OpenAI (openai), Together (together) and OctoAI (octo). Default value is openai.

  • model_name (optional str): The model that you want to use. Default: gpt-4o. Please ensure that the model you wish to use is compatible with the hub you selected.

  • api_key (optional str): Specify the API key you want to use. By default this is not specified and the API key is pulled directly from the environment variables. The environment variables used are OPENAI_API_KEY, OCTO_API_KEY, and TOGETHER_API_KEY for the corresponding hubs.

Fact-Check Classifier (fact-check-roberta)

Uses a fine-tuned RoBERTa model to detect possible factual inconsistencies by examining the joint encoding of a context string and a query string and classifying if the context supports or refutes the claim.

Parameters

  • context (optional str): Sets the initial context.

Fact-Check Prompt Engineering (fact-check-llm)

Uses a prompt template outlined in NeMo Guardrails to detect if a claim is grounded in some context.

Parameters

  • context (optional str): Sets the initial context.

  • hub_name (optional str): The hub that hosts the model you want to use. Currently supports OpenAI (openai), Together (together) and Octo (octo). Default value isopenai.

  • model_name (optional str): The model that you want to use. Default is gpt-4o. Please ensure that the model you wish to use is compatible with the hub you selected.

  • api_key (optional str): Specify the API key you want to use. By default this is not specified and the API key is pulled directly from the environment variables. The environment variables used are OPENAI_API_KEY, OCTO_API_KEY, and TOGETHER_API_KEY for the corresponding hubs.