UngroundedAttributesEvaluator Class
Definition
Important
Some information relates to prerelease product that may be substantially modified before it’s released. Microsoft makes no warranties, express or implied, with respect to the information provided here.
An IEvaluator that utilizes the Azure AI Foundry Evaluation service to evaluate responses produced by an AI model for presence of content that indicates ungrounded inference of human attributes.
public ref class UngroundedAttributesEvaluator sealed : Microsoft::Extensions::AI::Evaluation::Safety::ContentSafetyEvaluatorpublic sealed class UngroundedAttributesEvaluator : Microsoft.Extensions.AI.Evaluation.Safety.ContentSafetyEvaluatortype UngroundedAttributesEvaluator = class
inherit ContentSafetyEvaluatorPublic NotInheritable Class UngroundedAttributesEvaluator
Inherits ContentSafetyEvaluator- Inheritance
Remarks
The UngroundedAttributesEvaluator checks whether the response being evaluated is first, ungrounded based on the information present in the supplied GroundingContext. It then checks whether the response contains information about the protected class or emotional state of a person. It returns a BooleanMetric with a value of false indicating an excellent score, and a value of true indicating a poor score.
Note that UngroundedAttributesEvaluator does not support evaluation of multimodal content present in the evaluated responses. Images and other multimodal content present in the evaluated responses will be ignored. Also note that if a multi-turn conversation is supplied as input, UngroundedAttributesEvaluator will only evaluate the contents of the last conversation turn. The contents of previous conversation turns will be ignored.
The Azure AI Foundry Evaluation service uses a finetuned model to perform this evaluation which is expected to produce more accurate results than similar evaluations performed using a regular (non-finetuned) model.
Constructors
| UngroundedAttributesEvaluator() |
An IEvaluator that utilizes the Azure AI Foundry Evaluation service to evaluate responses produced by an AI model for presence of content that indicates ungrounded inference of human attributes. |
Properties
| EvaluationMetricNames |
Gets the Names of the EvaluationMetrics produced by this IEvaluator. (Inherited from ContentSafetyEvaluator) |
| UngroundedAttributesMetricName |
Gets the Name of the BooleanMetric returned by UngroundedAttributesEvaluator. |
Methods
| EvaluateAsync(IEnumerable<ChatMessage>, ChatResponse, ChatConfiguration, IEnumerable<EvaluationContext>, CancellationToken) |
Evaluates the supplied |
| EvaluateContentSafetyAsync(IChatClient, IEnumerable<ChatMessage>, ChatResponse, IEnumerable<EvaluationContext>, String, Boolean, CancellationToken) |
Evaluates the supplied |
| FilterAdditionalContext(IEnumerable<EvaluationContext>) |
Filters the EvaluationContexts supplied by the caller via |
