@arizeai/phoenix-evals/dist/esm/llm/createRefusalEvaluator.js

A library for running evaluations for AI use cases

import { REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG } from "../__generated__/default_templates/index.js";
import { createClassificationEvaluator } from "./createClassificationEvaluator.js";
/**
 * Creates a refusal evaluator function.
 *
 * This function returns an evaluator that detects when an LLM refuses,
 * declines, or avoids answering a user query. It is use-case agnostic:
 * it only detects whether a refusal occurred, not whether the refusal
 * was appropriate.
 *
 * @param args - The arguments for creating the refusal evaluator.
 * @param args.model - The model to use for classification.
 * @param args.choices - The possible classification choices (defaults to REFUSAL_CHOICES).
 * @param args.promptTemplate - The prompt template to use (defaults to REFUSAL_TEMPLATE).
 * @param args.telemetry - The telemetry to use for the evaluator.
 *
 * @returns An evaluator function that takes a {@link RefusalEvaluationRecord} and returns a classification result
 * indicating whether the output is a refusal or an answer.
 *
 * @example
 * ```ts
 * const evaluator = createRefusalEvaluator({ model: openai("gpt-4o-mini") });
 * const result = await evaluator.evaluate({
 *   input: "What is the capital of France?",
 *   output: "I'm sorry, I can only help with technical questions.",
 * });
 * console.log(result.label); // "refused" or "answered"
 * ```
 */
export function createRefusalEvaluator(args) {
    const { choices = REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.choices, promptTemplate = REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.template, optimizationDirection = REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.optimizationDirection, name = REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.name, ...rest } = args;
    return createClassificationEvaluator({
        ...rest,
        promptTemplate,
        choices,
        optimizationDirection,
        name,
    });
}
//# sourceMappingURL=createRefusalEvaluator.js.map