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

A library for running evaluations for AI use cases

import { TOXICITY_CLASSIFICATION_EVALUATOR_CONFIG } from "../__generated__/default_templates/index.js";
import { createClassificationEvaluator } from "./createClassificationEvaluator.js";
/**
 * Creates a toxicity evaluator function.
 *
 * This function returns an evaluator that determines whether a given text is
 * toxic — hateful, demeaning, abusive, or threatening.
 *
 * @param args - The arguments for creating the toxicity evaluator.
 * @param args.model - The model to use for classification.
 * @param args.choices - The possible classification choices (defaults to the config choices).
 * @param args.promptTemplate - The prompt template to use (defaults to the config template).
 * @param args.telemetry - The telemetry to use for the evaluator.
 *
 * @returns An evaluator function that takes a {@link ToxicityEvaluationRecord} and returns a classification result
 * indicating whether the text is toxic or non-toxic.
 *
 * @example
 * ```ts
 * const evaluator = createToxicityEvaluator({ model: openai("gpt-4o-mini") });
 * const result = await evaluator.evaluate({
 *   text: "You are a worthless idiot and everyone despises you.",
 * });
 * console.log(result.label); // "toxic" or "non-toxic"
 * ```
 */
export function createToxicityEvaluator(args) {
    const { choices = TOXICITY_CLASSIFICATION_EVALUATOR_CONFIG.choices, promptTemplate = TOXICITY_CLASSIFICATION_EVALUATOR_CONFIG.template, optimizationDirection = TOXICITY_CLASSIFICATION_EVALUATOR_CONFIG.optimizationDirection, name = TOXICITY_CLASSIFICATION_EVALUATOR_CONFIG.name, ...rest } = args;
    return createClassificationEvaluator({
        ...rest,
        promptTemplate,
        choices,
        optimizationDirection,
        name,
    });
}
//# sourceMappingURL=createToxicityEvaluator.js.map