evalz/dist/index.d.cts

Model graded evals with typescript

import { z } from 'zod';
import OpenAI from 'openai';

declare const BaseEvaluationDataItemSchema: z.ZodObject<{
    prompt: z.ZodOptional<z.ZodString>;
    completion: z.ZodString;
    expectedCompletion: z.ZodOptional<z.ZodString>;
    contexts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
    groundTruth: z.ZodOptional<z.ZodString>;
}, "strip", z.ZodTypeAny, {
    completion: string;
    prompt?: string | undefined;
    expectedCompletion?: string | undefined;
    contexts?: string[] | undefined;
    groundTruth?: string | undefined;
}, {
    completion: string;
    prompt?: string | undefined;
    expectedCompletion?: string | undefined;
    contexts?: string[] | undefined;
    groundTruth?: string | undefined;
}>;
declare const EvaluationDataItemSchema: z.ZodObject<{
    prompt: z.ZodOptional<z.ZodString>;
    completion: z.ZodString;
    expectedCompletion: z.ZodOptional<z.ZodString>;
    contexts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
    groundTruth: z.ZodOptional<z.ZodString>;
}, "strip", z.ZodTypeAny, {
    completion: string;
    prompt?: string | undefined;
    expectedCompletion?: string | undefined;
    contexts?: string[] | undefined;
    groundTruth?: string | undefined;
}, {
    completion: string;
    prompt?: string | undefined;
    expectedCompletion?: string | undefined;
    contexts?: string[] | undefined;
    groundTruth?: string | undefined;
}>;
declare const EvaluationDataItemResultSchema: z.ZodObject<{
    score: z.ZodNumber;
    scores: z.ZodOptional<z.ZodArray<z.ZodObject<{
        score: z.ZodNumber;
        evaluator: z.ZodString;
        evaluatorType: z.ZodString;
    }, "strip", z.ZodTypeAny, {
        score: number;
        evaluator: string;
        evaluatorType: string;
    }, {
        score: number;
        evaluator: string;
        evaluatorType: string;
    }>, "many">>;
    item: z.ZodObject<{
        prompt: z.ZodOptional<z.ZodString>;
        completion: z.ZodString;
        expectedCompletion: z.ZodOptional<z.ZodString>;
        contexts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
        groundTruth: z.ZodOptional<z.ZodString>;
    }, "strip", z.ZodTypeAny, {
        completion: string;
        prompt?: string | undefined;
        expectedCompletion?: string | undefined;
        contexts?: string[] | undefined;
        groundTruth?: string | undefined;
    }, {
        completion: string;
        prompt?: string | undefined;
        expectedCompletion?: string | undefined;
        contexts?: string[] | undefined;
        groundTruth?: string | undefined;
    }>;
}, "strip", z.ZodTypeAny, {
    score: number;
    item: {
        completion: string;
        prompt?: string | undefined;
        expectedCompletion?: string | undefined;
        contexts?: string[] | undefined;
        groundTruth?: string | undefined;
    };
    scores?: {
        score: number;
        evaluator: string;
        evaluatorType: string;
    }[] | undefined;
}, {
    score: number;
    item: {
        completion: string;
        prompt?: string | undefined;
        expectedCompletion?: string | undefined;
        contexts?: string[] | undefined;
        groundTruth?: string | undefined;
    };
    scores?: {
        score: number;
        evaluator: string;
        evaluatorType: string;
    }[] | undefined;
}>;
type ResultsType = "score" | "binary";
type BinaryResults = {
    trueCount: number;
    falseCount: number;
};
type AvgScoreResults = {
    value: number;
    individual?: Record<string, number>;
};
type EvaluationDataItem = z.infer<typeof EvaluationDataItemSchema>;
type EvaluationDataItemResult = z.infer<typeof EvaluationDataItemResultSchema>;
type EvaluationResponse<T extends ResultsType> = {
    results: EvaluationDataItemResult[];
} & (T extends "score" ? {
    scoreResults: AvgScoreResults;
} : {
    binaryResults: BinaryResults;
});
type ExecuteEvalParams = {
    data: EvaluationDataItem[];
};
interface EvalFunction extends Function {
    evalType: "model-graded" | "accuracy" | `context-${ContextEvaluatorType}` | "weighted";
}
type _Evaluator<T extends ResultsType> = ({ data }: ExecuteEvalParams) => Promise<EvaluationResponse<T>>;
interface Evaluator<T extends ResultsType> extends _Evaluator<T>, EvalFunction {
}
type ContextEvaluatorType = "entities-recall" | "precision" | "recall" | "relevance";
type ContextEvaluator = Evaluator<"score">;
type AccuracyEvaluator = Evaluator<"score">;

declare function createEvaluator<T extends ResultsType>({ resultsType, evaluationDescription, model, messages, client }: {
    resultsType?: T;
    evaluationDescription: string;
    model?: OpenAI.Model["id"];
    messages?: OpenAI.ChatCompletionMessageParam[];
    client: OpenAI;
}): Evaluator<T>;

/**
 * @name createWeightedEvaluator
 * @description
 * Create a weighted evaluator that combines the results of multiple evaluators
 * @param evaluators - A record of evaluators to combine
 * @param weights - A record of weights for each evaluator
 * @returns A weighted evaluator
 */
declare function createWeightedEvaluator({ evaluators, weights }: {
    evaluators: Record<string, Evaluator<"score"> | AccuracyEvaluator>;
    weights: Record<string, number>;
}): Evaluator<"score">;

declare function createAccuracyEvaluator({ model, weights }: {
    model?: OpenAI.Embeddings.EmbeddingCreateParams["model"];
    weights?: {
        factual: number;
        semantic: number;
    };
}): Evaluator<"score">;

declare function createContextEvaluator({ type, model }: {
    type: ContextEvaluatorType;
    model?: OpenAI.Embeddings.EmbeddingCreateParams["model"];
}): ContextEvaluator;

export { type AccuracyEvaluator, type AvgScoreResults, BaseEvaluationDataItemSchema, type BinaryResults, type ContextEvaluator, type ContextEvaluatorType, type EvaluationDataItem, type EvaluationDataItemResult, EvaluationDataItemResultSchema, EvaluationDataItemSchema, type EvaluationResponse, type Evaluator, type ExecuteEvalParams, type ResultsType, type _Evaluator, createAccuracyEvaluator, createContextEvaluator, createEvaluator, createWeightedEvaluator };