Type Alias LabeledCriteria

LabeledCriteria: {
    agentTools?: undefined | StructuredToolInterface<ZodObjectAny>[];
    chainOptions?: undefined | Partial<Omit<LLMEvalChainInput<EvalOutputType, BaseLanguageModelInterface<any, BaseLanguageModelCallOptions>>, "llm">>;
    criteria?: undefined | (CriteriaLike & (Record<string, string> | Criteria));
    distanceMetric?: undefined | EmbeddingDistanceType;
    embedding?: undefined | EmbeddingsInterface;
    evaluatorType: "labeled_criteria";
    feedbackKey: undefined | string;
    formatEvaluatorInputs: EvaluatorInputFormatter;
    llm?: undefined | BaseLanguageModelInterface<any, BaseLanguageModelCallOptions> & BaseLanguageModel<any, BaseLanguageModelCallOptions>;
}

Configuration to load a "LabeledCriteriaEvalChain" evaluator, which prompts an LLM to determine whether the model's prediction complies with the provided criteria and also provides a "ground truth" label for the evaluator to incorporate in its evaluation.

Type declaration

  • OptionalagentTools?: undefined | StructuredToolInterface<ZodObjectAny>[]

    A list of tools available to the agent, for TrajectoryEvalChain.

  • OptionalchainOptions?: undefined | Partial<Omit<LLMEvalChainInput<EvalOutputType, BaseLanguageModelInterface<any, BaseLanguageModelCallOptions>>, "llm">>
  • Optionalcriteria?: undefined | (CriteriaLike & (Record<string, string> | Criteria))

    The criteria to use for the evaluator.

  • OptionaldistanceMetric?: undefined | EmbeddingDistanceType

    The distance metric to use for comparing the embeddings.

  • Optionalembedding?: undefined | EmbeddingsInterface

    The embedding objects to vectorize the outputs.

  • evaluatorType: "labeled_criteria"
  • feedbackKey: undefined | string
  • formatEvaluatorInputs: EvaluatorInputFormatter
  • Optionalllm?: undefined | BaseLanguageModelInterface<any, BaseLanguageModelCallOptions> & BaseLanguageModel<any, BaseLanguageModelCallOptions>

Param: criteria

The criteria to use for the evaluator.

Param: llm

The language model to use for the evaluator.

Returns

The configuration for the evaluator.

Example

const evalConfig = {
  evaluators: [LabeledCriteria("correctness")],
};
@example
```ts
const evalConfig = {
  evaluators: [
    LabeledCriteria({
      "mentionsAllFacts": "Does the include all facts provided in the reference?"
    })
  ],

Example

const evalConfig = {
  evaluators: [{
    evaluatorType: "labeled_criteria",
    criteria: "correctness",
    formatEvaluatorInputs: ...
  }],
};

Example

const evalConfig = {
  evaluators: [{
    evaluatorType: "labeled_criteria",
    criteria: { "mentionsAllFacts": "Does the include all facts provided in the reference?" },
    formatEvaluatorInputs: ...
  }],

Settings

Member Visibility